* [PATCH 0/2] introduce PM QoS interface
@ 2024-03-20 10:55 Huisong Li
2024-03-20 10:55 ` [PATCH 1/2] power: " Huisong Li
` (10 more replies)
0 siblings, 11 replies; 56+ messages in thread
From: Huisong Li @ 2024-03-20 10:55 UTC (permalink / raw)
To: dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
Subject: [PATCH 0/2] introduce PM QoS interface
The system-wide CPU latency QoS limit has a positive impact on the idle
state selection in cpuidle governor.
Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
CPU latency QoS limit on system and send the QoS request for userspace.
Please see the PM QoS framework in the following link:
https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
This feature is supported by kernel-v2.6.25.
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
So this series introduce PM QoS interface.
Huisong Li (2):
power: introduce PM QoS interface
examples/l3fwd-power: add PM QoS request configuration
doc/guides/prog_guide/power_man.rst | 16 ++++
doc/guides/rel_notes/release_24_03.rst | 4 +
examples/l3fwd-power/main.c | 41 +++++++++-
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 98 ++++++++++++++++++++++++
lib/power/rte_power_qos.h | 101 +++++++++++++++++++++++++
lib/power/version.map | 4 +
7 files changed, 265 insertions(+), 1 deletion(-)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH 1/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
@ 2024-03-20 10:55 ` Huisong Li
2024-03-20 10:55 ` [PATCH 2/2] examples/l3fwd-power: add PM QoS request configuration Huisong Li
` (9 subsequent siblings)
10 siblings, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-03-20 10:55 UTC (permalink / raw)
To: dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
The system-wide CPU latency QoS limit has a positive impact on the idle
state selection in cpuidle governor.
Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
CPU latency QoS limit on system and send the QoS request for userspace.
Please see the PM QoS framework in the following link:
https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
This feature has beed supported by kernel-v2.6.25.
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
So this PM QoS API make it easy to obtain the CPU latency limit on system
and send the CPU latency QoS request for the application that need them.
The recommend usage method is as follows:
1) an application process first creates QoS request.
2) update the CPU latency request to zero when need.
3) back to the default value when no need(this step is optional).
4) release QoS request when process exit.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
doc/guides/prog_guide/power_man.rst | 16 ++++
doc/guides/rel_notes/release_24_03.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 98 ++++++++++++++++++++++++
lib/power/rte_power_qos.h | 101 +++++++++++++++++++++++++
lib/power/version.map | 4 +
6 files changed, 225 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..493c75bf9d 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,22 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+PM QoS API
+----------
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service threads are delay sensitive and very except
+the low resume time, like interrupt packet receiving mode.
+
+This PM QoS API is aimed to obtain the CPU latency limit on system and send the
+CPU latency QoS request for the application that need them.
+
+* ``rte_power_qos_get_curr_cpu_latency()`` is used to get the current CPU
+ latency limit on system.
+* For sending CPU latency QoS request, first call ``rte_power_create_qos_request()``
+ to create a QoS request, then update CPU latency value by calling
+ ``rte_power_qos_update_request()``. The ``rte_power_release_qos_request()`` is
+ used to release this QoS request when process exit.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_03.rst b/doc/guides/rel_notes/release_24_03.rst
index 14826ea08f..b5be724133 100644
--- a/doc/guides/rel_notes/release_24_03.rst
+++ b/doc/guides/rel_notes/release_24_03.rst
@@ -196,6 +196,10 @@ New Features
Added DMA producer mode to measure performance of ``OP_FORWARD`` mode
of event DMA adapter.
+* **Added CPU latency PM QoS support.**
+
+ Added the interface querying cpu latency PM QoS limit on system and
+ the interface sending cpu latency QoS request in power lib.
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..d2b55923a0
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,98 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define QOS_CPU_DMA_LATENCY_DEV "/dev/cpu_dma_latency"
+
+struct rte_power_qos_info {
+ /*
+ * Keep file descriptor to update QoS request until there are no
+ * necessary anymore.
+ */
+ int fd;
+ int cur_cpu_latency; /* unit microseconds */
+ };
+
+struct rte_power_qos_info g_qos = {
+ .fd = -1,
+ .cur_cpu_latency = -1,
+};
+
+int
+rte_power_qos_get_curr_cpu_latency(int *latency)
+{
+ int fd, ret;
+
+ fd = open(QOS_CPU_DMA_LATENCY_DEV, O_RDONLY);
+ if (fd < 0) {
+ POWER_LOG(ERR, "Failed to open %s", QOS_CPU_DMA_LATENCY_DEV);
+ return -1;
+ }
+
+ ret = read(fd, latency, sizeof(*latency));
+ if (ret == 0) {
+ POWER_LOG(ERR, "Failed to read %s", QOS_CPU_DMA_LATENCY_DEV);
+ return -1;
+ }
+ close(fd);
+
+ return 0;
+}
+
+int
+rte_power_qos_update_request(int latency)
+{
+ int ret;
+
+ if (g_qos.fd == -1) {
+ POWER_LOG(ERR, "please create QoS request first.");
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be non negative number.");
+ return -EINVAL;
+ }
+
+ if (g_qos.cur_cpu_latency != -1 && latency == g_qos.cur_cpu_latency)
+ return 0;
+
+ ret = write(g_qos.fd, &latency, sizeof(latency));
+ if (ret == 0) {
+ POWER_LOG(ERR, "Failed to write %s", QOS_CPU_DMA_LATENCY_DEV);
+ return -1;
+ }
+ g_qos.cur_cpu_latency = latency;
+
+ return 0;
+}
+
+int
+rte_power_create_qos_request(void)
+{
+ g_qos.fd = open(QOS_CPU_DMA_LATENCY_DEV, O_WRONLY);
+ if (g_qos.fd < 0) {
+ POWER_LOG(ERR, "Failed to open %s.", QOS_CPU_DMA_LATENCY_DEV);
+ return -1;
+ }
+
+ return 0;
+}
+
+void
+rte_power_release_qos_request(void)
+{
+ if (g_qos.fd != -1) {
+ close(g_qos.fd);
+ g_qos.fd = -1;
+ }
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..d39f5d0c0f
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,101 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The system-wide CPU latency QoS limit has a positive impact on the idle
+ * state selection in cpuidle governor.
+ *
+ * Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
+ * CPU latency QoS limit on system and send the QoS request for userspace.
+ * Please see the PM QoS framework in the following link:
+ * https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
+ *
+ * The deeper the idle state, the lower the power consumption, but the longer
+ * the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * So this PM QoS API make it easy to obtain the CPU latency limit on system and
+ * send the CPU latency QoS request for the application that need them.
+ *
+ * The recommend usage method is as follows:
+ * 1) an application process first creates QoS request.
+ * 2) update the CPU latency request to zero when need.
+ * 3) back to the default value @see PM_QOS_CPU_LATENCY_DEFAULT_VALUE when
+ * no need (this step is optional).
+ * 4)release QoS request when process exit.
+ */
+
+#define QOS_USEC_PER_SEC 1000000
+#define PM_QOS_CPU_LATENCY_DEFAULT_VALUE (2000 * QOS_USEC_PER_SEC)
+#define PM_QOS_STRICT_LATENCY_VALUE 0
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Create CPU latency QoS request and release this request by
+ * @see rte_power_release_qos_request.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_create_qos_request(void);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * release CPU latency QoS request.
+ */
+__rte_experimental
+void rte_power_release_qos_request(void);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current CPU latency QoS limit on system.
+ * The default value in kernel is @see PM_QOS_CPU_LATENCY_DEFAULT_VALUE.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_get_curr_cpu_latency(int *latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Update the CPU latency QoS request.
+ * Note: need to create QoS request first and then call this API.
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_update_request(int latency);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..42770762b1 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,8 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_create_qos_request;
+ rte_power_release_qos_request;
+ rte_power_qos_get_curr_cpu_latency;
+ rte_power_qos_update_request;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH 2/2] examples/l3fwd-power: add PM QoS request configuration
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
2024-03-20 10:55 ` [PATCH 1/2] power: " Huisong Li
@ 2024-03-20 10:55 ` Huisong Li
2024-03-20 14:05 ` [PATCH 0/2] introduce PM QoS interface Morten Brørup
` (8 subsequent siblings)
10 siblings, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-03-20 10:55 UTC (permalink / raw)
To: dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
Add PM QoS request configuration to declease the process resume latency.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
examples/l3fwd-power/main.c | 41 ++++++++++++++++++++++++++++++++++++-
1 file changed, 40 insertions(+), 1 deletion(-)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index f4adcf41b5..78f292ed02 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2232,12 +2233,48 @@ static int check_ptype(uint16_t portid)
}
+static int
+pm_qos_init(void)
+{
+ int cur_cpu_latency;
+ int ret;
+
+ ret = rte_power_qos_get_curr_cpu_latency(&cur_cpu_latency);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER, "failed to get current cpu latency.\n");
+ return ret;
+ }
+ RTE_LOG(INFO, L3FWD_POWER, "current cpu latency is %dus on system.\n",
+ (cur_cpu_latency / QOS_USEC_PER_SEC));
+
+ ret = rte_power_create_qos_request();
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER, "Failed to create power QoS request.\n");
+ return ret;
+ }
+
+ /*
+ * Set strict latency requirement to prevent service thread going into
+ * a deeper sleep state whose resume time is longer.
+ */
+ ret = rte_power_qos_update_request(PM_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0)
+ RTE_LOG(ERR, L3FWD_POWER, "Failed to change cpu latency to 0.\n");
+ return ret;
+}
+
static int
init_power_library(void)
{
enum power_management_env env;
unsigned int lcore_id;
- int ret = 0;
+ int ret;
+
+ ret = pm_qos_init();
+ if (ret != 0) {
+ RTE_LOG(ERR, L3FWD_POWER, "init power Qos failed.\n");
+ return ret;
+ }
RTE_LCORE_FOREACH(lcore_id) {
/* init power management library */
@@ -2268,6 +2305,8 @@ deinit_power_library(void)
unsigned int lcore_id, max_pkg, max_die, die, pkg;
int ret = 0;
+ rte_power_release_qos_request();
+
RTE_LCORE_FOREACH(lcore_id) {
/* deinit power management library */
ret = rte_power_exit(lcore_id);
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH 0/2] introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
2024-03-20 10:55 ` [PATCH 1/2] power: " Huisong Li
2024-03-20 10:55 ` [PATCH 2/2] examples/l3fwd-power: add PM QoS request configuration Huisong Li
@ 2024-03-20 14:05 ` Morten Brørup
2024-03-21 3:04 ` lihuisong (C)
2024-06-13 11:20 ` [PATCH v2 0/2] power: " Huisong Li
` (7 subsequent siblings)
10 siblings, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-03-20 14:05 UTC (permalink / raw)
To: Huisong Li, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> From: Huisong Li [mailto:lihuisong@huawei.com]
> Sent: Wednesday, 20 March 2024 11.55
>
> The system-wide CPU latency QoS limit has a positive impact on the idle
> state selection in cpuidle governor.
>
> Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
> CPU latency QoS limit on system and send the QoS request for userspace.
> Please see the PM QoS framework in the following link:
> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
> This feature is supported by kernel-v2.6.25.
>
> The deeper the idle state, the lower the power consumption, but the longer
> the resume time. Some service are delay sensitive and very except the low
> resume time, like interrupt packet receiving mode.
>
> So this series introduce PM QoS interface.
This looks like a 1:1 wrapper for a Linux kernel feature.
Does Windows or BSD offer something similar?
Furthermore, any high-res timing should use nanoseconds, not microseconds or milliseconds.
I realize that the Linux kernel only uses microseconds for these APIs, but the DPDK API should use nanoseconds.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-20 14:05 ` [PATCH 0/2] introduce PM QoS interface Morten Brørup
@ 2024-03-21 3:04 ` lihuisong (C)
2024-03-21 13:30 ` Morten Brørup
0 siblings, 1 reply; 56+ messages in thread
From: lihuisong (C) @ 2024-03-21 3:04 UTC (permalink / raw)
To: Morten Brørup, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
Hi Moren,
Thanks for your revew.
在 2024/3/20 22:05, Morten Brørup 写道:
>> From: Huisong Li [mailto:lihuisong@huawei.com]
>> Sent: Wednesday, 20 March 2024 11.55
>>
>> The system-wide CPU latency QoS limit has a positive impact on the idle
>> state selection in cpuidle governor.
>>
>> Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
>> CPU latency QoS limit on system and send the QoS request for userspace.
>> Please see the PM QoS framework in the following link:
>> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
>> This feature is supported by kernel-v2.6.25.
>>
>> The deeper the idle state, the lower the power consumption, but the longer
>> the resume time. Some service are delay sensitive and very except the low
>> resume time, like interrupt packet receiving mode.
>>
>> So this series introduce PM QoS interface.
> This looks like a 1:1 wrapper for a Linux kernel feature.
right
> Does Windows or BSD offer something similar?
How do we know Windows or BSD support this similar feature?
The DPDK power lib just work on Linux according to the meson.build under
lib/power.
If they support this features, they can open it.
>
> Furthermore, any high-res timing should use nanoseconds, not microseconds or milliseconds.
> I realize that the Linux kernel only uses microseconds for these APIs, but the DPDK API should use nanoseconds.
Nanoseconds is more precise, it's good.
But DPDK API how use nanoseconds as you said the the Linux kernel only
uses microseconds for these APIs.
Kernel interface just know an integer value with microseconds unit.
/BR
/Huisong
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH 0/2] introduce PM QoS interface
2024-03-21 3:04 ` lihuisong (C)
@ 2024-03-21 13:30 ` Morten Brørup
2024-03-22 8:54 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-03-21 13:30 UTC (permalink / raw)
To: lihuisong (C), dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> Sent: Thursday, 21 March 2024 04.04
>
> Hi Moren,
>
> Thanks for your revew.
>
> 在 2024/3/20 22:05, Morten Brørup 写道:
> >> From: Huisong Li [mailto:lihuisong@huawei.com]
> >> Sent: Wednesday, 20 March 2024 11.55
> >>
> >> The system-wide CPU latency QoS limit has a positive impact on the idle
> >> state selection in cpuidle governor.
> >>
> >> Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
> >> CPU latency QoS limit on system and send the QoS request for userspace.
> >> Please see the PM QoS framework in the following link:
> >> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
> >> This feature is supported by kernel-v2.6.25.
> >>
> >> The deeper the idle state, the lower the power consumption, but the longer
> >> the resume time. Some service are delay sensitive and very except the low
> >> resume time, like interrupt packet receiving mode.
> >>
> >> So this series introduce PM QoS interface.
> > This looks like a 1:1 wrapper for a Linux kernel feature.
> right
> > Does Windows or BSD offer something similar?
> How do we know Windows or BSD support this similar feature?
Ask Windows experts or research using Google.
> The DPDK power lib just work on Linux according to the meson.build under
> lib/power.
> If they support this features, they can open it.
The DPDK power lib currently only works on Linux, yes.
But its API should still be designed to be platform agnostic, so the functions can be implemented on other platforms in the future.
DPDK is on track to work across multiple platforms, including Windows.
We must always consider other platforms, and not design DPDK APIs as if they are for Linux/BSD only.
> >
> > Furthermore, any high-res timing should use nanoseconds, not microseconds or
> milliseconds.
> > I realize that the Linux kernel only uses microseconds for these APIs, but
> the DPDK API should use nanoseconds.
> Nanoseconds is more precise, it's good.
> But DPDK API how use nanoseconds as you said the the Linux kernel only
> uses microseconds for these APIs.
> Kernel interface just know an integer value with microseconds unit.
One solution is to expose nanoseconds in the DPDK API, and in the Linux specific implementation convert from/to microseconds.
You might also want to add a note to the in-line documentation of the relevant functions that the Linux implementation only uses microsecond resolution.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-21 13:30 ` Morten Brørup
@ 2024-03-22 8:54 ` lihuisong (C)
2024-03-22 12:35 ` Morten Brørup
2024-03-22 17:55 ` Tyler Retzlaff
0 siblings, 2 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-03-22 8:54 UTC (permalink / raw)
To: Morten Brørup, dev, Tyler Retzlaff, weh,
longli@microsoft.com >> Long Li, alan.elder
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
+Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
在 2024/3/21 21:30, Morten Brørup 写道:
>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>> Sent: Thursday, 21 March 2024 04.04
>>
>> Hi Moren,
>>
>> Thanks for your revew.
>>
>> 在 2024/3/20 22:05, Morten Brørup 写道:
>>>> From: Huisong Li [mailto:lihuisong@huawei.com]
>>>> Sent: Wednesday, 20 March 2024 11.55
>>>>
>>>> The system-wide CPU latency QoS limit has a positive impact on the idle
>>>> state selection in cpuidle governor.
>>>>
>>>> Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
>>>> CPU latency QoS limit on system and send the QoS request for userspace.
>>>> Please see the PM QoS framework in the following link:
>>>> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
>>>> This feature is supported by kernel-v2.6.25.
>>>>
>>>> The deeper the idle state, the lower the power consumption, but the longer
>>>> the resume time. Some service are delay sensitive and very except the low
>>>> resume time, like interrupt packet receiving mode.
>>>>
>>>> So this series introduce PM QoS interface.
>>> This looks like a 1:1 wrapper for a Linux kernel feature.
>> right
>>> Does Windows or BSD offer something similar?
>> How do we know Windows or BSD support this similar feature?
> Ask Windows experts or research using Google.
I download freebsd source code, I didn't find this similar feature.
They don't even support cpuidle feature(this QoS feature affects cpuilde.).
I don't find any useful about this on Windows from google.
@Tyler, @Alan, @Wei and @Long
Do you know windows support that userspace read and send CPU latency
which has an impact on deep level of CPU idle?
>> The DPDK power lib just work on Linux according to the meson.build under
>> lib/power.
>> If they support this features, they can open it.
> The DPDK power lib currently only works on Linux, yes.
> But its API should still be designed to be platform agnostic, so the functions can be implemented on other platforms in the future.
>
> DPDK is on track to work across multiple platforms, including Windows.
> We must always consider other platforms, and not design DPDK APIs as if they are for Linux/BSD only.
totally understand you.
>
>>> Furthermore, any high-res timing should use nanoseconds, not microseconds or
>> milliseconds.
>>> I realize that the Linux kernel only uses microseconds for these APIs, but
>> the DPDK API should use nanoseconds.
>> Nanoseconds is more precise, it's good.
>> But DPDK API how use nanoseconds as you said the the Linux kernel only
>> uses microseconds for these APIs.
>> Kernel interface just know an integer value with microseconds unit.
> One solution is to expose nanoseconds in the DPDK API, and in the Linux specific implementation convert from/to microseconds.
If so, we have to modify the implementation interface on Linux. This
change the input/output unit about the interface.
And DPDK also has to do this based on kernel version. It is not good.
The cpuidle governor select which idle state based on the worst-case
latency of idle state.
These the worst-case latency of Cstate reported by ACPI table is in
microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3. _LPI
(Low Power Idle States) in ACPI spec [1].
So it is probably not meaning to change this interface implementation.
For the case need PM QoS in DPDK, I think, it is better to set cpu
latency to zero to prevent service thread from the deeper the idle state.
> You might also want to add a note to the in-line documentation of the relevant functions that the Linux implementation only uses microsecond resolution.
>
[1]
https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH 0/2] introduce PM QoS interface
2024-03-22 8:54 ` lihuisong (C)
@ 2024-03-22 12:35 ` Morten Brørup
2024-03-26 2:11 ` lihuisong (C)
2024-03-22 17:55 ` Tyler Retzlaff
1 sibling, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-03-22 12:35 UTC (permalink / raw)
To: lihuisong (C), dev, Tyler Retzlaff, weh, longli, alan.elder
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> Sent: Friday, 22 March 2024 09.54
>
> +Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
>
> 在 2024/3/21 21:30, Morten Brørup 写道:
> >> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >> Sent: Thursday, 21 March 2024 04.04
> >>
> >> Hi Moren,
> >>
> >> Thanks for your revew.
> >>
> >> 在 2024/3/20 22:05, Morten Brørup 写道:
> >>>> From: Huisong Li [mailto:lihuisong@huawei.com]
> >>>> Sent: Wednesday, 20 March 2024 11.55
> >>>>
> >>>> The system-wide CPU latency QoS limit has a positive impact on the idle
> >>>> state selection in cpuidle governor.
> >>>>
> >>>> Linux creates a cpu_dma_latency device under '/dev' directory to obtain
> the
> >>>> CPU latency QoS limit on system and send the QoS request for userspace.
> >>>> Please see the PM QoS framework in the following link:
> >>>> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
> >>>> This feature is supported by kernel-v2.6.25.
> >>>>
> >>>> The deeper the idle state, the lower the power consumption, but the
> longer
> >>>> the resume time. Some service are delay sensitive and very except the low
> >>>> resume time, like interrupt packet receiving mode.
> >>>>
> >>>> So this series introduce PM QoS interface.
> >>> This looks like a 1:1 wrapper for a Linux kernel feature.
> >> right
> >>> Does Windows or BSD offer something similar?
> >> How do we know Windows or BSD support this similar feature?
> > Ask Windows experts or research using Google.
> I download freebsd source code, I didn't find this similar feature.
> They don't even support cpuidle feature(this QoS feature affects cpuilde.).
> I don't find any useful about this on Windows from google.
>
>
> @Tyler, @Alan, @Wei and @Long
>
> Do you know windows support that userspace read and send CPU latency
> which has an impact on deep level of CPU idle?
>
> >> The DPDK power lib just work on Linux according to the meson.build under
> >> lib/power.
> >> If they support this features, they can open it.
> > The DPDK power lib currently only works on Linux, yes.
> > But its API should still be designed to be platform agnostic, so the
> functions can be implemented on other platforms in the future.
> >
> > DPDK is on track to work across multiple platforms, including Windows.
> > We must always consider other platforms, and not design DPDK APIs as if they
> are for Linux/BSD only.
> totally understand you.
> >
> >>> Furthermore, any high-res timing should use nanoseconds, not microseconds
> or
> >> milliseconds.
> >>> I realize that the Linux kernel only uses microseconds for these APIs, but
> >> the DPDK API should use nanoseconds.
> >> Nanoseconds is more precise, it's good.
> >> But DPDK API how use nanoseconds as you said the the Linux kernel only
> >> uses microseconds for these APIs.
> >> Kernel interface just know an integer value with microseconds unit.
> > One solution is to expose nanoseconds in the DPDK API, and in the Linux
> specific implementation convert from/to microseconds.
> If so, we have to modify the implementation interface on Linux. This
> change the input/output unit about the interface.
> And DPDK also has to do this based on kernel version. It is not good.
> The cpuidle governor select which idle state based on the worst-case
> latency of idle state.
> These the worst-case latency of Cstate reported by ACPI table is in
> microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3. _LPI
> (Low Power Idle States) in ACPI spec [1].
> So it is probably not meaning to change this interface implementation.
OK... Since microsecond resolution is good enough for ACPI and Linux, you have me convinced that it's also good enough for DPDK (for this specific topic).
Thank you for the detailed reply!
>
> For the case need PM QoS in DPDK, I think, it is better to set cpu
> latency to zero to prevent service thread from the deeper the idle state.
It would defeat the purpose (i.e. not saving sufficient amounts of power) if the CPU cannot enter a deeper idle state.
Personally, I would think a wake-up latency of up to 10 microseconds should be fine for must purposes.
Default Linux timerslack is 50 microseconds, so you could also use that value.
> > You might also want to add a note to the in-line documentation of the
> relevant functions that the Linux implementation only uses microsecond
> resolution.
> >
> [1]
> https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-22 8:54 ` lihuisong (C)
2024-03-22 12:35 ` Morten Brørup
@ 2024-03-22 17:55 ` Tyler Retzlaff
2024-03-26 2:20 ` lihuisong (C)
1 sibling, 1 reply; 56+ messages in thread
From: Tyler Retzlaff @ 2024-03-22 17:55 UTC (permalink / raw)
To: lihuisong (C)
Cc: Morten Brørup, dev, weh,
longli@microsoft.com >> Long Li, alan.elder, thomas,
ferruh.yigit, anatoly.burakov, david.hunt, sivaprasad.tummala,
liuyonglong
On Fri, Mar 22, 2024 at 04:54:01PM +0800, lihuisong (C) wrote:
> +Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
>
> 在 2024/3/21 21:30, Morten Brørup 写道:
> >>From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >>Sent: Thursday, 21 March 2024 04.04
> >>
> >>Hi Moren,
> >>
> >>Thanks for your revew.
> >>
> >>在 2024/3/20 22:05, Morten Brørup 写道:
> >>>>From: Huisong Li [mailto:lihuisong@huawei.com]
> >>>>Sent: Wednesday, 20 March 2024 11.55
> >>>>
> >>>>The system-wide CPU latency QoS limit has a positive impact on the idle
> >>>>state selection in cpuidle governor.
> >>>>
> >>>>Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
> >>>>CPU latency QoS limit on system and send the QoS request for userspace.
> >>>>Please see the PM QoS framework in the following link:
> >>>>https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
> >>>>This feature is supported by kernel-v2.6.25.
> >>>>
> >>>>The deeper the idle state, the lower the power consumption, but the longer
> >>>>the resume time. Some service are delay sensitive and very except the low
> >>>>resume time, like interrupt packet receiving mode.
> >>>>
> >>>>So this series introduce PM QoS interface.
> >>>This looks like a 1:1 wrapper for a Linux kernel feature.
> >>right
> >>>Does Windows or BSD offer something similar?
> >>How do we know Windows or BSD support this similar feature?
> >Ask Windows experts or research using Google.
> I download freebsd source code, I didn't find this similar feature.
> They don't even support cpuidle feature(this QoS feature affects cpuilde.).
> I don't find any useful about this on Windows from google.
>
>
> @Tyler, @Alan, @Wei and @Long
>
> Do you know windows support that userspace read and send CPU latency
> which has an impact on deep level of CPU idle?
it is unlikely you'll find an api that let's you manage things in terms
of raw latency values as the linux knobs here do. windows more often employs
policy centric schemes to permit the system to abstract implementation detail.
powercfg is probably the closest thing you can use to tune the same
things on windows. where you select e.g. the 'performance' scheme but it
won't allow you to pick specific latency numbers.
https://learn.microsoft.com/en-us/windows-hardware/design/device-experiences/powercfg-command-line-options
>
> >>The DPDK power lib just work on Linux according to the meson.build under
> >>lib/power.
> >>If they support this features, they can open it.
> >The DPDK power lib currently only works on Linux, yes.
> >But its API should still be designed to be platform agnostic, so the functions can be implemented on other platforms in the future.
> >
> >DPDK is on track to work across multiple platforms, including Windows.
> >We must always consider other platforms, and not design DPDK APIs as if they are for Linux/BSD only.
> totally understand you.
since lib/power isn't built for windows at this time i don't think it's
appropriate to constrain your innovation. i do appreciate the engagement
though and would just offer general guidance that if you can design your
api with some kind of abstraction in mind that would be great and by all
means if you can figure out how to wrangle powercfg /Qh into satisfying the
api in a policy centric way it might be kind of nice.
i'll let other windows experts chime in here if they choose.
thanks!
> >
> >>>Furthermore, any high-res timing should use nanoseconds, not microseconds or
> >>milliseconds.
> >>>I realize that the Linux kernel only uses microseconds for these APIs, but
> >>the DPDK API should use nanoseconds.
> >>Nanoseconds is more precise, it's good.
> >>But DPDK API how use nanoseconds as you said the the Linux kernel only
> >>uses microseconds for these APIs.
> >>Kernel interface just know an integer value with microseconds unit.
> >One solution is to expose nanoseconds in the DPDK API, and in the Linux specific implementation convert from/to microseconds.
> If so, we have to modify the implementation interface on Linux. This
> change the input/output unit about the interface.
> And DPDK also has to do this based on kernel version. It is not good.
> The cpuidle governor select which idle state based on the worst-case
> latency of idle state.
> These the worst-case latency of Cstate reported by ACPI table is in
> microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3.
> _LPI (Low Power Idle States) in ACPI spec [1].
> So it is probably not meaning to change this interface implementation.
>
> For the case need PM QoS in DPDK, I think, it is better to set cpu
> latency to zero to prevent service thread from the deeper the idle
> state.
> >You might also want to add a note to the in-line documentation of the relevant functions that the Linux implementation only uses microsecond resolution.
> >
> [1] https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-22 12:35 ` Morten Brørup
@ 2024-03-26 2:11 ` lihuisong (C)
2024-03-26 8:27 ` Morten Brørup
0 siblings, 1 reply; 56+ messages in thread
From: lihuisong (C) @ 2024-03-26 2:11 UTC (permalink / raw)
To: Morten Brørup, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
在 2024/3/22 20:35, Morten Brørup 写道:
>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>> Sent: Friday, 22 March 2024 09.54
>>
>> +Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
>>
>> 在 2024/3/21 21:30, Morten Brørup 写道:
>>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>>>> Sent: Thursday, 21 March 2024 04.04
>>>>
>>>> Hi Moren,
>>>>
>>>> Thanks for your revew.
>>>>
>>>> 在 2024/3/20 22:05, Morten Brørup 写道:
>>>>>> From: Huisong Li [mailto:lihuisong@huawei.com]
>>>>>> Sent: Wednesday, 20 March 2024 11.55
>>>>>>
>>>>>> The system-wide CPU latency QoS limit has a positive impact on the idle
>>>>>> state selection in cpuidle governor.
>>>>>>
>>>>>> Linux creates a cpu_dma_latency device under '/dev' directory to obtain
>> the
>>>>>> CPU latency QoS limit on system and send the QoS request for userspace.
>>>>>> Please see the PM QoS framework in the following link:
>>>>>> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
>>>>>> This feature is supported by kernel-v2.6.25.
>>>>>>
>>>>>> The deeper the idle state, the lower the power consumption, but the
>> longer
>>>>>> the resume time. Some service are delay sensitive and very except the low
>>>>>> resume time, like interrupt packet receiving mode.
>>>>>>
>>>>>> So this series introduce PM QoS interface.
>>>>> This looks like a 1:1 wrapper for a Linux kernel feature.
>>>> right
>>>>> Does Windows or BSD offer something similar?
>>>> How do we know Windows or BSD support this similar feature?
>>> Ask Windows experts or research using Google.
>> I download freebsd source code, I didn't find this similar feature.
>> They don't even support cpuidle feature(this QoS feature affects cpuilde.).
>> I don't find any useful about this on Windows from google.
>>
>>
>> @Tyler, @Alan, @Wei and @Long
>>
>> Do you know windows support that userspace read and send CPU latency
>> which has an impact on deep level of CPU idle?
>>
>>>> The DPDK power lib just work on Linux according to the meson.build under
>>>> lib/power.
>>>> If they support this features, they can open it.
>>> The DPDK power lib currently only works on Linux, yes.
>>> But its API should still be designed to be platform agnostic, so the
>> functions can be implemented on other platforms in the future.
>>> DPDK is on track to work across multiple platforms, including Windows.
>>> We must always consider other platforms, and not design DPDK APIs as if they
>> are for Linux/BSD only.
>> totally understand you.
>>>>> Furthermore, any high-res timing should use nanoseconds, not microseconds
>> or
>>>> milliseconds.
>>>>> I realize that the Linux kernel only uses microseconds for these APIs, but
>>>> the DPDK API should use nanoseconds.
>>>> Nanoseconds is more precise, it's good.
>>>> But DPDK API how use nanoseconds as you said the the Linux kernel only
>>>> uses microseconds for these APIs.
>>>> Kernel interface just know an integer value with microseconds unit.
>>> One solution is to expose nanoseconds in the DPDK API, and in the Linux
>> specific implementation convert from/to microseconds.
>> If so, we have to modify the implementation interface on Linux. This
>> change the input/output unit about the interface.
>> And DPDK also has to do this based on kernel version. It is not good.
>> The cpuidle governor select which idle state based on the worst-case
>> latency of idle state.
>> These the worst-case latency of Cstate reported by ACPI table is in
>> microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3. _LPI
>> (Low Power Idle States) in ACPI spec [1].
>> So it is probably not meaning to change this interface implementation.
> OK... Since microsecond resolution is good enough for ACPI and Linux, you have me convinced that it's also good enough for DPDK (for this specific topic).
>
> Thank you for the detailed reply!
>
>> For the case need PM QoS in DPDK, I think, it is better to set cpu
>> latency to zero to prevent service thread from the deeper the idle state.
> It would defeat the purpose (i.e. not saving sufficient amounts of power) if the CPU cannot enter a deeper idle state.
Yes, it is not good for power.
AFAIS, PM QoS is just to decrease the influence for performance.
Anyway, if we set to zero, system can be into Cstates-0 at least.
>
> Personally, I would think a wake-up latency of up to 10 microseconds should be fine for must purposes.
> Default Linux timerslack is 50 microseconds, so you could also use that value.
How much CPU latency is ok. Maybe, we can give the decision to the
application.
Linux will collect all these QoS request and use the minimum latency.
what do you think, Morten?
>
>>> You might also want to add a note to the in-line documentation of the
>> relevant functions that the Linux implementation only uses microsecond
>> resolution.
>> [1]
>> https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-22 17:55 ` Tyler Retzlaff
@ 2024-03-26 2:20 ` lihuisong (C)
2024-03-26 16:04 ` Tyler Retzlaff
0 siblings, 1 reply; 56+ messages in thread
From: lihuisong (C) @ 2024-03-26 2:20 UTC (permalink / raw)
To: Tyler Retzlaff
Cc: Morten Brørup, dev, weh,
longli@microsoft.com >> Long Li, alan.elder, thomas,
ferruh.yigit, anatoly.burakov, david.hunt, sivaprasad.tummala,
liuyonglong
Hi Tyler,
在 2024/3/23 1:55, Tyler Retzlaff 写道:
> On Fri, Mar 22, 2024 at 04:54:01PM +0800, lihuisong (C) wrote:
>> +Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
>>
>> 在 2024/3/21 21:30, Morten Brørup 写道:
>>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>>>> Sent: Thursday, 21 March 2024 04.04
>>>>
>>>> Hi Moren,
>>>>
>>>> Thanks for your revew.
>>>>
>>>> 在 2024/3/20 22:05, Morten Brørup 写道:
>>>>>> From: Huisong Li [mailto:lihuisong@huawei.com]
>>>>>> Sent: Wednesday, 20 March 2024 11.55
>>>>>>
>>>>>> The system-wide CPU latency QoS limit has a positive impact on the idle
>>>>>> state selection in cpuidle governor.
>>>>>>
>>>>>> Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
>>>>>> CPU latency QoS limit on system and send the QoS request for userspace.
>>>>>> Please see the PM QoS framework in the following link:
>>>>>> https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
>>>>>> This feature is supported by kernel-v2.6.25.
>>>>>>
>>>>>> The deeper the idle state, the lower the power consumption, but the longer
>>>>>> the resume time. Some service are delay sensitive and very except the low
>>>>>> resume time, like interrupt packet receiving mode.
>>>>>>
>>>>>> So this series introduce PM QoS interface.
>>>>> This looks like a 1:1 wrapper for a Linux kernel feature.
>>>> right
>>>>> Does Windows or BSD offer something similar?
>>>> How do we know Windows or BSD support this similar feature?
>>> Ask Windows experts or research using Google.
>> I download freebsd source code, I didn't find this similar feature.
>> They don't even support cpuidle feature(this QoS feature affects cpuilde.).
>> I don't find any useful about this on Windows from google.
>>
>>
>> @Tyler, @Alan, @Wei and @Long
>>
>> Do you know windows support that userspace read and send CPU latency
>> which has an impact on deep level of CPU idle?
> it is unlikely you'll find an api that let's you manage things in terms
> of raw latency values as the linux knobs here do. windows more often employs
> policy centric schemes to permit the system to abstract implementation detail.
>
> powercfg is probably the closest thing you can use to tune the same
> things on windows. where you select e.g. the 'performance' scheme but it
> won't allow you to pick specific latency numbers.
>
> https://learn.microsoft.com/en-us/windows-hardware/design/device-experiences/powercfg-command-line-options
Thanks for your feedback. I will take a look at this tool.
>
>>>> The DPDK power lib just work on Linux according to the meson.build under
>>>> lib/power.
>>>> If they support this features, they can open it.
>>> The DPDK power lib currently only works on Linux, yes.
>>> But its API should still be designed to be platform agnostic, so the functions can be implemented on other platforms in the future.
>>>
>>> DPDK is on track to work across multiple platforms, including Windows.
>>> We must always consider other platforms, and not design DPDK APIs as if they are for Linux/BSD only.
>> totally understand you.
> since lib/power isn't built for windows at this time i don't think it's
> appropriate to constrain your innovation. i do appreciate the engagement
> though and would just offer general guidance that if you can design your
> api with some kind of abstraction in mind that would be great and by all
> means if you can figure out how to wrangle powercfg /Qh into satisfying the
> api in a policy centric way it might be kind of nice.
Testing this by using powercfg on Windows creates a very challenge for me.
So I don't plan to do this on Windows. If you need, you can add it, ok?
>
> i'll let other windows experts chime in here if they choose.
>
> thanks!
>
>>>>> Furthermore, any high-res timing should use nanoseconds, not microseconds or
>>>> milliseconds.
>>>>> I realize that the Linux kernel only uses microseconds for these APIs, but
>>>> the DPDK API should use nanoseconds.
>>>> Nanoseconds is more precise, it's good.
>>>> But DPDK API how use nanoseconds as you said the the Linux kernel only
>>>> uses microseconds for these APIs.
>>>> Kernel interface just know an integer value with microseconds unit.
>>> One solution is to expose nanoseconds in the DPDK API, and in the Linux specific implementation convert from/to microseconds.
>> If so, we have to modify the implementation interface on Linux. This
>> change the input/output unit about the interface.
>> And DPDK also has to do this based on kernel version. It is not good.
>> The cpuidle governor select which idle state based on the worst-case
>> latency of idle state.
>> These the worst-case latency of Cstate reported by ACPI table is in
>> microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3.
>> _LPI (Low Power Idle States) in ACPI spec [1].
>> So it is probably not meaning to change this interface implementation.
>>
>> For the case need PM QoS in DPDK, I think, it is better to set cpu
>> latency to zero to prevent service thread from the deeper the idle
>> state.
>>> You might also want to add a note to the in-line documentation of the relevant functions that the Linux implementation only uses microsecond resolution.
>>>
>> [1] https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH 0/2] introduce PM QoS interface
2024-03-26 2:11 ` lihuisong (C)
@ 2024-03-26 8:27 ` Morten Brørup
2024-03-26 12:15 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-03-26 8:27 UTC (permalink / raw)
To: lihuisong (C), dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> Sent: Tuesday, 26 March 2024 03.12
>
> 在 2024/3/22 20:35, Morten Brørup 写道:
> >> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >> Sent: Friday, 22 March 2024 09.54
[...]
> >> For the case need PM QoS in DPDK, I think, it is better to set cpu
> >> latency to zero to prevent service thread from the deeper the idle
> state.
> > It would defeat the purpose (i.e. not saving sufficient amounts of
> power) if the CPU cannot enter a deeper idle state.
> Yes, it is not good for power.
> AFAIS, PM QoS is just to decrease the influence for performance.
> Anyway, if we set to zero, system can be into Cstates-0 at least.
> >
> > Personally, I would think a wake-up latency of up to 10 microseconds
> should be fine for must purposes.
> > Default Linux timerslack is 50 microseconds, so you could also use
> that value.
> How much CPU latency is ok. Maybe, we can give the decision to the
> application.
Yes, the application should decide the acceptable worst-case latency.
> Linux will collect all these QoS request and use the minimum latency.
> what do you think, Morten?
For the example application, you could use a value of 50 microseconds and refer to this value also being the default timerslack in Linux.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-26 8:27 ` Morten Brørup
@ 2024-03-26 12:15 ` lihuisong (C)
2024-03-26 12:46 ` Morten Brørup
0 siblings, 1 reply; 56+ messages in thread
From: lihuisong (C) @ 2024-03-26 12:15 UTC (permalink / raw)
To: Morten Brørup, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
在 2024/3/26 16:27, Morten Brørup 写道:
>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>> Sent: Tuesday, 26 March 2024 03.12
>>
>> 在 2024/3/22 20:35, Morten Brørup 写道:
>>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>>>> Sent: Friday, 22 March 2024 09.54
> [...]
>
>>>> For the case need PM QoS in DPDK, I think, it is better to set cpu
>>>> latency to zero to prevent service thread from the deeper the idle
>> state.
>>> It would defeat the purpose (i.e. not saving sufficient amounts of
>> power) if the CPU cannot enter a deeper idle state.
>> Yes, it is not good for power.
>> AFAIS, PM QoS is just to decrease the influence for performance.
>> Anyway, if we set to zero, system can be into Cstates-0 at least.
>>> Personally, I would think a wake-up latency of up to 10 microseconds
>> should be fine for must purposes.
>>> Default Linux timerslack is 50 microseconds, so you could also use
>> that value.
>> How much CPU latency is ok. Maybe, we can give the decision to the
>> application.
> Yes, the application should decide the acceptable worst-case latency.
>
>> Linux will collect all these QoS request and use the minimum latency.
>> what do you think, Morten?
> For the example application, you could use a value of 50 microseconds and refer to this value also being the default timerslack in Linux.
There is a description for "/proc/<pid>/timerslack_ns" in Linux document [1]
"
This file provides the value of the task’s timerslack value in nanoseconds.
This value specifies an amount of time that normal timers may be
deferred in order to coalesce timers and avoid unnecessary wakeups.
This allows a task’s interactivity vs power consumption tradeoff to be
adjusted.
"
I cannot understand what the relationship is between the timerslack in
Linux and cpu latency to wake up.
It seems that timerslack is just to defer the timer in order to coalesce
timers and avoid unnecessary wakeups.
And it has not a lot to do with the CPU latency which is aimed to avoid
task to enter deeper idle state and satify application request.
>
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH 0/2] introduce PM QoS interface
2024-03-26 12:15 ` lihuisong (C)
@ 2024-03-26 12:46 ` Morten Brørup
2024-03-29 1:59 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-03-26 12:46 UTC (permalink / raw)
To: lihuisong (C), dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> Sent: Tuesday, 26 March 2024 13.15
>
> 在 2024/3/26 16:27, Morten Brørup 写道:
> >> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >> Sent: Tuesday, 26 March 2024 03.12
> >>
> >> 在 2024/3/22 20:35, Morten Brørup 写道:
> >>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >>>> Sent: Friday, 22 March 2024 09.54
> > [...]
> >
> >>>> For the case need PM QoS in DPDK, I think, it is better to set cpu
> >>>> latency to zero to prevent service thread from the deeper the idle
> >> state.
> >>> It would defeat the purpose (i.e. not saving sufficient amounts of
> >> power) if the CPU cannot enter a deeper idle state.
> >> Yes, it is not good for power.
> >> AFAIS, PM QoS is just to decrease the influence for performance.
> >> Anyway, if we set to zero, system can be into Cstates-0 at least.
> >>> Personally, I would think a wake-up latency of up to 10 microseconds
> >> should be fine for must purposes.
> >>> Default Linux timerslack is 50 microseconds, so you could also use
> >> that value.
> >> How much CPU latency is ok. Maybe, we can give the decision to the
> >> application.
> > Yes, the application should decide the acceptable worst-case latency.
> >
> >> Linux will collect all these QoS request and use the minimum latency.
> >> what do you think, Morten?
> > For the example application, you could use a value of 50 microseconds
> and refer to this value also being the default timerslack in Linux.
> There is a description for "/proc/<pid>/timerslack_ns" in Linux document
> [1]
> "
> This file provides the value of the task’s timerslack value in
> nanoseconds.
> This value specifies an amount of time that normal timers may be
> deferred in order to coalesce timers and avoid unnecessary wakeups.
> This allows a task’s interactivity vs power consumption tradeoff to be
> adjusted.
> "
> I cannot understand what the relationship is between the timerslack in
> Linux and cpu latency to wake up.
> It seems that timerslack is just to defer the timer in order to coalesce
> timers and avoid unnecessary wakeups.
> And it has not a lot to do with the CPU latency which is aimed to avoid
> task to enter deeper idle state and satify application request.
Correct. They control two different things.
However, both can cause latency for the application, so my rationale for the relationship was:
If the application accepts X us of latency caused by kernel scheduling delays (caused by timerslack), the application should accept the same amount of latency caused by CPU wake-up latency.
This also means that if you want lower latency than 50 us, you should not only set cpu wake-up latency, you should also set timerslack.
Obviously, if the application is only affected by one of the two, the application only needs to adjust that one of them.
As for the 50 us value, someone in the Linux kernel team decided that 50 us was an acceptable amount of latency for the kernel; we could use the same value, referring to that. Or we could choose some other value, and describe how we came up with our own value. And if necessary, also adjust timerslack accordingly.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-26 2:20 ` lihuisong (C)
@ 2024-03-26 16:04 ` Tyler Retzlaff
0 siblings, 0 replies; 56+ messages in thread
From: Tyler Retzlaff @ 2024-03-26 16:04 UTC (permalink / raw)
To: lihuisong (C)
Cc: Morten Brørup, dev, weh,
longli@microsoft.com >> Long Li, alan.elder, thomas,
ferruh.yigit, anatoly.burakov, david.hunt, sivaprasad.tummala,
liuyonglong
On Tue, Mar 26, 2024 at 10:20:45AM +0800, lihuisong (C) wrote:
> Hi Tyler,
>
> 在 2024/3/23 1:55, Tyler Retzlaff 写道:
> >On Fri, Mar 22, 2024 at 04:54:01PM +0800, lihuisong (C) wrote:
> >>+Tyler, +Alan, +Wei, +Long for asking this similar feature on Windows.
> >>
> >>在 2024/3/21 21:30, Morten Brørup 写道:
> >>>>From: lihuisong (C) [mailto:lihuisong@huawei.com]
> >>>>Sent: Thursday, 21 March 2024 04.04
> >>>>
> >>>>Hi Moren,
> >>>>
> >>>>Thanks for your revew.
> >>>>
> >>>>在 2024/3/20 22:05, Morten Brørup 写道:
> >>>>>>From: Huisong Li [mailto:lihuisong@huawei.com]
> >>>>>>Sent: Wednesday, 20 March 2024 11.55
> >>>>>>
> >>>>>>The system-wide CPU latency QoS limit has a positive impact on the idle
> >>>>>>state selection in cpuidle governor.
> >>>>>>
> >>>>>>Linux creates a cpu_dma_latency device under '/dev' directory to obtain the
> >>>>>>CPU latency QoS limit on system and send the QoS request for userspace.
> >>>>>>Please see the PM QoS framework in the following link:
> >>>>>>https://docs.kernel.org/power/pm_qos_interface.html?highlight=qos
> >>>>>>This feature is supported by kernel-v2.6.25.
> >>>>>>
> >>>>>>The deeper the idle state, the lower the power consumption, but the longer
> >>>>>>the resume time. Some service are delay sensitive and very except the low
> >>>>>>resume time, like interrupt packet receiving mode.
> >>>>>>
> >>>>>>So this series introduce PM QoS interface.
> >>>>>This looks like a 1:1 wrapper for a Linux kernel feature.
> >>>>right
> >>>>>Does Windows or BSD offer something similar?
> >>>>How do we know Windows or BSD support this similar feature?
> >>>Ask Windows experts or research using Google.
> >>I download freebsd source code, I didn't find this similar feature.
> >>They don't even support cpuidle feature(this QoS feature affects cpuilde.).
> >>I don't find any useful about this on Windows from google.
> >>
> >>
> >>@Tyler, @Alan, @Wei and @Long
> >>
> >>Do you know windows support that userspace read and send CPU latency
> >>which has an impact on deep level of CPU idle?
> >it is unlikely you'll find an api that let's you manage things in terms
> >of raw latency values as the linux knobs here do. windows more often employs
> >policy centric schemes to permit the system to abstract implementation detail.
> >
> >powercfg is probably the closest thing you can use to tune the same
> >things on windows. where you select e.g. the 'performance' scheme but it
> >won't allow you to pick specific latency numbers.
> >
> >https://learn.microsoft.com/en-us/windows-hardware/design/device-experiences/powercfg-command-line-options
>
> Thanks for your feedback. I will take a look at this tool.
>
> >
> >>>>The DPDK power lib just work on Linux according to the meson.build under
> >>>>lib/power.
> >>>>If they support this features, they can open it.
> >>>The DPDK power lib currently only works on Linux, yes.
> >>>But its API should still be designed to be platform agnostic, so the functions can be implemented on other platforms in the future.
> >>>
> >>>DPDK is on track to work across multiple platforms, including Windows.
> >>>We must always consider other platforms, and not design DPDK APIs as if they are for Linux/BSD only.
> >>totally understand you.
> >since lib/power isn't built for windows at this time i don't think it's
> >appropriate to constrain your innovation. i do appreciate the engagement
> >though and would just offer general guidance that if you can design your
> >api with some kind of abstraction in mind that would be great and by all
> >means if you can figure out how to wrangle powercfg /Qh into satisfying the
> >api in a policy centric way it might be kind of nice.
> Testing this by using powercfg on Windows creates a very challenge for me.
> So I don't plan to do this on Windows. If you need, you can add it, ok?
ordinarily i would say it is appropriate to, however in this
circumstance i agree. there is quite possibly significant porting work
to be done so i would have to address it if we ever include it for
windows.
thanks
> >
> >i'll let other windows experts chime in here if they choose.
> >
> >thanks!
> >
> >>>>>Furthermore, any high-res timing should use nanoseconds, not microseconds or
> >>>>milliseconds.
> >>>>>I realize that the Linux kernel only uses microseconds for these APIs, but
> >>>>the DPDK API should use nanoseconds.
> >>>>Nanoseconds is more precise, it's good.
> >>>>But DPDK API how use nanoseconds as you said the the Linux kernel only
> >>>>uses microseconds for these APIs.
> >>>>Kernel interface just know an integer value with microseconds unit.
> >>>One solution is to expose nanoseconds in the DPDK API, and in the Linux specific implementation convert from/to microseconds.
> >>If so, we have to modify the implementation interface on Linux. This
> >>change the input/output unit about the interface.
> >>And DPDK also has to do this based on kernel version. It is not good.
> >>The cpuidle governor select which idle state based on the worst-case
> >>latency of idle state.
> >>These the worst-case latency of Cstate reported by ACPI table is in
> >>microseconds as the section 8.4.1.1. _CST (C States) and 8.4.3.3.
> >>_LPI (Low Power Idle States) in ACPI spec [1].
> >>So it is probably not meaning to change this interface implementation.
> >>
> >>For the case need PM QoS in DPDK, I think, it is better to set cpu
> >>latency to zero to prevent service thread from the deeper the idle
> >>state.
> >>>You might also want to add a note to the in-line documentation of the relevant functions that the Linux implementation only uses microsecond resolution.
> >>>
> >>[1] https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html
> >.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH 0/2] introduce PM QoS interface
2024-03-26 12:46 ` Morten Brørup
@ 2024-03-29 1:59 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-03-29 1:59 UTC (permalink / raw)
To: Morten Brørup, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
在 2024/3/26 20:46, Morten Brørup 写道:
>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>> Sent: Tuesday, 26 March 2024 13.15
>>
>> 在 2024/3/26 16:27, Morten Brørup 写道:
>>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>>>> Sent: Tuesday, 26 March 2024 03.12
>>>>
>>>> 在 2024/3/22 20:35, Morten Brørup 写道:
>>>>>> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>>>>>> Sent: Friday, 22 March 2024 09.54
>>> [...]
>>>
>>>>>> For the case need PM QoS in DPDK, I think, it is better to set cpu
>>>>>> latency to zero to prevent service thread from the deeper the idle
>>>> state.
>>>>> It would defeat the purpose (i.e. not saving sufficient amounts of
>>>> power) if the CPU cannot enter a deeper idle state.
>>>> Yes, it is not good for power.
>>>> AFAIS, PM QoS is just to decrease the influence for performance.
>>>> Anyway, if we set to zero, system can be into Cstates-0 at least.
>>>>> Personally, I would think a wake-up latency of up to 10 microseconds
>>>> should be fine for must purposes.
>>>>> Default Linux timerslack is 50 microseconds, so you could also use
>>>> that value.
>>>> How much CPU latency is ok. Maybe, we can give the decision to the
>>>> application.
>>> Yes, the application should decide the acceptable worst-case latency.
>>>
>>>> Linux will collect all these QoS request and use the minimum latency.
>>>> what do you think, Morten?
>>> For the example application, you could use a value of 50 microseconds
>> and refer to this value also being the default timerslack in Linux.
>> There is a description for "/proc/<pid>/timerslack_ns" in Linux document
>> [1]
>> "
>> This file provides the value of the task’s timerslack value in
>> nanoseconds.
>> This value specifies an amount of time that normal timers may be
>> deferred in order to coalesce timers and avoid unnecessary wakeups.
>> This allows a task’s interactivity vs power consumption tradeoff to be
>> adjusted.
>> "
>> I cannot understand what the relationship is between the timerslack in
>> Linux and cpu latency to wake up.
>> It seems that timerslack is just to defer the timer in order to coalesce
>> timers and avoid unnecessary wakeups.
>> And it has not a lot to do with the CPU latency which is aimed to avoid
>> task to enter deeper idle state and satify application request.
> Correct. They control two different things.
>
> However, both can cause latency for the application, so my rationale for the relationship was:
> If the application accepts X us of latency caused by kernel scheduling delays (caused by timerslack), the application should accept the same amount of latency caused by CPU wake-up latency.
Understand, thanks for explain.
>
> This also means that if you want lower latency than 50 us, you should not only set cpu wake-up latency, you should also set timerslack.
>
> Obviously, if the application is only affected by one of the two, the application only needs to adjust that one of them.
Yes, I think it is.
>
> As for the 50 us value, someone in the Linux kernel team decided that 50 us was an acceptable amount of latency for the kernel; we could use the same value, referring to that. Or we could choose some other value, and describe how we came up with our own value. And if necessary, also adjust timerslack accordingly.
So how about use the default 50us of timerslack in l3fwd-power?
And we add some description about this in code or document, like,
suggest user also need to modify this process's timerslack if want a
more little latency.
>
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v2 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (2 preceding siblings ...)
2024-03-20 14:05 ` [PATCH 0/2] introduce PM QoS interface Morten Brørup
@ 2024-06-13 11:20 ` Huisong Li
2024-06-13 11:20 ` [PATCH v2 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-13 11:20 ` [PATCH v2 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
` (6 subsequent siblings)
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-13 11:20 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 22 +++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 29 +++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 116 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 70 +++++++++++++++
lib/power/version.map | 2 +
7 files changed, 245 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v2 1/2] power: introduce PM QoS API on CPU wide
2024-06-13 11:20 ` [PATCH v2 0/2] power: " Huisong Li
@ 2024-06-13 11:20 ` Huisong Li
2024-06-14 8:04 ` Morten Brørup
2024-06-13 11:20 ` [PATCH v2 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 1 reply; 56+ messages in thread
From: Huisong Li @ 2024-06-13 11:20 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
doc/guides/prog_guide/power_man.rst | 22 +++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 116 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 70 +++++++++++++++
lib/power/version.map | 2 +
6 files changed, 216 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..3ff46f06c1 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,28 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can effect the work
+CPU's idle state selection and just allow to enter the shallowest idle state
+if set to zero (strict resume latency) for this CPU.
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can obtain the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index e68a53d757..7c0d36e389 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -89,6 +89,10 @@ New Features
* Added SSE/NEON vector datapath.
+* **Introduce PM QoS interface.**
+
+ * Introduce PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..706f8432ee
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,116 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[BUFSIZ] = {0};
+ FILE *f;
+ int ret;
+
+ if (lcore_id >= RTE_MAX_LCORE) {
+ POWER_LOG(ERR, "Lcore id %u can not exceeds %u",
+ lcore_id, RTE_MAX_LCORE - 1U);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ sprintf(buf, "%s", "n/a");
+ else if (latency == PM_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ sprintf(buf, "%u", 0);
+ else
+ sprintf(buf, "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[BUFSIZ];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (lcore_id >= RTE_MAX_LCORE) {
+ POWER_LOG(ERR, "Lcore id %u can not exceeds %u",
+ lcore_id, RTE_MAX_LCORE - 1U);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? PM_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..1ba9568d1b
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,70 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define PM_QOS_STRICT_LATENCY_VALUE 0
+#define PM_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see PM_QOS_RESUME_LATENCY_NO_CONSTRAINT if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v2 2/2] examples/l3fwd-power: add PM QoS configuration
2024-06-13 11:20 ` [PATCH v2 0/2] power: " Huisong Li
2024-06-13 11:20 ` [PATCH v2 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-13 11:20 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-13 11:20 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
examples/l3fwd-power/main.c | 29 +++++++++++++++++++++++++++++
1 file changed, 29 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..96980352ee 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,25 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /*
+ * Set the work CPU with strict latency limit to allow the
+ * process running on the CPU can only enter the shallowest
+ * idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ PM_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2318,15 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ PM_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH v2 1/2] power: introduce PM QoS API on CPU wide
2024-06-13 11:20 ` [PATCH v2 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-14 8:04 ` Morten Brørup
2024-06-18 12:19 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Morten Brørup @ 2024-06-14 8:04 UTC (permalink / raw)
To: Huisong Li, dev, david.marchand
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
> +#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
> + "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
Is it OK to access this path using the lcore_id as CPU parameter to open_core_sysfs_file(), or must it be mapped through rte_lcore_to_cpu_id(lcore_id) first?
@David, do you know?
> +
> +int
> +rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
> +{
> + char buf[BUFSIZ] = {0};
> + FILE *f;
> + int ret;
> +
> + if (lcore_id >= RTE_MAX_LCORE) {
> + POWER_LOG(ERR, "Lcore id %u can not exceeds %u",
> + lcore_id, RTE_MAX_LCORE - 1U);
> + return -EINVAL;
> + }
The lcore_id could be a registered non-EAL thread.
You should probably fail in that case.
Same comment for rte_power_qos_get_cpu_resume_latency().
> +#define PM_QOS_STRICT_LATENCY_VALUE 0
> +#define PM_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
These definitions are in the public header file, and thus should be RTE_POWER_ prefixed and have comments describing them.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v2 1/2] power: introduce PM QoS API on CPU wide
2024-06-14 8:04 ` Morten Brørup
@ 2024-06-18 12:19 ` lihuisong (C)
2024-06-18 12:53 ` Morten Brørup
0 siblings, 1 reply; 56+ messages in thread
From: lihuisong (C) @ 2024-06-18 12:19 UTC (permalink / raw)
To: Morten Brørup, dev, david.marchand
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong
Hi Morten,
Thanks for your review.
在 2024/6/14 16:04, Morten Brørup 写道:
>> +#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
>> + "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
> Is it OK to access this path using the lcore_id as CPU parameter to open_core_sysfs_file(), or must it be mapped through rte_lcore_to_cpu_id(lcore_id) first?
The cpu_id getting by rte_lcore_to_cpu_id() is from
lcore_config[lcore_id].core_id which is from
"/sys/devices/system/cpu/cpuX/topology/core_id" file, please see the
function eal_cpu_core_id().
So I think the number in above "cpuX" must be the lcore_id in DPDK.
And the similar interface in power lib also directly use the locore_id.
>
> @David, do you know?
>
>> +
>> +int
>> +rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
>> +{
>> + char buf[BUFSIZ] = {0};
>> + FILE *f;
>> + int ret;
>> +
>> + if (lcore_id >= RTE_MAX_LCORE) {
>> + POWER_LOG(ERR, "Lcore id %u can not exceeds %u",
>> + lcore_id, RTE_MAX_LCORE - 1U);
>> + return -EINVAL;
>> + }
> The lcore_id could be a registered non-EAL thread.
> You should probably fail in that case.
right, how about use rte_lcore_is_enabled(locore_id)?
>
> Same comment for rte_power_qos_get_cpu_resume_latency().
>
>
>> +#define PM_QOS_STRICT_LATENCY_VALUE 0
>> +#define PM_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
> These definitions are in the public header file, and thus should be RTE_POWER_ prefixed and have comments describing them.
Ack
>
>
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH v2 1/2] power: introduce PM QoS API on CPU wide
2024-06-18 12:19 ` lihuisong (C)
@ 2024-06-18 12:53 ` Morten Brørup
0 siblings, 0 replies; 56+ messages in thread
From: Morten Brørup @ 2024-06-18 12:53 UTC (permalink / raw)
To: lihuisong (C), dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, liuyonglong, david.marchand
> From: lihuisong (C) [mailto:lihuisong@huawei.com]
>
> Hi Morten,
>
> Thanks for your review.
>
>
> 在 2024/6/14 16:04, Morten Brørup 写道:
> >> +#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
> >> + "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
> > Is it OK to access this path using the lcore_id as CPU parameter to
> open_core_sysfs_file(), or must it be mapped through
> rte_lcore_to_cpu_id(lcore_id) first?
> The cpu_id getting by rte_lcore_to_cpu_id() is from
> lcore_config[lcore_id].core_id which is from
> "/sys/devices/system/cpu/cpuX/topology/core_id" file, please see the
> function eal_cpu_core_id().
> So I think the number in above "cpuX" must be the lcore_id in DPDK.
> And the similar interface in power lib also directly use the locore_id.
Then it should be OK.
Thanks for the detailed answer.
> >
> > @David, do you know?
> >
> >> +
> >> +int
> >> +rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
> >> +{
> >> + char buf[BUFSIZ] = {0};
> >> + FILE *f;
> >> + int ret;
> >> +
> >> + if (lcore_id >= RTE_MAX_LCORE) {
> >> + POWER_LOG(ERR, "Lcore id %u can not exceeds %u",
> >> + lcore_id, RTE_MAX_LCORE - 1U);
> >> + return -EINVAL;
> >> + }
> > The lcore_id could be a registered non-EAL thread.
> > You should probably fail in that case.
> right, how about use rte_lcore_is_enabled(locore_id)?
I suppose setting latency for service cores should be forbidden too,
so using rte_lcore_is_enabled() to check for ROLE_RTE is correct.
> >
> > Same comment for rte_power_qos_get_cpu_resume_latency().
> >
> >
> >> +#define PM_QOS_STRICT_LATENCY_VALUE 0
> >> +#define PM_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
> > These definitions are in the public header file, and thus should be
> RTE_POWER_ prefixed and have comments describing them.
> Ack
> >
> >
> > .
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v3 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (3 preceding siblings ...)
2024-06-13 11:20 ` [PATCH v2 0/2] power: " Huisong Li
@ 2024-06-19 6:31 ` Huisong Li
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
` (2 more replies)
2024-06-27 6:00 ` [PATCH v4 " Huisong Li
` (5 subsequent siblings)
10 siblings, 3 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-19 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong, lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 22 +++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 29 +++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 71 +++++++++++++++
lib/power/version.map | 2 +
7 files changed, 244 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v3 1/2] power: introduce PM QoS API on CPU wide
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-06-19 6:31 ` Huisong Li
2024-06-19 14:56 ` Stephen Hemminger
2024-06-19 15:32 ` Thomas Monjalon
2024-06-19 6:31 ` [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-06-19 6:59 ` [PATCH v3 0/2] power: introduce PM QoS interface Morten Brørup
2 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-19 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong, lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
doc/guides/prog_guide/power_man.rst | 22 +++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 71 +++++++++++++++
lib/power/version.map | 2 +
6 files changed, 215 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..3ff46f06c1 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,28 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can effect the work
+CPU's idle state selection and just allow to enter the shallowest idle state
+if set to zero (strict resume latency) for this CPU.
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can obtain the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index e68a53d757..7c0d36e389 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -89,6 +89,10 @@ New Features
* Added SSE/NEON vector datapath.
+* **Introduce PM QoS interface.**
+
+ * Introduce PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..b131cf58e7
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[BUFSIZ] = {0};
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ sprintf(buf, "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ sprintf(buf, "%u", 0);
+ else
+ sprintf(buf, "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[BUFSIZ];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..2b25d0d4c1
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,71 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-19 6:31 ` Huisong Li
2024-06-19 14:54 ` Stephen Hemminger
2024-06-19 6:59 ` [PATCH v3 0/2] power: introduce PM QoS interface Morten Brørup
2 siblings, 1 reply; 56+ messages in thread
From: Huisong Li @ 2024-06-19 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong, lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
---
examples/l3fwd-power/main.c | 29 +++++++++++++++++++++++++++++
1 file changed, 29 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..d263cfef0a 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,25 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /*
+ * Set the work CPU with strict latency limit to allow the
+ * process running on the CPU can only enter the shallowest
+ * idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2318,15 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* RE: [PATCH v3 0/2] power: introduce PM QoS interface
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-19 6:31 ` [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
@ 2024-06-19 6:59 ` Morten Brørup
2 siblings, 0 replies; 56+ messages in thread
From: Morten Brørup @ 2024-06-19 6:59 UTC (permalink / raw)
To: Huisong Li, dev
Cc: thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
> From: Huisong Li [mailto:lihuisong@huawei.com]
> Sent: Wednesday, 19 June 2024 08.32
>
> The deeper the idle state, the lower the power consumption, but the longer
> the resume time. Some service are delay sensitive and very except the low
> resume time, like interrupt packet receiving mode.
>
> And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
> interface is used to set and get the resume latency limit on the cpuX for
> userspace. Please see the description in kernel document[1].
> Each cpuidle governor in Linux select which idle state to enter based on
> this CPU resume latency in their idle task.
>
> The per-CPU PM QoS API can be used to control this CPU's idle state
> selection and limit just enter the shallowest idle state to low the delay
> after sleep by setting strict resume latency (zero value).
>
> [1] https://www.kernel.org/doc/html/latest/admin-guide/abi-
> testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-
> resume-latency-us
>
> ---
> v3:
> - add RTE_POWER_xxx prefix for some macro in header
> - add the check for lcore_id with rte_lcore_is_enabled
> v2:
> - use PM QoS on CPU wide to replace the one on system wide
Series-acked-by: Morten Brørup <mb@smartsharesystems.com>
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration
2024-06-19 6:31 ` [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
@ 2024-06-19 14:54 ` Stephen Hemminger
2024-06-20 2:24 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Stephen Hemminger @ 2024-06-19 14:54 UTC (permalink / raw)
To: Huisong Li
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
On Wed, 19 Jun 2024 14:31:44 +0800
Huisong Li <lihuisong@huawei.com> wrote:
> + /*
> + * Set the work CPU with strict latency limit to allow the
> + * process running on the CPU can only enter the shallowest
> + * idle state.
> + */
Wording of that comment is awkward to read.
Suggest:
Set the worker lcore's to have strict latency limit to allow the
CPU to enter the shallowest idle state.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 1/2] power: introduce PM QoS API on CPU wide
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-19 14:56 ` Stephen Hemminger
2024-06-20 2:22 ` lihuisong (C)
2024-06-19 15:32 ` Thomas Monjalon
1 sibling, 1 reply; 56+ messages in thread
From: Stephen Hemminger @ 2024-06-19 14:56 UTC (permalink / raw)
To: Huisong Li
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
On Wed, 19 Jun 2024 14:31:43 +0800
Huisong Li <lihuisong@huawei.com> wrote:
> +PM QoS
> +------
> +
> +The deeper the idle state, the lower the power consumption, but the longer
> +the resume time. Some service are delay sensitive and very except the low
> +resume time, like interrupt packet receiving mode.
> +
> +And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
> +interface is used to set and get the resume latency limit on the cpuX for
> +userspace. Each cpuidle governor in Linux select which idle state to enter
> +based on this CPU resume latency in their idle task.
> +
> +The per-CPU PM QoS API can be used to set and get the CPU resume latency.
> +
> +The ``rte_power_qos_set_cpu_resume_latency()`` function can effect the work
> +CPU's idle state selection and just allow to enter the shallowest idle state
> +if set to zero (strict resume latency) for this CPU.
> +
> +The ``rte_power_qos_get_cpu_resume_latency()`` function can obtain the resume
> +latency on specified CPU.
> +
Wording of this is hard to read and needs to be reworded for clarity.
Explain more what PM QoS is to the user.
Also, not sure if details about sysfs implementation is helpful.
Should also say this is Linux only.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 1/2] power: introduce PM QoS API on CPU wide
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-19 14:56 ` Stephen Hemminger
@ 2024-06-19 15:32 ` Thomas Monjalon
2024-06-20 2:32 ` lihuisong (C)
1 sibling, 1 reply; 56+ messages in thread
From: Thomas Monjalon @ 2024-06-19 15:32 UTC (permalink / raw)
To: Huisong Li
Cc: dev, mb, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong, lihuisong
19/06/2024 08:31, Huisong Li:
> --- /dev/null
> +++ b/lib/power/rte_power_qos.h
> @@ -0,0 +1,71 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2024 HiSilicon Limited
> + */
> +
> +#ifndef RTE_POWER_QOS_H
> +#define RTE_POWER_QOS_H
> +
> +#include <rte_compat.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +/**
> + * @file rte_power_qos.h
> + *
> + * PM QoS API.
> + *
> + * The CPU-wide resume latency limit has a positive impact on this CPU's idle
> + * state selection in each cpuidle governor.
> + * Please see the PM QoS on CPU wide in the following link:
> + * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
> + *
> + * The deeper the idle state, the lower the power consumption, but the
> + * longer the resume time. Some service are delay sensitive and very except the
> + * low resume time, like interrupt packet receiving mode.
> + *
> + * In these case, per-CPU PM QoS API can be used to control this CPU's idle
> + * state selection and limit just enter the shallowest idle state to low the
> + * delay after sleep by setting strict resume latency (zero value).
> + */
> +
> +#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
> +#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
stdint.h include is missing
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 1/2] power: introduce PM QoS API on CPU wide
2024-06-19 14:56 ` Stephen Hemminger
@ 2024-06-20 2:22 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-06-20 2:22 UTC (permalink / raw)
To: Stephen Hemminger
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
Hi Stephen,
Thanks for your review.
在 2024/6/19 22:56, Stephen Hemminger 写道:
> On Wed, 19 Jun 2024 14:31:43 +0800
> Huisong Li <lihuisong@huawei.com> wrote:
>
>> +PM QoS
>> +------
>> +
>> +The deeper the idle state, the lower the power consumption, but the longer
>> +the resume time. Some service are delay sensitive and very except the low
>> +resume time, like interrupt packet receiving mode.
>> +
>> +And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
>> +interface is used to set and get the resume latency limit on the cpuX for
>> +userspace. Each cpuidle governor in Linux select which idle state to enter
>> +based on this CPU resume latency in their idle task.
>> +
>> +The per-CPU PM QoS API can be used to set and get the CPU resume latency.
>> +
>> +The ``rte_power_qos_set_cpu_resume_latency()`` function can effect the work
>> +CPU's idle state selection and just allow to enter the shallowest idle state
>> +if set to zero (strict resume latency) for this CPU.
>> +
>> +The ``rte_power_qos_get_cpu_resume_latency()`` function can obtain the resume
>> +latency on specified CPU.
>> +
> Wording of this is hard to read and needs to be reworded for clarity.
> Explain more what PM QoS is to the user.
Yes, it's very important. How do you feel about the following description?
-->
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to set and get the CPU resume latency
based
on this sysfs.
The ``rte_power_qos_set_cpu_resume_latency()`` function can control the
CPU's
idle state selection in Linux and limit just to enter the shallowest
idle state
to low the delay of resuming service after sleeping by setting strict resume
latency (zero value).
The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
latency on specified CPU.
> Also, not sure if details about sysfs implementation is helpful.
It's just a short background.
IMO, it is helpful for user to make sense of this API.
> Should also say this is Linux only.
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration
2024-06-19 14:54 ` Stephen Hemminger
@ 2024-06-20 2:24 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-06-20 2:24 UTC (permalink / raw)
To: Stephen Hemminger
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
在 2024/6/19 22:54, Stephen Hemminger 写道:
> On Wed, 19 Jun 2024 14:31:44 +0800
> Huisong Li <lihuisong@huawei.com> wrote:
>
>> + /*
>> + * Set the work CPU with strict latency limit to allow the
>> + * process running on the CPU can only enter the shallowest
>> + * idle state.
>> + */
> Wording of that comment is awkward to read.
>
> Suggest:
>
> Set the worker lcore's to have strict latency limit to allow the
> CPU to enter the shallowest idle state.
Ack. Thanks for your suggestion.
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v3 1/2] power: introduce PM QoS API on CPU wide
2024-06-19 15:32 ` Thomas Monjalon
@ 2024-06-20 2:32 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-06-20 2:32 UTC (permalink / raw)
To: Thomas Monjalon
Cc: dev, mb, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
在 2024/6/19 23:32, Thomas Monjalon 写道:
> 19/06/2024 08:31, Huisong Li:
>> --- /dev/null
>> +++ b/lib/power/rte_power_qos.h
>> @@ -0,0 +1,71 @@
>> +/* SPDX-License-Identifier: BSD-3-Clause
>> + * Copyright(c) 2024 HiSilicon Limited
>> + */
>> +
>> +#ifndef RTE_POWER_QOS_H
>> +#define RTE_POWER_QOS_H
>> +
>> +#include <rte_compat.h>
>> +
>> +#ifdef __cplusplus
>> +extern "C" {
>> +#endif
>> +
>> +/**
>> + * @file rte_power_qos.h
>> + *
>> + * PM QoS API.
>> + *
>> + * The CPU-wide resume latency limit has a positive impact on this CPU's idle
>> + * state selection in each cpuidle governor.
>> + * Please see the PM QoS on CPU wide in the following link:
>> + * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
>> + *
>> + * The deeper the idle state, the lower the power consumption, but the
>> + * longer the resume time. Some service are delay sensitive and very except the
>> + * low resume time, like interrupt packet receiving mode.
>> + *
>> + * In these case, per-CPU PM QoS API can be used to control this CPU's idle
>> + * state selection and limit just enter the shallowest idle state to low the
>> + * delay after sleep by setting strict resume latency (zero value).
>> + */
>> +
>> +#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
>> +#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
> stdint.h include is missing
Yes, it desn't satisfy self-contained header files.
will add it in next version, thanks.
>
>
>
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v4 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (4 preceding siblings ...)
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-06-27 6:00 ` Huisong Li
2024-06-27 6:00 ` [PATCH v4 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-27 6:00 ` [PATCH v4 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-02 3:50 ` [PATCH v5 0/2] power: introduce PM QoS interface Huisong Li
` (4 subsequent siblings)
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-27 6:00 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 28 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
7 files changed, 247 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v4 1/2] power: introduce PM QoS API on CPU wide
2024-06-27 6:00 ` [PATCH v4 " Huisong Li
@ 2024-06-27 6:00 ` Huisong Li
2024-06-27 15:06 ` Stephen Hemminger
2024-06-27 6:00 ` [PATCH v4 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 1 reply; 56+ messages in thread
From: Huisong Li @ 2024-06-27 6:00 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
6 files changed, 219 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index e68a53d757..4de96f60ac 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -89,6 +89,10 @@ New Features
* Added SSE/NEON vector datapath.
+* **Introduce PM QoS interface.**
+
+ * Introduce per-CPU PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..b131cf58e7
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[BUFSIZ] = {0};
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ sprintf(buf, "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ sprintf(buf, "%u", 0);
+ else
+ sprintf(buf, "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[BUFSIZ];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v4 2/2] examples/l3fwd-power: add PM QoS configuration
2024-06-27 6:00 ` [PATCH v4 " Huisong Li
2024-06-27 6:00 ` [PATCH v4 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-27 6:00 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-06-27 6:00 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 28 ++++++++++++++++++++++++++++
1 file changed, 28 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..74a07afc6c 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,24 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2317,15 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v4 1/2] power: introduce PM QoS API on CPU wide
2024-06-27 6:00 ` [PATCH v4 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-06-27 15:06 ` Stephen Hemminger
2024-06-28 4:07 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Stephen Hemminger @ 2024-06-27 15:06 UTC (permalink / raw)
To: Huisong Li
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
On Thu, 27 Jun 2024 14:00:10 +0800
Huisong Li <lihuisong@huawei.com> wrote:
> + char buf[BUFSIZ] = {0};
BUFSIZ is 4K and you probably don't need all of that.
And initializing to 0 here should not be needed.
Why not:
char buf[LINE_MAX];
> + if (latency == 0)
> + sprintf(buf, "%s", "n/a");
> + else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
> + sprintf(buf, "%u", 0);
> + else
> + sprintf(buf, "%u", latency);
Use snprintf instead.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v4 1/2] power: introduce PM QoS API on CPU wide
2024-06-27 15:06 ` Stephen Hemminger
@ 2024-06-28 4:07 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-06-28 4:07 UTC (permalink / raw)
To: Stephen Hemminger
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
在 2024/6/27 23:06, Stephen Hemminger 写道:
> On Thu, 27 Jun 2024 14:00:10 +0800
> Huisong Li <lihuisong@huawei.com> wrote:
>
>> + char buf[BUFSIZ] = {0};
> BUFSIZ is 4K and you probably don't need all of that.
I rember the maximum buffer length of sysfs show in Linux is BUFSIZ.
Just from the same size, here is ok to receive the data from Linux.
But LINE_MAX is also enough to use.
> And initializing to 0 here should not be needed.
Ack
>
> Why not:
> char buf[LINE_MAX];
Thanks for your suggestion. use it in next version.
>> + if (latency == 0)
>> + sprintf(buf, "%s", "n/a");
>> + else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
>> + sprintf(buf, "%u", 0);
>> + else
>> + sprintf(buf, "%u", latency);
> Use snprintf instead.
Ack
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v5 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (5 preceding siblings ...)
2024-06-27 6:00 ` [PATCH v4 " Huisong Li
@ 2024-07-02 3:50 ` Huisong Li
2024-07-02 3:50 ` [PATCH v5 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-02 3:50 ` [PATCH v5 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 2:29 ` [PATCH v6 0/2] power: introduce PM QoS interface Huisong Li
` (3 subsequent siblings)
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-02 3:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v5:
- use LINE_MAX to replace BUFSIZ, and use snprintf to replace sprintf.
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 28 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
7 files changed, 247 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v5 1/2] power: introduce PM QoS API on CPU wide
2024-07-02 3:50 ` [PATCH v5 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-02 3:50 ` Huisong Li
2024-07-03 1:32 ` zhoumin
2024-07-02 3:50 ` [PATCH v5 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 1 reply; 56+ messages in thread
From: Huisong Li @ 2024-07-02 3:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
6 files changed, 219 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index e68a53d757..4de96f60ac 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -89,6 +89,10 @@ New Features
* Added SSE/NEON vector datapath.
+* **Introduce PM QoS interface.**
+
+ * Introduce per-CPU PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..375746f832
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[LINE_MAX];
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ snprintf(buf, sizeof(buf), "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ snprintf(buf, sizeof(buf), "%u", 0);
+ else
+ snprintf(buf, sizeof(buf), "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[LINE_MAX];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v5 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-02 3:50 ` [PATCH v5 0/2] power: introduce PM QoS interface Huisong Li
2024-07-02 3:50 ` [PATCH v5 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-07-02 3:50 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-02 3:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 28 ++++++++++++++++++++++++++++
1 file changed, 28 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..74a07afc6c 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,24 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2317,15 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v5 1/2] power: introduce PM QoS API on CPU wide
2024-07-02 3:50 ` [PATCH v5 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-07-03 1:32 ` zhoumin
2024-07-03 2:52 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: zhoumin @ 2024-07-03 1:32 UTC (permalink / raw)
To: Huisong Li, dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong
Hi Huisong,
I knew that this patchset was based on the *dpdk-next-net* repository in
our previous communication. However, maybe it's better to rebase this
pactchset on the *dpdk* repository. Because the CI system is not smart
enough to recognize the patchset as a change for the *dpdk-next-net*
repository. I personally feel this patchset is more likely a change for
the *dpdk* repository because it modified the `lib` directory which is
the infrastructure of DPDK instead of a feature for *dpdk-next-net*.
Best regard,
Min Zhou
On Tue, July 2, 2024 at 11:50AM, Huisong Li wrote:
> The deeper the idle state, the lower the power consumption, but the longer
> the resume time. Some service are delay sensitive and very except the low
> resume time, like interrupt packet receiving mode.
>
> And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
> interface is used to set and get the resume latency limit on the cpuX for
> userspace. Each cpuidle governor in Linux select which idle state to enter
> based on this CPU resume latency in their idle task.
>
> The per-CPU PM QoS API can be used to control this CPU's idle state
> selection and limit just enter the shallowest idle state to low the delay
> after sleep by setting strict resume latency (zero value).
>
> Signed-off-by: Huisong Li <lihuisong@huawei.com>
> Acked-by: Morten Brørup <mb@smartsharesystems.com>
> ---
> doc/guides/prog_guide/power_man.rst | 24 ++++++
> doc/guides/rel_notes/release_24_07.rst | 4 +
> lib/power/meson.build | 2 +
> lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
> lib/power/rte_power_qos.h | 73 ++++++++++++++++
> lib/power/version.map | 2 +
> 6 files changed, 219 insertions(+)
> create mode 100644 lib/power/rte_power_qos.c
> create mode 100644 lib/power/rte_power_qos.h
<snip>
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v5 1/2] power: introduce PM QoS API on CPU wide
2024-07-03 1:32 ` zhoumin
@ 2024-07-03 2:52 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-07-03 2:52 UTC (permalink / raw)
To: zhoumin, dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong
Hi
在 2024/7/3 9:32, zhoumin 写道:
> Hi Huisong,
>
> I knew that this patchset was based on the *dpdk-next-net* repository
> in our previous communication. However, maybe it's better to rebase
> this pactchset on the *dpdk* repository. Because the CI system is not
> smart enough to recognize the patchset as a change for the
> *dpdk-next-net* repository. I personally feel this patchset is more
> likely a change for the *dpdk* repository because it modified the
> `lib` directory which is the infrastructure of DPDK instead of a
> feature for *dpdk-next-net*.
>
Generally, the 'dpdk-next-net' repo is ahead of 'dpdk' repo. And we make
some development based on 'dpdk-next-net' repo.
I found that there are some patches also happen to the same issue that
CI applt patch failed.
So I think the reason why this series trigger warning should be analyzed
and resolved.
>
> On Tue, July 2, 2024 at 11:50AM, Huisong Li wrote:
>> The deeper the idle state, the lower the power consumption, but the
>> longer
>> the resume time. Some service are delay sensitive and very except the
>> low
>> resume time, like interrupt packet receiving mode.
>>
>> And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us"
>> sysfs
>> interface is used to set and get the resume latency limit on the cpuX
>> for
>> userspace. Each cpuidle governor in Linux select which idle state to
>> enter
>> based on this CPU resume latency in their idle task.
>>
>> The per-CPU PM QoS API can be used to control this CPU's idle state
>> selection and limit just enter the shallowest idle state to low the
>> delay
>> after sleep by setting strict resume latency (zero value).
>>
>> Signed-off-by: Huisong Li <lihuisong@huawei.com>
>> Acked-by: Morten Brørup <mb@smartsharesystems.com>
>> ---
>> doc/guides/prog_guide/power_man.rst | 24 ++++++
>> doc/guides/rel_notes/release_24_07.rst | 4 +
>> lib/power/meson.build | 2 +
>> lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
>> lib/power/rte_power_qos.h | 73 ++++++++++++++++
>> lib/power/version.map | 2 +
>> 6 files changed, 219 insertions(+)
>> create mode 100644 lib/power/rte_power_qos.c
>> create mode 100644 lib/power/rte_power_qos.h
> <snip>
>
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v6 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (6 preceding siblings ...)
2024-07-02 3:50 ` [PATCH v5 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 2:29 ` Huisong Li
2024-07-09 2:29 ` [PATCH v6 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 2:29 ` [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 6:31 ` [PATCH v7 0/2] power: introduce PM QoS interface Huisong Li
` (2 subsequent siblings)
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 2:29 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v6:
- update release_24_07.rst based on dpdk repo to resolve CI warning.
v5:
- use LINE_MAX to replace BUFSIZ, and use snprintf to replace sprintf.
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 28 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
7 files changed, 247 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v6 1/2] power: introduce PM QoS API on CPU wide
2024-07-09 2:29 ` [PATCH v6 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 2:29 ` Huisong Li
2024-07-09 2:29 ` [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 2:29 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
6 files changed, 219 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index 1dd842df3a..af6fd82a3c 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -155,6 +155,10 @@ New Features
Added an API that allows the user to reclaim the defer queue with RCU.
+* **Introduce per-CPU PM QoS interface.**
+
+ * Introduce per-CPU PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..375746f832
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[LINE_MAX];
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ snprintf(buf, sizeof(buf), "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ snprintf(buf, sizeof(buf), "%u", 0);
+ else
+ snprintf(buf, sizeof(buf), "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[LINE_MAX];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-09 2:29 ` [PATCH v6 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 2:29 ` [PATCH v6 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-07-09 2:29 ` Huisong Li
2024-07-09 3:07 ` Stephen Hemminger
1 sibling, 1 reply; 56+ messages in thread
From: Huisong Li @ 2024-07-09 2:29 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 28 ++++++++++++++++++++++++++++
1 file changed, 28 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..74a07afc6c 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,24 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2317,15 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ if (rte_lcore_is_enabled(lcore_id) == 0)
+ continue;
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-09 2:29 ` [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
@ 2024-07-09 3:07 ` Stephen Hemminger
2024-07-09 3:18 ` lihuisong (C)
0 siblings, 1 reply; 56+ messages in thread
From: Stephen Hemminger @ 2024-07-09 3:07 UTC (permalink / raw)
To: Huisong Li
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
On Tue, 9 Jul 2024 10:29:27 +0800
Huisong Li <lihuisong@huawei.com> wrote:
> + RTE_LCORE_FOREACH(lcore_id) {
> + if (rte_lcore_is_enabled(lcore_id) == 0)
> + continue;
>
Why do you need this check? RTE_LCORE_FOREACH calls rte_next_lcore which
already skips lcores that are not enabled.
^ permalink raw reply [flat|nested] 56+ messages in thread
* Re: [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-09 3:07 ` Stephen Hemminger
@ 2024-07-09 3:18 ` lihuisong (C)
0 siblings, 0 replies; 56+ messages in thread
From: lihuisong (C) @ 2024-07-09 3:18 UTC (permalink / raw)
To: Stephen Hemminger
Cc: dev, mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, david.marchand, liuyonglong
在 2024/7/9 11:07, Stephen Hemminger 写道:
> On Tue, 9 Jul 2024 10:29:27 +0800
> Huisong Li <lihuisong@huawei.com> wrote:
>
>> + RTE_LCORE_FOREACH(lcore_id) {
>> + if (rte_lcore_is_enabled(lcore_id) == 0)
>> + continue;
>>
> Why do you need this check? RTE_LCORE_FOREACH calls rte_next_lcore which
> already skips lcores that are not enabled.
Yes, it is dead code. will delete it in next version. Thanks.
> .
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v7 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (7 preceding siblings ...)
2024-07-09 2:29 ` [PATCH v6 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 6:31 ` Huisong Li
2024-07-09 6:31 ` [PATCH v7 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 6:31 ` [PATCH v7 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 7:25 ` [PATCH v8 0/2] power: introduce PM QoS interface Huisong Li
2024-08-09 9:50 ` [PATCH v9 0/2] power: introduce PM QoS interface Huisong Li
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v7:
- remove a dead code rte_lcore_is_enabled in patch[2/2]
v6:
- update release_24_07.rst based on dpdk repo to resolve CI warning.
v5:
- use LINE_MAX to replace BUFSIZ, and use snprintf to replace sprintf.
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 24 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
7 files changed, 243 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v7 1/2] power: introduce PM QoS API on CPU wide
2024-07-09 6:31 ` [PATCH v7 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 6:31 ` Huisong Li
2024-07-09 6:31 ` [PATCH v7 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
6 files changed, 219 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index 1dd842df3a..af6fd82a3c 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -155,6 +155,10 @@ New Features
Added an API that allows the user to reclaim the defer queue with RCU.
+* **Introduce per-CPU PM QoS interface.**
+
+ * Introduce per-CPU PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..375746f832
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[LINE_MAX];
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ snprintf(buf, sizeof(buf), "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ snprintf(buf, sizeof(buf), "%u", 0);
+ else
+ snprintf(buf, sizeof(buf), "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[LINE_MAX];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v7 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-09 6:31 ` [PATCH v7 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 6:31 ` [PATCH v7 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-07-09 6:31 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 6:31 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..d518e19467 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,22 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2315,13 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v8 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (8 preceding siblings ...)
2024-07-09 6:31 ` [PATCH v7 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 7:25 ` Huisong Li
2024-07-09 7:25 ` [PATCH v8 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 7:25 ` [PATCH v8 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-08-09 9:50 ` [PATCH v9 0/2] power: introduce PM QoS interface Huisong Li
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 7:25 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v8:
- update the latest code to resolve CI warning
v7:
- remove a dead code rte_lcore_is_enabled in patch[2/2]
v6:
- update release_24_07.rst based on dpdk repo to resolve CI warning.
v5:
- use LINE_MAX to replace BUFSIZ, and use snprintf to replace sprintf.
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
examples/l3fwd-power/main.c | 24 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
7 files changed, 243 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v8 1/2] power: introduce PM QoS API on CPU wide
2024-07-09 7:25 ` [PATCH v8 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-07-09 7:25 ` Huisong Li
2024-07-09 7:25 ` [PATCH v8 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 7:25 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_07.rst | 4 +
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 2 +
6 files changed, 219 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_07.rst b/doc/guides/rel_notes/release_24_07.rst
index 50ffc1f74a..e771868d9f 100644
--- a/doc/guides/rel_notes/release_24_07.rst
+++ b/doc/guides/rel_notes/release_24_07.rst
@@ -156,6 +156,10 @@ New Features
* Added defer queue reclamation via RCU.
* Added SVE support for bulk lookup.
+* **Introduce per-CPU PM QoS interface.**
+
+ * Introduce per-CPU PM QoS interface to low the delay after sleep.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..375746f832
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[LINE_MAX];
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ snprintf(buf, sizeof(buf), "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ snprintf(buf, sizeof(buf), "%u", 0);
+ else
+ snprintf(buf, sizeof(buf), "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[LINE_MAX];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index ad92a65f91..81b8ff11b7 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,6 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v8 2/2] examples/l3fwd-power: add PM QoS configuration
2024-07-09 7:25 ` [PATCH v8 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 7:25 ` [PATCH v8 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-07-09 7:25 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-07-09 7:25 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fba11da7ca..d518e19467 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2259,6 +2260,22 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2298,6 +2315,13 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v9 0/2] power: introduce PM QoS interface
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
` (9 preceding siblings ...)
2024-07-09 7:25 ` [PATCH v8 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-08-09 9:50 ` Huisong Li
2024-08-09 9:50 ` [PATCH v9 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-08-09 9:50 ` [PATCH v9 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
10 siblings, 2 replies; 56+ messages in thread
From: Huisong Li @ 2024-08-09 9:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Please see the description in kernel document[1].
Each cpuidle governor in Linux select which idle state to enter based on
this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
[1] https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
---
v9:
- move new feature description from release_24_07.rst to release_24_11.rst.
v8:
- update the latest code to resolve CI warning
v7:
- remove a dead code rte_lcore_is_enabled in patch[2/2]
v6:
- update release_24_07.rst based on dpdk repo to resolve CI warning.
v5:
- use LINE_MAX to replace BUFSIZ, and use snprintf to replace sprintf.
v4:
- fix some comments basd on Stephen
- add stdint.h include
- add Acked-by Morten Brørup <mb@smartsharesystems.com>
v3:
- add RTE_POWER_xxx prefix for some macro in header
- add the check for lcore_id with rte_lcore_is_enabled
v2:
- use PM QoS on CPU wide to replace the one on system wide
Huisong Li (2):
power: introduce PM QoS API on CPU wide
examples/l3fwd-power: add PM QoS configuration
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_11.rst | 5 ++
examples/l3fwd-power/main.c | 24 ++++++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 4 +
7 files changed, 246 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v9 1/2] power: introduce PM QoS API on CPU wide
2024-08-09 9:50 ` [PATCH v9 0/2] power: introduce PM QoS interface Huisong Li
@ 2024-08-09 9:50 ` Huisong Li
2024-08-09 9:50 ` [PATCH v9 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-08-09 9:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
The deeper the idle state, the lower the power consumption, but the longer
the resume time. Some service are delay sensitive and very except the low
resume time, like interrupt packet receiving mode.
And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
interface is used to set and get the resume latency limit on the cpuX for
userspace. Each cpuidle governor in Linux select which idle state to enter
based on this CPU resume latency in their idle task.
The per-CPU PM QoS API can be used to control this CPU's idle state
selection and limit just enter the shallowest idle state to low the delay
after sleep by setting strict resume latency (zero value).
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
doc/guides/prog_guide/power_man.rst | 24 ++++++
doc/guides/rel_notes/release_24_11.rst | 5 ++
lib/power/meson.build | 2 +
lib/power/rte_power_qos.c | 114 +++++++++++++++++++++++++
lib/power/rte_power_qos.h | 73 ++++++++++++++++
lib/power/version.map | 4 +
6 files changed, 222 insertions(+)
create mode 100644 lib/power/rte_power_qos.c
create mode 100644 lib/power/rte_power_qos.h
diff --git a/doc/guides/prog_guide/power_man.rst b/doc/guides/prog_guide/power_man.rst
index f6674efe2d..faa32b4320 100644
--- a/doc/guides/prog_guide/power_man.rst
+++ b/doc/guides/prog_guide/power_man.rst
@@ -249,6 +249,30 @@ Get Num Pkgs
Get Num Dies
Get the number of die's on a given package.
+
+PM QoS
+------
+
+The deeper the idle state, the lower the power consumption, but the longer
+the resume time. Some service are delay sensitive and very except the low
+resume time, like interrupt packet receiving mode.
+
+And the "/sys/devices/system/cpu/cpuX/power/pm_qos_resume_latency_us" sysfs
+interface is used to set and get the resume latency limit on the cpuX for
+userspace. Each cpuidle governor in Linux select which idle state to enter
+based on this CPU resume latency in their idle task.
+
+The per-CPU PM QoS API can be used to set and get the CPU resume latency based
+on this sysfs.
+
+The ``rte_power_qos_set_cpu_resume_latency()`` function can control the CPU's
+idle state selection in Linux and limit just to enter the shallowest idle state
+to low the delay of resuming service after sleeping by setting strict resume
+latency (zero value).
+
+The ``rte_power_qos_get_cpu_resume_latency()`` function can get the resume
+latency on specified CPU.
+
References
----------
diff --git a/doc/guides/rel_notes/release_24_11.rst b/doc/guides/rel_notes/release_24_11.rst
index 0ff70d9057..bd72d0a595 100644
--- a/doc/guides/rel_notes/release_24_11.rst
+++ b/doc/guides/rel_notes/release_24_11.rst
@@ -55,6 +55,11 @@ New Features
Also, make sure to start the actual text at the margin.
=======================================================
+* **Introduce per-CPU PM QoS interface.**
+
+ * Add per-CPU PM QoS interface to low the delay after sleep by controlling
+ CPU idle state selection.
+
Removed Items
-------------
diff --git a/lib/power/meson.build b/lib/power/meson.build
index b8426589b2..8222e178b0 100644
--- a/lib/power/meson.build
+++ b/lib/power/meson.build
@@ -23,12 +23,14 @@ sources = files(
'rte_power.c',
'rte_power_uncore.c',
'rte_power_pmd_mgmt.c',
+ 'rte_power_qos.c',
)
headers = files(
'rte_power.h',
'rte_power_guest_channel.h',
'rte_power_pmd_mgmt.h',
'rte_power_uncore.h',
+ 'rte_power_qos.h',
)
if cc.has_argument('-Wno-cast-qual')
cflags += '-Wno-cast-qual'
diff --git a/lib/power/rte_power_qos.c b/lib/power/rte_power_qos.c
new file mode 100644
index 0000000000..375746f832
--- /dev/null
+++ b/lib/power/rte_power_qos.c
@@ -0,0 +1,114 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_lcore.h>
+#include <rte_log.h>
+
+#include "power_common.h"
+#include "rte_power_qos.h"
+
+#define PM_QOS_SYSFILE_RESUME_LATENCY_US \
+ "/sys/devices/system/cpu/cpu%u/power/pm_qos_resume_latency_us"
+
+int
+rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency)
+{
+ char buf[LINE_MAX];
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ if (latency < 0) {
+ POWER_LOG(ERR, "latency should be greater than and equal to 0");
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "w", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different input string.
+ * 1> the resume latency is 0 if the input is "n/a".
+ * 2> the resume latency is no constraint if the input is "0".
+ * 3> the resume latency is the actual value to be set.
+ */
+ if (latency == 0)
+ snprintf(buf, sizeof(buf), "%s", "n/a");
+ else if (latency == RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT)
+ snprintf(buf, sizeof(buf), "%u", 0);
+ else
+ snprintf(buf, sizeof(buf), "%u", latency);
+
+ ret = write_core_sysfs_s(f, buf);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to write "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return ret;
+}
+
+int
+rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id)
+{
+ char buf[LINE_MAX];
+ int latency = -1;
+ FILE *f;
+ int ret;
+
+ if (!rte_lcore_is_enabled(lcore_id)) {
+ POWER_LOG(ERR, "lcore id %u is not enabled", lcore_id);
+ return -EINVAL;
+ }
+
+ ret = open_core_sysfs_file(&f, "r", PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to open "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ return ret;
+ }
+
+ ret = read_core_sysfs_s(f, buf, sizeof(buf));
+ if (ret != 0) {
+ POWER_LOG(ERR, "Failed to read "PM_QOS_SYSFILE_RESUME_LATENCY_US, lcore_id);
+ goto out;
+ }
+
+ /*
+ * Based on the sysfs interface pm_qos_resume_latency_us under
+ * @PM_QOS_SYSFILE_RESUME_LATENCY_US directory in kernel, their meanning
+ * is as follows for different output string.
+ * 1> the resume latency is 0 if the output is "n/a".
+ * 2> the resume latency is no constraint if the output is "0".
+ * 3> the resume latency is the actual value in used for other string.
+ */
+ if (strcmp(buf, "n/a") == 0)
+ latency = 0;
+ else {
+ latency = strtoul(buf, NULL, 10);
+ latency = latency == 0 ? RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT : latency;
+ }
+
+out:
+ if (f != NULL)
+ fclose(f);
+
+ return latency != -1 ? latency : ret;
+}
diff --git a/lib/power/rte_power_qos.h b/lib/power/rte_power_qos.h
new file mode 100644
index 0000000000..990c488373
--- /dev/null
+++ b/lib/power/rte_power_qos.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2024 HiSilicon Limited
+ */
+
+#ifndef RTE_POWER_QOS_H
+#define RTE_POWER_QOS_H
+
+#include <stdint.h>
+
+#include <rte_compat.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @file rte_power_qos.h
+ *
+ * PM QoS API.
+ *
+ * The CPU-wide resume latency limit has a positive impact on this CPU's idle
+ * state selection in each cpuidle governor.
+ * Please see the PM QoS on CPU wide in the following link:
+ * https://www.kernel.org/doc/html/latest/admin-guide/abi-testing.html?highlight=pm_qos_resume_latency_us#abi-sys-devices-power-pm-qos-resume-latency-us
+ *
+ * The deeper the idle state, the lower the power consumption, but the
+ * longer the resume time. Some service are delay sensitive and very except the
+ * low resume time, like interrupt packet receiving mode.
+ *
+ * In these case, per-CPU PM QoS API can be used to control this CPU's idle
+ * state selection and limit just enter the shallowest idle state to low the
+ * delay after sleep by setting strict resume latency (zero value).
+ */
+
+#define RTE_POWER_QOS_STRICT_LATENCY_VALUE 0
+#define RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT ((int)(UINT32_MAX >> 1))
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * @param lcore_id
+ * target logical core id
+ *
+ * @param latency
+ * The latency should be greater than and equal to zero in microseconds unit.
+ *
+ * @return
+ * 0 on success. Otherwise negative value is returned.
+ */
+__rte_experimental
+int rte_power_qos_set_cpu_resume_latency(uint16_t lcore_id, int latency);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the current resume latency of this logical core.
+ * The default value in kernel is @see RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT
+ * if don't set it.
+ *
+ * @return
+ * Negative value on failure.
+ * >= 0 means the actual resume latency limit on this core.
+ */
+__rte_experimental
+int rte_power_qos_get_cpu_resume_latency(uint16_t lcore_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* RTE_POWER_QOS_H */
diff --git a/lib/power/version.map b/lib/power/version.map
index c9a226614e..4e4955a4cf 100644
--- a/lib/power/version.map
+++ b/lib/power/version.map
@@ -51,4 +51,8 @@ EXPERIMENTAL {
rte_power_set_uncore_env;
rte_power_uncore_freqs;
rte_power_unset_uncore_env;
+
+ # added in 24.11
+ rte_power_qos_set_cpu_resume_latency;
+ rte_power_qos_get_cpu_resume_latency;
};
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
* [PATCH v9 2/2] examples/l3fwd-power: add PM QoS configuration
2024-08-09 9:50 ` [PATCH v9 0/2] power: introduce PM QoS interface Huisong Li
2024-08-09 9:50 ` [PATCH v9 1/2] power: introduce PM QoS API on CPU wide Huisong Li
@ 2024-08-09 9:50 ` Huisong Li
1 sibling, 0 replies; 56+ messages in thread
From: Huisong Li @ 2024-08-09 9:50 UTC (permalink / raw)
To: dev
Cc: mb, thomas, ferruh.yigit, anatoly.burakov, david.hunt,
sivaprasad.tummala, stephen, david.marchand, liuyonglong,
lihuisong
Add PM QoS configuration to declease the delay after sleep in case of
entering deeper idle state.
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
examples/l3fwd-power/main.c | 24 ++++++++++++++++++++++++
1 file changed, 24 insertions(+)
diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index 2bb6b092c3..9b386c3710 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -47,6 +47,7 @@
#include <rte_telemetry.h>
#include <rte_power_pmd_mgmt.h>
#include <rte_power_uncore.h>
+#include <rte_power_qos.h>
#include "perf_core.h"
#include "main.h"
@@ -2260,6 +2261,22 @@ init_power_library(void)
return -1;
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /*
+ * Set the worker lcore's to have strict latency limit to allow
+ * the CPU to enter the shallowest idle state.
+ */
+ ret = rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_STRICT_LATENCY_VALUE);
+ if (ret < 0) {
+ RTE_LOG(ERR, L3FWD_POWER,
+ "Failed to set strict resume latency on CPU%u.\n",
+ lcore_id);
+ return ret;
+ }
+ }
+
return ret;
}
@@ -2299,6 +2316,13 @@ deinit_power_library(void)
}
}
}
+
+ RTE_LCORE_FOREACH(lcore_id) {
+ /* Restore the original value in kernel. */
+ rte_power_qos_set_cpu_resume_latency(lcore_id,
+ RTE_POWER_QOS_RESUME_LATENCY_NO_CONSTRAINT);
+ }
+
return ret;
}
--
2.22.0
^ permalink raw reply [flat|nested] 56+ messages in thread
end of thread, other threads:[~2024-08-09 9:59 UTC | newest]
Thread overview: 56+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-03-20 10:55 [PATCH 0/2] introduce PM QoS interface Huisong Li
2024-03-20 10:55 ` [PATCH 1/2] power: " Huisong Li
2024-03-20 10:55 ` [PATCH 2/2] examples/l3fwd-power: add PM QoS request configuration Huisong Li
2024-03-20 14:05 ` [PATCH 0/2] introduce PM QoS interface Morten Brørup
2024-03-21 3:04 ` lihuisong (C)
2024-03-21 13:30 ` Morten Brørup
2024-03-22 8:54 ` lihuisong (C)
2024-03-22 12:35 ` Morten Brørup
2024-03-26 2:11 ` lihuisong (C)
2024-03-26 8:27 ` Morten Brørup
2024-03-26 12:15 ` lihuisong (C)
2024-03-26 12:46 ` Morten Brørup
2024-03-29 1:59 ` lihuisong (C)
2024-03-22 17:55 ` Tyler Retzlaff
2024-03-26 2:20 ` lihuisong (C)
2024-03-26 16:04 ` Tyler Retzlaff
2024-06-13 11:20 ` [PATCH v2 0/2] power: " Huisong Li
2024-06-13 11:20 ` [PATCH v2 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-14 8:04 ` Morten Brørup
2024-06-18 12:19 ` lihuisong (C)
2024-06-18 12:53 ` Morten Brørup
2024-06-13 11:20 ` [PATCH v2 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-06-19 6:31 ` [PATCH v3 0/2] power: introduce PM QoS interface Huisong Li
2024-06-19 6:31 ` [PATCH v3 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-19 14:56 ` Stephen Hemminger
2024-06-20 2:22 ` lihuisong (C)
2024-06-19 15:32 ` Thomas Monjalon
2024-06-20 2:32 ` lihuisong (C)
2024-06-19 6:31 ` [PATCH v3 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-06-19 14:54 ` Stephen Hemminger
2024-06-20 2:24 ` lihuisong (C)
2024-06-19 6:59 ` [PATCH v3 0/2] power: introduce PM QoS interface Morten Brørup
2024-06-27 6:00 ` [PATCH v4 " Huisong Li
2024-06-27 6:00 ` [PATCH v4 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-06-27 15:06 ` Stephen Hemminger
2024-06-28 4:07 ` lihuisong (C)
2024-06-27 6:00 ` [PATCH v4 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-02 3:50 ` [PATCH v5 0/2] power: introduce PM QoS interface Huisong Li
2024-07-02 3:50 ` [PATCH v5 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-03 1:32 ` zhoumin
2024-07-03 2:52 ` lihuisong (C)
2024-07-02 3:50 ` [PATCH v5 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 2:29 ` [PATCH v6 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 2:29 ` [PATCH v6 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 2:29 ` [PATCH v6 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 3:07 ` Stephen Hemminger
2024-07-09 3:18 ` lihuisong (C)
2024-07-09 6:31 ` [PATCH v7 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 6:31 ` [PATCH v7 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 6:31 ` [PATCH v7 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-07-09 7:25 ` [PATCH v8 0/2] power: introduce PM QoS interface Huisong Li
2024-07-09 7:25 ` [PATCH v8 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-07-09 7:25 ` [PATCH v8 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
2024-08-09 9:50 ` [PATCH v9 0/2] power: introduce PM QoS interface Huisong Li
2024-08-09 9:50 ` [PATCH v9 1/2] power: introduce PM QoS API on CPU wide Huisong Li
2024-08-09 9:50 ` [PATCH v9 2/2] examples/l3fwd-power: add PM QoS configuration Huisong Li
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).