From: luca.boccassi@gmail.com
To: Gagandeep Singh <g.singh@nxp.com>
Cc: Hemant Agrawal <hemant.agrawal@nxp.com>, dpdk stable <stable@dpdk.org>
Subject: patch 'bus/dpaa: fix PFDRs leaks due to FQRNIs' has been queued to stable release 22.11.7
Date: Wed, 23 Oct 2024 22:16:36 +0100 [thread overview]
Message-ID: <20241023211704.1216956-56-luca.boccassi@gmail.com> (raw)
In-Reply-To: <20241023211704.1216956-1-luca.boccassi@gmail.com>
Hi,
FYI, your patch has been queued to stable release 22.11.7
Note it hasn't been pushed to http://dpdk.org/browse/dpdk-stable yet.
It will be pushed if I get no objections before 10/25/24. So please
shout if anyone has objections.
Also note that after the patch there's a diff of the upstream commit vs the
patch applied to the branch. This will indicate if there was any rebasing
needed to apply to the stable branch. If there were code changes for rebasing
(ie: not only metadata diffs), please double check that the rebase was
correctly done.
Queued patches are on a temporary branch at:
https://github.com/bluca/dpdk-stable
This queued commit can be viewed at:
https://github.com/bluca/dpdk-stable/commit/0c1fd86cee21c314cecba532a87d0ca766f0f13a
Thanks.
Luca Boccassi
---
From 0c1fd86cee21c314cecba532a87d0ca766f0f13a Mon Sep 17 00:00:00 2001
From: Gagandeep Singh <g.singh@nxp.com>
Date: Tue, 1 Oct 2024 16:33:08 +0530
Subject: [PATCH] bus/dpaa: fix PFDRs leaks due to FQRNIs
[ upstream commit b292acc3c4a8fd5104cfdfa5c6d3d0df95b6543b ]
When a Retire FQ command is executed on a FQ in the
Tentatively Scheduled or Parked states, in that case FQ
is retired immediately and a FQRNI (Frame Queue Retirement
Notification Immediate) message is generated. Software
must read this message from MR and consume it to free
the memory used by it.
Although it is not mentioned about which memory to be used
by FQRNIs in the RM but through experiments it is proven
that it can use PFDRs. So if these messages are allowed to
build up indefinitely then PFDR resources can become exhausted
and cause enqueues to stall. Therefore software must consume
these MR messages on a regular basis to avoid depleting
the available PFDR resources.
This is the PFDRs leak issue which user can experience while
using the DPDK crypto driver and creating and destroying the
sessions multiple times. On a session destroy, DPDK calls the
qman_retire_fq() for each FQ used by the session, but it does
not handle the FQRNIs generated and allowed them to build up
indefinitely in MR.
This patch fixes this issue by consuming the FQRNIs received
from MR immediately after FQ retire by calling drain_mr_fqrni().
Please note that this drain_mr_fqrni() only look for
FQRNI type messages to consume. If there are other type of messages
like FQRN, FQRL, FQPN, ERN etc. also coming on MR then those
messages need to be handled separately.
Fixes: c47ff048b99a ("bus/dpaa: add QMAN driver core routines")
Signed-off-by: Gagandeep Singh <g.singh@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
---
drivers/bus/dpaa/base/qbman/qman.c | 46 ++++++++++++++++--------------
1 file changed, 25 insertions(+), 21 deletions(-)
diff --git a/drivers/bus/dpaa/base/qbman/qman.c b/drivers/bus/dpaa/base/qbman/qman.c
index 83db0a534e..f06992ca48 100644
--- a/drivers/bus/dpaa/base/qbman/qman.c
+++ b/drivers/bus/dpaa/base/qbman/qman.c
@@ -294,10 +294,32 @@ static inline void qman_stop_dequeues_ex(struct qman_portal *p)
qm_dqrr_set_maxfill(&p->p, 0);
}
+static inline void qm_mr_pvb_update(struct qm_portal *portal)
+{
+ register struct qm_mr *mr = &portal->mr;
+ const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi);
+
+#ifdef RTE_LIBRTE_DPAA_HWDEBUG
+ DPAA_ASSERT(mr->pmode == qm_mr_pvb);
+#endif
+ /* when accessing 'verb', use __raw_readb() to ensure that compiler
+ * inlining doesn't try to optimise out "excess reads".
+ */
+ if ((__raw_readb(&res->ern.verb) & QM_MR_VERB_VBIT) == mr->vbit) {
+ mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1);
+ if (!mr->pi)
+ mr->vbit ^= QM_MR_VERB_VBIT;
+ mr->fill++;
+ res = MR_INC(res);
+ }
+ dcbit_ro(res);
+}
+
static int drain_mr_fqrni(struct qm_portal *p)
{
const struct qm_mr_entry *msg;
loop:
+ qm_mr_pvb_update(p);
msg = qm_mr_current(p);
if (!msg) {
/*
@@ -319,6 +341,7 @@ loop:
do {
now = mfatb();
} while ((then + 10000) > now);
+ qm_mr_pvb_update(p);
msg = qm_mr_current(p);
if (!msg)
return 0;
@@ -481,27 +504,6 @@ static inline int qm_mr_init(struct qm_portal *portal,
return 0;
}
-static inline void qm_mr_pvb_update(struct qm_portal *portal)
-{
- register struct qm_mr *mr = &portal->mr;
- const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi);
-
-#ifdef RTE_LIBRTE_DPAA_HWDEBUG
- DPAA_ASSERT(mr->pmode == qm_mr_pvb);
-#endif
- /* when accessing 'verb', use __raw_readb() to ensure that compiler
- * inlining doesn't try to optimise out "excess reads".
- */
- if ((__raw_readb(&res->ern.verb) & QM_MR_VERB_VBIT) == mr->vbit) {
- mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1);
- if (!mr->pi)
- mr->vbit ^= QM_MR_VERB_VBIT;
- mr->fill++;
- res = MR_INC(res);
- }
- dcbit_ro(res);
-}
-
struct qman_portal *
qman_init_portal(struct qman_portal *portal,
const struct qm_portal_config *c,
@@ -1825,6 +1827,8 @@ int qman_retire_fq(struct qman_fq *fq, u32 *flags)
}
out:
FQUNLOCK(fq);
+ /* Draining FQRNIs, if any */
+ drain_mr_fqrni(&p->p);
return rval;
}
--
2.45.2
---
Diff of the applied patch vs upstream commit (please double-check if non-empty:
---
--- - 2024-10-23 22:16:42.644755490 +0100
+++ 0056-bus-dpaa-fix-PFDRs-leaks-due-to-FQRNIs.patch 2024-10-23 22:16:40.515943363 +0100
@@ -1 +1 @@
-From b292acc3c4a8fd5104cfdfa5c6d3d0df95b6543b Mon Sep 17 00:00:00 2001
+From 0c1fd86cee21c314cecba532a87d0ca766f0f13a Mon Sep 17 00:00:00 2001
@@ -5,0 +6,2 @@
+[ upstream commit b292acc3c4a8fd5104cfdfa5c6d3d0df95b6543b ]
+
@@ -37 +38,0 @@
-Cc: stable@dpdk.org
@@ -46 +47 @@
-index 301057723e..9c90ee25a6 100644
+index 83db0a534e..f06992ca48 100644
@@ -49 +50 @@
-@@ -292,10 +292,32 @@ static inline void qman_stop_dequeues_ex(struct qman_portal *p)
+@@ -294,10 +294,32 @@ static inline void qman_stop_dequeues_ex(struct qman_portal *p)
@@ -82 +83 @@
-@@ -317,6 +339,7 @@ loop:
+@@ -319,6 +341,7 @@ loop:
@@ -90 +91 @@
-@@ -479,27 +502,6 @@ static inline int qm_mr_init(struct qm_portal *portal,
+@@ -481,27 +504,6 @@ static inline int qm_mr_init(struct qm_portal *portal,
@@ -118 +119 @@
-@@ -1794,6 +1796,8 @@ int qman_retire_fq(struct qman_fq *fq, u32 *flags)
+@@ -1825,6 +1827,8 @@ int qman_retire_fq(struct qman_fq *fq, u32 *flags)
next prev parent reply other threads:[~2024-10-23 21:20 UTC|newest]
Thread overview: 79+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-10-23 21:15 patch 'devtools: fix forbidden token check with multiple files' " luca.boccassi
2024-10-23 21:15 ` patch 'eal/x86: fix 32-bit write combining store' " luca.boccassi
2024-10-23 21:15 ` patch 'examples/eventdev: fix queue crash with generic pipeline' " luca.boccassi
2024-10-23 21:15 ` patch 'crypto/dpaa2_sec: fix memory leak' " luca.boccassi
2024-10-23 21:15 ` patch 'common/dpaax/caamflib: fix PDCP SNOW-ZUC watchdog' " luca.boccassi
2024-10-23 21:15 ` patch 'dev: fix callback lookup when unregistering device' " luca.boccassi
2024-10-23 21:15 ` patch 'examples/ipsec-secgw: fix dequeue count from cryptodev' " luca.boccassi
2024-10-23 21:15 ` patch 'bpf: fix free function mismatch if convert fails' " luca.boccassi
2024-10-23 21:15 ` patch 'baseband/la12xx: fix use after free in modem config' " luca.boccassi
2024-10-23 21:15 ` patch 'crypto/bcmfs: fix free function mismatch' " luca.boccassi
2024-10-23 21:15 ` patch 'dma/idxd: fix free function mismatch in device probe' " luca.boccassi
2024-10-23 21:15 ` patch 'event/cnxk: fix free function mismatch in port config' " luca.boccassi
2024-10-23 21:15 ` patch 'net/e1000: fix use after free in filter flush' " luca.boccassi
2024-10-23 21:15 ` patch 'net/nfp: fix double free in flow destroy' " luca.boccassi
2024-10-23 21:15 ` patch 'net/sfc: fix use after free in debug logs' " luca.boccassi
2024-10-23 21:15 ` patch 'raw/ifpga/base: fix use after free' " luca.boccassi
2024-10-23 21:15 ` patch 'raw/ifpga: fix free function mismatch in interrupt config' " luca.boccassi
2024-10-23 21:15 ` patch 'examples/vhost: fix free function mismatch' " luca.boccassi
2024-10-23 21:15 ` patch 'net/nfb: fix use after free' " luca.boccassi
2024-10-23 21:16 ` patch 'power: enable CPPC' " luca.boccassi
2024-10-23 21:16 ` patch 'fib6: add runtime checks in AVX512 lookup' " luca.boccassi
2024-10-23 21:16 ` patch 'app/dumpcap: fix handling of jumbo frames' " luca.boccassi
2024-10-23 21:16 ` patch 'net/cnxk: fix Rx timestamp handling for VF' " luca.boccassi
2024-10-23 21:16 ` patch 'net/cnxk: fix Rx offloads to handle timestamp' " luca.boccassi
2024-10-23 21:16 ` patch 'event/cnxk: fix Rx timestamp handling' " luca.boccassi
2024-10-23 21:16 ` patch 'common/cnxk: fix CPT HW word size for outbound SA' " luca.boccassi
2024-10-23 21:16 ` patch 'common/cnxk: fix base log level' " luca.boccassi
2024-10-23 21:16 ` patch 'baseband/acc: fix access to deallocated mem' " luca.boccassi
2024-10-23 21:16 ` patch 'vhost: fix offset while mapping log base address' " luca.boccassi
2024-10-23 21:16 ` patch 'vdpa: update used flags in used ring relay' " luca.boccassi
2024-10-23 21:16 ` patch 'net/virtio-user: reset used index counter' " luca.boccassi
2024-10-23 21:16 ` patch 'fib: fix AVX512 lookup' " luca.boccassi
2024-10-23 21:16 ` patch 'net/e1000: fix link status crash in secondary process' " luca.boccassi
2024-10-23 21:16 ` patch 'net/iavf: fix crash when link is unstable' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ice/base: fix link speed for 200G' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ice/base: fix iteration of TLVs in Preserved Fields Area' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ixgbe/base: fix unchecked return value' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix setting flags in init function' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix misleading debug logs and comments' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix blinking X722 with X557 PHY' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix DDP loading with reserved track ID' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix repeated register dumps' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix unchecked return value' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e/base: fix loop bounds' " luca.boccassi
2024-10-23 21:16 ` patch 'net/i40e: fix AVX-512 pointer copy on 32-bit' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ice: " luca.boccassi
2024-10-23 21:16 ` patch 'net/iavf: " luca.boccassi
2024-10-23 21:16 ` patch 'net/tap: avoid memcpy with null argument' " luca.boccassi
2024-10-23 21:16 ` patch 'app/testpmd: remove unnecessary cast' " luca.boccassi
2024-10-23 21:16 ` patch 'net/pcap: set live interface as non-blocking' " luca.boccassi
2024-10-23 21:16 ` patch 'net/mana: support rdma-core via pkg-config' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ena: revert redefining memcpy' " luca.boccassi
2024-10-23 21:16 ` patch 'net/hns3: remove some basic address dump' " luca.boccassi
2024-10-23 21:16 ` patch 'net/hns3: fix dump counter of registers' " luca.boccassi
2024-10-23 21:16 ` patch 'ethdev: fix overflow in descriptor count' " luca.boccassi
2024-10-23 21:16 ` luca.boccassi [this message]
2024-10-23 21:16 ` patch 'net/dpaa: fix typecasting channel ID' " luca.boccassi
2024-10-23 21:16 ` patch 'bus/dpaa: fix VSP for 1G fm1-mac9 and 10' " luca.boccassi
2024-10-23 21:16 ` patch 'bus/dpaa: fix the fman details status' " luca.boccassi
2024-10-23 21:16 ` patch 'net/dpaa: fix reallocate mbuf handling' " luca.boccassi
2024-10-23 21:16 ` patch 'net/memif: fix buffer overflow in zero copy Rx' " luca.boccassi
2024-10-23 21:16 ` patch 'net/tap: restrict maximum number of MP FDs' " luca.boccassi
2024-10-23 21:16 ` patch 'ethdev: verify queue ID in Tx done cleanup' " luca.boccassi
2024-10-23 21:16 ` patch 'net/hns3: verify reset type from firmware' " luca.boccassi
2024-10-23 21:16 ` patch 'net/nfp: fix link change return value' " luca.boccassi
2024-10-23 21:16 ` patch 'net/pcap: fix blocking Rx' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ice/base: add bounds check' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ice/base: fix VLAN replay after reset' " luca.boccassi
2024-10-23 21:16 ` patch 'net/iavf: preserve MAC address with i40e PF Linux driver' " luca.boccassi
2024-10-23 21:16 ` patch 'net/mlx5: workaround list management of Rx queue control' " luca.boccassi
2024-10-23 21:16 ` patch 'net/mlx5: fix number of supported flex parsers' " luca.boccassi
2024-10-23 21:16 ` patch 'app/testpmd: remove flex item init command leftover' " luca.boccassi
2024-10-23 21:16 ` patch 'net/mlx5: fix next protocol validation after flex item' " luca.boccassi
2024-10-23 21:16 ` patch 'build: remove version check on compiler links function' " luca.boccassi
2024-10-23 21:16 ` patch 'hash: fix thash LFSR initialization' " luca.boccassi
2024-10-23 21:16 ` patch 'dmadev: fix potential null pointer access' " luca.boccassi
2024-10-23 21:16 ` patch 'net/gve/base: fix build with Fedora Rawhide' " luca.boccassi
2024-10-23 21:16 ` patch 'power: fix mapped lcore ID' " luca.boccassi
2024-10-23 21:16 ` patch 'net/ionic: fix build on Fedora Rawhide' " luca.boccassi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241023211704.1216956-56-luca.boccassi@gmail.com \
--to=luca.boccassi@gmail.com \
--cc=g.singh@nxp.com \
--cc=hemant.agrawal@nxp.com \
--cc=stable@dpdk.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).