From: zhoumin <zhoumin@loongson.cn>
To: David Marchand <david.marchand@redhat.com>
Cc: dev@dpdk.org, olivier.matz@6wind.com, ferruh.yigit@amd.com,
kaisenx.you@intel.com,
Anatoly Burakov <anatoly.burakov@intel.com>,
Bruce Richardson <bruce.richardson@intel.com>
Subject: Re: [PATCH] malloc: enhance NUMA affinity heuristic
Date: Tue, 27 Dec 2022 17:00:19 +0800 [thread overview]
Message-ID: <0572b450-609d-0053-6fe3-beab118e7020@loongson.cn> (raw)
In-Reply-To: <CAJFAV8xtwMc+931v2QeW_JGhLBn6KL6xkWKKKYEMz=dnP9x60A@mail.gmail.com>
Hi David,
First of all, I sincerely apologize for the late reply.
I had checked this issue carefully and had some useful findings.
On Wed, Dec 21, 2022 at 22:57 PM, David Marchand wrote:
> Hello Min,
>
> On Wed, Dec 21, 2022 at 11:49 AM David Marchand
> <david.marchand@redhat.com> wrote:
>> Trying to allocate memory on the first detected numa node has less
>> chance to find some memory actually available rather than on the main
>> lcore numa node (especially when the DPDK application is started only
>> on one numa node).
>>
>> Signed-off-by: David Marchand <david.marchand@redhat.com>
> I see a failure in the loongarch CI.
>
> Running binary with
> argv[]:'/home/zhoumin/dpdk/build/app/test/dpdk-test'
> '--file-prefix=eal_flags_c_opt_autotest' '--proc-type=secondary'
> '--lcores' '0-1,2@(5-7),(3-5)@(0,2),(0,6),7'
> Error - process did not run ok with valid corelist value
> Test Failed
>
> The logs don't give the full picture (though it is not LoongArch CI fault).
>
> I tried to read back on past mail exchanges about the loongarch
> server, but I did not find the info.
> I suspect cores 5 to 7 belong to different numa nodes, can you confirm?
The cores 5 to 7 belong to the same numa node (NUMA node1) on the
Loongson-3C5000LL CPU on which LoongArch DPDK CI runs.
>
> I'll post a new revision to account for this case.
>
The LoongArch DPDK CI uses the core 0-7 to run all the DPDK unit tests
by adding the arg '-l 0-7' in the meson test args. In the above test
case, the arg '--lcores' '0-1,2@(5-7),(3-5)@(0,2),(0,6),7' will make the
lcore 0 and 6 to run on the core 0 or 6. The logs of EAL will make it
more clear when I set the log level of EAL to debug as follows:
EAL: Main lcore 0 is ready (tid=fff3ee18f0;cpuset=[0,6])
EAL: lcore 1 is ready (tid=fff2de4cf0;cpuset=[1])
EAL: lcore 2 is ready (tid=fff25e0cf0;cpuset=[5,6,7])
EAL: lcore 5 is ready (tid=fff0dd4cf0;cpuset=[0,2])
EAL: lcore 4 is ready (tid=fff15d8cf0;cpuset=[0,2])
EAL: lcore 3 is ready (tid=fff1ddccf0;cpuset=[0,2])
EAL: lcore 7 is ready (tid=ffdb7f8cf0;cpuset=[7])
EAL: lcore 6 is ready (tid=ffdbffccf0;cpuset=[0,6])
However, The cores 0 and 6 belong to different numa nodes on the
Loongson-3C5000LL CPU. The core 0 belongs to NUMA node 0 and the core 6
belongs to NUMA node 1 as follows:
$ lscpu
Architecture: loongarch64
Byte Order: Little Endian
CPU(s): 32
On-line CPU(s) list: 0-31
Thread(s) per core: 1
Core(s) per socket: 4
Socket(s): 8
NUMA node(s): 8
...
NUMA node0 CPU(s): 0-3
NUMA node1 CPU(s): 4-7
NUMA node2 CPU(s): 8-11
NUMA node3 CPU(s): 12-15
NUMA node4 CPU(s): 16-19
NUMA node5 CPU(s): 20-23
NUMA node6 CPU(s): 24-27
NUMA node7 CPU(s): 28-31
...
So the socket_id for the lcore 0 and 6 will be set to -1 which can be
seen from the thread_update_affinity(). Meanwhile, I print out the
socket_id for the lcore 0 to RTE_MAX_LCORE - 1 as follows:
lcore_config[*].socket_id: -1 0 1 0 0 0 -1 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5
5 5 6 6 6 6 7 7 7 7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0
In this test case, the modified malloc_get_numa_socket() will return -1
which caused a memory allocation failure.
Whether it is acceptable in DPDK that the socket_id for a lcore is -1?
If it's ok, maybe we can check the socket_id of main lcore before using
it, such as:
diff --git a/lib/eal/common/malloc_heap.c b/lib/eal/common/malloc_heap.c
index d7c410b786..3ee19aee15 100644
--- a/lib/eal/common/malloc_heap.c
+++ b/lib/eal/common/malloc_heap.c
@@ -717,6 +717,10 @@ malloc_get_numa_socket(void)
return socket_id;
}
+ socket_id = rte_lcore_to_socket_id(rte_get_main_lcore());
+ if (socket_id != (unsigned int)SOCKET_ID_ANY)
+ return socket_id;
+
return rte_socket_id_by_idx(0);
}
next prev parent reply other threads:[~2022-12-27 9:00 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-21 10:48 David Marchand
2022-12-21 11:16 ` Bruce Richardson
2022-12-21 13:50 ` Ferruh Yigit
2022-12-21 14:57 ` David Marchand
2022-12-27 9:00 ` zhoumin [this message]
2023-01-03 10:56 ` David Marchand
2023-01-03 10:58 ` [PATCH v2] " David Marchand
2023-01-03 13:32 ` [PATCH v3] " David Marchand
2023-01-31 3:23 ` You, KaisenX
2023-01-31 15:05 ` [PATCH v4] net/iavf:enhance " Kaisen You
2023-01-31 16:05 ` Thomas Monjalon
2023-02-01 5:32 ` You, KaisenX
2023-02-01 12:20 ` [PATCH v5] enhance " Kaisen You
2023-02-01 10:52 ` Jiale, SongX
2023-02-15 14:22 ` Burakov, Anatoly
2023-02-15 14:47 ` Burakov, Anatoly
2023-02-16 2:50 ` You, KaisenX
2023-03-03 14:07 ` Thomas Monjalon
2023-03-09 1:58 ` You, KaisenX
2023-04-13 0:56 ` You, KaisenX
2023-04-19 12:16 ` Thomas Monjalon
2023-04-21 2:34 ` You, KaisenX
2023-04-21 8:12 ` Thomas Monjalon
2023-04-23 6:52 ` You, KaisenX
2023-04-23 8:57 ` You, KaisenX
2023-04-23 13:19 ` Thomas Monjalon
2023-04-25 5:16 ` [PATCH v6] " Kaisen You
2023-04-27 6:57 ` Thomas Monjalon
2023-05-16 5:19 ` You, KaisenX
2023-05-23 2:50 ` [PATCH v7] " Kaisen You
2023-05-23 10:44 ` Burakov, Anatoly
2023-05-26 6:44 ` You, KaisenX
2023-05-23 12:45 ` Burakov, Anatoly
2023-05-26 6:50 ` [PATCH v8] " Kaisen You
2023-05-26 8:45 ` Kaisen You
2023-05-26 14:44 ` Burakov, Anatoly
2023-05-26 17:50 ` Stephen Hemminger
2023-05-29 10:37 ` Burakov, Anatoly
2023-06-01 14:42 ` David Marchand
2023-06-06 14:04 ` Thomas Monjalon
2023-06-12 9:36 ` Burakov, Anatoly
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=0572b450-609d-0053-6fe3-beab118e7020@loongson.cn \
--to=zhoumin@loongson.cn \
--cc=anatoly.burakov@intel.com \
--cc=bruce.richardson@intel.com \
--cc=david.marchand@redhat.com \
--cc=dev@dpdk.org \
--cc=ferruh.yigit@amd.com \
--cc=kaisenx.you@intel.com \
--cc=olivier.matz@6wind.com \
/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).