RE: [PATCH] net/iavf:fix slow memory allocation

["You, KaisenX" <kaisenx.you@intel.com>]

> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> Sent: 2023年1月10日 18:17
> To: You, KaisenX <kaisenx.you@intel.com>
> Cc: Ferruh Yigit <ferruh.yigit@amd.com>; dev@dpdk.org; Burakov, Anatoly
> <anatoly.burakov@intel.com>; stable@dpdk.org; Yang, Qiming
> <qiming.yang@intel.com>; Zhou, YidingX <yidingx.zhou@intel.com>; Wu,
> Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>; Zhang,
> Qi Z <qi.z.zhang@intel.com>; Luca Boccassi <bluca@debian.org>; Mcnamara,
> John <john.mcnamara@intel.com>; Kevin Traynor <ktraynor@redhat.com>
> Subject: Re: [PATCH] net/iavf:fix slow memory allocation
> 
> Hello,
> 
> On Tue, Dec 27, 2022 at 7:06 AM You, KaisenX <kaisenx.you@intel.com>
> wrote:
> > > > > > > > I tried to play a bit with a E810 nic on a dual numa and I
> > > > > > > > can't see anything wrong for now.
> > > > > > > > Can you provide a simple and small reproducer of your issue?
> > > > > > > >
> > > > > > > > Thanks.
> > > > > > > >
> > > > > > > This is my environment:
> > > > > > > Enter "lscpu" on the command line:
> > > > > > > NUMA:
> > > > > > >         NUMA node(s): 2
> > > > > > >         NUMA node0 CPU(S) : 0-27,56-83
> > > > > > >         NUMA node1 CPU(S) : 28-55,84-111
> > > > > > >
> > > > > > > List the steps to reproduce the issue:
> > > > > > >
> > > > > > > 1. create vf and blind to dpdk echo 1 >
> > > > > > > /sys/bus/pci/devices/0000\:ca\:00.0/sriov_ numvfs
> > > > > > > ./usertools/dpdk-devbind. py -b vfio-pci 0000:ca:01.0 2.
> > > > > > > launch testpmd ./x86_
> > > > > > > 64-native-linuxapp-clang/app/dpdk-testpmd -l
> > > > > > > 28-48 -n 4 -a 0000:ca:01.0 --file-prefix=dpdk_ 525342_
> > > > > > > 20221104042659 -- -i
> > > > > > > --rxq=256 --txq=256
> > > > > > > --total-num-mbufs=500000
> > > > > > >
> > > > > > > Parameter Description:
> > > > > > >  "-l 28-48":The range of parameter values after "-l" must be
> > > > > > > on "NUMA
> > > > > > node1 CPU(S)"
> > > > > > >  "0000:ca:01.0":inset on node1
> > > > > > - Back to your topic.
> > > > > > Can you try this simple hack:
> > > > > >
> > > > > > diff --git a/lib/eal/common/eal_common_thread.c
> > > > > > b/lib/eal/common/eal_common_thread.c
> > > > > > index c5d8b4327d..92160c7fa6 100644
> > > > > > --- a/lib/eal/common/eal_common_thread.c
> > > > > > +++ b/lib/eal/common/eal_common_thread.c
> > > > > > @@ -253,6 +253,7 @@ static void *ctrl_thread_init(void *arg)
> > > > > >         void *routine_arg = params->arg;
> > > > > >
> > > > > >         __rte_thread_init(rte_lcore_id(), cpuset);
> > > > > > +       RTE_PER_LCORE(_socket_id) = SOCKET_ID_ANY;
> > > > > >         params->ret = pthread_setaffinity_np(pthread_self(),
> > > > sizeof(*cpuset),
> > > > > >                 cpuset);
> > > > > >         if (params->ret != 0) {
> > > > > >
> > > > > Thanks for your advice.
> > > > >
> > > > > But this issue still exists after I tried.
> > > >
> > > > Ok, I think I understand what is wrong... but I am still guessing
> > > > as I am not sure what your "issue" is.
> > > > Can you have a try with:
> > > >
> https://patchwork.dpdk.org/project/dpdk/patch/20221221104858.29653
> > > > 0-
> > > 1-
> > > > david.marchand@redhat.com/
> > > >
> > > > Thanks.
> > > >
> > > I think this issue is similar to the description in the patch you gave me.
> > >
> > > when the DPDK application is started only on one numa node,
> > > Interrupt thread find memory on another numa node. This leads to a
> > > whole set of memory allocation/release operations every time when
> "rte_malloc" is called.
> > > This is the root cause of this issue.
> > >
> > > This issue can be solved after I tried.
> > > Thanks for your advice.
> >
> > After further testing in a different environment, we found the issue
> > still existed in your last patch.  After troubleshooting, it is found
> > that in the "malloc_get_numa_socket()" API, if the return value of
> "rte_socket_id()"
> > is "SOCKET_ID_ANY" (- 1), the API will return "rte_lcore_to_socket_id
> > (rte_get_main_lcore())"; Otherwise, "malloc_get_numa_socket()" API
> > will directly return "the return value of rte_socket_id()",in this
> > case, the issue cannot be solved.
> >
> > And the return value of "rte_socket_id()" is modified by the solution
> > you suggested in your last email (RTE_PER_LCORE (_socket_id)=SOCKET_
> ID_ ANY;).
> > Therefore, I think merging your two suggestions together could completely
> solve this issue.
> >
> > Can you please update your accordingly?
> 
> Please try the last revision and report back.
> https://patchwork.dpdk.org/project/dpdk/list/?series=26362
> 
> Thanks.
> 
Still failed.

Reason:
/lib/eal/common/malloc_heap.c
malloc_get_numa_socket(void)
{
        const struct internal_config *conf = eal_get_internal_configuration();
        unsigned int socket_id = rte_socket_id();   // The return value of "rte_socket_id()" is 1
        unsigned int idx;

        if (socket_id != (unsigned int)SOCKET_ID_ANY)
                return socket_id;    //so return here

So this patch cannot completely solve this issue.
Because the interrupt thread is on socket 1 when the thread is initialized (rte_crl_thread_create).
I think the following modifications should be added in this patch.

diff --git a/lib/eal/common/eal_common_thread.c
b/lib/eal/common/eal_common_thread.c
index c5d8b4327d..92160c7fa6 100644
--- a/lib/eal/common/eal_common_thread.c
+++ b/lib/eal/common/eal_common_thread.c
@@ -253,6 +253,7 @@ static void *ctrl_thread_init(void *arg)
        void *routine_arg = params->arg;

        __rte_thread_init(rte_lcore_id(), cpuset);
 +     RTE_PER_LCORE(_socket_id) = SOCKET_ID_ANY;
        params->ret = pthread_setaffinity_np(pthread_self(), sizeof(*cpuset),
                cpuset);
        if (params->ret != 0) {


> --
> David Marchand