From: Yasin CANER <yasinncaner@gmail.com>
To: Ferruh Yigit <ferruh.yigit@amd.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>, users@dpdk.org
Subject: Re: DPDK 22.11 - How to fix memory leak for KNI - How to debug
Date: Fri, 19 May 2023 20:47:28 +0300 [thread overview]
Message-ID: <CAP5epcPKD-RWGh8dz2S7SESvYojyMLw+1xyen44gm-Y377Rgzw@mail.gmail.com> (raw)
In-Reply-To: <4f53f3be-0bae-e204-5737-7735b4a2ba5b@amd.com>
[-- Attachment #1: Type: text/plain, Size: 15224 bytes --]
Hello,
I tested all day both before and after patching.
I could not understand that it is a memory leak or not. Maybe it needs
optimization. You lead, I follow.
1-) You are right, alloc_q is never bigger than 1024. But it always
allocates 32 units then more than 1024 are being freed. Maybe it takes
time, I don't know.
2-) I tested tx_rs_thresh via ping. After 210 sec , allocated memories are
back to mempool (most of them). (driver virtio and eth-devices are binded
via igb_uio) . It really takes time. So it is better to increase the size
of the mempool. (https://doc.dpdk.org/guides/prog_guide/poll_mode_drv.html)
3-) try to list mempool state in randomly
Test -1 -) (old code) ICMP testing. The whole mempool size is about 10350.
So after FIFO reaches max-size -1024, %10 of the size of the mempool is in
use. But little by little memory is waiting in use and doesn't go back to
the pool. I could not find the reason.
MBUF_POOL 448 9,951
4.31% [|....................]
MBUF_POOL 1,947 8,452
18.72% [||||.................]
MBUF_POOL 1,803 8,596
17.34% [||||.................]
MBUF_POOL 1,941 8,458
18.67% [||||.................]
MBUF_POOL 1,900 8,499
18.27% [||||.................]
MBUF_POOL 1,999 8,400
19.22% [||||.................]
MBUF_POOL 1,724 8,675
16.58% [||||.................]
MBUF_POOL 1,811 8,588
17.42% [||||.................]
MBUF_POOL 1,978 8,421
19.02% [||||.................]
MBUF_POOL 2,008 8,391
19.31% [||||.................]
MBUF_POOL 1,854 8,545
17.83% [||||.................]
MBUF_POOL 1,922 8,477
18.48% [||||.................]
MBUF_POOL 1,892 8,507
18.19% [||||.................]
MBUF_POOL 1,957 8,442
18.82% [||||.................]
Test-2 -) (old code) run iperf3 udp testing that from Kernel to eth device.
Waited to see what happens in 4 min. memory doesn't go back to the mempool.
little by little, memory usage increases.
MBUF_POOL 512 9,887
4.92% [|....................]
MBUF_POOL 1,411 8,988
13.57% [|||..................]
MBUF_POOL 1,390 9,009
13.37% [|||..................]
MBUF_POOL 1,558 8,841
14.98% [|||..................]
MBUF_POOL 1,453 8,946
13.97% [|||..................]
MBUF_POOL 1,525 8,874
14.66% [|||..................]
MBUF_POOL 1,592 8,807
15.31% [||||.................]
MBUF_POOL 1,639 8,760
15.76% [||||.................]
MBUF_POOL 1,624 8,775
15.62% [||||.................]
MBUF_POOL 1,618 8,781
15.56% [||||.................]
MBUF_POOL 1,708 8,691
16.42% [||||.................]
iperf is STOPPED to tx_fresh for 4 min
MBUF_POOL 1,709 8,690
16.43% [||||.................]
iperf is STOPPED to tx_fresh for 4 min
MBUF_POOL 1,709 8,690
16.43% [||||.................]
MBUF_POOL 1,683 8,716
16.18% [||||.................]
MBUF_POOL 1,563 8,836
15.03% [||||.................]
MBUF_POOL 1,726 8,673
16.60% [||||.................]
MBUF_POOL 1,589 8,810
15.28% [||||.................]
MBUF_POOL 1,556 8,843
14.96% [|||..................]
MBUF_POOL 1,610 8,789
15.48% [||||.................]
MBUF_POOL 1,616 8,783
15.54% [||||.................]
MBUF_POOL 1,709 8,690
16.43% [||||.................]
MBUF_POOL 1,740 8,659
16.73% [||||.................]
MBUF_POOL 1,546 8,853
14.87% [|||..................]
MBUF_POOL 1,710 8,689
16.44% [||||.................]
MBUF_POOL 1,787 8,612
17.18% [||||.................]
MBUF_POOL 1,579 8,820
15.18% [||||.................]
MBUF_POOL 1,780 8,619
17.12% [||||.................]
MBUF_POOL 1,679 8,720
16.15% [||||.................]
MBUF_POOL 1,604 8,795
15.42% [||||.................]
MBUF_POOL 1,761 8,638
16.93% [||||.................]
MBUF_POOL 1,773 8,626
17.05% [||||.................]
Test-3 -) (after patching) run iperf3 udp testing that from Kernel to eth
device. looks stable.
After patching ,
MBUF_POOL 76 10,323
0.73% [|....................]
MBUF_POOL 193 10,206
1.86% [|....................]
MBUF_POOL 96 10,303
0.92% [|....................]
MBUF_POOL 269 10,130
2.59% [|....................]
MBUF_POOL 102 10,297
0.98% [|....................]
MBUF_POOL 235 10,164
2.26% [|....................]
MBUF_POOL 87 10,312
0.84% [|....................]
MBUF_POOL 293 10,106
2.82% [|....................]
MBUF_POOL 99 10,300
0.95% [|....................]
MBUF_POOL 296 10,103
2.85% [|....................]
MBUF_POOL 90 10,309
0.87% [|....................]
MBUF_POOL 299 10,100
2.88% [|....................]
MBUF_POOL 86 10,313
0.83% [|....................]
MBUF_POOL 262 10,137
2.52% [|....................]
MBUF_POOL 81 10,318
0.78% [|....................]
MBUF_POOL 81 10,318
0.78% [|....................]
MBUF_POOL 87 10,312
0.84% [|....................]
MBUF_POOL 252 10,147
2.42% [|....................]
MBUF_POOL 97 10,302
0.93% [|....................]
iperf is STOPPED to tx_fresh for 4 min
MBUF_POOL 296 10,103
2.85% [|....................]
MBUF_POOL 95 10,304
0.91% [|....................]
MBUF_POOL 269 10,130
2.59% [|....................]
MBUF_POOL 302 10,097
2.90% [|....................]
MBUF_POOL 88 10,311
0.85% [|....................]
MBUF_POOL 305 10,094
2.93% [|....................]
MBUF_POOL 88 10,311
0.85% [|....................]
MBUF_POOL 290 10,109
2.79% [|....................]
MBUF_POOL 84 10,315
0.81% [|....................]
MBUF_POOL 85 10,314
0.82% [|....................]
MBUF_POOL 291 10,108
2.80% [|....................]
MBUF_POOL 303 10,096
2.91% [|....................]
MBUF_POOL 92 10,307
0.88% [|....................]
Best regards.
Ferruh Yigit <ferruh.yigit@amd.com>, 18 May 2023 Per, 17:56 tarihinde şunu
yazdı:
> On 5/18/2023 9:14 AM, Yasin CANER wrote:
> > Hello Ferruh,
> >
> > Thanks for your kind response. Also thanks to Stephen.
> >
> > Even if 1 packet is consumed from the kernel , each time rx_kni
> > allocates another 32 units. After a while all mempool is used in alloc_q
> > from kni. there is not any room for it.
> >
>
> What you described continues until 'alloc_q' is full, by default fifo
> length is 1024 (KNI_FIFO_COUNT_MAX), do you allocate less mbuf in your
> mempool?
>
> You can consider either increasing mempool size, or decreasing 'alloc_q'
> fifo length, but reducing fifo size may cause performance issues so you
> need to evaluate that option.
>
> > Do you think my mistake is using one and common mempool usage both kni
> > and eth?
> >
>
> Using same mempool for both is fine.
>
> > If it needs a separate mempool , i'd like to note in docs.
> >
> > Best regards.
> >
> > Ferruh Yigit <ferruh.yigit@amd.com <mailto:ferruh.yigit@amd.com>>, 17
> > May 2023 Çar, 20:53 tarihinde şunu yazdı:
> >
> > On 5/9/2023 12:13 PM, Yasin CANER wrote:
> > > Hello,
> > >
> > > I draw a flow via asciiflow to explain myself better. Problem is
> after
> > > transmitting packets(mbufs) , it never puts in the kni->free_q to
> back
> > > to the original pool. Each cycle, it allocates another 32 units
> that
> > > cause leaks. Or I am missing something.
> > >
> > > I already tried the rte_eth_tx_done_cleanup() function but it
> > didn't fix
> > > anything.
> > >
> > > I am working on a patch to fix this issue but I am not sure if
> there
> > > is another way.
> > >
> > > Best regards.
> > >
> > > https://pastebin.ubuntu.com/p/s4h5psqtgZ/
> > <https://pastebin.ubuntu.com/p/s4h5psqtgZ/>
> > > <https://pastebin.ubuntu.com/p/s4h5psqtgZ/
> > <https://pastebin.ubuntu.com/p/s4h5psqtgZ/>>
> > >
> > >
> > > unsigned
> > > rte_kni_rx_burst(struct rte_kni *kni, struct rte_mbuf **mbufs,
> > unsigned
> > > int num)
> > > {
> > > unsigned int ret = kni_fifo_get(kni->tx_q, (void **)mbufs, num);
> > >
> > > /* If buffers removed, allocate mbufs and then put them into
> > alloc_q */
> > > /* Question, how to test buffers is removed or not?*/
> > > if (ret)
> > > kni_allocate_mbufs(kni);
> > >
> > > return ret;
> > > }
> > >
> >
> > Selam Yasin,
> >
> >
> > You can expect 'kni->alloc_q' fifo to be full, this is not a memory
> > leak.
> >
> > As you pointed out, number of mbufs consumed by kernel from 'alloc_q'
> > and number of mbufs added to 'alloc_q' is not equal and this is
> > expected.
> >
> > Target here is to prevent buffer underflow from kernel perspective,
> so
> > it will always have available mbufs for new packets.
> > That is why new mbufs are added to 'alloc_q' at worst same or
> sometimes
> > higher rate than it is consumed.
> >
> > You should calculate your mbuf requirement with the assumption that
> > 'kni->alloc_q' will be full of mbufs.
> >
> >
> > 'kni->alloc_q' is freed when kni is removed.
> > Since 'alloc_q' holds physical address of the mbufs, it is a little
> > challenging to free them in the userspace, that is why first kernel
> > tries to move mbufs to 'kni->free_q' fifo, please check
> > 'kni_net_release_fifo_phy()' for it.
> >
> > If all moved to 'free_q' fifo, nothing left to in 'alloc_q', but if
> not,
> > userspace frees 'alloc_q' in 'rte_kni_release()', with following
> call:
> > `kni_free_fifo_phy(kni->pktmbuf_pool, kni->alloc_q);`
> >
> >
> > I can see you have submitted fixes for this issue, although as I
> > explained above I don't think a defect exist, I will review them
> > today/tomorrow.
> >
> > Regards,
> > Ferruh
> >
> >
> > > Stephen Hemminger <stephen@networkplumber.org
> > <mailto:stephen@networkplumber.org>
> > > <mailto:stephen@networkplumber.org
> > <mailto:stephen@networkplumber.org>>>, 8 May 2023 Pzt, 19:18
> tarihinde
> > > şunu yazdı:
> > >
> > > On Mon, 8 May 2023 09:01:41 +0300
> > > Yasin CANER <yasinncaner@gmail.com
> > <mailto:yasinncaner@gmail.com> <mailto:yasinncaner@gmail.com
> > <mailto:yasinncaner@gmail.com>>>
> > > wrote:
> > >
> > > > Hello Stephen,
> > > >
> > > > Thank you for response, it helps me a lot. I understand
> problem
> > > better.
> > > >
> > > > After reading mbuf library (
> > > > https://doc.dpdk.org/guides/prog_guide/mempool_lib.html
> > <https://doc.dpdk.org/guides/prog_guide/mempool_lib.html>
> > > <https://doc.dpdk.org/guides/prog_guide/mempool_lib.html
> > <https://doc.dpdk.org/guides/prog_guide/mempool_lib.html>>) i
> > > realized that
> > > > 31 units allocation memory slot doesn't return to pool!
> > >
> > > If receive burst returns 1 mbuf, the other 31 pointers in the
> > array
> > > are not valid. They do not point to mbufs.
> > >
> > > > 1 unit mbuf can be freed via rte_pktmbuf_free so it can back
> > to pool.
> > > >
> > > > Main problem is that allocation doesn't return to original
> pool,
> > > act as
> > > > used. So, after following rte_pktmbuf_free
> > > >
> > >
> > <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a1215458932900b7cd5192326fa4a6902
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a1215458932900b7cd5192326fa4a6902>
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a1215458932900b7cd5192326fa4a6902
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a1215458932900b7cd5192326fa4a6902
> >>>
> > > > function,
> > > > i realized that there is 2 function to helps to mbufs back
> > to pool.
> > > >
> > > > These are rte_mbuf_raw_free
> > > >
> > >
> > <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a9f188d53834978aca01ea101576d7432
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a9f188d53834978aca01ea101576d7432>
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a9f188d53834978aca01ea101576d7432
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a9f188d53834978aca01ea101576d7432
> >>>
> > > > and rte_pktmbuf_free_seg
> > > >
> > >
> > <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a006ee80357a78fbb9ada2b0432f82f37
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a006ee80357a78fbb9ada2b0432f82f37>
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a006ee80357a78fbb9ada2b0432f82f37
> <
> http://doc.dpdk.org/api/rte__mbuf_8h.html#a006ee80357a78fbb9ada2b0432f82f37
> >>>.
> > > > I will focus on them.
> > > >
> > > > If there is another suggestion, I will be very pleased.
> > > >
> > > > Best regards.
> > > >
> > > > Yasin CANER
> > > > Ulak
> > >
> >
>
>
[-- Attachment #2: Type: text/html, Size: 24249 bytes --]
next prev parent reply other threads:[~2023-05-19 17:49 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-08 6:01 Yasin CANER
2023-05-08 16:18 ` Stephen Hemminger
2023-05-09 11:13 ` Yasin CANER
2023-05-11 14:14 ` Yasin CANER
2023-05-17 17:53 ` Ferruh Yigit
2023-05-18 8:14 ` Yasin CANER
2023-05-18 14:56 ` Ferruh Yigit
2023-05-19 17:47 ` Yasin CANER [this message]
2023-05-19 18:43 ` Ferruh Yigit
2023-05-29 6:33 ` Yasin CANER
-- strict thread matches above, loose matches on Subject: below --
2023-05-04 7:32 Yasin CANER
2023-05-04 13:00 ` Yasin CANER
2023-05-04 16:14 ` Stephen Hemminger
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=CAP5epcPKD-RWGh8dz2S7SESvYojyMLw+1xyen44gm-Y377Rgzw@mail.gmail.com \
--to=yasinncaner@gmail.com \
--cc=ferruh.yigit@amd.com \
--cc=stephen@networkplumber.org \
--cc=users@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).