linux-rdma.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Max Gurtovoy <maxg@mellanox.com>
To: Krishnamraju Eraparaju <krishna2@chelsio.com>
Cc: Sagi Grimberg <sagi@grimberg.me>,
	"linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
	Nirranjan Kirubaharan <nirranjan@chelsio.com>,
	Potnuri Bharat Teja <bharat@chelsio.com>
Subject: Re: iSERT SQ overflow with single target and multi luns
Date: Tue, 7 Jul 2020 21:53:11 +0300	[thread overview]
Message-ID: <3db0b668-fd0d-c31a-0db7-9adb46c7a0a1@mellanox.com> (raw)
In-Reply-To: <20200707171627.GA2352@chelsio.com>

great news.

Sagi,

can you review the below before I send the formal patch ?

BTW, we solved a similar issue with NVMf/RDMA target recently..

On 7/7/2020 8:16 PM, Krishnamraju Eraparaju wrote:
> Hi Max,
>
> Thanks for the quick response!
>
> I just tested your patch, it's working without any issue.
>
>
> Just for reference, here is what I observed with Chelsio iWARP adapter,
> after applying your patch:
>
> attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS:138 +1 + (rdma_ctxs:4096 *
> factor:3) = 12427
>
>
> rdma_ctxs:4096 comes from  ==>
> rdma_rw_mr_factor(ISCSI_ISER_MAX_SG_TABLESIZE:4096)=32 *
> ISCSI_DEF_XMIT_CMDS_MAX:128     where mr_pages is 128
>
>
> Finally, 12427 size got capped to 8190 due to "dev->attrs.max_qp_wr" in
> rdma_rw_init_qp().
>
> So final value of attr.cap.max_send_wr is 8190 for Chelsio adapters.
>
>
>
> Thanks,
> Krishna.
> On Tuesday, July 07/07/20, 2020 at 16:23:30 +0300, Max Gurtovoy wrote:
>> Hi Krishna,
>>
>> thanks for debugging this.
>>
>> please try the following untested patch:
>>
>>
>> diff --git a/drivers/infiniband/ulp/isert/ib_isert.c
>> b/drivers/infiniband/ulp/isert/ib_isert.c
>> index b7df38e..49f5f05 100644
>> --- a/drivers/infiniband/ulp/isert/ib_isert.c
>> +++ b/drivers/infiniband/ulp/isert/ib_isert.c
>> @@ -119,7 +119,7 @@
>>   {
>>          struct isert_device *device = isert_conn->device;
>>          struct ib_qp_init_attr attr;
>> -       int ret;
>> +       int ret, factor;
>>
>>          memset(&attr, 0, sizeof(struct ib_qp_init_attr));
>>          attr.event_handler = isert_qp_event_callback;
>> @@ -128,7 +128,9 @@
>>          attr.recv_cq = comp->cq;
>>          attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS + 1;
>>          attr.cap.max_recv_wr = ISERT_QP_MAX_RECV_DTOS + 1;
>> -       attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX;
>> +       factor = rdma_rw_mr_factor(device->ib_device, cma_id->port_num,
>> +                                  ISCSI_ISER_MAX_SG_TABLESIZE);
>> +       attr.cap.max_rdma_ctxs = ISCSI_DEF_XMIT_CMDS_MAX * factor;
>>          attr.cap.max_send_sge = device->ib_device->attrs.max_send_sge;
>>          attr.cap.max_recv_sge = 1;
>>          attr.sq_sig_type = IB_SIGNAL_REQ_WR;
>> diff --git a/drivers/infiniband/ulp/isert/ib_isert.h
>> b/drivers/infiniband/ulp/isert/ib_isert.h
>> index 3b296ba..c9ccf1d 100644
>> --- a/drivers/infiniband/ulp/isert/ib_isert.h
>> +++ b/drivers/infiniband/ulp/isert/ib_isert.h
>> @@ -63,7 +63,8 @@
>>                  (ISER_RX_PAYLOAD_SIZE + sizeof(u64) + sizeof(struct
>> ib_sge) + \
>>                   sizeof(struct ib_cqe) + sizeof(bool)))
>>
>> -#define ISCSI_ISER_SG_TABLESIZE                256
>> +/* Maximum support is 16MB I/O size */
>> +#define ISCSI_ISER_MAX_SG_TABLESIZE    4096
>>
>>   enum isert_desc_type {
>>          ISCSI_TX_CONTROL,
>>
>>
>> On 7/7/2020 3:36 PM, Krishnamraju Eraparaju wrote:
>>> Looks like the commit 07173c3e(block: enable multipage bvecs) has
>>> uncovered iSER SQ sizing issue.
>>>
>>> Here is how I hit the issue:
>>> Created two luns under single target, then run the below script on each
>>> lun(parallelly).
>>>
>>>    while [ 1 ]
>>>    do
>>>    iozone -i 0 -i 1 -I -+d -s 100000 -r 16384 -w
>>>    done
>>>
>>>
>>> Then failures like below are logged in dmesg output, due to iw_cxgb4 SQ
>>> getting full at iSER target.
>>>     "isert: isert_rdma_rw_ctx_post: Cmd: 00000000cb75342a failed to post
>>> RDMA res"
>>>
>>>
>>> This issue won't occur if luns are created on seperate targets.
>>> Also, the issue won't occur if I revert the multipage bvecs(07173c3e)
>>> changes at initator.
>>>
>>>
>>> Currently SQ is being sized this way:
>>> attr.cap.max_send_wr = ISERT_QP_MAX_REQ_DTOS:138 +1 +
>>> (ISCSI_DEF_XMIT_CMDS_MAX:128 * factor:3) = 523.
>>> I tried increaseing the SQ size and observed that the issue is not
>>> occuring when attr.cap.max_send_wr is 562.
>>>
>>>
>>> Looks like the avg length of RDMA READ/WRITE operations has increased
>>> after "multipage bvecs" changes.
>>> Queueing many large sized RDMA READ/WRITE WRs may cause backpressure and
>>> increases the chances of SQ getting full at provider driver.
>>> Notice the length(0x7f000 & 0x2000) of each RDMA READ operation below,
>>> for Before and After case.
>>>
>>> Before "multipage bvecs" RDMA READ:
>>> [  +0.001903] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x78]
>>> [  +0.000007] iser: iser_fast_reg_mr: lkey=0x8a41 rkey=0x8a41
>>> addr=0x446166000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:120 READ tags
>>> RKEY:0X8A41 VA:0X446166000
>>> [  +0.000007] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x6f]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x13b51 rkey=0x13b51
>>> addr=0x443b25000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:111 READ tags
>>> RKEY:0X13B51 VA:0X443B25000
>>> [  +0.000022] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0xe]
>>> [  +0.000001] iser: iser_fast_reg_mr: lkey=0xa371 rkey=0xa371
>>> addr=0x4461a4000 length=0x2000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:14 READ tags
>>> RKEY:0XA371 VA:0X4461A4000
>>> [  +0.000004] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x79]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x12f4f rkey=0x12f4f
>>> addr=0x4461a9000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:121 READ tags
>>> RKEY:0X12F4F VA:0X4461A9000
>>> [  +0.000005] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7d]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0xe040 rkey=0xe040
>>> addr=0x447e67000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:125 READ tags
>>> RKEY:0XE040 VA:0X447E67000
>>> [  +0.000021] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7b]
>>> [  +0.000001] iser: iser_fast_reg_mr: lkey=0xb149 rkey=0xb149
>>> addr=0x3d0366000 length=0x2000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:123 READ tags
>>> RKEY:0XB149 VA:0X3D0366000
>>> [  +0.000004] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0xb]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x1014c rkey=0x1014c
>>> addr=0x3d0368000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:11 READ tags
>>> RKEY:0X1014C VA:0X3D0368000
>>> [  +0.000007] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x62]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x7c3b rkey=0x7c3b
>>> addr=0x3d03e7000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:98 READ tags
>>> RKEY:0X7C3B VA:0X3D03E7000
>>> [  +0.000021] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x11]
>>> [  +0.000001] iser: iser_fast_reg_mr: lkey=0x11752 rkey=0x11752
>>> addr=0x3d6de6000 length=0x2000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:17 READ tags
>>> RKEY:0X11752 VA:0X3D6DE6000
>>> [  +0.000004] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x77]
>>>
>>>
>>> After "multipage bvecs" RDMA READ:
>>> [  +0.002455] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7d]
>>> [  +0.000006] iser: iser_fast_reg_mr: lkey=0x7991 rkey=0x7991
>>> addr=0x3d2819000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:125 READ tags
>>> RKEY:0X7991 VA:0X3D2819000
>>> [  +0.000005] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7e]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x8c9b rkey=0x8c9b
>>> addr=0x3d2898000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:126 READ tags
>>> RKEY:0X8C9B VA:0X3D2898000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7f]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x856d rkey=0x856d
>>> addr=0x3d2917000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:127 READ tags
>>> RKEY:0X856D VA:0X3D2917000
>>> [  +0.000004] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x1]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0x9b55 rkey=0x9b55
>>> addr=0x3d2999000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:1 READ tags
>>> RKEY:0X9B55 VA:0X3D2999000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x2]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0x86cf rkey=0x86cf
>>> addr=0x3d2018000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:2 READ tags
>>> RKEY:0X86CF VA:0X3D2018000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x3]
>>> [  +0.000003] iser: iser_fast_reg_mr: lkey=0x8062 rkey=0x8062
>>> addr=0x3d2097000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:3 READ tags
>>> RKEY:0X8062 VA:0X3D2097000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x4]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0xc34b rkey=0xc34b
>>> addr=0x3d2116000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:4 READ tags
>>> RKEY:0XC34B VA:0X3D2116000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x5]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0x8b6d rkey=0x8b6d
>>> addr=0x3d2195000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:5 READ tags
>>> RKEY:0X8B6D VA:0X3D2195000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x6]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0xce56 rkey=0xce56
>>> addr=0x3d0e14000 length=0x7f000
>>> [  +0.000000] iser: iser_prepare_read_cmd: Cmd itt:6 READ tags
>>> RKEY:0XCE56 VA:0X3D0E14000
>>> [  +0.000003] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x7]
>>> [  +0.000002] iser: iser_fast_reg_mr: lkey=0xba45 rkey=0xba45
>>> addr=0x3d0e93000 length=0x7f000
>>> [  +0.000001] iser: iser_prepare_read_cmd: Cmd itt:7 READ tags
>>> RKEY:0XBA45 VA:0X3D0E93000
>>> [  +0.000002] iser: iscsi_iser_task_xmit: ctask xmit [cid 0 itt 0x8]
>>>
>>> Hence, I feel iSER target SQ is undersized and needs to be sized
>>> properly to hold max possible entries. I might be wrong.
>>>
>>> Please take a look.
>>>
>>> Thanks,
>>> Krishna.

      reply	other threads:[~2020-07-07 18:53 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-07-07 12:36 iSERT SQ overflow with single target and multi luns Krishnamraju Eraparaju
2020-07-07 13:23 ` Max Gurtovoy
2020-07-07 17:16   ` Krishnamraju Eraparaju
2020-07-07 18:53     ` Max Gurtovoy [this message]

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=3db0b668-fd0d-c31a-0db7-9adb46c7a0a1@mellanox.com \
    --to=maxg@mellanox.com \
    --cc=bharat@chelsio.com \
    --cc=krishna2@chelsio.com \
    --cc=linux-rdma@vger.kernel.org \
    --cc=nirranjan@chelsio.com \
    --cc=sagi@grimberg.me \
    /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).