From: Zhu Yanjun <yanjun.zhu@linux.dev>
To: Andrew Stellman <astellman@stellman-greene.com>,
"yanjun.zhu@linux.dev" <yanjun.zhu@linux.dev>
Cc: Zhu Yanjun <zyjzyj2000@gmail.com>, Jason Gunthorpe <jgg@ziepe.ca>,
Leon Romanovsky <leon@kernel.org>,
linux-rdma@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH rdma-next v2] RDMA/rxe: Require a responder resource for FLUSH
Date: Sun, 22 Nov 2026 12:13:46 -0800 [thread overview]
Message-ID: <9ce6bd33-f3a8-43dc-b238-d9f936ec5c54@linux.dev> (raw)
In-Reply-To: <CAChPuV9vPzV8k4SR-nn4O6N3Ld4g0uxA4qdahDUhJxZWnzNn7Q@mail.gmail.com>
在 2026/11/22 11:51, Andrew Stellman 写道:
> Hi Zhu,
>
> Thanks for looking at it.
>
>> (1) RXE_FLUSH_MASK takes the same error/NAK path here without later
>> responder states assuming a valid qp->resp.res or non-ZERO_SIZE_PTR
>> qp->resp.resources, and
> Yes. With the change, check_resource() returns
> RESPST_ERR_TOO_MANY_RDMA_ATM_REQ for a FLUSH, the same state READ and
> ATOMIC use. That state calls do_class_ac_error() and moves to
> RESPST_COMPLETE and then acknowledge(), which sends the NAK because
> aeth_syndrome is set. None of those read qp->resp.res or
> qp->resp.resources, and process_flush() and rxe_prepare_res() are not
> reached. A retransmitted FLUSH goes to duplicate_request(), where
> find_resource() loops over max_dest_rd_atomic entries, so with 0 it
> returns NULL and the packet is dropped without touching the array.
>
> In the QEMU test with the fix, the requester's FLUSH completed with
> status 9 (IB_WC_REM_INV_REQ_ERR), and the test program then exited
> normally (closing its QPs) with no oops or warning. Without the fix,
> closing the QP hit the __rxe_cleanup() timeout warning and hung.
>
>> (2) no other RXE_READ_OR_ATOMIC_MASK / RXE_ATOMIC_WRITE_MASK checks in
>> rxe_resp.c for max_dest_rd_atomic or responder resources missed
>> RXE_FLUSH_MASK.
> check_resource() is the only place in rxe_resp.c that checks those
> masks against max_dest_rd_atomic. The other places that test them
> (check_op_valid(), rxe_resp_check_length(), execute() and
> acknowledge()) already handle RXE_FLUSH_MASK, and the only users of
> qp->resp.resources are rxe_prepare_res() and find_resource().
Thanks a lot for your quick reply. The above explanations answer my
questions.
I am fine with this commit.
Reviewed-by: Zhu Yanjun <yanjun.zhu@linux.dev>
Zhu Yanjun
>
> Andrew
>
>
> On Sun, Nov 22, 2026 at 2:42 PM Zhu Yanjun <yanjun.zhu@linux.dev> wrote:
>> 在 2026/11/22 9:05, Andrew Stellman 写道:
>>> process_flush() keeps its state in a responder resource from
>>> rxe_prepare_res(), as READ and ATOMIC requests do. check_resource()
>>> refuses those requests when max_dest_rd_atomic is 0, but it does not
>>> check FLUSH. With max_dest_rd_atomic 0, qp->resp.resources is
>>> ZERO_SIZE_PTR, and rxe_prepare_res() writes through it in
>>> free_rd_atomic_resource().
>>>
>>> A FLUSH sent to an RC QP with max_dest_rd_atomic 0 oopses the
>>> responder:
>>>
>>> Oops: general protection fault, probably for non-canonical address 0xdffffc0000000002: 0000 [#1] SMP KASAN NOPTI
>>> KASAN: null-ptr-deref in range [0x0000000000000010-0x0000000000000017]
>>> CPU: 0 UID: 0 PID: 13 Comm: kworker/u8:0 Not tainted 7.4.0-rc3-gaebb72744764 #1 PREEMPT(lazy)
>>> Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
>>> Workqueue: rxe_wq do_work
>>> RIP: 0010:free_rd_atomic_resource+0x15/0x50
>>> Code: 00 00 00 00 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 0f 1f 40 d6 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <0f> b6 04 02 84 c0 74 04 3c 03 7e 0b c7 07 00 00 00 00 c3 cc cc cc
>>> RSP: 0018:ff110000013479f8 EFLAGS: 00000202
>>> RAX: dffffc0000000000 RBX: 0000000000000010 RCX: ff110000059f9760
>>> RDX: 0000000000000002 RSI: 0000000000000000 RDI: 0000000000000010
>>> RBP: 0000000000000001 R08: ff110000024a2180 R09: ffe21c0000268f2c
>>> R10: 0000000000000003 R11: ff110000059f9298 R12: ff1100000592dc00
>>> R13: 0000000000000040 R14: ff110000059f9000 R15: ff110000024a2168
>>> FS: 0000000000000000(0000) GS:ff110000e18da000(0000) knlGS:0000000000000000
>>> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: dffffc0000000002 CR3: 00000000052ca000 CR4: 0000000000751ef0
>>> PKRU: 55555554
>>> Call Trace:
>>> <TASK>
>>> rxe_receiver+0x5f6e/0x9a20
>>> ? flush_workqueue_prep_pwqs+0x1a4/0xe80
>>> ? kthread+0x2b0/0x3b0
>>> ? __pfx_rxe_receiver+0x10/0x10
>>> ? update_cfs_rq_load_avg+0x5a/0x560
>>> ? pick_task_fair+0x468/0x1a60
>>> ? __pfx_rxe_receiver+0x10/0x10
>>> ? _raw_spin_lock_irqsave+0x86/0xe0
>>> ? __pfx_rxe_receiver+0x10/0x10
>>> do_work+0x144/0x470
>>> process_one_work+0x6fc/0x1180
>>> ? __pfx___schedule+0x10/0x10
>>> ? __pfx_process_one_work+0x10/0x10
>>> ? _raw_spin_lock_irq+0x81/0xe0
>>> ? __pfx_do_work+0x10/0x10
>>> ? assign_work+0x11d/0x410
>>> worker_thread+0x45b/0xd10
>>> ? __pfx_worker_thread+0x10/0x10
>>> kthread+0x2c8/0x3b0
>>> ? recalc_sigpending+0x15c/0x1e0
>>> ? __pfx_kthread+0x10/0x10
>>> ret_from_fork+0x36e/0x5a0
>>> ? __pfx_ret_from_fork+0x10/0x10
>>> ? __switch_to+0x572/0xdc0
>>> ? __switch_to_asm+0x33/0x70
>>> ? __pfx_kthread+0x10/0x10
>>> ret_from_fork_asm+0x1a/0x30
>>> </TASK>
>>> Modules linked in:
>>> ---[ end trace 0000000000000000 ]---
>>> RIP: 0010:free_rd_atomic_resource+0x15/0x50
>>> Code: 00 00 00 00 00 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 0f 1f 40 d6 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <0f> b6 04 02 84 c0 74 04 3c 03 7e 0b c7 07 00 00 00 00 c3 cc cc cc
>>> RSP: 0018:ff110000013479f8 EFLAGS: 00000202
>>> RAX: dffffc0000000000 RBX: 0000000000000010 RCX: ff110000059f9760
>>> RDX: 0000000000000002 RSI: 0000000000000000 RDI: 0000000000000010
>>> RBP: 0000000000000001 R08: ff110000024a2180 R09: ffe21c0000268f2c
>>> R10: 0000000000000003 R11: ff110000059f9298 R12: ff1100000592dc00
>>> R13: 0000000000000040 R14: ff110000059f9000 R15: ff110000024a2168
>>> FS: 0000000000000000(0000) GS:ff110000e18da000(0000) knlGS:0000000000000000
>>> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: dffffc0000000002 CR3: 00000000052ca000 CR4: 0000000000751ef0
>>> PKRU: 55555554
>>>
>>> ------------[ cut here ]------------
>>> WARNING: drivers/infiniband/sw/rxe/rxe_pool.c:212 at __rxe_cleanup+0x21b/0x290, CPU#0: t028/87
>>> Modules linked in:
>>> CPU: 0 UID: 0 PID: 87 Comm: t028 Tainted: G D 7.4.0-rc3-gaebb72744764 #1 PREEMPT(lazy)
>>> Tainted: [D]=DIE
>>> Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
>>> RIP: 0010:__rxe_cleanup+0x21b/0x290
>>> Code: f7 fe ff ff 4c 89 e7 e8 43 99 ad fd 84 c0 0f 85 6f ff ff ff be 50 c3 00 00 4c 89 e7 e8 9e 22 f3 00 85 c0 0f 85 5a ff ff ff 90 <0f> 0b 90 eb 04 90 0f 0b 90 bd 92 ff ff ff e9 48 ff ff ff 90 0f 0b
>>> RSP: 0018:ff110000053f78f8 EFLAGS: 00000246
>>> RAX: 0000000000000000 RBX: ff110000059f9160 RCX: ffe21c0000a7ef08
>>> RDX: 0000000000000000 RSI: ff110000053f7860 RDI: ff110000059f9190
>>> RBP: 0000000000000001 R08: 0000000000000001 R09: ffe21c0000a7eef5
>>> R10: 0000000000000003 R11: ffffffff89c14700 R12: ff110000059f9188
>>> R13: ff110000059f9170 R14: ff11000004794ea0 R15: 1fe2200000a7ef1f
>>> FS: 0000000000000000(0000) GS:ff110000e18da000(0000) knlGS:0000000000000000
>>> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>>> CR2: dffffc0000000002 CR3: 000000001b872000 CR4: 0000000000751ef0
>>> PKRU: 55555554
>>> Call Trace:
>>> <TASK>
>>> ? __pfx___rxe_cleanup+0x10/0x10
>>> ? rxe_qp_chk_destroy+0x61/0xc0
>>> rxe_destroy_qp+0x2f/0x180
>>> ib_destroy_qp_user+0x30f/0x4f0
>>> ? __pfx_ib_destroy_qp_user+0x10/0x10
>>> ? stack_trace_save+0x8e/0xc0
>>> uverbs_free_qp+0x7c/0x150
>>> ? put_files_struct+0x181/0x310
>>> destroy_hw_idr_uobject+0x7e/0x130
>>> ? entry_SYSCALL_64_after_hwframe+0x77/0x7f
>>> uverbs_destroy_uobject+0xdf/0x770
>>> __uverbs_cleanup_ufile+0x18b/0x390
>>> ? __pfx___uverbs_cleanup_ufile+0x10/0x10
>>> ? __pfx_down_write+0x10/0x10
>>> uverbs_destroy_ufile_hw+0x6d/0x260
>>> ib_uverbs_close+0x46/0x270
>>> __fput+0x2f8/0x9e0
>>> fput_close_sync+0xe2/0x190
>>> ? __pfx_fput_close_sync+0x10/0x10
>>> filp_close_sync+0x18/0x30
>>> put_files_struct+0x181/0x310
>>> do_exit+0x6ab/0x25f0
>>> ? __pfx_do_exit+0x10/0x10
>>> ? zap_other_threads+0xb9/0x300
>>> ? __pfx_zap_other_threads+0x10/0x10
>>> ? handle_mm_fault+0x13d/0x420
>>> do_group_exit+0xab/0x220
>>> __x64_sys_exit_group+0x39/0x40
>>> x64_sys_call+0x1962/0x1970
>>> do_syscall_64+0xdf/0x4c0
>>> entry_SYSCALL_64_after_hwframe+0x77/0x7f
>>> RIP: 0033:0x7f34f928a21d
>>> Code: Unable to access opcode bytes at 0x7f34f928a1f3.
>>> RSP: 002b:00007ffeed584088 EFLAGS: 00000202 ORIG_RAX: 00000000000000e7
>>> RAX: ffffffffffffffda RBX: 00007f34f93a0fa8 RCX: 00007f34f928a21d
>>> RDX: 00000000000000e7 RSI: ffffffffffffff88 RDI: 0000000000000000
>>> RBP: 00007ffeed5840e0 R08: 00007ffeed584030 R09: 0000000000000000
>>> R10: 000055d8b71c93e0 R11: 0000000000000202 R12: 0000000000000001
>>> R13: 0000000000000000 R14: 0000000000000000 R15: 00007f34f93a0fc0
>>> </TASK>
>>> ---[ end trace 0000000000000000 ]---
>>>
>>> The responder task dies, and when the test exits, destroying the QP
>>> warns on a timeout in __rxe_cleanup() (the WARNING above) and then
>>> hangs.
>>>
>>> Add RXE_FLUSH_MASK to the check, so a FLUSH gets the same Invalid
>>> Request NAK a READ would, and the requester sees IB_WC_REM_INV_REQ_ERR.
>>>
>>> The issue was found by Claude Opus 5.5 running Quality Playbook, an
>>> LLM-driven code review tool:
>>> https://github.com/andrewstellman/quality-playbook
>>> The fix and this changelog were drafted with Claude Opus 5.5 and
>>> reviewed by me.
>>>
>>> Fixes: ea1bb00ee9a5 ("RDMA/rxe: Implement flush execution in responder side")
>>> Signed-off-by: Andrew Stellman <astellman@stellman-greene.com>
>>> ---
>>>
>>> v2: the full kernel log of the oops and the later WARNING, as Leon
>>> asked (timestamps removed). The code change is the same as v1.
>>> v1: https://lore.kernel.org/r/20261119151332.95466-1-astellman@stellman-greene.com
>>>
>>> Tested in QEMU with KASAN on v7.4-rc3, and on v7.4-rc3 with the RDMA
>>> code from rdma for-next (7404c82cbf12). The reproducer is a short
>>> libibverbs program; I can send it off-list.
>>>
>>> drivers/infiniband/sw/rxe/rxe_resp.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c
>>> index 1b7f31737225..38b5b0c2abe3 100644
>>> --- a/drivers/infiniband/sw/rxe/rxe_resp.c
>>> +++ b/drivers/infiniband/sw/rxe/rxe_resp.c
>>> @@ -340,7 +340,8 @@ static enum resp_states check_resource(struct rxe_qp *qp,
>>> {
>>> struct rxe_srq *srq = qp->srq;
>>>
>>> - if (pkt->mask & (RXE_READ_OR_ATOMIC_MASK | RXE_ATOMIC_WRITE_MASK)) {
>>> + if (pkt->mask & (RXE_READ_OR_ATOMIC_MASK | RXE_ATOMIC_WRITE_MASK |
>>> + RXE_FLUSH_MASK)) {
>> This looks sensible to me—adding RXE_FLUSH_MASK here avoids
>> rxe_prepare_res() / free_rd_atomic_resource() touching ZERO_SIZE_PTR
>> when max_dest_rd_atomic == 0 and returns the expected Invalid Request NAK.
>>
>> Just worth confirming:
>> (1) RXE_FLUSH_MASK takes the same error/NAK path here without later
>> responder states assuming a valid qp->resp.res or non-ZERO_SIZE_PTR
>> qp->resp.resources, and
>>
>> (2) no other RXE_READ_OR_ATOMIC_MASK / RXE_ATOMIC_WRITE_MASK checks in
>> rxe_resp.c for max_dest_rd_atomic or responder resources missed
>> RXE_FLUSH_MASK.
>>
>> Thanks a lot.
>>
>> Zhu Yanjun
>>
>>> /* it is the requesters job to not send
>>> * too many read/atomic ops, we just
>>> * recycle the responder resource queue
>>>
>>> base-commit: 7404c82cbf1246e0e84e2ec21834d23c26f9a8ed
--
Best Regards,
Yanjun.Zhu
prev parent reply other threads:[~2026-11-22 20:13 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-22 17:05 Andrew Stellman
2026-11-22 19:42 ` Zhu Yanjun
2026-11-22 19:51 ` Andrew Stellman
2026-11-22 20:13 ` Zhu Yanjun [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=9ce6bd33-f3a8-43dc-b238-d9f936ec5c54@linux.dev \
--to=yanjun.zhu@linux.dev \
--cc=astellman@stellman-greene.com \
--cc=jgg@ziepe.ca \
--cc=leon@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=zyjzyj2000@gmail.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
all inboxes | Powered by JetHome®