On 5/6/22 19:29, Yanjun Zhu wrote:
> 在 2022/5/7 8:10, Bart Van Assche 写道:
>> On 5/6/22 11:11, Bob Pearson wrote:
>>> Before the most recent kernel update I had blktests running OK on rdma_rxe. Since we went on to 5.18.0-rc1+
>>> I have been experiencing hangs. All of this is with the 'revert scsi-debug' patch which addressed the
>>> 3 min timeout related to modprobe -r scsi-debug.
>>>
>>> You suggested checking with siw and I finally got around to this and the behavior is exactly the same.
>>>
>>> Specifically here is a run and dmesgs from that run:
>>>
>>> root@u-22:/home/bob/src/blktests# use_siw=1 ./check srp
>>>
>>> srp/001 (Create and remove LUNs) [passed]
>>>
>>> runtime 3.388s ... 3.501s
>>>
>>> srp/002 (File I/O on top of multipath concurrently with logout and login (mq))
>>>
>>> runtime 54.689s ...
>>> <HANGS HERE>
>>>
>>> I had to reboot to recover.
>>>
>>> The dmesg output is attached in a long file called out.
>>> The output looks normal until line 1875 where it hangs at an "Already connected ..." message.
>>> This is the same as the other hangs I have been seeing.
>>> This is followed by a splat warning that a cpu has hung for 120 seconds.
>>>
>>> Since this is behaving the same for rxe and siw I am going to stop chasing this bug since
>>> it is most likely outside of the the rxe driver.
>>
>> Hi Bob,
>>
>> What I see on my test setup is that the SRP tests from the blktests suite pass with
>> the SoftiWARP driver (kernel v5.18-rc5 / commit 4b97bac0756a):
>>
>> # (cd blktests && use_siw=1 ./check -q srp)
>> srp/001 (Create and remove LUNs) [passed]
>> runtime 5.781s ... 5.464s
>> srp/002 (File I/O on top of multipath concurrently with logout and login (mq)) [passed]time 40.772s ...
>> runtime 40.772s ... 42.039s
>> srp/003 (File I/O on top of multipath concurrently with logout and login (sq)) [not run]
>> legacy device mapper support is missing
>> srp/004 (File I/O on top of multipath concurrently with logout and login (sq-on-srp/004 (File I/O on top of multipath concurrently with logout and login (sq-on-mq)) [not run]
>> legacy device mapper support is missing
>> srp/005 (Direct I/O with large transfer sizes, cmd_sg_entries=255 and bs=4M) [passed]untime 17.870s ...
>> runtime 17.870s ... 17.016s
>> srp/006 (Direct I/O with large transfer sizes, cmd_sg_entries=255 and bs=8M) [passed]untime 16.369s ...
>> runtime 16.369s ... 17.315s
>> srp/007 (Direct I/O with large transfer sizes, cmd_sg_entries=1 and bs=4M) [passed] runtime 16.729s ...
>> runtime 16.729s ... 17.409s
>> srp/008 (Direct I/O with large transfer sizes, cmd_sg_entries=1 and bs=8M) [passed] runtime 16.823s ...
>> runtime 16.823s ... 16.453s
>> srp/009 (Buffered I/O with large transfer sizes, cmd_sg_entries=255 and bs=4M) [passed]time 17.304s ...
>> runtime 17.304s ... 17.838s
>> srp/010 (Buffered I/O with large transfer sizes, cmd_sg_entries=255 and bs=8M) [passed]time 17.191s ...
>> runtime 17.191s ... 17.117s
>> srp/011 (Block I/O on top of multipath concurrently with logout and login) [passed] runtime 40.835s ...
>> runtime 40.835s ... 38.728s
>> srp/012 (dm-mpath on top of multiple I/O schedulers) [passed]
>> runtime 23.703s ... 24.763s
>> srp/013 (Direct I/O using a discontiguous buffer) [passed]
>> runtime 11.279s ... 9.265s
>> srp/014 (Run sg_reset while I/O is ongoing) [passed]
>> runtime 39.110s ... 37.929s
>> srp/015 (File I/O on top of multipath concurrently with logout and login (mq) ussrp/015
>> (File I/O on top of multipath concurrently with logout and login (mq) using the SoftiWARP (siw) driver) [passed]
>> runtime 40.027s ... 40.220s
>>
>> If I try to run the SRP test 002 with the soft-RoCE driver, the following appears:
>>
>> [ 749.901966] ================================
>> [ 749.903638] WARNING: inconsistent lock state
>> [ 749.905376] 5.18.0-rc5-dbg+ #1 Not tainted
>> [ 749.907039] --------------------------------
>> [ 749.908699] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
>> [ 749.910646] ksoftirqd/5/40 [HC0[0]:SC1[1]:HE0:SE0] takes:
>> [ 749.912499] ffff88818244d350 (&xa->xa_lock#14){+.?.}-{2:2}, at: rxe_pool_get_index+0x73/0x170 [rdma_rxe]
>> [ 749.914691] {SOFTIRQ-ON-W} state was registered at:
>> [ 749.916648] __lock_acquire+0x45b/0xce0
>> [ 749.918599] lock_acquire+0x18a/0x450
>> [ 749.920480] _raw_spin_lock+0x34/0x50
>> [ 749.922580] __rxe_add_to_pool+0xcc/0x140 [rdma_rxe]
>> [ 749.924583] rxe_alloc_pd+0x2d/0x40 [rdma_rxe]
>> [ 749.926394] __ib_alloc_pd+0xa3/0x270 [ib_core]
>> [ 749.928579] ib_mad_port_open+0x44a/0x790 [ib_core]
>> [ 749.930640] ib_mad_init_device+0x8e/0x110 [ib_core]
>> [ 749.932495] add_client_context+0x26a/0x330 [ib_core]
>> [ 749.934302] enable_device_and_get+0x169/0x2b0 [ib_core]
>> [ 749.936217] ib_register_device+0x26f/0x330 [ib_core]
>> [ 749.938020] rxe_register_device+0x1b4/0x1d0 [rdma_rxe]
>> [ 749.939794] rxe_add+0x8c/0xc0 [rdma_rxe]
>> [ 749.941552] rxe_net_add+0x5b/0x90 [rdma_rxe]
>> [ 749.943356] rxe_newlink+0x71/0x80 [rdma_rxe]
>> [ 749.945182] nldev_newlink+0x21e/0x370 [ib_core]
>> [ 749.946917] rdma_nl_rcv_msg+0x200/0x410 [ib_core]
>> [ 749.948657] rdma_nl_rcv+0x140/0x220 [ib_core]
>> [ 749.950373] netlink_unicast+0x307/0x460
>> [ 749.952063] netlink_sendmsg+0x422/0x750
>> [ 749.953672] __sys_sendto+0x1c2/0x250
>> [ 749.955281] __x64_sys_sendto+0x7f/0x90
>> [ 749.956849] do_syscall_64+0x35/0x80
>> [ 749.958353] entry_SYSCALL_64_after_hwframe+0x44/0xae
>> [ 749.959942] irq event stamp: 1411849
>> [ 749.961517] hardirqs last enabled at (1411848): [<ffffffff810cdb28>] __local_bh_enable_ip+0x88/0xf0
>> [ 749.963338] hardirqs last disabled at (1411849): [<ffffffff81ebf24d>] _raw_spin_lock_irqsave+0x5d/0x60
>> [ 749.965214] softirqs last enabled at (1411838): [<ffffffff82200467>] __do_softirq+0x467/0x6e1
>> [ 749.967027] softirqs last disabled at (1411843): [<ffffffff810cd947>] run_ksoftirqd+0x37/0x60
> To this, Please use this patch series news://nntp.lore.kernel.org:119/20220422194416.983549-1-yanjun.zhu@xxxxxxxxx
>
> Zhu Yanjun
>>
>> I think the above is strong evidence that there is something wrong with the
>> soft-RoCE driver.
>>
>> Thanks,
>>
>> Bart.
>
I was showing siw results not rxe results. When I have run srp on rxe I use a patch similar to the
one Zhu suggested to fix the lockdep warnings.
Bob
