在 2022/4/8 17:10, Yi Zhang 写道:
On Fri, Apr 8, 2022 at 1:09 PM Yanjun Zhu <yanjun.zhu@xxxxxxxxx> wrote:
Hi, all
In 5.18-rc1, this commit "[PATCH for-next] RDMA/rxe: Revert changes from
irqsave to bh locks" does not exist.
The link is
https://patchwork.kernel.org/project/linux-rdma/patch/20220215194448.44369-1-rpearsonhpe@xxxxxxxxx/
Hi Yanjun
I tried rdma/for-next which already included this commit, and this
issue still can be reproduced.
Hi, Yi
I delved into the source code. And I found the followings:
xa_lock first is acquired in this:
[ 296.655588] {SOFTIRQ-ON-W} state was registered at:
[ 296.660467] lock_acquire+0x1d2/0x5a0
[ 296.664221] _raw_spin_lock+0x33/0x80
[ 296.667972] __rxe_add_to_pool+0x183/0x230 [rdma_rxe]
[ 296.673112] __ib_alloc_pd+0xf9/0x550 [ib_core]
[ 296.677758] ib_mad_init_device+0x2d9/0xd20 [ib_core]
[ 296.682924] add_client_context+0x2fa/0x450 [ib_core]
[ 296.688088] enable_device_and_get+0x1b7/0x350 [ib_core]
[ 296.693503] ib_register_device+0x757/0xaf0 [ib_core]
[ 296.698660] rxe_register_device+0x2eb/0x390 [rdma_rxe]
[ 296.703973] rxe_net_add+0x83/0xc0 [rdma_rxe]
[ 296.708421] rxe_newlink+0x76/0x90 [rdma_rxe]
[ 296.712865] nldev_newlink+0x245/0x3e0 [ib_core]
[ 296.717597] rdma_nl_rcv_msg+0x2d4/0x790 [ib_core]
[ 296.722492] rdma_nl_rcv+0x1ca/0x3f0 [ib_core]
[ 296.727042] netlink_unicast+0x43b/0x640
[ 296.731056] netlink_sendmsg+0x7eb/0xc40
[ 296.735069] sock_sendmsg+0xe0/0x110
[ 296.738734] __sys_sendto+0x1d7/0x2b0
[ 296.742486] __x64_sys_sendto+0xdd/0x1b0
[ 296.746500] do_syscall_64+0x37/0x80
[ 296.750166] entry_SYSCALL_64_after_hwframe+0x44/0xae
[ 296.755304] irq event stamp: 25305
[ 296.758710] hardirqs last enabled at (25304): [<ffffffff89815de8>]
__local_bh_enable_ip+0xa8/0x110
[ 296.767750] hardirqs last disabled at (25305): [<ffffffff8b970219>]
_raw_spin_lock_irqsave+0x69/0x90
[ 296.776875] softirqs last enabled at (25294): [<ffffffff8bc0064a>]
__do_softirq+0x64a/0xa4c
[ 296.785307] softirqs last disabled at (25299): [<ffffffff898159f2>]
run_ksoftirqd+0x32/0x60
[ 296.793654]
xa_lock then is acquired in this:
{IN-SOFTIRQ-W}:
stack backtrace:
[ 296.835686] CPU: 30 PID: 188 Comm: ksoftirqd/30 Tainted: G S
I 5.18.0-rc1 #1
[ 296.843859] Hardware name: Dell Inc. PowerEdge R640/06NR82, BIOS
2.11.2 004/21/2021
[ 296.851509] Call Trace:
[ 296.853964] <TASK>
[ 296.856070] dump_stack_lvl+0x44/0x57
[ 296.859734] mark_lock.part.52.cold.79+0x3c/0x46
[ 296.913364] __lock_acquire+0x1565/0x34a0
[ 296.926529] lock_acquire+0x1d2/0x5a0
[ 296.948247] _raw_spin_lock_irqsave+0x42/0x90
[ 296.957830] rxe_pool_get_index+0x72/0x1d0 [rdma_rxe]
[ 296.972554] rxe_get_av+0x168/0x2a0 [rdma_rxe]
[ 296.980761] rxe_requester+0x75b/0x4a90 [rdma_rxe]
[ 297.023764] rxe_do_task+0x134/0x230 [rdma_rxe]
[ 297.028297] tasklet_action_common.isra.12+0x1f7/0x2d0
[ 297.033435] __do_softirq+0x1ea/0xa4c
[ 297.040941] run_ksoftirqd+0x32/0x60
[ 297.044518] smpboot_thread_fn+0x503/0x860
[ 297.052110] kthread+0x29b/0x340
[ 297.060137] ret_from_fork+0x1f/0x30
[ 297.063720] </TASK>
From the above, in the function __rxe_add_to_pool, xa_lock is acquired.
Then the function __rxe_add_to_pool is interrupted by softirq. The
function rxe_pool_get_index will also acquire xa_lock.
Finally, the dead lock appears.
[ 296.806097] CPU0
[ 296.808550] ----
[ 296.811003] lock(&xa->xa_lock#15); <----- __rxe_add_to_pool
[ 296.814583] <Interrupt>
[ 296.817209] lock(&xa->xa_lock#15); <---- rxe_pool_get_index
[ 296.820961]
*** DEADLOCK ***
I disable softirq in the function __rxe_add_to_pool. Please make tests
with the following.
Please let me know the test results. Thanks a lot.
diff --git a/drivers/infiniband/sw/rxe/rxe_pool.c
b/drivers/infiniband/sw/rxe/rxe_pool.c
index 87066d04ed18..9a8f83787d61 100644
--- a/drivers/infiniband/sw/rxe/rxe_pool.c
+++ b/drivers/infiniband/sw/rxe/rxe_pool.c
@@ -166,10 +166,14 @@ int __rxe_add_to_pool(struct rxe_pool *pool,
struct rxe_pool_elem *elem)
elem->obj = (u8 *)elem - pool->elem_offset;
kref_init(&elem->ref_cnt);
+ local_bh_disable();
err = xa_alloc_cyclic(&pool->xa, &elem->index, elem, pool->limit,
&pool->next, GFP_KERNEL);
- if (err)
+ if (err) {
+ local_bh_enable();
goto err_cnt;
+ }
+ local_bh_enable();
return 0;
Zhu Yanjun
Please apply this commit and make tests again.
Please let me know the test result.
Best Regards,
Zhu Yanjun
在 2022/4/6 11:08, Yi Zhang 写道:
Hello
Below WARNING triggered during blktests srp/ tests with 5.18.0-rc1
debug kernel, pls help check it, let me know if you need any info for
it, thanks.
[ 290.308984] run blktests srp/001 at 2022-04-05 23:01:02
[ 290.999913] null_blk: module loaded
[ 291.260285] device-mapper: multipath service-time: version 0.3.0 loaded
[ 291.262990] device-mapper: table: 253:3: multipath: error getting
device (-EBUSY)
[ 291.270535] device-mapper: ioctl: error adding target to table
[ 291.362284] rdma_rxe: loaded
[ 291.428444] infiniband eno1_rxe: set active
[ 291.428462] infiniband eno1_rxe: added eno1
[ 291.467142] eno2 speed is unknown, defaulting to 1000
[ 291.472327] eno2 speed is unknown, defaulting to 1000
[ 291.477680] eno2 speed is unknown, defaulting to 1000
[ 291.516123] infiniband eno2_rxe: set down
[ 291.516130] infiniband eno2_rxe: added eno2
[ 291.516649] eno2 speed is unknown, defaulting to 1000
[ 291.542551] eno2 speed is unknown, defaulting to 1000
[ 291.558995] eno3 speed is unknown, defaulting to 1000
[ 291.564127] eno3 speed is unknown, defaulting to 1000
[ 291.569462] eno3 speed is unknown, defaulting to 1000
[ 291.607876] infiniband eno3_rxe: set down
[ 291.607883] infiniband eno3_rxe: added eno3
[ 291.608430] eno3 speed is unknown, defaulting to 1000
[ 291.632891] eno2 speed is unknown, defaulting to 1000
[ 291.638180] eno3 speed is unknown, defaulting to 1000
[ 291.655089] eno4 speed is unknown, defaulting to 1000
[ 291.660213] eno4 speed is unknown, defaulting to 1000
[ 291.665569] eno4 speed is unknown, defaulting to 1000
[ 291.703975] infiniband eno4_rxe: set down
[ 291.703982] infiniband eno4_rxe: added eno4
[ 291.704642] eno4 speed is unknown, defaulting to 1000
[ 291.822650] scsi_debug:sdebug_add_store: dif_storep 524288 bytes @
ffffc90001801000
[ 291.825441] scsi_debug:sdebug_driver_probe: scsi_debug: trim
poll_queues to 0. poll_q/nr_hw = (0/1)
[ 291.834505] scsi_debug:sdebug_driver_probe: host protection DIF3 DIX3
[ 291.834513] scsi host15: scsi_debug: version 0191 [20210520]
dev_size_mb=32, opts=0x0, submit_queues=1, statistics=0
[ 291.837267] scsi 15:0:0:0: Direct-Access Linux scsi_debug
0191 PQ: 0 ANSI: 7
[ 291.839793] sd 15:0:0:0: Power-on or device reset occurred
[ 291.840110] sd 15:0:0:0: Attached scsi generic sg1 type 0
[ 291.845521] sd 15:0:0:0: [sdb] Enabling DIF Type 3 protection
[ 291.845878] sd 15:0:0:0: [sdb] 65536 512-byte logical blocks: (33.6
MB/32.0 MiB)
[ 291.845939] sd 15:0:0:0: [sdb] Write Protect is off
[ 291.845954] sd 15:0:0:0: [sdb] Mode Sense: 73 00 10 08
[ 291.846049] sd 15:0:0:0: [sdb] Write cache: enabled, read cache:
enabled, supports DPO and FUA
[ 291.846254] sd 15:0:0:0: [sdb] Optimal transfer size 524288 bytes
[ 291.859380] sd 15:0:0:0: [sdb] Enabling DIX T10-DIF-TYPE3-CRC protection
[ 291.859398] sd 15:0:0:0: [sdb] DIF application tag size 6
[ 291.860158] sd 15:0:0:0: [sdb] Attached SCSI disk
[ 292.666414] eno2 speed is unknown, defaulting to 1000
[ 292.771984] eno3 speed is unknown, defaulting to 1000
[ 292.876762] eno4 speed is unknown, defaulting to 1000
[ 293.033291] Rounding down aligned max_sectors from 4294967295 to 4294967288
[ 293.102261] ib_srpt:srpt_add_one: ib_srpt device = 0000000047a39f45
[ 293.102363] ib_srpt:srpt_use_srq: ib_srpt srpt_use_srq(eno1_rxe):
use_srq = 0; ret = 0
[ 293.102369] ib_srpt:srpt_add_one: ib_srpt Target login info:
id_ext=f2d4e2fffee6e1e0,ioc_guid=f2d4e2fffee6e1e0,pkey=ffff,service_id=f2d4e2fffee6e1e0
[ 293.102680] ib_srpt:srpt_add_one: ib_srpt added eno1_rxe.
[ 293.102692] ib_srpt:srpt_add_one: ib_srpt device = 00000000b2dfcbe9
[ 293.102725] ib_srpt:srpt_use_srq: ib_srpt srpt_use_srq(eno2_rxe):
use_srq = 0; ret = 0
[ 293.102730] ib_srpt:srpt_add_one: ib_srpt Target login info:
id_ext=f2d4e2fffee6e1e0,ioc_guid=f2d4e2fffee6e1e0,pkey=ffff,service_id=f2d4e2fffee6e1e0
[ 293.102741] eno2 speed is unknown, defaulting to 1000
[ 293.107884] ib_srpt:srpt_add_one: ib_srpt added eno2_rxe.
[ 293.107893] ib_srpt:srpt_add_one: ib_srpt device = 0000000061e03247
[ 293.107922] ib_srpt:srpt_use_srq: ib_srpt srpt_use_srq(eno3_rxe):
use_srq = 0; ret = 0
[ 293.107926] ib_srpt:srpt_add_one: ib_srpt Target login info:
id_ext=f2d4e2fffee6e1e0,ioc_guid=f2d4e2fffee6e1e0,pkey=ffff,service_id=f2d4e2fffee6e1e0
[ 293.107936] eno3 speed is unknown, defaulting to 1000
[ 293.113038] ib_srpt:srpt_add_one: ib_srpt added eno3_rxe.
[ 293.113046] ib_srpt:srpt_add_one: ib_srpt device = 00000000c0e3d43d
[ 293.113081] ib_srpt:srpt_use_srq: ib_srpt srpt_use_srq(eno4_rxe):
use_srq = 0; ret = 0
[ 293.113085] ib_srpt:srpt_add_one: ib_srpt Target login info:
id_ext=f2d4e2fffee6e1e0,ioc_guid=f2d4e2fffee6e1e0,pkey=ffff,service_id=f2d4e2fffee6e1e0
[ 293.113096] eno4 speed is unknown, defaulting to 1000
[ 293.118198] ib_srpt:srpt_add_one: ib_srpt added eno4_rxe.
[ 293.584001] Rounding down aligned max_sectors from 255 to 248
[ 293.654030] Rounding down aligned max_sectors from 255 to 248
[ 293.724363] Rounding down aligned max_sectors from 4294967295 to 4294967288
[ 296.450772] ib_srp:srp_add_one: ib_srp: srp_add_one:
18446744073709551615 / 4096 = 4503599627370495 <> 512
[ 296.450783] ib_srp:srp_add_one: ib_srp: eno1_rxe: mr_page_shift =
12, device->max_mr_size = 0xffffffffffffffff,
device->max_fast_reg_page_list_len = 512, max_pages_per_mr = 512,
mr_max_size = 0x200000
[ 296.451222] ib_srp:srp_add_one: ib_srp: srp_add_one:
18446744073709551615 / 4096 = 4503599627370495 <> 512
[ 296.451229] ib_srp:srp_add_one: ib_srp: eno2_rxe: mr_page_shift =
12, device->max_mr_size = 0xffffffffffffffff,
device->max_fast_reg_page_list_len = 512, max_pages_per_mr = 512,
mr_max_size = 0x200000
[ 296.451517] ib_srp:srp_add_one: ib_srp: srp_add_one:
18446744073709551615 / 4096 = 4503599627370495 <> 512
[ 296.451523] ib_srp:srp_add_one: ib_srp: eno3_rxe: mr_page_shift =
12, device->max_mr_size = 0xffffffffffffffff,
device->max_fast_reg_page_list_len = 512, max_pages_per_mr = 512,
mr_max_size = 0x200000
[ 296.451769] ib_srp:srp_add_one: ib_srp: srp_add_one:
18446744073709551615 / 4096 = 4503599627370495 <> 512
[ 296.451774] ib_srp:srp_add_one: ib_srp: eno4_rxe: mr_page_shift =
12, device->max_mr_size = 0xffffffffffffffff,
device->max_fast_reg_page_list_len = 512, max_pages_per_mr = 512,
mr_max_size = 0x200000
[ 296.605925] ib_srp:srp_parse_in: ib_srp: 10.16.221.116 -> 10.16.221.116:0
[ 296.605956] ib_srp:srp_parse_in: ib_srp: 10.16.221.116:5555 ->
10.16.221.116:5555
[ 296.606014] ib_srp:add_target_store: ib_srp: max_sectors = 1024;
max_pages_per_mr = 512; mr_page_size = 4096; max_sectors_per_mr =
4096; mr_per_cmd = 2
[ 296.606021] ib_srp:srp_max_it_iu_len: ib_srp: max_iu_len = 8260
[ 296.616660] ================================
[ 296.620931] WARNING: inconsistent lock state
[ 296.625207] 5.18.0-rc1 #1 Tainted: G S I
[ 296.630259] --------------------------------
[ 296.634531] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[ 296.640535] ksoftirqd/30/188 [HC0[0]:SC1[1]:HE0:SE0] takes:
[ 296.646106] ffff888151491468 (&xa->xa_lock#15){+.?.}-{2:2}, at:
rxe_pool_get_index+0x72/0x1d0 [rdma_rxe]
[ 296.655588] {SOFTIRQ-ON-W} state was registered at:
[ 296.660467] lock_acquire+0x1d2/0x5a0
[ 296.664221] _raw_spin_lock+0x33/0x80
[ 296.667972] __rxe_add_to_pool+0x183/0x230 [rdma_rxe]
[ 296.673112] __ib_alloc_pd+0xf9/0x550 [ib_core]
[ 296.677758] ib_mad_init_device+0x2d9/0xd20 [ib_core]
[ 296.682924] add_client_context+0x2fa/0x450 [ib_core]
[ 296.688088] enable_device_and_get+0x1b7/0x350 [ib_core]
[ 296.693503] ib_register_device+0x757/0xaf0 [ib_core]
[ 296.698660] rxe_register_device+0x2eb/0x390 [rdma_rxe]
[ 296.703973] rxe_net_add+0x83/0xc0 [rdma_rxe]
[ 296.708421] rxe_newlink+0x76/0x90 [rdma_rxe]
[ 296.712865] nldev_newlink+0x245/0x3e0 [ib_core]
[ 296.717597] rdma_nl_rcv_msg+0x2d4/0x790 [ib_core]
[ 296.722492] rdma_nl_rcv+0x1ca/0x3f0 [ib_core]
[ 296.727042] netlink_unicast+0x43b/0x640
[ 296.731056] netlink_sendmsg+0x7eb/0xc40
[ 296.735069] sock_sendmsg+0xe0/0x110
[ 296.738734] __sys_sendto+0x1d7/0x2b0
[ 296.742486] __x64_sys_sendto+0xdd/0x1b0
[ 296.746500] do_syscall_64+0x37/0x80
[ 296.750166] entry_SYSCALL_64_after_hwframe+0x44/0xae
[ 296.755304] irq event stamp: 25305
[ 296.758710] hardirqs last enabled at (25304): [<ffffffff89815de8>]
__local_bh_enable_ip+0xa8/0x110
[ 296.767750] hardirqs last disabled at (25305): [<ffffffff8b970219>]
_raw_spin_lock_irqsave+0x69/0x90
[ 296.776875] softirqs last enabled at (25294): [<ffffffff8bc0064a>]
__do_softirq+0x64a/0xa4c
[ 296.785307] softirqs last disabled at (25299): [<ffffffff898159f2>]
run_ksoftirqd+0x32/0x60
[ 296.793654]
other info that might help us debug this:
[ 296.800177] Possible unsafe locking scenario:
[ 296.806097] CPU0
[ 296.808550] ----
[ 296.811003] lock(&xa->xa_lock#15);
[ 296.814583] <Interrupt>
[ 296.817209] lock(&xa->xa_lock#15);
[ 296.820961]
*** DEADLOCK ***
[ 296.826880] no locks held by ksoftirqd/30/188.
[ 296.831326]
stack backtrace:
[ 296.835686] CPU: 30 PID: 188 Comm: ksoftirqd/30 Tainted: G S
I 5.18.0-rc1 #1
[ 296.843859] Hardware name: Dell Inc. PowerEdge R640/06NR82, BIOS
2.11.2 004/21/2021
[ 296.851509] Call Trace:
[ 296.853964] <TASK>
[ 296.856070] dump_stack_lvl+0x44/0x57
[ 296.859734] mark_lock.part.52.cold.79+0x3c/0x46
[ 296.864355] ? lock_chain_count+0x20/0x20
[ 296.868367] ? rxe_do_task+0x26/0x230 [rdma_rxe]
[ 296.872994] ? orc_find.part.4+0x310/0x310
[ 296.877096] ? __module_text_address.part.55+0x13/0x140
[ 296.882326] ? rxe_do_task+0x26/0x230 [rdma_rxe]
[ 296.886947] ? rxe_do_task+0x26/0x230 [rdma_rxe]
[ 296.891566] ? is_module_text_address+0x41/0x60
[ 296.896098] ? rxe_do_task+0x26/0x230 [rdma_rxe]
[ 296.900718] ? kernel_text_address+0x13/0xd0
[ 296.904991] ? create_prof_cpu_mask+0x20/0x20
[ 296.909351] ? sched_clock_cpu+0x15/0x200
[ 296.913364] __lock_acquire+0x1565/0x34a0
[ 296.917377] ? rcu_read_lock_sched_held+0xaf/0xe0
[ 296.922079] ? rcu_read_lock_bh_held+0xc0/0xc0
[ 296.926529] lock_acquire+0x1d2/0x5a0
[ 296.930193] ? rxe_pool_get_index+0x72/0x1d0 [rdma_rxe]
[ 296.935429] ? rcu_read_unlock+0x50/0x50
[ 296.939353] ? mark_lock.part.52+0xa3d/0x1c00
[ 296.943712] ? _raw_spin_lock_irqsave+0x69/0x90
[ 296.948247] _raw_spin_lock_irqsave+0x42/0x90
[ 296.952603] ? rxe_pool_get_index+0x72/0x1d0 [rdma_rxe]
[ 296.957830] rxe_pool_get_index+0x72/0x1d0 [rdma_rxe]
[ 296.962882] ? __rxe_add_to_pool+0x230/0x230 [rdma_rxe]
[ 296.968108] ? __rxe_get+0xc1/0x140 [rdma_rxe]
[ 296.972554] rxe_get_av+0x168/0x2a0 [rdma_rxe]
[ 296.977007] ? lockdep_lock+0xcb/0x1c0
[ 296.980761] rxe_requester+0x75b/0x4a90 [rdma_rxe]
[ 296.985557] ? rxe_do_task+0xe2/0x230 [rdma_rxe]
[ 296.990183] ? sched_clock_cpu+0x15/0x200
[ 296.994193] ? find_held_lock+0x3a/0x1c0
[ 296.998119] ? rnr_nak_timer+0x80/0x80 [rdma_rxe]
[ 297.002826] ? lock_release+0x42f/0xc90
[ 297.006664] ? lock_downgrade+0x6b0/0x6b0
[ 297.010676] ? lock_acquired+0x262/0xb10
[ 297.014605] ? __local_bh_enable_ip+0xa8/0x110
[ 297.019051] ? rnr_nak_timer+0x80/0x80 [rdma_rxe]
[ 297.023764] rxe_do_task+0x134/0x230 [rdma_rxe]
[ 297.028297] tasklet_action_common.isra.12+0x1f7/0x2d0
[ 297.033435] __do_softirq+0x1ea/0xa4c
[ 297.037100] ? tasklet_kill+0x1c0/0x1c0
[ 297.040941] run_ksoftirqd+0x32/0x60
[ 297.044518] smpboot_thread_fn+0x503/0x860
[ 297.048618] ? sort_range+0x20/0x20
[ 297.052110] kthread+0x29b/0x340
[ 297.055342] ? kthread_complete_and_exit+0x20/0x20
[ 297.060137] ret_from_fork+0x1f/0x30
[ 297.063720] </TASK>
