scsi: mpt3sas: avoid watchdog issue while releasing chain buffers
Subsystems (2)
PATCH 1/1 scsi: mpt3sas: avoid watchdog issue while releasing chain buffers
From: Martin Wilck <[email protected]>
I observe the watchdog timer being triggered while unloading the mpt3sas driver:
Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_detach Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_free_resources Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_make_ioc_ready Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: sending message unit reset !! Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: message unit reset: SUCCESS Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_unmap_resources Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: _base_release_memory_pools Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: request_pool(0x00000000144b1531): free Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: sense_pool(0x000000009665c238): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_pool(0x000000005c5e0fa5): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_free_pool(0x000000006f897f6c): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_post_free_pool(0x00000000d1edc4aa): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: config_page(0x000000009f651842): free Jan 12 12:26:23 tegmen kernel: watchdog: BUG: soft lockup - CPU#27 stuck for 26s! [rmmod:2594] Jan 12 12:26:23 tegmen kernel: Hardware name: HP ProLiant DL560 Gen8, BIOS P77 05/24/2019 Jan 12 12:26:23 tegmen kernel: RIP: 0010:_raw_spin_unlock_irqrestore+0x26/0x2e Jan 12 12:26:23 tegmen kernel: Code: 1f 44 00 00 0f 1f 44 00 00 c6 07 00 0f 1f 40 00 f7 c6 00 02 00 00 75 0b 65 ff 0d 05 ce a1 5f 74 0> Jan 12 12:26:23 tegmen kernel: RSP: 0018:ffffab1546bdfcc8 EFLAGS: 00000206 Jan 12 12:26:23 tegmen kernel: RAX: 0000000000000c80 RBX: ffff8d82b0f16700 RCX: 0000000000000d00 Jan 12 12:26:23 tegmen kernel: RDX: 0000000453642d00 RSI: 0000000000000282 RDI: ffff8d8292075f90 Jan 12 12:26:23 tegmen kernel: RBP: ffff8d8292075f80 R08: 0000000000000000 R09: 0000000000000001 Jan 12 12:26:23 tegmen kernel: R10: 0000000000000003 R11: ffff8d8284256a00 R12: ffff8d8293642d00 Jan 12 12:26:23 tegmen kernel: R13: ffff8d8292075f90 R14: 0000000000000282 R15: 0000000000000d00 Jan 12 12:26:23 tegmen kernel: FS: 00007fbd96388740(0000) GS:ffff8d8e7f6c0000(0000) knlGS:0000000000000000 Jan 12 12:26:23 tegmen kernel: CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 Jan 12 12:26:23 tegmen kernel: CR2: 000055bbd50f9918 CR3: 0000000c80b0c001 CR4: 00000000000606e0 Jan 12 12:26:23 tegmen kernel: Call Trace: Jan 12 12:26:23 tegmen kernel: <TASK> Jan 12 12:26:23 tegmen kernel: dma_pool_free+0xc1/0x100 Jan 12 12:26:23 tegmen kernel: _base_release_memory_pools+0x343/0x4c0 [mpt3sas 6ff0715b1f6f07c16051cb2772836069b2821b01] Jan 12 12:26:23 tegmen kernel: mpt3sas_base_detach+0x2e/0x130 [mpt3sas 6ff0715b1f6f07c16051cb2772836069b2821b01]
When the driver is unloaded during system shutdown, this may actually cause a kernel panic triggered by the watchdog.
The problem is that with the hardware in question, the driver allocates a very large number of DMA buffers for chain lookup (scsiio_depth = 29868, chains_needed_per_io = 15, total number of buffers = 448020). The loop that frees all DMA buffers takes ~30s to execute. By adding a cond_resched() in the loop, the watchdog is avoided.
Note: This is the 2nd issue I saw with this controller and the reported can_queue value after https://lore.kernel.org/linux-scsi/Ydug9nWg4loEVkJw@T590/T/
--- drivers/scsi/mpt3sas/mpt3sas_base.c | 1 + 1 file changed, 1 insertion(+)
drivers/scsi/mpt3sas/mpt3sas_base.c +1 -0
@@ -5715,6 +5715,7 @@ _base_release_memory_pools(struct MPT3SAS_ADAPTER *ioc) ct->chain_buffer_dma); }kfree (ioc->chain_lookup[i].chains_per_smid); +cond_resched (); }dma_pool_destroy (ioc->chain_dma_pool);kfree (ioc->chain_lookup);
Signature · 1 line
2.34.1
Quoted 109 lines · On Wed, 2022-01-12 at 15:01 +0100, [email protected] wrote:
From: Martin Wilck <[email protected]>
I observe the watchdog timer being triggered while unloading the mpt3sas driver:
Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_detach Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_free_resources Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_make_ioc_ready Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: sending message unit reset !! Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: message unit reset: SUCCESS Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: mpt3sas_base_unmap_resources Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: _base_release_memory_pools Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: request_pool(0x00000000144b1531): free Jan 12 12:25:51 tegmen kernel: mpt2sas_cm0: sense_pool(0x000000009665c238): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_pool(0x000000005c5e0fa5): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_free_pool(0x000000006f897f6c): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: reply_post_free_pool(0x00000000d1edc4aa): free Jan 12 12:25:52 tegmen kernel: mpt2sas_cm0: config_page(0x000000009f651842): free Jan 12 12:26:23 tegmen kernel: watchdog: BUG: soft lockup - CPU#27 stuck for 26s! [rmmod:2594] Jan 12 12:26:23 tegmen kernel: Hardware name: HP ProLiant DL560 Gen8, BIOS P77 05/24/2019 Jan 12 12:26:23 tegmen kernel: RIP: 0010:_raw_spin_unlock_irqrestore+0x26/0x2e Jan 12 12:26:23 tegmen kernel: Code: 1f 44 00 00 0f 1f 44 00 00 c6 07 00 0f 1f 40 00 f7 c6 00 02 00 00 75 0b 65 ff 0d 05 ce a1 5f 74 0> Jan 12 12:26:23 tegmen kernel: RSP: 0018:ffffab1546bdfcc8 EFLAGS: 00000206 Jan 12 12:26:23 tegmen kernel: RAX: 0000000000000c80 RBX: ffff8d82b0f16700 RCX: 0000000000000d00 Jan 12 12:26:23 tegmen kernel: RDX: 0000000453642d00 RSI: 0000000000000282 RDI: ffff8d8292075f90 Jan 12 12:26:23 tegmen kernel: RBP: ffff8d8292075f80 R08: 0000000000000000 R09: 0000000000000001 Jan 12 12:26:23 tegmen kernel: R10: 0000000000000003 R11: ffff8d8284256a00 R12: ffff8d8293642d00 Jan 12 12:26:23 tegmen kernel: R13: ffff8d8292075f90 R14: 0000000000000282 R15: 0000000000000d00 Jan 12 12:26:23 tegmen kernel: FS: 00007fbd96388740(0000) GS:ffff8d8e7f6c0000(0000) knlGS:0000000000000000 Jan 12 12:26:23 tegmen kernel: CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 Jan 12 12:26:23 tegmen kernel: CR2: 000055bbd50f9918 CR3: 0000000c80b0c001 CR4: 00000000000606e0 Jan 12 12:26:23 tegmen kernel: Call Trace: Jan 12 12:26:23 tegmen kernel: <TASK> Jan 12 12:26:23 tegmen kernel: dma_pool_free+0xc1/0x100 Jan 12 12:26:23 tegmen kernel: _base_release_memory_pools+0x343/0x4c0 [mpt3sas 6ff0715b1f6f07c16051cb2772836069b2821b01] Jan 12 12:26:23 tegmen kernel: mpt3sas_base_detach+0x2e/0x130[mpt3sas
6ff0715b1f6f07c16051cb2772836069b2821b01]When the driver is unloaded during system shutdown, this may actually cause a
kernel panic triggered by the watchdog.The problem is that with the hardware in question, the driver allocates a very
large number of DMA buffers for chain lookup (scsiio_depth= 29868,chains_needed_per_io= 15, total number of buffers = 448020). The loop that
frees all DMA buffers takes ~30s to execute. By adding acond_resched()in the
loop, the watchdog is avoided.Note: This is the 2nd issue I saw with this controller and the reported
can_queue
value after
https://lore.kernel.org/linux-scsi/Ydug9nWg4loEVkJw@T590/T/Fixes 93204b782a88 ("scsi: mpt3sas: Lockless access for chainbuffers.")
Signed-off-by Martin Wilck <[email protected]>CC Sathya Prakash <[email protected]>Cc Sreekanth Reddy <[email protected]>Cc Suganath Prabu Subramani <[email protected]>---
drivers/scsi/mpt3sas/mpt3sas_base.c | 1 +1 file changed, 1 insertion(+)drivers/scsi/mpt3sas/mpt3sas_base.c +0
-0@@ -5715,6 +5715,7 @@ _base_release_memory_pools(struct MPT3SAS_ADAPTER *ioc)ct-
Quoted 1 line
chain_buffer_dma);}
kfree(ioc->chain_lookup[i].chains_per_smid); + cond_resched();
}
dma_pool_destroy(ioc->chain_dma_pool);kfree(ioc->chain_lookup);