On 2/13/24 17:22, Don Brace wrote:
> Correct blk-mq registration issue with module parameter
> disable_managed_interrupts enabled.
>
> When we turn off the default PCI_IRQ_AFFINITY flag, the driver needs to
> register with blk-mq using blk_mq_map_queues(). The driver is currently
> calling blk_mq_pci_map_queues() which results in a stack trace and
> possibly undefined behavior.
>
> Stack Trace:
> [ 7.860089] scsi host2: smartpqi
> [ 7.871934] WARNING: CPU: 0 PID: 238 at block/blk-mq-pci.c:52 blk_mq_pci_map_queues+0xca/0xd0
> [ 7.889231] Modules linked in: sd_mod t10_pi sg uas smartpqi(+) crc32c_intel scsi_transport_sas usb_storage dm_mirror dm_region_hash dm_log dm_mod ipmi_devintf ipmi_msghandler fuse
> [ 7.924755] CPU: 0 PID: 238 Comm: kworker/0:3 Not tainted 4.18.0-372.88.1.el8_6_smartpqi_test.x86_64 #1
> [ 7.944336] Hardware name: HPE ProLiant DL380 Gen10/ProLiant DL380 Gen10, BIOS U30 03/08/2022
> [ 7.963026] Workqueue: events work_for_cpu_fn
> [ 7.978275] RIP: 0010:blk_mq_pci_map_queues+0xca/0xd0
> [ 7.978278] Code: 48 89 de 89 c7 e8 f6 0f 4f 00 3b 05 c4 b7 8e 01 72 e1 5b 31 c0 5d 41 5c 41 5d 41 5e 41 5f e9 7d df 73 00 31 c0 e9 76 df 73 00 <0f> 0b eb bc 90 90 0f 1f 44 00 00 41 57 49 89 ff 41 56 41 55 41 54
> [ 7.978280] RSP: 0018:ffffa95fc3707d50 EFLAGS: 00010216
> [ 7.978283] RAX: 00000000ffffffff RBX: 0000000000000000 RCX: 0000000000000010
> [ 7.978284] RDX: 0000000000000004 RSI: 0000000000000000 RDI: ffff9190c32d4310
> [ 7.978286] RBP: 0000000000000000 R08: ffffa95fc3707d38 R09: ffff91929b81ac00
> [ 7.978287] R10: 0000000000000001 R11: ffffa95fc3707ac0 R12: 0000000000000000
> [ 7.978288] R13: ffff9190c32d4000 R14: 00000000ffffffff R15: ffff9190c4c950a8
> [ 7.978290] FS: 0000000000000000(0000) GS:ffff9193efc00000(0000) knlGS:0000000000000000
> [ 7.978292] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 8.172814] CR2: 000055d11166c000 CR3: 00000002dae10002 CR4: 00000000007706f0
> [ 8.172816] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [ 8.172817] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> [ 8.172818] PKRU: 55555554
> [ 8.172819] Call Trace:
> [ 8.172823] blk_mq_alloc_tag_set+0x12e/0x310
> [ 8.264339] scsi_add_host_with_dma.cold.9+0x30/0x245
> [ 8.279302] pqi_ctrl_init+0xacf/0xc8e [smartpqi]
> [ 8.294085] ? pqi_pci_probe+0x480/0x4c8 [smartpqi]
> [ 8.309015] pqi_pci_probe+0x480/0x4c8 [smartpqi]
> [ 8.323286] local_pci_probe+0x42/0x80
> [ 8.337855] work_for_cpu_fn+0x16/0x20
> [ 8.351193] process_one_work+0x1a7/0x360
> [ 8.364462] ? create_worker+0x1a0/0x1a0
> [ 8.379252] worker_thread+0x1ce/0x390
> [ 8.392623] ? create_worker+0x1a0/0x1a0
> [ 8.406295] kthread+0x10a/0x120
> [ 8.418428] ? set_kthread_struct+0x50/0x50
> [ 8.431532] ret_from_fork+0x1f/0x40
> [ 8.444137] ---[ end trace 1bf0173d39354506 ]---
This patch fixes the issue on my machine.
Reviewed-by: Tomas Henzl <thenzl@xxxxxxxxxx>
>
> Fixes: ("cf15c3e734e8 scsi: smartpqi: Add module param to disable managed ints")
>
> Tested-by: Yogesh Chandra Pandey <YogeshChandra.Pandey@xxxxxxxxxxxxx>
> Reviewed-by: Scott Benesh <scott.benesh@xxxxxxxxxxxxx>
> Reviewed-by: Scott Teel <scott.teel@xxxxxxxxxxxxx>
> Reviewed-by: Mahesh Rajashekhara <mahesh.rajashekhara@xxxxxxxxxxxxx>
> Reviewed-by: Mike McGowen <mike.mcgowen@xxxxxxxxxxxxx>
> Reviewed-by: Kevin Barnett <kevin.barnett@xxxxxxxxxxxxx>
> Signed-off-by: Don Brace <don.brace@xxxxxxxxxxxxx>
> ---
> drivers/scsi/smartpqi/smartpqi_init.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/scsi/smartpqi/smartpqi_init.c b/drivers/scsi/smartpqi/smartpqi_init.c
> index ceff1ec13f9e..385180c98be4 100644
> --- a/drivers/scsi/smartpqi/smartpqi_init.c
> +++ b/drivers/scsi/smartpqi/smartpqi_init.c
> @@ -6533,8 +6533,11 @@ static void pqi_map_queues(struct Scsi_Host *shost)
> {
> struct pqi_ctrl_info *ctrl_info = shost_to_hba(shost);
>
> - blk_mq_pci_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT],
> + if (!ctrl_info->disable_managed_interrupts)
> + return blk_mq_pci_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT],
> ctrl_info->pci_dev, 0);
> + else
> + return blk_mq_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT]);
> }
>
> static inline bool pqi_is_tape_changer_device(struct pqi_scsi_dev *device)
