Re: [PATCH v1] mmc: core: check R1_STATUS for erase/trim/discard

[Florian Fainelli <f.fainelli@gmail.com>]

On 4/23/24 13:02, Kamal Dasu wrote:
> When erase/trim/discard completion was converted to mmc_poll_for_busy(),
> optional ->card_busy() host ops support was added. sdhci card->busy()
> could return busy for long periods to cause mmc_do_erase() to block during
> discard operation as shown below during mkfs.f2fs :
> 
> Info: [/dev/mmcblk1p9] Discarding device
> [   39.597258] sysrq: Show Blocked State
> [   39.601183] task:mkfs.f2fs       state:D stack:0     pid:1561  tgid:1561  ppid:1542   flags:0x0000000d
> [   39.610609] Call trace:
> [   39.613098]  __switch_to+0xd8/0xf4
> [   39.616582]  __schedule+0x440/0x4f4
> [   39.620137]  schedule+0x2c/0x48
> [   39.623341]  schedule_hrtimeout_range_clock+0xe0/0x114
> [   39.628562]  schedule_hrtimeout_range+0x10/0x18
> [   39.633169]  usleep_range_state+0x5c/0x90
> [   39.637253]  __mmc_poll_for_busy+0xec/0x128
> [   39.641514]  mmc_poll_for_busy+0x48/0x70
> [   39.645511]  mmc_do_erase+0x1ec/0x210
> [   39.649237]  mmc_erase+0x1b4/0x1d4
> [   39.652701]  mmc_blk_mq_issue_rq+0x35c/0x6ac
> [   39.657037]  mmc_mq_queue_rq+0x18c/0x214
> [   39.661022]  blk_mq_dispatch_rq_list+0x3a8/0x528
> [   39.665722]  __blk_mq_sched_dispatch_requests+0x3a0/0x4ac
> [   39.671198]  blk_mq_sched_dispatch_requests+0x28/0x5c
> [   39.676322]  blk_mq_run_hw_queue+0x11c/0x12c
> [   39.680668]  blk_mq_flush_plug_list+0x200/0x33c
> [   39.685278]  blk_add_rq_to_plug+0x68/0xd8
> [   39.689365]  blk_mq_submit_bio+0x3a4/0x458
> [   39.693539]  __submit_bio+0x1c/0x80
> [   39.697096]  submit_bio_noacct_nocheck+0x94/0x174
> [   39.701875]  submit_bio_noacct+0x1b0/0x22c
> [   39.706042]  submit_bio+0xac/0xe8
> [   39.709424]  blk_next_bio+0x4c/0x5c
> [   39.712973]  blkdev_issue_secure_erase+0x118/0x170
> [   39.717835]  blkdev_common_ioctl+0x374/0x728
> [   39.722175]  blkdev_ioctl+0x8c/0x2b0
> [   39.725816]  vfs_ioctl+0x24/0x40
> [   39.729117]  __arm64_sys_ioctl+0x5c/0x8c
> [   39.733114]  invoke_syscall+0x68/0xec
> [   39.736839]  el0_svc_common.constprop.0+0x70/0xd8
> [   39.741609]  do_el0_svc+0x18/0x20
> [   39.744981]  el0_svc+0x68/0x94
> [   39.748107]  el0t_64_sync_handler+0x88/0x124
> [   39.752455]  el0t_64_sync+0x168/0x16c
> 
> Fix skips the card->busy() and uses MMC_SEND_STATUS and R1_STATUS
> check for MMC_ERASE_BUSY busy_cmd case in the mmc_busy_cb() function.
> 
> Fixes: 0d84c3e6a5b2 ("mmc: core: Convert to mmc_poll_for_busy() for erase/trim/discard")
> Signed-off-by: Kamal Dasu <kamal.dasu@broadcom.com>

Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>

Thanks!
-- 
Florian