scsi: core: Fix a regression triggered by scsi_host_busy()
Subsystems (2)
Bart Van Assche
reviewerPATCH 1/1 scsi: core: Fix a regression triggered by scsi_host_busy()
Commit 995412e23bb2 ("blk-mq: Replace tags->lock with SRCU for tag iterators") introduced the following regression:
Call trace: __srcu_read_lock+0x30/0x80 (P) blk_mq_tagset_busy_iter+0x44/0x300 scsi_host_busy+0x38/0x70 ufshcd_print_host_state+0x34/0x1bc ufshcd_link_startup.constprop.0+0xe4/0x2e0 ufshcd_init+0x944/0xf80 ufshcd_pltfrm_init+0x504/0x820 ufs_rockchip_probe+0x2c/0x88 platform_probe+0x5c/0xa4 really_probe+0xc0/0x38c __driver_probe_device+0x7c/0x150 driver_probe_device+0x40/0x120 __driver_attach+0xc8/0x1e0 bus_for_each_dev+0x7c/0xdc driver_attach+0x24/0x30 bus_add_driver+0x110/0x230 driver_register+0x68/0x130 __platform_driver_register+0x20/0x2c ufs_rockchip_pltform_init+0x1c/0x28 do_one_initcall+0x60/0x1e0 kernel_init_freeable+0x248/0x2c4 kernel_init+0x20/0x140 ret_from_fork+0x10/0x20
Fix this regression by making scsi_host_busy() check whether the SCSI host tag set has already been initialized. tag_set->ops is set by scsi_mq_setup_tags() just before blk_mq_alloc_tag_set() is called. This fix is based on the assumption that scsi_host_busy() and scsi_mq_setup_tags() calls are serialized. This is the case in the UFS driver.
--- drivers/scsi/hosts.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/scsi/hosts.c +3 -2
@@ -611,8 +611,9 @@ int scsi_host_busy(struct Scsi_Host *shost) {int cnt =0 ; -blk_mq_tagset_busy_iter (&shost->tag_set, - scsi_host_check_in_flight, &cnt); +if (shost->tag_set.ops) +blk_mq_tagset_busy_iter (&shost->tag_set, + scsi_host_check_in_flight, &cnt);return cnt; }EXPORT_SYMBOL (scsi_host_busy);
Quoted 51 lines · Bart Van Assche
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Call trace: __srcu_read_lock+0x30/0x80 (P) blk_mq_tagset_busy_iter+0x44/0x300 scsi_host_busy+0x38/0x70 ufshcd_print_host_state+0x34/0x1bc ufshcd_link_startup.constprop.0+0xe4/0x2e0 ufshcd_init+0x944/0xf80 ufshcd_pltfrm_init+0x504/0x820 ufs_rockchip_probe+0x2c/0x88 platform_probe+0x5c/0xa4 really_probe+0xc0/0x38c __driver_probe_device+0x7c/0x150 driver_probe_device+0x40/0x120 __driver_attach+0xc8/0x1e0 bus_for_each_dev+0x7c/0xdc driver_attach+0x24/0x30 bus_add_driver+0x110/0x230 driver_register+0x68/0x130 __platform_driver_register+0x20/0x2c ufs_rockchip_pltform_init+0x1c/0x28 do_one_initcall+0x60/0x1e0 kernel_init_freeable+0x248/0x2c4 kernel_init+0x20/0x140 ret_from_fork+0x10/0x20Fix this regression by making
scsi_host_busy()check whether the SCSI host tag set has already been initialized.tag_set->opsis set byscsi_mq_setup_tags()just beforeblk_mq_alloc_tag_set()is called. This fix is based on the assumption thatscsi_host_busy()andscsi_mq_setup_tags()calls are serialized. This is the case in the UFS driver.Reported-by Sebastian Reichel <[email protected]>Closes https://lore.kernel.org/linux-block/pnezafputodmqlpumwfbn644ohjybouveehcjhz2hmhtcf2rka@sdhoiivync4y/ open hereCc Ming Lei <[email protected]>Cc Jens Axboe <[email protected]>Signed-off-by Bart Van Assche <[email protected]>--- drivers/scsi/hosts.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/scsi/hosts.c +0
-0@@ -611,8 +611,9 @@ int scsi_host_busy(struct Scsi_Host *shost) {
drivers/scsi/hosts.c int scsi_host_busy(struct Scsi_Host *shost) · line 619
Quoted 10 lines
drivers/scsi/hosts.c +3
-2int cnt =0 ; -blk_mq_tagset_busy_iter (&shost->tag_set, - scsi_host_check_in_flight, &cnt); +if (shost->tag_set.ops) +blk_mq_tagset_busy_iter (&shost->tag_set, + scsi_host_check_in_flight, &cnt);return cnt; }EXPORT_SYMBOL (scsi_host_busy);
Long term, the UFS driver need to be fixed, this or most of scsi core APIs should
have been called after the scsi host is initialized.
Thanks,
Ming
Quoted 3 lines · Ming Lei
Long term, the UFS driver need to be fixed, this or most of scsi core APIs should
have been called after the scsi host is initialized.
Please fix ufs now. Otherwise we'll be stuck with these hacks forever as they never get cleaned up.
Hi,
Quoted 7 lines · Christoph Hellwig
Quoted 3 lines · Ming Lei
Long term, the UFS driver need to be fixed, this or most of scsi core APIs should
have been called after the scsi host is initialized.Please fix ufs now. Otherwise we'll be stuck with these hacks forever as they never get cleaned up.
The Rock 4D has a pluggable UFS module. UFS is described in the system's device-tree, but the affected system itself does not have a module plugged in resulting in UFS erroring out early.
To have an idea - the following is logged at error level because no UFS module is plugged:
[ 2.731721] ufshcd-rockchip 2a2d0000.ufshc: link startup failed 1 [ 2.732311] ufshcd-rockchip 2a2d0000.ufshc: UFS Host state=0 [ 2.732812] ufshcd-rockchip 2a2d0000.ufshc: 0 outstanding reqs, tasks=0x0 [ 2.733410] ufshcd-rockchip 2a2d0000.ufshc: saved_err=0x0, saved_uic_err=0x0 [ 2.734030] ufshcd-rockchip 2a2d0000.ufshc: Device power mode=1, UIC link state=0 [ 2.734687] ufshcd-rockchip 2a2d0000.ufshc: PM in progress=0, sys. suspended=0 [ 2.735320] ufshcd-rockchip 2a2d0000.ufshc: Auto BKOPS=0, Host self-block=0 [ 2.735931] ufshcd-rockchip 2a2d0000.ufshc: Clk gate=0 [ 2.736396] ufshcd-rockchip 2a2d0000.ufshc: last_hibern8_exit_tstamp at 0 us, hibern8_exit_cnt=0 [ 2.737166] ufshcd-rockchip 2a2d0000.ufshc: error handling flags=0x0, req. abort count=0 [ 2.737875] ufshcd-rockchip 2a2d0000.ufshc: hba->ufs_version=0x200, Host capabilities=0x187011f, caps=0x48c [ 2.738727] ufshcd-rockchip 2a2d0000.ufshc: quirks=0x2000, dev. quirks=0x0 [ 2.739349] host_regs: 00000000: 0187011f 00000000 00000200 00000000 [ 2.739910] host_regs: 00000010: 00000000 000005e6 00000000 00000000 [ 2.740480] host_regs: 00000020: 00000000 00000470 00000000 00000000 [ 2.741039] host_regs: 00000030: 00000008 00000001 00000000 00000000 [ 2.741598] host_regs: 00000040: 00000000 00000000 00000000 00000000 [ 2.742156] host_regs: 00000050: 00000000 00000000 00000000 00000000 [ 2.742715] host_regs: 00000060: 00000000 00000000 00000000 00000000 [ 2.743273] host_regs: 00000070: 00000000 00000000 00000000 00000000 [ 2.743831] host_regs: 00000080: 00000000 00000000 00000000 00000000 [ 2.744398] host_regs: 00000090: 00000016 00000000 00000001 00000000 [ 2.744959] ufshcd-rockchip 2a2d0000.ufshc: No record of pa_err [ 2.745480] ufshcd-rockchip 2a2d0000.ufshc: No record of dl_err [ 2.746000] ufshcd-rockchip 2a2d0000.ufshc: No record of nl_err [ 2.746520] ufshcd-rockchip 2a2d0000.ufshc: No record of tl_err [ 2.747040] ufshcd-rockchip 2a2d0000.ufshc: No record of dme_err [ 2.747568] ufshcd-rockchip 2a2d0000.ufshc: No record of auto_hibern8_err [ 2.748174] ufshcd-rockchip 2a2d0000.ufshc: No record of fatal_err [ 2.748716] ufshcd-rockchip 2a2d0000.ufshc: link_startup_fail[0] = 0x1 at 2731713 us [ 2.749398] ufshcd-rockchip 2a2d0000.ufshc: link_startup_fail: total cnt=1 [ 2.750001] ufshcd-rockchip 2a2d0000.ufshc: No record of resume_fail [ 2.750559] ufshcd-rockchip 2a2d0000.ufshc: No record of suspend_fail [ 2.751125] ufshcd-rockchip 2a2d0000.ufshc: No record of wlun resume_fail [ 2.751721] ufshcd-rockchip 2a2d0000.ufshc: No record of wlun suspend_fail [ 2.752332] ufshcd-rockchip 2a2d0000.ufshc: dev_reset[0] = 0x0 at 2297266 us [ 2.752953] ufshcd-rockchip 2a2d0000.ufshc: dev_reset: total cnt=1 [ 2.753497] ufshcd-rockchip 2a2d0000.ufshc: No record of host_reset [ 2.754047] ufshcd-rockchip 2a2d0000.ufshc: No record of task_abort [ 2.755475] ufshcd-rockchip 2a2d0000.ufshc: error 000000000dbbd97c: Initialization failed with error 1 [ 2.756358] ufshcd-rockchip 2a2d0000.ufshc: error 000000000dbbd97c: ufshcd_pltfrm_init failed [ 2.757115] ufshcd-rockchip 2a2d0000.ufshc: probe with driver ufshcd-rockchip failed with error 1
I think the best solution would be to avoid this huge error print in the first place for a completley missing device, but I haven't checked the details.
Greetings,
-- Sebastian
Bart Van Assche
reviewerQuoted 50 lines · Sebastian Reichel
The Rock 4D has a pluggable UFS module. UFS is described in the system's device-tree, but the affected system itself does not have a module plugged in resulting in UFS erroring out early.
To have an idea - the following is logged at error level because no UFS module is plugged:
[ 2.731721] ufshcd-rockchip 2a2d0000.ufshc: link startup failed 1 [ 2.732311] ufshcd-rockchip 2a2d0000.ufshc: UFS Host state=0 [ 2.732812] ufshcd-rockchip 2a2d0000.ufshc: 0 outstanding reqs, tasks=0x0 [ 2.733410] ufshcd-rockchip 2a2d0000.ufshc: saved_err=0x0, saved_uic_err=0x0 [ 2.734030] ufshcd-rockchip 2a2d0000.ufshc: Device power mode=1, UIC link state=0 [ 2.734687] ufshcd-rockchip 2a2d0000.ufshc: PM in progress=0, sys. suspended=0 [ 2.735320] ufshcd-rockchip 2a2d0000.ufshc: Auto BKOPS=0, Host self-block=0 [ 2.735931] ufshcd-rockchip 2a2d0000.ufshc: Clk gate=0 [ 2.736396] ufshcd-rockchip 2a2d0000.ufshc: last_hibern8_exit_tstamp at 0 us, hibern8_exit_cnt=0 [ 2.737166] ufshcd-rockchip 2a2d0000.ufshc: error handling flags=0x0, req. abort count=0 [ 2.737875] ufshcd-rockchip 2a2d0000.ufshc: hba->ufs_version=0x200, Host capabilities=0x187011f, caps=0x48c [ 2.738727] ufshcd-rockchip 2a2d0000.ufshc: quirks=0x2000, dev. quirks=0x0 [ 2.739349] host_regs: 00000000: 0187011f 00000000 00000200 00000000 [ 2.739910] host_regs: 00000010: 00000000 000005e6 00000000 00000000 [ 2.740480] host_regs: 00000020: 00000000 00000470 00000000 00000000 [ 2.741039] host_regs: 00000030: 00000008 00000001 00000000 00000000 [ 2.741598] host_regs: 00000040: 00000000 00000000 00000000 00000000 [ 2.742156] host_regs: 00000050: 00000000 00000000 00000000 00000000 [ 2.742715] host_regs: 00000060: 00000000 00000000 00000000 00000000 [ 2.743273] host_regs: 00000070: 00000000 00000000 00000000 00000000 [ 2.743831] host_regs: 00000080: 00000000 00000000 00000000 00000000 [ 2.744398] host_regs: 00000090: 00000016 00000000 00000001 00000000 [ 2.744959] ufshcd-rockchip 2a2d0000.ufshc: No record of pa_err [ 2.745480] ufshcd-rockchip 2a2d0000.ufshc: No record of dl_err [ 2.746000] ufshcd-rockchip 2a2d0000.ufshc: No record of nl_err [ 2.746520] ufshcd-rockchip 2a2d0000.ufshc: No record of tl_err [ 2.747040] ufshcd-rockchip 2a2d0000.ufshc: No record of dme_err [ 2.747568] ufshcd-rockchip 2a2d0000.ufshc: No record of auto_hibern8_err [ 2.748174] ufshcd-rockchip 2a2d0000.ufshc: No record of fatal_err [ 2.748716] ufshcd-rockchip 2a2d0000.ufshc: link_startup_fail[0] = 0x1 at 2731713 us [ 2.749398] ufshcd-rockchip 2a2d0000.ufshc: link_startup_fail: total cnt=1 [ 2.750001] ufshcd-rockchip 2a2d0000.ufshc: No record of resume_fail [ 2.750559] ufshcd-rockchip 2a2d0000.ufshc: No record of suspend_fail [ 2.751125] ufshcd-rockchip 2a2d0000.ufshc: No record of wlun resume_fail [ 2.751721] ufshcd-rockchip 2a2d0000.ufshc: No record of wlun suspend_fail [ 2.752332] ufshcd-rockchip 2a2d0000.ufshc: dev_reset[0] = 0x0 at 2297266 us [ 2.752953] ufshcd-rockchip 2a2d0000.ufshc: dev_reset: total cnt=1 [ 2.753497] ufshcd-rockchip 2a2d0000.ufshc: No record of host_reset [ 2.754047] ufshcd-rockchip 2a2d0000.ufshc: No record of task_abort [ 2.755475] ufshcd-rockchip 2a2d0000.ufshc: error 000000000dbbd97c: Initialization failed with error 1 [ 2.756358] ufshcd-rockchip 2a2d0000.ufshc: error 000000000dbbd97c: ufshcd_pltfrm_init failed [ 2.757115] ufshcd-rockchip 2a2d0000.ufshc: probe with driver ufshcd-rockchip failed with error 1
Quoted 3 lines
I think the best solution would be to avoid this huge error print in the first place for a completley missing device, but I haven't checked the details.
How about reducing link startup logging with the patch below?
Thanks,
Bart.
Subject: [PATCH] ufs: core: Reduce link startup failure logging
Some systems, e.g. Rock 4D, have a pluggable UFS module. Link startup fails systematically on these systems. If no UFS module has been plugged in, more than fourty lines are logged after the "link startup failed" message. Avoid this by reducing link startup failure logging.
An intended side effect of this patch is that scsi_host_busy() is not called before scsi_add_host() is called.
--- drivers/ufs/core/ufshcd.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/ufs/core/ufshcd.c +1 -5
@@ -5117,12 +5117,8 @@ static int ufshcd_link_startup(struct ufs_hba *hba)ufshcd_readl (hba,REG_UIC_ERROR_CODE_PHY_ADAPTER_LAYER ); ret =ufshcd_make_hba_operational (hba); out: -if (ret) { +if (ret)dev_err (hba->dev,"link startup failed %d\n" , ret); -ufshcd_print_host_state (hba); -ufshcd_print_pwr_info (hba); -ufshcd_print_evt_hist (hba); - }return ret; }
Bart Van Assche
reviewerQuoted 7 lines · Christoph Hellwig
Quoted 3 lines · Ming Lei
Long term, the UFS driver need to be fixed, this or most of scsi core APIs should
have been called after the scsi host is initialized.Please fix ufs now. Otherwise we'll be stuck with these hacks forever as they never get cleaned up.
Although it is easy to make sure that the UFS driver doesn't callscsi_host_busy() before the SCSI host is added, this is not just a UFS
driver issue. The advansys driver calls scsi_host_busy() as follows before scsi_add_host() is called:
advansys_board_found()
ASC_DBG_PRT_SCSI_HOST()
scsi_host_busy()
Thanks,
Bart.
Quoted 7 lines · Bart Van Assche
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Call trace: __srcu_read_lock+0x30/0x80 (P) blk_mq_tagset_busy_iter+0x44/0x300 scsi_host_busy+0x38/0x70
this value is readable from sysfs, so why even print it?
Quoted 2 lines
ufshcd_print_host_state+0x34/0x1bc ufshcd_link_startup.constprop.0+0xe4/0x2e0
Bart Van Assche
reviewerQuoted 10 lines · John Garry
Quoted 7 lines · Bart Van Assche
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Call trace: __srcu_read_lock+0x30/0x80 (P) blk_mq_tagset_busy_iter+0x44/0x300 scsi_host_busy+0x38/0x70this value is readable from sysfs, so why even print it?
That's an interesting question. The infrastructure for reporting Android bugs includes the kernel logs in a bug report but not the values of sysfs attributes. Hence, if something unexpected happens, having state information available in the kernel log helps more than having it available in sysfs. That being said, I think that the UFS kernel driver reports more information than it should when an unexpected condition is met, e.g. a link startup failure or starting the UFS error handler.
Bart.
Hi,
Quoted 35 lines · Bart Van Assche
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Call trace: __srcu_read_lock+0x30/0x80 (P) blk_mq_tagset_busy_iter+0x44/0x300 scsi_host_busy+0x38/0x70 ufshcd_print_host_state+0x34/0x1bc ufshcd_link_startup.constprop.0+0xe4/0x2e0 ufshcd_init+0x944/0xf80 ufshcd_pltfrm_init+0x504/0x820 ufs_rockchip_probe+0x2c/0x88 platform_probe+0x5c/0xa4 really_probe+0xc0/0x38c __driver_probe_device+0x7c/0x150 driver_probe_device+0x40/0x120 __driver_attach+0xc8/0x1e0 bus_for_each_dev+0x7c/0xdc driver_attach+0x24/0x30 bus_add_driver+0x110/0x230 driver_register+0x68/0x130 __platform_driver_register+0x20/0x2c ufs_rockchip_pltform_init+0x1c/0x28 do_one_initcall+0x60/0x1e0 kernel_init_freeable+0x248/0x2c4 kernel_init+0x20/0x140 ret_from_fork+0x10/0x20Fix this regression by making
scsi_host_busy()check whether the SCSI host tag set has already been initialized.tag_set->opsis set byscsi_mq_setup_tags()just beforeblk_mq_alloc_tag_set()is called. This fix is based on the assumption thatscsi_host_busy()andscsi_mq_setup_tags()calls are serialized. This is the case in the UFS driver.
Quoted 6 lines
Reported-by Sebastian Reichel <[email protected]>Closes https://lore.kernel.org/linux-block/pnezafputodmqlpumwfbn644ohjybouveehcjhz2hmhtcf2rka@sdhoiivync4y/ open hereCc Ming Lei <[email protected]>Cc Jens Axboe <[email protected]>Signed-off-by Bart Van Assche <[email protected]>---
-- Sebastian
Quoted 19 lines
drivers/scsi/hosts.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/scsi/hosts.c +3
-2@@ -611,8 +611,9 @@ int scsi_host_busy(struct Scsi_Host *shost) {int cnt =0 ; -blk_mq_tagset_busy_iter (&shost->tag_set, - scsi_host_check_in_flight, &cnt); +if (shost->tag_set.ops) +blk_mq_tagset_busy_iter (&shost->tag_set, + scsi_host_check_in_flight, &cnt);return cnt; }EXPORT_SYMBOL (scsi_host_busy);
Martin K. Petersen
AppliedBart,
Quoted 2 lines
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:
Applied to 6.18/scsi-fixes, thanks!
Signature · 1 line
Martin K. Petersen
Quoted 8 lines · Martin K. Petersen
Bart,
Quoted 2 lines
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Applied to 6.18/scsi-fixes, thanks!
I don't think that we should call scsi_host_busy() on a shost which has not been added, so it would be nice to have a plan to fix the LLDs also.
Bart Van Assche
reviewerQuoted 7 lines · John Garry
Quoted 4 lines · Martin K. Petersen
Quoted 2 lines
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Applied to 6.18/scsi-fixes, thanks!
I don't think that we should call
scsi_host_busy()on a shost which has not been added, so it would be nice to have a plan to fix the LLDs also.
A fix for the UFS driver has already been merged by Martin. See also "scsi: ufs: core: Reduce link startup failure logging".
Thanks,
Bart.
Quoted 7 lines · Bart Van Assche
Quoted 4 lines
Quoted 2 lines
Applied to 6.18/scsi-fixes, thanks!
I don't think that we should call
scsi_host_busy()on a shost which has not been added, so it would be nice to have a plan to fix the LLDs also.A fix for the UFS driver has already been merged by Martin. See also "scsi: ufs: core: Reduce link startup failure logging".
Good, so do you know which other drivers need to be fixed? I saw the advansys driver mentioned. I can look at sending a fix for that.
Bart Van Assche
reviewerQuoted 1 line · John Garry
Good, so do you know which other drivers need to be fixed? I saw the advansys driver mentioned. I can look at sending a fix for that.
The advansys driver is the only driver of which I could determine quickly that it needs to be fixed. There may be other drivers that need to be fixed but I'm not sure of this.
Bart.
Quoted 11 lines · Bart Van Assche
Quoted 7 lines · John Garry
Quoted 4 lines · Martin K. Petersen
Quoted 2 lines
Commit
995412e23bb2("blk-mq: Replacetags->lockwith SRCU for tag iterators") introduced the following regression:Applied to 6.18/scsi-fixes, thanks!
I don't think that we should call
scsi_host_busy()on a shost which has not been added, so it would be nice to have a plan to fix the LLDs also.A fix for the UFS driver has already been merged by Martin. See also "scsi: ufs: core: Reduce link startup failure logging".
Hi Bart,
I am auditing all the other scsi_host_busy() callsites to see if it is possible to backout the change in this patch (which checks tagset ops is non-NULL before calling the tagset iter function).
Can you confirm that all paths in the ufs driver are safe, as there are still some references and they involve workqueues (so not so callchains are not so obvious)?
So far it seems that only scsi_debug may have a hidden problem, but I am not overly concerned with that.
John
Bart Van Assche
reviewerQuoted 5 lines · John Garry
I am auditing all the other
scsi_host_busy()callsites to see if it is possible to backout the change in this patch (which checks tagset ops is non-NULL before calling the tagset iter function).Can you confirm that all paths in the ufs driver are safe, as there are still some references and they involve workqueues (so not so callchains are not so obvious)?
So far it seems that only
scsi_debugmay have a hidden problem, but I am not overly concerned with that.
Hi John,
The patch series "Optimize the hot path in the UFS driver" makes the scsi_add_host() call happen before most scsi_host_busy() calls. However, a few UIC commands are sent before scsi_add_host() is called and these may trigger the UFS error handler. The UFS error handler may trigger a call to ufshcd_print_host_state(). Something like this patch may be required (I will post this as an official patch after having tested it):
drivers/ufs/core/ufshcd.c +2 -1
@@ -678,7 +678,8 @@ static void ufshcd_print_host_state(struct ufs_hba *hba)dev_err (hba->dev,"UFS Host state=%d\n" , hba->ufshcd_state);dev_err (hba->dev,"%d outstanding reqs, tasks=0x%lx\n" , -scsi_host_busy (hba->host), hba->outstanding_tasks); + hba->scsi_host_added ?scsi_host_busy (hba->host) :0 , + hba->outstanding_tasks);dev_err (hba->dev,"saved_err=0x%x, saved_uic_err=0x%x\n" , hba->saved_err, hba->saved_uic_err); dev_err(hba->dev, "Device power mode=%d, UIC link state=%d\n",
Bart.
Quoted 10 lines · Bart Van Assche
Quoted 1 line
So far it seems that only
scsi_debugmay have a hidden problem, but I am not overly concerned with that.Hi John,
The patch series "Optimize the hot path in the UFS driver" makes the
scsi_add_host()call happen before mostscsi_host_busy()calls. However, a few UIC commands are sent beforescsi_add_host()is called and these may trigger the UFS error handler. The UFS error handler may trigger a call toufshcd_print_host_state(). Something like this patch may be required (I will post this as an official patch after having tested it):
Hi Bart,
Any update on this?
Thanks!
Quoted 16 lines
drivers/ufs/core/ufshcd.c +2
-1@@ -678,7 +678,8 @@ static void ufshcd_print_host_state(struct ufs_hba *hba) dev_err(hba->dev, "UFS Host state=%d\n", hba->ufshcd_state); dev_err(hba->dev, "%d outstanding reqs, tasks=0x%lx\n", - scsi_host_busy(hba->host), hba->outstanding_tasks); + hba->scsi_host_added ? scsi_host_busy(hba->host) : 0, + hba->outstanding_tasks); dev_err(hba->dev, "saved_err=0x%x, saved_uic_err=0x%x\n", hba->saved_err, hba->saved_uic_err);dev_err(hba->dev, "Device power mode=%d, UIC link state=%d\n",
Bart Van Assche
reviewerQuoted 1 line · John Garry
Any update on this?
Hi John,
Thanks for the reminder. I have been OoO during the past two weeks and that's why the promised work has not yet resulted in the publication of a new patch series. Anyway, I plan to do that this week.
Bart.