Two UFS bug fixes
v3 cover Two UFS bug fixes
Change log from v2:
- fix build warning
PATCH v3 1/2 scsi: ufs: fix livelock of ufshcd_clear_ua_wluns
When gate_work/ungate_work gets an error during hibern8_enter or exit,
ufshcd_err_handler()
ufshcd_scsi_block_requests()
ufshcd_reset_and_restore()
ufshcd_clear_ua_wluns() -> stuck
ufshcd_scsi_unblock_requests()
In order to avoid it, ufshcd_clear_ua_wluns() can be called per recovery flows such as suspend/resume, link_recovery, and error_handler.
--- drivers/scsi/ufs/ufshcd.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
drivers/scsi/ufs/ufshcd.c +10 -5
@@ -3996,6 +3996,8 @@ int ufshcd_link_recovery(struct ufs_hba *hba)if (ret)dev_err (hba->dev,"%s: link recovery failed, err %d" , __func__, ret); +else +ufshcd_clear_ua_wluns (hba);return ret; }
@@ -6003,6 +6005,9 @@ static void ufshcd_err_handler(struct work_struct *work)ufshcd_scsi_unblock_requests (hba);ufshcd_err_handling_unprepare (hba);up (&hba->eh_sem); + +if (!err && needs_reset) +ufshcd_clear_ua_wluns (hba); }/**
@@ -6940,14 +6945,11 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)ufshcd_set_clk_freq (hba, true); err =ufshcd_hba_enable (hba); -if (err) - goto out;/* Establish the link again and restore the device */ - err =ufshcd_probe_hba (hba, false);if (!err) -ufshcd_clear_ua_wluns (hba); -out: + err =ufshcd_probe_hba (hba, false); +if (err)dev_err (hba->dev,"%s: Host init failed %d\n" , __func__, err);ufshcd_update_evt_hist (hba,UFS_EVT_HOST_RESET , (u32 )err);
@@ -8777,6 +8779,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)ufshcd_resume_clkscaling (hba); hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false; +ufshcd_clear_ua_wluns (hba);ufshcd_release (hba); out:if (hba->dev_info.b_rpm_dev_flush_capable) {
@@ -8887,6 +8890,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op)cancel_delayed_work (&hba->rpm_dev_flush_recheck_work); } +ufshcd_clear_ua_wluns (hba); +/* Schedule clock gating in case of no access to UFS device yet */ ufshcd_release (hba);
Signature · 1 line
2.29.2.729.g45daf8777d-goog
Re: [PATCH v3 1/2] scsi: ufs: fix livelock of ufshcd_clear_ua_wluns
Quoted 21 lines · Jaegeuk Kim
When gate_work/ungate_work gets an error during
hibern8_enteror exit,
ufshcd_err_handler()
ufshcd_scsi_block_requests()
ufshcd_reset_and_restore()ufshcd_clear_ua_wluns() -> stuck
ufshcd_scsi_unblock_requests()In order to avoid it,
ufshcd_clear_ua_wluns()can be called per recovery flows such as suspend/resume,link_recovery, anderror_handler.Fixes 1918651f2d7e ("scsi: ufs: Clear UAC for RPMB after ufshcd resets")Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)drivers/scsi/ufs/ufshcd.c +0
-0
drivers/scsi/ufs/ufshcd.c int ufshcd_link_recovery(struct ufs_hba *hba) · line 4000
Quoted 5 lines
drivers/scsi/ufs/ufshcd.c +2
-0@@ -3996,6 +3996,8 @@ int ufshcd_link_recovery(struct ufs_hba *hba)if (ret)dev_err (hba->dev,"%s: link recovery failed, err %d" , __func__, ret); +else +ufshcd_clear_ua_wluns (hba);
Can we put it right after ufshcd_scsi_add_wlus() in ufshcd_add_lus()?
Thanks,
Can Guo.
Quoted 50 lines
drivers/scsi/ufs/ufshcd.c +3
-0return ret; }@@ -6003,6 +6005,9 @@ static void ufshcd_err_handler(struct work_struct *work)ufshcd_scsi_unblock_requests (hba);ufshcd_err_handling_unprepare (hba);up (&hba->eh_sem); + +if (!err && needs_reset) +ufshcd_clear_ua_wluns (hba); }/**
drivers/scsi/ufs/ufshcd.c +2
-5@@ -6940,14 +6945,11 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)ufshcd_set_clk_freq (hba, true); err =ufshcd_hba_enable (hba); -if (err) - goto out;/* Establish the link again and restore the device */ - err =ufshcd_probe_hba (hba, false);if (!err) -ufshcd_clear_ua_wluns (hba); -out: + err =ufshcd_probe_hba (hba, false); +if (err)dev_err (hba->dev,"%s: Host init failed %d\n" , __func__, err);drivers/scsi/ufs/ufshcd.c +1
-0ufshcd_update_evt_hist (hba,UFS_EVT_HOST_RESET , (u32 )err);@@ -8777,6 +8779,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)ufshcd_resume_clkscaling (hba ); hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false; +ufshcd_clear_ua_wluns (hba);ufshcd_release (hba); out:drivers/scsi/ufs/ufshcd.c +2
-0if (hba->dev_info.b_rpm_dev_flush_capable) {@@ -8887,6 +8890,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) cancel_delayed_work(&hba->rpm_dev_flush_recheck_work); } +ufshcd_clear_ua_wluns (hba); +/* Schedule clock gating in case of no access to UFS device yet */ ufshcd_release (hba);
Re: [PATCH v3 1/2] scsi: ufs: fix livelock of ufshcd_clear_ua_wluns
Quoted 19 lines · Can Guo
Quoted 18 lines · Jaegeuk Kim
When gate_work/ungate_work gets an error during
hibern8_enteror exit,
ufshcd_err_handler()
ufshcd_scsi_block_requests()
ufshcd_reset_and_restore()ufshcd_clear_ua_wluns() -> stuck
ufshcd_scsi_unblock_requests()In order to avoid it,
ufshcd_clear_ua_wluns()can be called per recovery flows
such as suspend/resume,link_recovery, anderror_handler.Fixes 1918651f2d7e ("scsi: ufs: Clear UAC for RPMB after ufshcdresets")
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)
Quoted 12 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_link_recovery(struct ufs_hba *hba) · line 4000
Quoted 10 lines
drivers/scsi/ufs/ufshcd.c +2
-0@@ -3996,6 +3996,8 @@ int ufshcd_link_recovery(struct ufs_hba *hba)if (ret)dev_err (hba->dev,"%s: link recovery failed, err %d" , __func__, ret); +else +ufshcd_clear_ua_wluns (hba);Can we put it right after
ufshcd_scsi_add_wlus()inufshcd_add_lus()?
May I ask the reason? We'll call it after ufshcd_add_lus() later tho.
Quoted 54 lines
Thanks,
Can Guo.Quoted 50 lines
drivers/scsi/ufs/ufshcd.c +3
-0return ret; }@@ -6003,6 +6005,9 @@ static void ufshcd_err_handler(struct work_struct *work)ufshcd_scsi_unblock_requests (hba);ufshcd_err_handling_unprepare (hba);up (&hba->eh_sem); + +if (!err && needs_reset) +ufshcd_clear_ua_wluns (hba); }/**
drivers/scsi/ufs/ufshcd.c +2
-5@@ -6940,14 +6945,11 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)ufshcd_set_clk_freq (hba, true); err =ufshcd_hba_enable (hba); -if (err) - goto out;/* Establish the link again and restore the device */ - err =ufshcd_probe_hba (hba, false);if (!err) -ufshcd_clear_ua_wluns (hba); -out: + err =ufshcd_probe_hba (hba, false); +if (err)dev_err (hba->dev,"%s: Host init failed %d\n" , __func__, err);drivers/scsi/ufs/ufshcd.c +1
-0ufshcd_update_evt_hist (hba,UFS_EVT_HOST_RESET , (u32 )err);@@ -8777,6 +8779,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)ufshcd_resume_clkscaling (hba ); hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false; +ufshcd_clear_ua_wluns (hba);ufshcd_release (hba); out:drivers/scsi/ufs/ufshcd.c +2
-0if (hba->dev_info.b_rpm_dev_flush_capable) {@@ -8887,6 +8890,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) cancel_delayed_work(&hba->rpm_dev_flush_recheck_work); } +ufshcd_clear_ua_wluns (hba); +/* Schedule clock gating in case of no access to UFS device yet */ ufshcd_release (hba);
Re: [PATCH v3 1/2] scsi: ufs: fix livelock of ufshcd_clear_ua_wluns
Quoted 33 lines · Jaegeuk Kim
Quoted 19 lines · Can Guo
Quoted 18 lines · Jaegeuk Kim
When gate_work/ungate_work gets an error during
hibern8_enteror exit,
ufshcd_err_handler()
ufshcd_scsi_block_requests()
ufshcd_reset_and_restore()ufshcd_clear_ua_wluns() -> stuck
ufshcd_scsi_unblock_requests()In order to avoid it,
ufshcd_clear_ua_wluns()can be called per recovery flows
such as suspend/resume,link_recovery, anderror_handler.Fixes 1918651f2d7e ("scsi: ufs: Clear UAC for RPMB after ufshcdresets")
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)Quoted 12 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_link_recovery(struct ufs_hba *hba) · line 4000
Quoted 10 lines
drivers/scsi/ufs/ufshcd.c +2
-0@@ -3996,6 +3996,8 @@ int ufshcd_link_recovery(struct ufs_hba *hba)if (ret)dev_err (hba->dev,"%s: link recovery failed, err %d" , __func__, ret); +else +ufshcd_clear_ua_wluns (hba);Can we put it right after
ufshcd_scsi_add_wlus()inufshcd_add_lus()?
Quoted 2 lines
May I ask the reason? We'll call it after
ufshcd_add_lus()later tho.
I think the code will be more readable - we do all the LU related stuffs in one func, just nit-picking though. I found this because I am planning to move the devfreq init codes out of ufshcd_add_lus() due to it is inappropriate to init devfreq in there by its naming, but it might be a good place for ufshcd_clear_ua_wluns().
Thanks,
Can Guo.
Quoted 54 lines
Quoted 54 lines
Thanks,
Can Guo.Quoted 50 lines
drivers/scsi/ufs/ufshcd.c +3
-0return ret; }@@ -6003,6 +6005,9 @@ static void ufshcd_err_handler(struct work_struct *work)ufshcd_scsi_unblock_requests (hba);ufshcd_err_handling_unprepare (hba);up (&hba->eh_sem); + +if (!err && needs_reset) +ufshcd_clear_ua_wluns (hba); }/**
drivers/scsi/ufs/ufshcd.c +2
-5@@ -6940,14 +6945,11 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)ufshcd_set_clk_freq (hba, true); err =ufshcd_hba_enable (hba); -if (err) - goto out;/* Establish the link again and restore the device */ - err =ufshcd_probe_hba (hba, false);if (!err) -ufshcd_clear_ua_wluns (hba); -out: + err =ufshcd_probe_hba (hba, false); +if (err)dev_err (hba->dev,"%s: Host init failed %d\n" , __func__, err);drivers/scsi/ufs/ufshcd.c +1
-0ufshcd_update_evt_hist (hba,UFS_EVT_HOST_RESET , (u32 )err);@@ -8777,6 +8779,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)ufshcd_resume_clkscaling (hba ); hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false; +ufshcd_clear_ua_wluns (hba);ufshcd_release (hba); out:drivers/scsi/ufs/ufshcd.c +2
-0if (hba->dev_info.b_rpm_dev_flush_capable) {@@ -8887,6 +8890,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) cancel_delayed_work(&hba->rpm_dev_flush_recheck_work); } +ufshcd_clear_ua_wluns (hba); +/* Schedule clock gating in case of no access to UFS device yet */ ufshcd_release (hba);
Re: [PATCH v3 1/2] scsi: ufs: fix livelock of ufshcd_clear_ua_wluns
Quoted 34 lines · Can Guo
Quoted 33 lines · Jaegeuk Kim
Quoted 31 lines · Can Guo
Quoted 23 lines · Jaegeuk Kim
When gate_work/ungate_work gets an error during
hibern8_enteror exit,
ufshcd_err_handler()
ufshcd_scsi_block_requests()
ufshcd_reset_and_restore()ufshcd_clear_ua_wluns() -> stuck
ufshcd_scsi_unblock_requests()In order to avoid it,
ufshcd_clear_ua_wluns()can be called per recovery flows
such as suspend/resume,link_recovery, anderror_handler.Fixes 1918651f2d7e ("scsi: ufs: Clear UAC for RPMB after ufshcdresets")
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-)drivers/scsi/ufs/ufshcd.c +0
-0drivers/scsi/ufs/ufshcd.c int ufshcd_link_recovery(struct ufs_hba *hba) · line 4000
Quoted 5 lines
drivers/scsi/ufs/ufshcd.c +2
-0@@ -3996,6 +3996,8 @@ int ufshcd_link_recovery(struct ufs_hba *hba)if (ret)dev_err (hba->dev,"%s: link recovery failed, err %d" , __func__, ret); +else +ufshcd_clear_ua_wluns (hba);Can we put it right after
ufshcd_scsi_add_wlus()inufshcd_add_lus()?
Quoted 8 lines
Quoted 2 lines
May I ask the reason? We'll call it after
ufshcd_add_lus()later tho.I think the code will be more readable - we do all the LU related stuffs in one func, just nit-picking though. I found this because I am planning to move the devfreq init codes out of
ufshcd_add_lus()due to it is inappropriate to init devfreq in there by its naming, but it might be a good place forufshcd_clear_ua_wluns().
Ok, that looks good to me. Thanks.
Quoted 58 lines
Thanks,
Can Guo.Quoted 54 lines
Quoted 54 lines
Thanks,
Can Guo.Quoted 50 lines
drivers/scsi/ufs/ufshcd.c +3
-0return ret; }@@ -6003,6 +6005,9 @@ static void ufshcd_err_handler(struct work_struct *work)ufshcd_scsi_unblock_requests (hba);ufshcd_err_handling_unprepare (hba);up (&hba->eh_sem); + +if (!err && needs_reset) +ufshcd_clear_ua_wluns (hba); }/**
drivers/scsi/ufs/ufshcd.c +2
-5@@ -6940,14 +6945,11 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)ufshcd_set_clk_freq (hba, true); err =ufshcd_hba_enable (hba); -if (err) - goto out;/* Establish the link again and restore the device */ - err =ufshcd_probe_hba (hba, false);if (!err) -ufshcd_clear_ua_wluns (hba); -out: + err =ufshcd_probe_hba (hba, false); +if (err)dev_err (hba->dev,"%s: Host init failed %d\n" , __func__, err);drivers/scsi/ufs/ufshcd.c +1
-0ufshcd_update_evt_hist (hba,UFS_EVT_HOST_RESET , (u32 )err);@@ -8777,6 +8779,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op)ufshcd_resume_clkscaling (hba ); hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false; +ufshcd_clear_ua_wluns (hba);ufshcd_release (hba); out:drivers/scsi/ufs/ufshcd.c +2
-0if (hba->dev_info.b_rpm_dev_flush_capable) {@@ -8887,6 +8890,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op) cancel_delayed_work(&hba->rpm_dev_flush_recheck_work); } +ufshcd_clear_ua_wluns (hba); +/* Schedule clock gating in case of no access to UFS device yet */ ufshcd_release (hba);
PATCH v3 2/2 scsi: ufs: handle LINERESET with correct tm_cmd
In reply to Jaegeuk KimFrom: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in __ufshcd_issue_tm_cmd.
WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46c
And, in ufshcd_err_handler(), we can avoid to send tm_cmd before aborting outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd 0x80 timed-out
--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)
drivers/scsi/ufs/ufshcd.c +32 -4
@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */
@@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct *work)* check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err) hba->saved_err &= ~UIC_ERROR ;
@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling; }
@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { +dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n" , + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " ); }
@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba,* Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT ); +if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);
@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) hba->tmf_tag_set = (struct blk_mq_tag_set ) { .nr_hw_queues =1 , + .reserved_tags =1 , .queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops, .flags =BLK_MQ_F_NO_SCHED ,
Signature · 1 line
2.29.2.729.g45daf8777d-goog
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
Hi Jaegeuk,
Quoted 32 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cAnd, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outSigned-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +3
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */
drivers/scsi/ufs/ufshcd.c static void ufshcd_err_handler(struct work_struct · line 5906
Quoted 5 lines
drivers/scsi/ufs/ufshcd.c +1
-0@@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get ();
I don't see the connection btw line-reset and following tmf cmd. My point is that line-reset is not the only non-fatal error which leads us to the following tmf cmd. So the wait should be outside of this check - just put it right before clearing outstanding reqs.
Thanks,
Can Guo.
Quoted 68 lines
drivers/scsi/ufs/ufshcd.c +1
-0+ hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work) if (ufshcd_is_pwr_mode_restore_needed(hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );drivers/scsi/ufs/ufshcd.c +5
-1}@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT ); +if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +1
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) hba->tmf_tag_set = (struct blk_mq_tag_set) { .nr_hw_queues =1 , + .reserved_tags =1 , .queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
Quoted 37 lines · Can Guo
Hi Jaegeuk,
Quoted 34 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cAnd, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting
outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outSigned-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +3
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */
Quoted 11 lines
drivers/scsi/ufs/ufshcd.c static void ufshcd_err_handler(struct work_struct · line 5906
Quoted 6 lines
drivers/scsi/ufs/ufshcd.c +1
-0@@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get ();I don't see the connection btw line-reset and following tmf cmd. My point is that line-reset is not the only non-fatal error which leads us to the following tmf cmd. So the wait should be outside of this check - just put it right before clearing outstanding reqs.
Ok. Let me move it in v4.
Quoted 72 lines
Thanks,
Can Guo.Quoted 68 lines
drivers/scsi/ufs/ufshcd.c +1
-0+ hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );drivers/scsi/ufs/ufshcd.c +5
-1}@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT ); +if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +1
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) hba->tmf_tag_set = (struct blk_mq_tag_set) { .nr_hw_queues =1 , + .reserved_tags =1 , .queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
In reply to Jaegeuk KimQuoted 4 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.
Quoted 8 lines
WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cAnd, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-out
Would you mind add a Fixes tag?
Quoted 71 lines
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +5
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */ @@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work) if (ufshcd_is_pwr_mode_restore_needed(hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );drivers/scsi/ufs/ufshcd.c +0
-0}
drivers/scsi/ufs/ufshcd.c static int __ufshcd_issue_tm_cmd(struct ufs_hba · line 6376
Quoted 7 lines
drivers/scsi/ufs/ufshcd.c +2
-1@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT );
Sorry that I didn't pay much attention to this part of code before. May I know why must we use the BLK_MQ_REQ_RESERVED flag?
Thanks,
Can Guo.
drivers/scsi/ufs/ufshcd.c int ufshcd_init(struct ufs_hba *hba, void · line 9386
Quoted 12 lines
drivers/scsi/ufs/ufshcd.c +3
-0+if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +1
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) hba->tmf_tag_set = (struct blk_mq_tag_set) { .nr_hw_queues =1 , + .reserved_tags =1 ,
If we give reserved_tags as 1 and always ask for a tm requst with BLK_MQ_REQ_RESERVED flag set, then the tag shall only be allocated from the reserved sbitmap_queue, whose depth is set to 1 here. UFS supports tm queue depth as 8, but here is allowing only one tm req at a time. Why? Please correct me if my understanding is wrong.
Thanks,
Can Guo.
Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +0
-0.queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
Quoted 10 lines · Can Guo
Quoted 9 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46c
Quoted 8 lines
Quoted 6 lines
And, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting
outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outWould you mind add a Fixes tag?
Ok.
Quoted 71 lines
Quoted 71 lines
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +5
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */ @@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );}
Quoted 11 lines
drivers/scsi/ufs/ufshcd.c static int __ufshcd_issue_tm_cmd(struct ufs_hba · line 6376
Quoted 8 lines
drivers/scsi/ufs/ufshcd.c +2
-1@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT );Sorry that I didn't pay much attention to this part of code before. May I know why must we use the
BLK_MQ_REQ_RESERVEDflag?
What I understood is the reserved tag is used when aborting outstanding IOs when all the 32 tags were used.
Quoted 13 lines
Thanks,
Can Guo.Quoted 9 lines
drivers/scsi/ufs/ufshcd.c +3
-0+if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +0
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
Quoted 9 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) · line 9385
Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +1
-0hba->tmf_tag_set = (struct blk_mq_tag_set ) { .nr_hw_queues =1 , + .reserved_tags =1 ,If we give
reserved_tagsas 1 and always ask for a tm requst withBLK_MQ_REQ_RESERVEDflag set, then the tag shall only be allocated from the reservedsbitmap_queue, whose depth is set to 1 here. UFS supports tm queue depth as 8, but here is allowing only one tm req at a time. Why? Please correct me if my understanding is wrong.
I couldn't find tm can be issued in parallel, so thought it was issued one at a time. If we set 8, then we can use 24 for IOs, IIUC.
Please correct me as well. I'm still trying to understand the flow.
Quoted 7 lines
Thanks,
Can Guo.Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +0
-0.queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
Quoted 105 lines · Jaegeuk Kim
Quoted 10 lines · Can Guo
Quoted 9 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cQuoted 8 lines
Quoted 6 lines
And, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting
outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outWould you mind add a Fixes tag?
Ok.
Quoted 71 lines
Quoted 71 lines
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +5
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */ @@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );}
Quoted 11 lines
drivers/scsi/ufs/ufshcd.c static int __ufshcd_issue_tm_cmd(struct ufs_hba · line 6376
Quoted 8 lines
drivers/scsi/ufs/ufshcd.c +2
-1@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT );Sorry that I didn't pay much attention to this part of code before. May I know why must we use the
BLK_MQ_REQ_RESERVEDflag?
Quoted 3 lines
What I understood is the reserved tag is used when aborting outstanding IOs when all the 32 tags were used.
No, the tm requests and I/O requests are on two different tag sets:
tm requests come from hba->tmf_tag_set, while I/O requests come from hba->shost->tag_set. Meaning they don't share tags with each other.
Quoted 23 lines
Quoted 13 lines
Thanks,
Can Guo.Quoted 9 lines
drivers/scsi/ufs/ufshcd.c +3
-0+if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +0
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)Quoted 9 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) · line 9385
Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +1
-0hba->tmf_tag_set = (struct blk_mq_tag_set ) { .nr_hw_queues =1 , + .reserved_tags =1 ,If we give
reserved_tagsas 1 and always ask for a tm requst withBLK_MQ_REQ_RESERVEDflag set, then the tag shall only be allocated from the reservedsbitmap_queue, whose depth is set to 1 here. UFS supports tm queue depth as 8, but here is allowing only one tm req at a time. Why? Please correct me if my understanding is wrong.
Quoted 5 lines
I couldn't find tm can be issued in parallel, so thought it was issued one at a time. If we set 8, then we can use 24 for IOs, IIUC.
Please correct me as well. I'm still trying to understand the flow.
UFS allows a queue depth as 8, which means it support sending multiple tm requests at the same time. You can check commit 69a6c269c097d780a2 - before this change, we used to use below func to allocate tags for tm reqs, which can tell you the true story.
So I am thinking why don't we just we remove the BLK_MQ_REQ_RESERVED flag? Removing it can also fix the warning I suppose. What do you think?
-static bool ufshcd_get_tm_free_slot(struct ufs_hba *hba, int *free_slot) -{
- int tag;
- bool ret = false;
-
- if (!free_slot)
- goto out;
-
- do {
- tag = find_first_zero_bit(&hba->tm_slots_in_use, hba->nutmrs);
- if (tag >= hba->nutmrs)
- goto out;
- } while (test_and_set_bit_lock(tag, &hba->tm_slots_in_use));
-
- *free_slot = tag;
- ret = true;
-out:
- return ret;
-}
Thanks,
Can Guo.
Quoted 7 lines
Quoted 7 lines
Thanks,
Can Guo.Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +0
-0.queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
Quoted 106 lines · Can Guo
Quoted 105 lines · Jaegeuk Kim
Quoted 10 lines · Can Guo
Quoted 9 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cQuoted 8 lines
Quoted 6 lines
And, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting
outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outWould you mind add a Fixes tag?
Ok.
Quoted 82 lines
Quoted 72 lines
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +5
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */ @@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );drivers/scsi/ufs/ufshcd.c +0
-0}drivers/scsi/ufs/ufshcd.c static int __ufshcd_issue_tm_cmd(struct ufs_hba · line 6376
Quoted 7 lines
drivers/scsi/ufs/ufshcd.c +2
-1@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT );Sorry that I didn't pay much attention to this part of code before. May I know why must we use the
BLK_MQ_REQ_RESERVEDflag?
Quoted 7 lines
Quoted 3 lines
What I understood is the reserved tag is used when aborting outstanding IOs when all the 32 tags were used.
No, the tm requests and I/O requests are on two different tag sets:
tm requests come fromhba->tmf_tag_set, while I/O requests come fromhba->shost->tag_set. Meaning they don't share tags with each other.
Add they are issued on two different HW queues - one for tm reqs, one for I/O reqs, which is why two different tag sets are created.
Quoted 70 lines
Quoted 24 lines
Quoted 14 lines
Thanks,
Can Guo.Quoted 9 lines
drivers/scsi/ufs/ufshcd.c +3
-0+if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +0
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)Quoted 9 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) · line 9385
Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +1
-0hba->tmf_tag_set = (struct blk_mq_tag_set ) { .nr_hw_queues =1 , + .reserved_tags =1 ,If we give
reserved_tagsas 1 and always ask for a tm requst withBLK_MQ_REQ_RESERVEDflag set, then the tag shall only be allocated from the reservedsbitmap_queue, whose depth is set to 1 here. UFS supports tm queue depth as 8, but here is allowing only one tm req at a time. Why? Please correct me if my understanding is wrong.Quoted 5 lines
I couldn't find tm can be issued in parallel, so thought it was issued one at a time. If we set 8, then we can use 24 for IOs, IIUC.
Please correct me as well. I'm still trying to understand the flow.
UFS allows a queue depth as 8, which means it support sending multiple tm requests at the same time. You can check commit
69a6c269c097d780a2- before this change, we used to use below func to allocate tags for tm reqs, which can tell you the true story.So I am thinking why don't we just we remove the
BLK_MQ_REQ_RESERVEDflag? Removing it can also fix the warning I suppose. What do you think?-static bool
ufshcd_get_tm_free_slot(struct ufs_hba *hba, int *free_slot)(quoted diff) +0
-9-{ - int tag; - bool ret = false; - - if (!free_slot) - goto out; - - do { - tag = find_first_zero_bit(&hba->tm_slots_in_use,
hba->nutmrs);(quoted diff) +0
-9- if (tag >= hba->nutmrs) - goto out; - } while (test_and_set_bit_lock(tag, &hba->tm_slots_in_use)); - - *free_slot = tag; - ret = true; -out: - return ret; -}Thanks,
Can Guo.Quoted 7 lines
Quoted 7 lines
Thanks,
Can Guo.Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +0
-0.queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,
Re: [PATCH v3 2/2] scsi: ufs: handle LINERESET with correct tm_cmd
In reply to Can GuoQuoted 106 lines · Can Guo
Quoted 105 lines · Jaegeuk Kim
Quoted 10 lines · Can Guo
Quoted 9 lines · Jaegeuk Kim
From: Jaegeuk Kim <[email protected]>
This fixes a warning caused by wrong reserve tag usage in
__ufshcd_issue_tm_cmd.WARNING: CPU: 7 PID: 7 at block/blk-core.c:630 blk_get_request+0x68/0x70 WARNING: CPU: 4 PID: 157 at block/blk-mq-tag.c:82 blk_mq_get_tag+0x438/0x46cQuoted 8 lines
Quoted 6 lines
And, in
ufshcd_err_handler(), we can avoid to sendtm_cmdbefore aborting
outstanding commands by waiting a bit for IO completion like this.
__ufshcd_issue_tm_cmd: task management cmd0x80timed-outWould you mind add a Fixes tag?
Ok.
Quoted 82 lines
Quoted 72 lines
Signed-off-by Jaegeuk Kim <[email protected]>--- drivers/scsi/ufs/ufshcd.c | 36 ++++++++++++++++++++++++++++++++---- 1 file changed, 32 insertions(+), 4 deletions(-)drivers/scsi/ufs/ufshcd.c +5
-0@@ -44,6 +44,9 @@/* Query request timeout */ #define QUERY_REQ_TIMEOUT 1500/* 1.5 seconds */ +/* LINERESET TIME OUT */ +#define LINERESET_IO_TIMEOUT_MS (30000)/* 30 sec */ +/* Task management command timeout */ #define TM_CMD_TIMEOUT 100/* msecs */ @@ -5899,6 +5902,8 @@ static void ufshcd_err_handler(struct work_struct*work) * check if power mode restore is needed. */ if (hba->saved_uic_err &UFSHCD_UIC_PA_GENERIC_ERROR ) { +ktime_t start =ktime_get (); + hba->saved_uic_err &= ~UFSHCD_UIC_PA_GENERIC_ERROR ;if (!hba->saved_uic_err)drivers/scsi/ufs/ufshcd.c +14
-0hba->saved_err &= ~UIC_ERROR ;@@ -5906,6 +5911,20 @@ static void ufshcd_err_handler(struct work_struct *work)if (ufshcd_is_pwr_mode_restore_needed (hba)) needs_restore = true;spin_lock_irqsave (hba->host->host_lock, flags); +/* Wait for IO completion to avoid aborting IOs */ +while (hba->outstanding_reqs) { +ufshcd_complete_requests (hba); +spin_unlock_irqrestore (hba->host->host_lock, flags); +schedule (); +spin_lock_irqsave (hba->host->host_lock, flags); +if (ktime_to_ms (ktime_sub (ktime_get (), start)) > +LINERESET_IO_TIMEOUT_MS ) { +dev_err (hba->dev,"%s: timeout, outstanding=0x%lx\n" , + __func__, hba->outstanding_reqs); +break ; + } + } +if (!hba->saved_err && !needs_restore) goto skip_err_handling;drivers/scsi/ufs/ufshcd.c +7
-3}@@ -6302,9 +6321,13 @@ static irqreturn_t ufshcd_intr(int irq, void *__hba) intr_status =ufshcd_readl (hba,REG_INTERRUPT_STATUS ); } -if (enabled_intr_status && retval ==IRQ_NONE ) { -dev_err (hba->dev,"%s: Unhandled interrupt 0x%08x\n" , - __func__, intr_status); +if (enabled_intr_status && retval ==IRQ_NONE && + !ufshcd_eh_in_progress (hba)) { + dev_err(hba->dev, "%s: Unhandled interrupt 0x%08x (0x%08x, 0x%08x)\n", + __func__, + intr_status, + hba->ufs_stats.last_intr_status, + enabled_intr_status);ufshcd_dump_regs (hba,0 ,UFSHCI_REG_SPACE_SIZE ,"host_regs: " );drivers/scsi/ufs/ufshcd.c +0
-0}drivers/scsi/ufs/ufshcd.c static int __ufshcd_issue_tm_cmd(struct ufs_hba · line 6376
Quoted 7 lines
drivers/scsi/ufs/ufshcd.c +2
-1@@ -6348,7 +6371,11 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba*hba, * Even though we use wait_event() which sleeps indefinitely, * the maximum wait time is bounded by %TM_CMD_TIMEOUT. */ - req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED ); + req =blk_get_request (q,REQ_OP_DRV_OUT ,BLK_MQ_REQ_RESERVED | +BLK_MQ_REQ_NOWAIT );Sorry that I didn't pay much attention to this part of code before. May I know why must we use the
BLK_MQ_REQ_RESERVEDflag?
Quoted 7 lines
Quoted 3 lines
What I understood is the reserved tag is used when aborting outstanding IOs when all the 32 tags were used.
No, the tm requests and I/O requests are on two different tag sets:
tm requests come fromhba->tmf_tag_set, while I/O requests come fromhba->shost->tag_set. Meaning they don't share tags with each other.
I see. :)
Quoted 30 lines
Quoted 24 lines
Quoted 14 lines
Thanks,
Can Guo.Quoted 9 lines
drivers/scsi/ufs/ufshcd.c +3
-0+if (IS_ERR (req)) +return PTR_ERR (req); + req->end_io_data = &wait; free_slot = req->tag;drivers/scsi/ufs/ufshcd.c +0
-0WARN_ON_ONCE (free_slot <0 || free_slot >= hba->nutmrs);@@ -9355,6 +9382,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)Quoted 9 lines
drivers/scsi/ufs/ufshcd.c int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq) · line 9385
Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +1
-0hba->tmf_tag_set = (struct blk_mq_tag_set ) { .nr_hw_queues =1 , + .reserved_tags =1 ,If we give
reserved_tagsas 1 and always ask for a tm requst withBLK_MQ_REQ_RESERVEDflag set, then the tag shall only be allocated from the reservedsbitmap_queue, whose depth is set to 1 here. UFS supports tm queue depth as 8, but here is allowing only one tm req at a time. Why? Please correct me if my understanding is wrong.Quoted 5 lines
I couldn't find tm can be issued in parallel, so thought it was issued one at a time. If we set 8, then we can use 24 for IOs, IIUC.
Please correct me as well. I'm still trying to understand the flow.
Quoted 7 lines
UFS allows a queue depth as 8, which means it support sending multiple tm requests at the same time. You can check commit
69a6c269c097d780a2- before this change, we used to use below func to allocate tags for tm reqs, which can tell you the true story.So I am thinking why don't we just we remove the
BLK_MQ_REQ_RESERVEDflag? Removing it can also fix the warning I suppose. What do you think?
Yeah, I believe it won't give a warning. Okay, let me check it out.
Quoted 32 lines
(quoted diff) +0
-10-static bool ufshcd_get_tm_free_slot(struct ufs_hba *hba, int *free_slot) -{ - int tag; - bool ret = false; - - if (!free_slot) - goto out; - - do { - tag = find_first_zero_bit(&hba->tm_slots_in_use,
hba->nutmrs);(quoted diff) +0
-9- if (tag >= hba->nutmrs) - goto out; - } while (test_and_set_bit_lock(tag, &hba->tm_slots_in_use)); - - *free_slot = tag; - ret = true; -out: - return ret; -}Thanks,
Can Guo.Quoted 7 lines
Quoted 7 lines
Thanks,
Can Guo.Quoted 3 lines
drivers/scsi/ufs/ufshcd.c +0
-0.queue_depth = hba->nutmrs, .ops = &ufshcd_tmf_ops,.flags =
BLK_MQ_F_NO_SCHED,