Skip to contentSkip to conversation

scsi: ufs: fix no clkgating due to tag being alive

PATCH In review Jaegeuk Kim 2 participants 5 messages Oct 16, 2020 – Oct 20, 2020 SCSI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 scsi: ufs: fix no clkgating due to tag being alive

The below call stack prevents clk_gating at every IO completion. We can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work will check it again.

ufshcd_complete_requests(struct ufs_hba *hba)
ufshcd_transfer_req_compl()

    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();
Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Can Guo <[email protected]>
---
 drivers/scsi/ufs/ufshcd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/ufs/ufshcd.c +1 -1
index 1d157ff58d81..22fecb9a4ec2 100644
@@ -1727,7 +1727,7 @@ static void __ufshcd_release(struct ufs_hba *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
+		|| hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
 		|| ufshcd_eh_in_progress(hba))
 		return;
Signature · 1 line
2.29.0.rc1.297.gfa9743e501-goog
#1 lore raw

Re: [PATCH v2] scsi: ufs: fix clkgating on/off correctly

The below call stack prevents clk_gating at every IO completion. We can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work will check it again.

ufshcd_complete_requests(struct ufs_hba *hba)
ufshcd_transfer_req_compl()

    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();

In addition, we have to avoid gate_work, if it was disabled.

Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Can Guo <[email protected]>

---

Change log from v1:
- change the patch subject
- fix clkgate.is_enable to work

 drivers/scsi/ufs/ufshcd.c | 5 +++--
 drivers/scsi/ufs/ufshcd.h | 5 +++++
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +3 -2
index a2db8182663d..75e8a76f20c7 100644
@@ -1729,9 +1729,10 @@ static void __ufshcd_release(struct ufs_hba *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
+		|| hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
-		|| ufshcd_eh_in_progress(hba))
+		|| ufshcd_eh_in_progress(hba)
+		|| ufshcd_is_clkgating_enabled(hba))
 		return;
 
 	hba->clk_gating.state = REQ_CLKS_OFF;
drivers/scsi/ufs/ufshcd.h +5 -0
index 8344d8cb3678..09e59cb86e69 100644
@@ -814,6 +814,11 @@ static inline bool ufshcd_is_auto_hibern8_supported(struct ufs_hba *hba)
 		!(hba->quirks & UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8);
 }
 
+static inline bool ufshcd_is_clkgating_enabled(struct ufs_hba *hba)
+{
+	return hba->clk_gating.is_enabled;
+}
+
 static inline bool ufshcd_is_auto_hibern8_enabled(struct ufs_hba *hba)
 {
 	return FIELD_GET(UFSHCI_AHIBERN8_TIMER_MASK, hba->ahit) ? true : false;
Signature · 1 line
2.29.0.rc1.297.gfa9743e501-goog
#2 lore raw

Re: [PATCH v2] scsi: ufs: fix clkgating on/off correctly

Quoted 3 lines · On 2020-10-17 05:18, [email protected] wrote:

The below call stack prevents clk_gating at every IO completion. We can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work will check it again.

Thanks for fixing it, actually we have noticed this for a while. It used to work well, please add a Fixes tag or mention the commit which breaks clk gating functionality - the commit which introduces func ufshcd_any_tag_in_use().

Regards,

Can Guo.

Quoted 62 lines

ufshcd_complete_requests(struct ufs_hba *hba)
ufshcd_transfer_req_compl()

    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();

In addition, we have to avoid gate_work, if it was disabled.

Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Can Guo <[email protected]>

---

Change log from v1:
- change the patch subject
- fix clkgate.is_enable to work

 drivers/scsi/ufs/ufshcd.c | 5 +++--
 drivers/scsi/ufs/ufshcd.h | 5 +++++
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +3 -2
index a2db8182663d..75e8a76f20c7 100644
@@ -1729,9 +1729,10 @@ static void __ufshcd_release(struct ufs_hba 
 *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
+		|| hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
-		|| ufshcd_eh_in_progress(hba))
+		|| ufshcd_eh_in_progress(hba)
+		|| ufshcd_is_clkgating_enabled(hba))
 		return;
 
drivers/scsi/ufs/ufshcd.c +0 -0
 	hba->clk_gating.state = REQ_CLKS_OFF;
drivers/scsi/ufs/ufshcd.h +5 -0
index 8344d8cb3678..09e59cb86e69 100644
@@ -814,6 +814,11 @@ static inline bool
 ufshcd_is_auto_hibern8_supported(struct ufs_hba *hba)
 		!(hba->quirks & UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8);
 }
 
+static inline bool ufshcd_is_clkgating_enabled(struct ufs_hba *hba)
+{
+	return hba->clk_gating.is_enabled;
+}
+
 static inline bool ufshcd_is_auto_hibern8_enabled(struct ufs_hba *hba)
 {

return FIELD_GET(UFSHCI_AHIBERN8_TIMER_MASK, hba->ahit) ? true :
false;

#3 lore raw

Re: [PATCH v2] scsi: ufs: fix clkgating on/off correctly

In reply to [email protected]
Quoted 34 lines · On 2020-10-17 05:18, [email protected] wrote:

The below call stack prevents clk_gating at every IO completion. We can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work will check it again.

ufshcd_complete_requests(struct ufs_hba *hba)
ufshcd_transfer_req_compl()

    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();

In addition, we have to avoid gate_work, if it was disabled.

Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Can Guo <[email protected]>

---

Change log from v1:
- change the patch subject
- fix clkgate.is_enable to work

 drivers/scsi/ufs/ufshcd.c | 5 +++--
 drivers/scsi/ufs/ufshcd.h | 5 +++++
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +0 -0
index a2db8182663d..75e8a76f20c7 100644

drivers/scsi/ufs/ufshcd.c static void __ufshcd_release(struct ufs_hba · line 1736

Quoted 10 lines
drivers/scsi/ufs/ufshcd.c +3 -2
@@ -1729,9 +1729,10 @@ static void __ufshcd_release(struct ufs_hba 
 *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
+		|| hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
-		|| ufshcd_eh_in_progress(hba))
+		|| ufshcd_eh_in_progress(hba)
+		|| ufshcd_is_clkgating_enabled(hba))

I guess you want it to be "!ufshcd_is_clkgating_enabled(hba)" - if clk gating is enabled, we should let gating happen but not bail, right?

That said, I don't think we need to check whether clk gating is enabled or not here, since ufshcd_clkgating_store() manipulates clk_gating->active_reqs in an atomic way. If someone disables clk gating -> clk_gating->active_reqs++, this check becomes TURE on the very first condition.

So the very fisrt patch looks better to me.

Thanks,

Can Guo.

Quoted 21 lines
drivers/scsi/ufs/ufshcd.c +0 -0
 		return;
 
drivers/scsi/ufs/ufshcd.c +0 -0
 	hba->clk_gating.state = REQ_CLKS_OFF;
drivers/scsi/ufs/ufshcd.h +5 -0
index 8344d8cb3678..09e59cb86e69 100644
@@ -814,6 +814,11 @@ static inline bool
 ufshcd_is_auto_hibern8_supported(struct ufs_hba *hba)
 		!(hba->quirks & UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8);
 }
 
+static inline bool ufshcd_is_clkgating_enabled(struct ufs_hba *hba)
+{
+	return hba->clk_gating.is_enabled;
+}
+
 static inline bool ufshcd_is_auto_hibern8_enabled(struct ufs_hba *hba)
 {

return FIELD_GET(UFSHCI_AHIBERN8_TIMER_MASK, hba->ahit) ? true :
false;

#4 lore raw

Re: [PATCH v2] scsi: ufs: fix clkgating on/off correctly

Quoted 49 lines · Can Guo
Quoted 35 lines · On 2020-10-17 05:18, [email protected] wrote:

The below call stack prevents clk_gating at every IO completion. We can remove the condition, ufshcd_any_tag_in_use(), since clkgating_work
will check it again.

ufshcd_complete_requests(struct ufs_hba *hba)
ufshcd_transfer_req_compl()

    __ufshcd_transfer_req_compl()
      __ufshcd_release(hba)
        if (ufshcd_any_tag_in_use() == 1)
           return;
  ufshcd_tmc_handler(hba);
    blk_mq_tagset_busy_iter();

In addition, we have to avoid gate_work, if it was disabled.

Cc Alim Akhtar <[email protected]>
Cc Avri Altman <[email protected]>
Cc Can Guo <[email protected]>

---

Change log from v1:
- change the patch subject
- fix clkgate.is_enable to work

 drivers/scsi/ufs/ufshcd.c | 5 +++--
 drivers/scsi/ufs/ufshcd.h | 5 +++++
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/ufs/ufshcd.c +0 -0
index a2db8182663d..75e8a76f20c7 100644

drivers/scsi/ufs/ufshcd.c static void __ufshcd_release(struct ufs_hba *hba) · line 1735

Quoted 9 lines
drivers/scsi/ufs/ufshcd.c +3 -2
@@ -1729,9 +1729,10 @@ static void __ufshcd_release(struct ufs_hba *hba)
 
 	if (hba->clk_gating.active_reqs || hba->clk_gating.is_suspended
 		|| hba->ufshcd_state != UFSHCD_STATE_OPERATIONAL
-		|| ufshcd_any_tag_in_use(hba) || hba->outstanding_tasks
+		|| hba->outstanding_tasks
 		|| hba->active_uic_cmd || hba->uic_async_done
-		|| ufshcd_eh_in_progress(hba))
+		|| ufshcd_eh_in_progress(hba)
+		|| ufshcd_is_clkgating_enabled(hba))

I guess you want it to be "!ufshcd_is_clkgating_enabled(hba)" - if clk gating is enabled, we should let gating happen but not bail, right?

Quoted 5 lines

That said, I don't think we need to check whether clk gating is enabled or not here, since ufshcd_clkgating_store() manipulates clk_gating->active_reqs in an atomic way. If someone disables clk gating -> clk_gating->active_reqs++,
this check becomes TURE on the very first condition.

Yeah, I found the above and and was testing it locally. Anyway, it seems you're right. Let me give it a try back again.

Thanks,

Quoted 28 lines

So the very fisrt patch looks better to me.

Thanks,

Can Guo.

Quoted 21 lines
drivers/scsi/ufs/ufshcd.c +0 -0
 		return;
 
 	hba->clk_gating.state = REQ_CLKS_OFF;
drivers/scsi/ufs/ufshcd.h +5 -0
index 8344d8cb3678..09e59cb86e69 100644
@@ -814,6 +814,11 @@ static inline bool
 ufshcd_is_auto_hibern8_supported(struct ufs_hba *hba)
 		!(hba->quirks & UFSHCD_QUIRK_BROKEN_AUTO_HIBERN8);
 }
 
+static inline bool ufshcd_is_clkgating_enabled(struct ufs_hba *hba)
+{
+	return hba->clk_gating.is_enabled;
+}
+
 static inline bool ufshcd_is_auto_hibern8_enabled(struct ufs_hba *hba)
 {

return FIELD_GET(UFSHCI_AHIBERN8_TIMER_MASK, hba->ahit) ? true :
false;