Skip to contentSkip to conversation

scsi: allow auto suspend override by low-level driver

PATCH Superseded Stanley Chu 2 participants 8 messages Sep 12, 2019 – Sep 16, 2019

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 scsi: allow auto suspend override by low-level driver

Until now the scsi mid-layer forbids runtime suspend till userspace enables it. This is mainly to quarantine some disks with broken runtime power management or have high latencies executing suspend resume callbacks. If the userspace doesn't enable the runtime suspend the underlying hardware will be always on even when it is not doing any useful work and thus wasting power.

Some low-level drivers for the controllers can efficiently use runtime power management to reduce power consumption and improve battery life.

This patchset allows runtime suspend parameters override within the LLD itself instead of waiting for userspace to control the power management, and make UFS as the first user of this capability.

v1 => v2:
- Allow "zero" sdev->rpm_autosuspend_delay (Avri)
- Fix format of some lines (Avri)

Stanley Chu (3):
scsi: core: allow auto suspend override by low-level driver
scsi: ufs: override auto suspend tunables for ufs
scsi: ufs-mediatek: enable auto suspend capability

 drivers/scsi/scsi_scan.c        |  6 ++++++
 drivers/scsi/scsi_sysfs.c       |  3 ++-
 drivers/scsi/sd.c               |  4 ++++
 drivers/scsi/ufs/ufs-mediatek.c |  7 +++++++
 drivers/scsi/ufs/ufshcd.c       |  8 ++++++++
 drivers/scsi/ufs/ufshcd.h       | 10 ++++++++++
 include/scsi/scsi_device.h      |  2 +-
 7 files changed, 38 insertions(+), 2 deletions(-)
Signature · 1 line
2.18.0
#1 lore raw

PATCH v2 1/3 scsi: core: allow auto suspend override by low-level driver

Rework from previous work by:
Sujit Reddy Thumma <[email protected]>

Until now the scsi mid-layer forbids runtime suspend till userspace enables it. This is mainly to quarantine some disks with broken runtime power management or have high latencies executing suspend resume callbacks. If the userspace doesn't enable the runtime suspend the underlying hardware will be always on even when it is not doing any useful work and thus wasting power.

Some low-level drivers for the controllers can efficiently use runtime power management to reduce power consumption and improve battery life. Allow runtime suspend parameters override within the LLD itself instead of waiting for userspace to control the power management.

---
 drivers/scsi/scsi_scan.c   | 6 ++++++
 drivers/scsi/scsi_sysfs.c  | 3 ++-
 drivers/scsi/sd.c          | 4 ++++
 include/scsi/scsi_device.h | 2 +-
 4 files changed, 13 insertions(+), 2 deletions(-)
drivers/scsi/scsi_scan.c +6 -0
index 058079f915f1..caf700a6b7c8 100644
@@ -280,6 +280,12 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
 	scsi_change_queue_depth(sdev, sdev->host->cmd_per_lun ?
 					sdev->host->cmd_per_lun : 1);
 
+	/*
+	 * Keep autosuspend disabled by default unless LLDD specifically
+	 * enables it in slave_configure.
+	 */
+	sdev->rpm_autosuspend_delay = -1;
+
 	scsi_sysfs_device_initialize(sdev);
 
 	if (shost->hostt->slave_alloc) {
drivers/scsi/scsi_sysfs.c +2 -1
index 64c96c7828ee..461aafadd208 100644
@@ -1300,7 +1300,8 @@ int scsi_sysfs_add_sdev(struct scsi_device *sdev)
 	device_enable_async_suspend(&sdev->sdev_gendev);
 	scsi_autopm_get_target(starget);
 	pm_runtime_set_active(&sdev->sdev_gendev);
-	pm_runtime_forbid(&sdev->sdev_gendev);
+	if (sdev->rpm_autosuspend_delay < 0)
+		pm_runtime_forbid(&sdev->sdev_gendev);
 	pm_runtime_enable(&sdev->sdev_gendev);
 	scsi_autopm_put_target(starget);
 
drivers/scsi/sd.c +4 -0
index 149d406aacc9..de410b272158 100644
@@ -3371,6 +3371,10 @@ static int sd_probe(struct device *dev)
 	}
 
 	blk_pm_runtime_init(sdp->request_queue, dev);
+	if (sdp->rpm_autosuspend_delay >= 0) {
+		pm_runtime_set_autosuspend_delay(dev,
+						 sdp->rpm_autosuspend_delay);
+	}
 	device_add_disk(dev, gd, NULL);
 	if (sdkp->capacity)
 		sd_dif_config_host(sdkp);
include/scsi/scsi_device.h +1 -1
index 202f4d6a4342..133b282fae5a 100644
@@ -199,7 +199,7 @@ struct scsi_device {
 	unsigned broken_fua:1;		/* Don't set FUA bit */
 	unsigned lun_in_cdb:1;		/* Store LUN bits in CDB[1] */
 	unsigned unmap_limit_for_ws:1;	/* Use the UNMAP limit for WRITE SAME */
-
+	int rpm_autosuspend_delay;
 	atomic_t disk_events_disable_depth; /* disable depth for disk events */
 
 	DECLARE_BITMAP(supported_events, SDEV_EVT_MAXBITS); /* supported events */
Signature · 1 line
2.18.0
#2 lore raw

Re: [PATCH v2 1/3] scsi: core: allow auto suspend override by low-level driver

drivers/scsi/scsi_sysfs.c int scsi_sysfs_add_sdev(struct scsi_device *sdev) · line 1306

Quoted 13 lines · Stanley Chu
drivers/scsi/scsi_sysfs.c +2 -1
index 64c96c7828ee..461aafadd208 100644
@@ -1300,7 +1300,8 @@ int scsi_sysfs_add_sdev(struct scsi_device *sdev)
 	device_enable_async_suspend(&sdev->sdev_gendev);
 	scsi_autopm_get_target(starget);
 	pm_runtime_set_active(&sdev->sdev_gendev);
-	pm_runtime_forbid(&sdev->sdev_gendev);
+	if (sdev->rpm_autosuspend_delay < 0)
+		pm_runtime_forbid(&sdev->sdev_gendev);
 	pm_runtime_enable(&sdev->sdev_gendev);
 	scsi_autopm_put_target(starget);

So we have a single new struct member, rpm_autosuspend_delay, that controls two different behaviors: (a) whether or not runtime suspend is enabled at device creation time and (b) the power management autosuspend delay. I don't like this. Should two separate variables be introduced instead of using a single variable to control both behaviors?

Quoted 6 lines
include/scsi/scsi_device.h +0 -0
index 202f4d6a4342..133b282fae5a 100644

include/scsi/scsi_device.h struct scsi_device { · line 205

Quoted 8 lines
include/scsi/scsi_device.h +1 -1
@@ -199,7 +199,7 @@ struct scsi_device {
 	unsigned broken_fua:1;		/* Don't set FUA bit */
 	unsigned lun_in_cdb:1;		/* Store LUN bits in CDB[1] */
 	unsigned unmap_limit_for_ws:1;	/* Use the UNMAP limit for WRITE SAME */
-
+	int rpm_autosuspend_delay;
 	atomic_t disk_events_disable_depth; /* disable depth for disk events */
 
 	DECLARE_BITMAP(supported_events, SDEV_EVT_MAXBITS); /* supported events */

Since the default value for the autosuspend delay is the same for all SCSI devices attached to a SCSI host is the same, please add a variable with the same name in the SCSI host template and use that value as the default value for SCSI devices. If the rpm_autosuspend_delay variable only occurs in struct scsi_device then LLD authors are forced to introduce a slave_configure function. Introducing such a function can be avoided if the default autosuspend delay can be specified in the host template.

Bart.

#3 lore raw

Re: [PATCH v2 1/3] scsi: core: allow auto suspend override by low-level driver

Hi Bart,

Quoted 12 lines

drivers/scsi/scsi_sysfs.c int scsi_sysfs_add_sdev(struct scsi_device *sdev) · line 1306

Quoted 5 lines
drivers/scsi/scsi_sysfs.c +2 -1
-	pm_runtime_forbid(&sdev->sdev_gendev);
+	if (sdev->rpm_autosuspend_delay < 0)
+		pm_runtime_forbid(&sdev->sdev_gendev);
 	pm_runtime_enable(&sdev->sdev_gendev);
 	scsi_autopm_put_target(starget);

So we have a single new struct member, rpm_autosuspend_delay, that controls two different behaviors: (a) whether or not runtime suspend is enabled at device creation time and (b) the power management autosuspend delay. I don't like this. Should two separate variables be introduced instead of using a single variable to control both behaviors?

OK! Will try to separate different variables to control different things in next version.

Quoted 12 lines
Quoted 12 lines
include/scsi/scsi_device.h +1 -1
index 202f4d6a4342..133b282fae5a 100644
@@ -199,7 +199,7 @@ struct scsi_device {
 	unsigned broken_fua:1;		/* Don't set FUA bit */
 	unsigned lun_in_cdb:1;		/* Store LUN bits in CDB[1] */
 	unsigned unmap_limit_for_ws:1;	/* Use the UNMAP limit for WRITE SAME */
-
+	int rpm_autosuspend_delay;
 	atomic_t disk_events_disable_depth; /* disable depth for disk events */
 
Quoted 12 lines

include/scsi/scsi_device.h struct scsi_device { · line 205

Quoted 2 lines
include/scsi/scsi_device.h +0 -0
 	DECLARE_BITMAP(supported_events, SDEV_EVT_MAXBITS); /* supported events */

Since the default value for the autosuspend delay is the same for all SCSI devices attached to a SCSI host is the same, please add a variable with the same name in the SCSI host template and use that value as the default value for SCSI devices. If the rpm_autosuspend_delay variable only occurs in struct scsi_device then LLD authors are forced to introduce a slave_configure function. Introducing such a function can be avoided if the default autosuspend delay can be specified in the host template.

Sounds reasonable. Will create a member indicating autosuspend delay for the same SCSI host in SCSI host template in next version.

Quoted 2 lines

Bart.

Thanks,
Stanley

#4 lore raw

PATCH v2 2/3 scsi: ufs: override auto suspend tunables for ufs

In reply to Stanley Chu

Rework from previous work by:
Sujit Reddy Thumma <[email protected]>

Override auto suspend tunables for UFS device LUNs during initialization so as to efficiently manage background operations and the power consumption.

---
 drivers/scsi/ufs/ufshcd.c |  8 ++++++++
 drivers/scsi/ufs/ufshcd.h | 10 ++++++++++
 2 files changed, 18 insertions(+)
drivers/scsi/ufs/ufshcd.c +8 -0
index 30b752c61b97..d253a018a73b 100644
@@ -88,6 +88,9 @@
 /* Interrupt aggregation default timeout, unit: 40us */
 #define INT_AGGR_DEF_TO	0x02
 
+/* default delay of autosuspend: 2000 ms */
+#define RPM_AUTOSUSPEND_DELAY_MS 2000
+
 #define ufshcd_toggle_vreg(_dev, _vreg, _on)				\
 	({                                                              \
 		int _ret;                                               \
@@ -4612,9 +4615,14 @@ static int ufshcd_change_queue_depth(struct scsi_device *sdev, int depth)
  */
 static int ufshcd_slave_configure(struct scsi_device *sdev)
 {
+	struct ufs_hba *hba = shost_priv(sdev->host);
 	struct request_queue *q = sdev->request_queue;
 
 	blk_queue_update_dma_pad(q, PRDT_DATA_BYTE_COUNT_PAD - 1);
+
+	if (ufshcd_is_rpm_autosuspend_allowed(hba))
+		sdev->rpm_autosuspend_delay = RPM_AUTOSUSPEND_DELAY_MS;
+
 	return 0;
 }
 
drivers/scsi/ufs/ufshcd.h +10 -0
index a43c7135f33d..99ea416519af 100644
@@ -714,6 +714,12 @@ struct ufs_hba {
 	 * the performance of ongoing read/write operations.
 	 */
 #define UFSHCD_CAP_KEEP_AUTO_BKOPS_ENABLED_EXCEPT_SUSPEND (1 << 5)
+	/*
+	 * This capability allows host controller driver to automatically
+	 * enable runtime power management by itself instead of waiting
+	 * for userspace to control the power management.
+	 */
+#define UFSHCD_CAP_RPM_AUTOSUSPEND (1 << 6)
 
 	struct devfreq *devfreq;
 	struct ufs_clk_scaling clk_scaling;
@@ -747,6 +753,10 @@ static inline bool ufshcd_can_autobkops_during_suspend(struct ufs_hba *hba)
 {
 	return hba->caps & UFSHCD_CAP_AUTO_BKOPS_SUSPEND;
 }
+static inline bool ufshcd_is_rpm_autosuspend_allowed(struct ufs_hba *hba)
+{
+	return hba->caps & UFSHCD_CAP_RPM_AUTOSUSPEND;
+}
 
 static inline bool ufshcd_is_intr_aggr_allowed(struct ufs_hba *hba)
 {
Signature · 1 line
2.18.0
#5 lore raw

PATCH v2 3/3 scsi: ufs-mediatek: enable auto suspend capability

In reply to Stanley Chu

Enable auto suspend capability in MediaTek UFS driver.

---
 drivers/scsi/ufs/ufs-mediatek.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/scsi/ufs/ufs-mediatek.c +7 -0
index 0f6ff33ce52e..b7b177c6194c 100644
@@ -117,6 +117,11 @@ static int ufs_mtk_setup_clocks(struct ufs_hba *hba, bool on,
 	return ret;
 }
 
+static void ufs_mtk_set_caps(struct ufs_hba *hba)
+{
+	hba->caps |= UFSHCD_CAP_RPM_AUTOSUSPEND;
+}
+
 /**
  * ufs_mtk_init - find other essential mmio bases
  * @hba: host controller instance
@@ -147,6 +152,8 @@ static int ufs_mtk_init(struct ufs_hba *hba)
 	if (err)
 		goto out_variant_clear;
 
+	ufs_mtk_set_caps(hba);
+
 	/*
 	 * ufshcd_vops_init() is invoked after
 	 * ufshcd_setup_clock(true) in ufshcd_hba_init() thus
Signature · 1 line
2.18.0
#6 lore raw

Re: [PATCH v2 3/3] scsi: ufs-mediatek: enable auto suspend capability

Quoted 25 lines · Stanley Chu

Enable auto suspend capability in MediaTek UFS driver.

---
 drivers/scsi/ufs/ufs-mediatek.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/scsi/ufs/ufs-mediatek.c +5 -0
index 0f6ff33ce52e..b7b177c6194c 100644
@@ -117,6 +117,11 @@ static int ufs_mtk_setup_clocks(struct ufs_hba *hba, bool on,
 	return ret;
 }
 
+static void ufs_mtk_set_caps(struct ufs_hba *hba)
+{
+	hba->caps |= UFSHCD_CAP_RPM_AUTOSUSPEND;
+}
+
 /**
  * ufs_mtk_init - find other essential mmio bases
  * @hba: host controller instance

drivers/scsi/ufs/ufs-mediatek.c static int ufs_mtk_init(struct ufs_hba *hba) · line 159

Quoted 8 lines
drivers/scsi/ufs/ufs-mediatek.c +2 -0
@@ -147,6 +152,8 @@ static int ufs_mtk_init(struct ufs_hba *hba)
 	if (err)
 		goto out_variant_clear;
 
+	ufs_mtk_set_caps(hba);
+
 	/*
 	 * ufshcd_vops_init() is invoked after
 	 * ufshcd_setup_clock(true) in ufshcd_hba_init() thus

Please inline the ufs_mtk_set_caps() function. Introducing single line functions like is done in this patch doesn't improve readability.

Thanks,

Bart.

#7 lore raw

Re: [PATCH v2 3/3] scsi: ufs-mediatek: enable auto suspend capability

Hi Bart,

Quoted 4 lines
Quoted 4 lines
drivers/scsi/ufs/ufs-mediatek.c +0 -0
@@ -147,6 +152,8 @@ static int ufs_mtk_init(struct ufs_hba *hba)
 	if (err)
 		goto out_variant_clear;
 
Quoted 9 lines

drivers/scsi/ufs/ufs-mediatek.c static int ufs_mtk_init(struct ufs_hba *hba) · line 159

Quoted 5 lines
drivers/scsi/ufs/ufs-mediatek.c +2 -0
+	ufs_mtk_set_caps(hba);
+
 	/*
 	 * ufshcd_vops_init() is invoked after
 	 * ufshcd_setup_clock(true) in ufshcd_hba_init() thus

Please inline the ufs_mtk_set_caps() function. Introducing single line functions like is done in this patch doesn't improve readability.

OK! Will be fixed in next version.

Thanks,
Stanley