mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v2 0/2] scsi: ufs-mediatek: fix va09 regulator lifecycle
@ 2026-11-18  3:20 Xiangwu Lai
  2026-11-18  3:20 ` [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator Xiangwu Lai
  2026-11-18  3:20 ` [PATCH v2 2/2] scsi: ufs-mediatek: keep mphy state in sync when va09 disable fails Xiangwu Lai
  0 siblings, 2 replies; 4+ messages in thread
From: Xiangwu Lai @ 2026-11-18  3:20 UTC (permalink / raw)
  To: peter.wang, chaotian.jing
  Cc: chu.stanley, alim.akhtar, bvanassche, stanley.chu, sashiko-bot,
	linux-scsi, linux-mediatek, linux-kernel, Xiangwu Lai

This is v2 of the va09 regulator reference fix.  Two issues surfaced
while reviewing the devm conversion (many thanks to the Sashiko AI
review of v1):

- the devm release alone would trip
  WARN_ON(regulator->enable_count) in _regulator_put() on unbind,
  because ufs_mtk_init() powers the M-PHY (and thus va09) on and
  nothing powers it off again on the way out;

- the M-PHY power-off path can leave host->mphy_powered_on set when
  regulator_disable() fails, so the next power-on bails out early
  and never restores the PHY (pre-existing).

Patch 1 keeps the devm conversion and adds an .exit callback which
balances the enable count before the devm release, in the same spirit
as the .exit callbacks of ufs-exynos and ufs-qcom.  Patch 2 fixes
the pre-existing state desync.

Changes in v2:
- 1/2: additionally add an .exit callback powering the M-PHY off so
  the va09 enable count is balanced before devm releases the
  regulator.  Without it, _regulator_put() would warn on every
  driver unbind (Sashiko AI review, [Medium]).
- 2/2: new patch fixing the mphy state desync when va09 disable
  fails (Sashiko AI review, [High], pre-existing).

Link: https://lore.kernel.org/r/20261118014234.424-1-laixiangwu@cmss.chinamobile.com

Xiangwu Lai (2):
  scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator
  scsi: ufs-mediatek: keep mphy state in sync when va09 disable fails

 drivers/ufs/host/ufs-mediatek.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

-- 
2.50.1.windows.1




^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator
  2026-11-18  3:20 [PATCH v2 0/2] scsi: ufs-mediatek: fix va09 regulator lifecycle Xiangwu Lai
@ 2026-11-18  3:20 ` Xiangwu Lai
  2026-11-19  3:54   ` Peter Wang
  2026-11-18  3:20 ` [PATCH v2 2/2] scsi: ufs-mediatek: keep mphy state in sync when va09 disable fails Xiangwu Lai
  1 sibling, 1 reply; 4+ messages in thread
From: Xiangwu Lai @ 2026-11-18  3:20 UTC (permalink / raw)
  To: peter.wang, chaotian.jing
  Cc: chu.stanley, alim.akhtar, bvanassche, stanley.chu, sashiko-bot,
	linux-scsi, linux-mediatek, linux-kernel, Xiangwu Lai

ufs_mtk_init_va09_pwr_ctrl() obtains the va09 regulator with
regulator_get() but the reference is never released with
regulator_put(), neither on driver unbind nor on probe failure. The
other regulators handled by this driver (dvfsrc-vcore) are already
obtained with devm_regulator_get_optional().

Use devm_regulator_get() so that the reference is released
automatically when the UFS device is detached.

This alone is not enough, though: ufs_mtk_init() powers the M-PHY on,
which enables the va09 regulator, and nothing powers it off again on
the way out -- the variant operations have no .exit callback and
ufshcd_pltfrm_remove() only resumes the device.  The devm release
would then hit WARN_ON(regulator->enable_count) in _regulator_put()
and leak the enable reference.

Balance the use count by adding an .exit callback which powers the
M-PHY off, in the same spirit as the .exit callbacks of ufs-exynos
and ufs-qcom.  ufshcd_hba_exit() invokes it both on driver unbind and
on ufshcd_init() failure after the variant init, so the enable count
is always balanced by the time devm releases the regulator.  It
relies on the state tracking in ufs_mtk_mphy_power_on() and is a
no-op if the M-PHY was never powered on.

Fixes: cf137b3ea49a ("scsi: ufs-mediatek: Support VA09 regulator operations")
Signed-off-by: Xiangwu Lai <laixiangwu@cmss.chinamobile.com>
---

Changes in v2:
- Keep devm_regulator_get() and additionally add an .exit callback
  that powers the M-PHY off, so the va09 enable count is balanced
  before the devm release.  Without it, _regulator_put() would hit
  WARN_ON(regulator->enable_count) on every driver unbind, since
  ufs_mtk_init() leaves the regulator enabled (reported by Sashiko
  AI review).

Link: https://lore.kernel.org/r/20261118014234.424-1-laixiangwu@cmss.chinamobile.com

 drivers/ufs/host/ufs-mediatek.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/ufs/host/ufs-mediatek.c b/drivers/ufs/host/ufs-mediatek.c
index 5b6bff2cf..ebbc017d0 100644
--- a/drivers/ufs/host/ufs-mediatek.c
+++ b/drivers/ufs/host/ufs-mediatek.c
@@ -692,7 +692,7 @@ static void ufs_mtk_init_va09_pwr_ctrl(struct ufs_hba *hba)
 {
 	struct ufs_mtk_host *host = ufshcd_get_variant(hba);
 
-	host->reg_va09 = regulator_get(hba->dev, "va09");
+	host->reg_va09 = devm_regulator_get(hba->dev, "va09");
 	if (IS_ERR(host->reg_va09))
 		dev_info(hba->dev, "failed to get va09");
 	else
@@ -1331,6 +1331,14 @@ static int ufs_mtk_init(struct ufs_hba *hba)
 	return err;
 }
 
+static void ufs_mtk_exit(struct ufs_hba *hba)
+{
+	struct ufs_mtk_host *host = ufshcd_get_variant(hba);
+
+	if (host)
+		ufs_mtk_mphy_power_on(hba, false);
+}
+
 static int ufs_mtk_negotiate_pwr_mode(struct ufs_hba *hba,
 				      const struct ufs_pa_layer_attr *dev_max_params,
 				      struct ufs_pa_layer_attr *dev_req_params)
@@ -2353,6 +2361,7 @@ static void ufs_mtk_config_scsi_dev(struct scsi_device *sdev)
 static const struct ufs_hba_variant_ops ufs_hba_mtk_vops = {
 	.name                = "mediatek.ufshci",
 	.init                = ufs_mtk_init,
+	.exit                = ufs_mtk_exit,
 	.get_ufs_hci_version = ufs_mtk_get_ufs_hci_version,
 	.setup_clocks        = ufs_mtk_setup_clocks,
 	.hce_enable_notify   = ufs_mtk_hce_enable_notify,
-- 
2.50.1.windows.1




^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v2 2/2] scsi: ufs-mediatek: keep mphy state in sync when va09 disable fails
  2026-11-18  3:20 [PATCH v2 0/2] scsi: ufs-mediatek: fix va09 regulator lifecycle Xiangwu Lai
  2026-11-18  3:20 ` [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator Xiangwu Lai
@ 2026-11-18  3:20 ` Xiangwu Lai
  1 sibling, 0 replies; 4+ messages in thread
From: Xiangwu Lai @ 2026-11-18  3:20 UTC (permalink / raw)
  To: peter.wang, chaotian.jing
  Cc: chu.stanley, alim.akhtar, bvanassche, stanley.chu, sashiko-bot,
	linux-scsi, linux-mediatek, linux-kernel, Xiangwu Lai

In ufs_mtk_mphy_power_on(), the power-off path calls phy_power_off()
before disabling the va09 regulator.  If regulator_disable() fails,
the error is returned while host->mphy_powered_on is left set,
although the M-PHY is physically powered off already.  The next call
with on = true then bails out through the !(on ^ host->mphy_powered_on)
check without powering the M-PHY back on, leaving the UFS link dead
(I/O timeouts, controller lockups).

The M-PHY is off regardless of the va09 disable result, so update the
state flag in the power-off path and let the next power-on retry the
whole sequence.

Found by code inspection while reviewing the va09 error paths of the
driver; the failure has not been observed on real hardware.

Link: https://lore.kernel.org/r/sashiko-outbox-219916@kernel.org
Fixes: cf137b3ea49a ("scsi: ufs-mediatek: Support VA09 regulator operations")
Signed-off-by: Xiangwu Lai <laixiangwu@cmss.chinamobile.com>
---
 drivers/ufs/host/ufs-mediatek.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/ufs/host/ufs-mediatek.c b/drivers/ufs/host/ufs-mediatek.c
index ebbc017d0..8d4b84d58 100644
--- a/drivers/ufs/host/ufs-mediatek.c
+++ b/drivers/ufs/host/ufs-mediatek.c
@@ -542,6 +542,14 @@ static int ufs_mtk_mphy_power_on(struct ufs_hba *hba, bool on)
 			ufs_mtk_va09_pwr_ctrl(res, 0);
 			ret = regulator_disable(host->reg_va09);
 		}
+		/*
+		 * The M-PHY is off regardless of the va09 result, so keep
+		 * the state in sync with the hardware.  Otherwise the next
+		 * power-on bails out through the
+		 * !(on ^ host->mphy_powered_on) check without powering the
+		 * M-PHY back on.
+		 */
+		host->mphy_powered_on = false;
 	}
 out:
 	if (ret) {
-- 
2.50.1.windows.1




^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator
  2026-11-18  3:20 ` [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator Xiangwu Lai
@ 2026-11-19  3:54   ` Peter Wang
  0 siblings, 0 replies; 4+ messages in thread
From: Peter Wang @ 2026-11-19  3:54 UTC (permalink / raw)
  To: Xiangwu Lai, chaotian.jing
  Cc: chu.stanley, alim.akhtar, bvanassche, stanley.chu, sashiko-bot,
	linux-scsi, linux-mediatek, linux-kernel

On Wed, 2026-11-18 at 11:20 +0800, Xiangwu Lai wrote:
> ufs_mtk_init_va09_pwr_ctrl() obtains the va09 regulator with
> regulator_get() but the reference is never released with
> regulator_put(), neither on driver unbind nor on probe failure. The
> other regulators handled by this driver (dvfsrc-vcore) are already
> obtained with devm_regulator_get_optional().
> 
> Use devm_regulator_get() so that the reference is released
> automatically when the UFS device is detached.
> 
> This alone is not enough, though: ufs_mtk_init() powers the M-PHY on,
> which enables the va09 regulator, and nothing powers it off again on
> the way out -- the variant operations have no .exit callback and
> ufshcd_pltfrm_remove() only resumes the device.  The devm release
> would then hit WARN_ON(regulator->enable_count) in _regulator_put()
> and leak the enable reference.
> 
> Balance the use count by adding an .exit callback which powers the
> M-PHY off, in the same spirit as the .exit callbacks of ufs-exynos
> and ufs-qcom.  ufshcd_hba_exit() invokes it both on driver unbind and
> on ufshcd_init() failure after the variant init, so the enable count
> is always balanced by the time devm releases the regulator.  It
> relies on the state tracking in ufs_mtk_mphy_power_on() and is a
> no-op if the M-PHY was never powered on.
> 
> Fixes: cf137b3ea49a ("scsi: ufs-mediatek: Support VA09 regulator
> operations")
> Signed-off-by: Xiangwu Lai <laixiangwu@cmss.chinamobile.com>
> ---

Hi Xiangwu,

Thanks for fix this bug.
Reviewed-by: Peter Wang <peter.wang@mediatek.com>



^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-11-19  3:54 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-18  3:20 [PATCH v2 0/2] scsi: ufs-mediatek: fix va09 regulator lifecycle Xiangwu Lai
2026-11-18  3:20 ` [PATCH v2 1/2] scsi: ufs-mediatek: use devm_regulator_get() for the va09 regulator Xiangwu Lai
2026-11-19  3:54   ` Peter Wang
2026-11-18  3:20 ` [PATCH v2 2/2] scsi: ufs-mediatek: keep mphy state in sync when va09 disable fails Xiangwu Lai

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®