[PATCH 1/1] ufs: core: add fatal errors check for LINERESET

[PATCH 1/1] ufs: core: add fatal errors check for LINERESET

[naomi.chu]

From: Naomi Chu <naomi.chu@mediatek.com>

The current error handling flow directly checks the power mode after a
LINERESET occurs. However, if a PA_INIT_ERR occurs after the LINERESET,
checking the power mode may require waiting for three DME_GET timeouts,
which wastes 1.5 seconds.

To improve efficiency, when a LINERESET occurs, wait for PA_INIT to
complete and check for any other errors. After a LINERESET, PA_INIT
takes up to 75.4ms to complete. Therefore, a 100ms delay is added.

Signed-off-by: Naomi Chu <naomi.chu@mediatek.com>
---
 drivers/ufs/core/ufshcd.c | 45 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index 7735421e3991..c7cc62429fc9 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -6359,6 +6359,46 @@ static inline bool ufshcd_is_saved_err_fatal(struct ufs_hba *hba)
 	       (hba->saved_err & (INT_FATAL_ERRORS | UFSHCD_UIC_HIBERN8_MASK));
 }
 
+static bool ufshcd_is_linereset_fatal(struct ufs_hba *hba)
+{
+	bool needs_reset = true;
+	unsigned long flags;
+	int err;
+
+	spin_lock_irqsave(hba->host->host_lock, flags);
+
+	if (ufshcd_is_saved_err_fatal(hba)) {
+		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		goto out;
+	}
+
+	/*
+	 * Wait for 100ms to ensure the PA_INIT flow is complete,
+	 * and check for PA_INIT_ERR or other fatal errors.
+	 */
+	spin_unlock_irqrestore(hba->host->host_lock, flags);
+	msleep(100);
+	spin_lock_irqsave(hba->host->host_lock, flags);
+
+	if (ufshcd_is_saved_err_fatal(hba)) {
+		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		goto out;
+	}
+
+	/*
+	 * PA_INIT_ERR on the device side will not trigger UIC
+	 * error interrupt.
+	 * Send NOP to check if the link is alive.
+	 */
+	spin_unlock_irqrestore(hba->host->host_lock, flags);
+	err = ufshcd_verify_dev_init(hba);
+	if (!err)
+		needs_reset = false;
+
+out:
+	return needs_reset;
+}
+
 void ufshcd_schedule_eh_work(struct ufs_hba *hba)
 {
 	lockdep_assert_held(hba->host->host_lock);
@@ -6655,6 +6695,11 @@ static void ufshcd_err_handler(struct work_struct *work)
 		if (!hba->saved_uic_err)
 			hba->saved_err &= ~UIC_ERROR;
 		spin_unlock_irqrestore(hba->host->host_lock, flags);
+		if (ufshcd_is_linereset_fatal(hba)) {
+			needs_reset = true;
+			spin_lock_irqsave(hba->host->host_lock, flags);
+			goto do_reset;
+		}
 		if (ufshcd_is_pwr_mode_restore_needed(hba))
 			needs_restore = true;
 		spin_lock_irqsave(hba->host->host_lock, flags);
-- 
2.45.2

Re: [PATCH 1/1] ufs: core: add fatal errors check for LINERESET

[Bart Van Assche]

On 5/23/25 3:10 AM, naomi.chu@mediatek.com wrote:
> +static bool ufshcd_is_linereset_fatal(struct ufs_hba *hba)
> +{
> +	bool needs_reset = true;
> +	unsigned long flags;
> +	int err;
> +
> +	spin_lock_irqsave(hba->host->host_lock, flags);
> +
> +	if (ufshcd_is_saved_err_fatal(hba)) {
> +		spin_unlock_irqrestore(hba->host->host_lock, flags);
> +		goto out;
> +	}
> +
> +	/*
> +	 * Wait for 100ms to ensure the PA_INIT flow is complete,
> +	 * and check for PA_INIT_ERR or other fatal errors.
> +	 */
> +	spin_unlock_irqrestore(hba->host->host_lock, flags);

Please use scoped_guard(spinlock_irqsave) instead of calling 
spin_lock_irqsave() and spin_unlock_irqrestore() directly.

 > +	msleep(100);

Can ufshcd_is_linereset_fatal() be called from IRQ context? If so,
calling msleep() is not allowed. If it can't be called from IRQ
context, saving and restoring 'flags' is not necessary.

Bart.