2023-12-13 02:26:12

by Chanwoo Lee

[permalink] [raw]
Subject: [PATCH] scsi: ufs: qcom: Re-fix for error handling

From: ChanWoo Lee <[email protected]>

I modified the code to handle errors.

The error handling code has been changed from the patch below.
-'commit 031312dbc695 ("scsi: ufs: ufs-qcom: Remove unnecessary goto statements")'

What I have confirmed are three cases.
1) ufs_qcom_host_reset -> 'reset_control_deassert' error -> return 0;
2) ufs_qcom_clk_scale_notify -> 'ufs_qcom_clk_scale_up_/down_pre_change' error -> return 0;
3) ufs_qcom_init_lane_clks -> 'ufs_qcom_host_clk_get(tx_lane1_sync_clk)' error -> return 0;

It is unknown whether the above commit was intended to change error handling.
However, if it is not an intended fix, a patch may be needed.

Signed-off-by: ChanWoo Lee <[email protected]>
---
drivers/ufs/host/ufs-qcom.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
index 96cb8b5b4e66..8a93d93ab08f 100644
--- a/drivers/ufs/host/ufs-qcom.c
+++ b/drivers/ufs/host/ufs-qcom.c
@@ -313,6 +313,8 @@ static int ufs_qcom_init_lane_clks(struct ufs_qcom_host *host)

err = ufs_qcom_host_clk_get(dev, "tx_lane1_sync_clk",
&host->tx_l1_sync_clk, true);
+ if (err)
+ return err;
}

return 0;
@@ -404,9 +406,11 @@ static int ufs_qcom_host_reset(struct ufs_hba *hba)
usleep_range(200, 210);

ret = reset_control_deassert(host->core_reset);
- if (ret)
+ if (ret) {
dev_err(hba->dev, "%s: core_reset deassert failed, err = %d\n",
__func__, ret);
+ return ret;
+ }

usleep_range(1000, 1100);

@@ -415,7 +419,7 @@ static int ufs_qcom_host_reset(struct ufs_hba *hba)
hba->is_irq_enabled = true;
}

- return 0;
+ return ret;
}

static u32 ufs_qcom_get_hs_gear(struct ufs_hba *hba)
@@ -1535,7 +1539,7 @@ static int ufs_qcom_clk_scale_notify(struct ufs_hba *hba,
ufshcd_uic_hibern8_exit(hba);
}

- return 0;
+ return err;
}

static void ufs_qcom_enable_test_bus(struct ufs_qcom_host *host)
--
2.29.0


2023-12-13 17:17:41

by Andrew Halaney

[permalink] [raw]
Subject: Re: [PATCH] scsi: ufs: qcom: Re-fix for error handling

On Wed, Dec 13, 2023 at 11:25:00AM +0900, Chanwoo Lee wrote:
> From: ChanWoo Lee <[email protected]>
>
> I modified the code to handle errors.
>
> The error handling code has been changed from the patch below.
> -'commit 031312dbc695 ("scsi: ufs: ufs-qcom: Remove unnecessary goto statements")'
>
> What I have confirmed are three cases.
> 1) ufs_qcom_host_reset -> 'reset_control_deassert' error -> return 0;
> 2) ufs_qcom_clk_scale_notify -> 'ufs_qcom_clk_scale_up_/down_pre_change' error -> return 0;
> 3) ufs_qcom_init_lane_clks -> 'ufs_qcom_host_clk_get(tx_lane1_sync_clk)' error -> return 0;
>
> It is unknown whether the above commit was intended to change error handling.
> However, if it is not an intended fix, a patch may be needed.

I think you're right, these were not intentionally changed. There's a
patch series in flight right now that cleans up some of this driver and
inadvertently tackles some of the problems below.

>
> Signed-off-by: ChanWoo Lee <[email protected]>
> ---
> drivers/ufs/host/ufs-qcom.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
> index 96cb8b5b4e66..8a93d93ab08f 100644
> --- a/drivers/ufs/host/ufs-qcom.c
> +++ b/drivers/ufs/host/ufs-qcom.c
> @@ -313,6 +313,8 @@ static int ufs_qcom_init_lane_clks(struct ufs_qcom_host *host)
>
> err = ufs_qcom_host_clk_get(dev, "tx_lane1_sync_clk",
> &host->tx_l1_sync_clk, true);
> + if (err)
> + return err;

This patch cleans this up: https://lore.kernel.org/linux-arm-msm/[email protected]/

> }
>
> return 0;
> @@ -404,9 +406,11 @@ static int ufs_qcom_host_reset(struct ufs_hba *hba)
> usleep_range(200, 210);
>
> ret = reset_control_deassert(host->core_reset);
> - if (ret)
> + if (ret) {
> dev_err(hba->dev, "%s: core_reset deassert failed, err = %d\n",
> __func__, ret);
> + return ret;
> + }

This patch cleans this up: https://lore.kernel.org/linux-arm-msm/[email protected]/#t

>
> usleep_range(1000, 1100);
>
> @@ -415,7 +419,7 @@ static int ufs_qcom_host_reset(struct ufs_hba *hba)
> hba->is_irq_enabled = true;
> }
>
> - return 0;
> + return ret;

If I'm reading right returning ret is pointless here with your change
above (it already returns ret, and it is no longer updated right so the
only possible value here is 0?

> }
>
> static u32 ufs_qcom_get_hs_gear(struct ufs_hba *hba)
> @@ -1535,7 +1539,7 @@ static int ufs_qcom_clk_scale_notify(struct ufs_hba *hba,
> ufshcd_uic_hibern8_exit(hba);
> }
>
> - return 0;
> + return err;
> }

I think you could move this one up into the PRE_CHANGE block and leave
return 0 here? I believe this is the only case not yet covered by the
patch series I linked. Good catch!

>
> static void ufs_qcom_enable_test_bus(struct ufs_qcom_host *host)
> --
> 2.29.0
>
>