2021-01-12 12:13:54

by John Garry

Subject: Re: [PATCH v2 1/2] scsi: hisi_sas: Remove unnecessary devm_kfree

On 11/01/2021 23:10, Bean Huo wrote:
> From: Bean Huo <[email protected]>
>
> The memory allocated with devm_kzalloc() is freed automatically
> no need to explicitly call devm_kfree.
>

This change is not right - we use devm_kfree() to manually release the
devm-allocated debugfs memories upon memory allocation failure for
driver debugfs feature during probe. The reason is that we allow the
driver probe can still continue (for this failure).

Thanks,
John

> Signed-off-by: Bean Huo <[email protected]>
> ---
> drivers/scsi/hisi_sas/hisi_sas_v3_hw.c | 28 +-------------------------
> 1 file changed, 1 insertion(+), 27 deletions(-)
>
> diff --git a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> index 91a7286e8102..5600411a0820 100644
> --- a/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> +++ b/drivers/scsi/hisi_sas/hisi_sas_v3_hw.c
> @@ -4172,30 +4172,6 @@ static void debugfs_work_handler_v3_hw(struct work_struct *work)
> hisi_hba->debugfs_dump_index++;
> }
>
> -static void debugfs_release_v3_hw(struct hisi_hba *hisi_hba, int dump_index)
> -{
> - struct device *dev = hisi_hba->dev;
> - int i;
> -
> - devm_kfree(dev, hisi_hba->debugfs_iost_cache[dump_index].cache);
> - devm_kfree(dev, hisi_hba->debugfs_itct_cache[dump_index].cache);
> - devm_kfree(dev, hisi_hba->debugfs_iost[dump_index].iost);
> - devm_kfree(dev, hisi_hba->debugfs_itct[dump_index].itct);
> -
> - for (i = 0; i < hisi_hba->queue_count; i++)
> - devm_kfree(dev, hisi_hba->debugfs_dq[dump_index][i].hdr);
> -
> - for (i = 0; i < hisi_hba->queue_count; i++)
> - devm_kfree(dev,
> - hisi_hba->debugfs_cq[dump_index][i].complete_hdr);
> -
> - for (i = 0; i < DEBUGFS_REGS_NUM; i++)
> - devm_kfree(dev, hisi_hba->debugfs_regs[dump_index][i].data);
> -
> - for (i = 0; i < hisi_hba->n_phy; i++)
> - devm_kfree(dev, hisi_hba->debugfs_port_reg[dump_index][i].data);
> -}
> -
> static const struct hisi_sas_debugfs_reg *debugfs_reg_array_v3_hw[DEBUGFS_REGS_NUM] = {
> [DEBUGFS_GLOBAL] = &debugfs_global_reg,
> [DEBUGFS_AXI] = &debugfs_axi_reg,
> @@ -4206,7 +4182,7 @@ static int debugfs_alloc_v3_hw(struct hisi_hba *hisi_hba, int dump_index)
> {
> const struct hisi_sas_hw *hw = hisi_hba->hw;
> struct device *dev = hisi_hba->dev;
> - int p, c, d, r, i;
> + int p, c, d, r;
> size_t sz;
>
> for (r = 0; r < DEBUGFS_REGS_NUM; r++) {
> @@ -4286,8 +4262,6 @@ static int debugfs_alloc_v3_hw(struct hisi_hba *hisi_hba, int dump_index)
>
> return 0;
> fail:
> - for (i = 0; i < hisi_sas_debugfs_dump_count; i++)
> - debugfs_release_v3_hw(hisi_hba, i);
> return -ENOMEM;
> }
>
>

2021-01-12 12:20:07

by Bean Huo

Subject: Re: [PATCH v2 1/2] scsi: hisi_sas: Remove unnecessary devm_kfree

On Tue, 2021-01-12 at 09:03 +0000, John Garry wrote:
> On 11/01/2021 23:10, Bean Huo wrote:
> > From: Bean Huo <[email protected]>
> >
> > The memory allocated with devm_kzalloc() is freed automatically
> > no need to explicitly call devm_kfree.
> >
>
> This change is not right - we use devm_kfree() to manually release
> the
> devm-allocated debugfs memories upon memory allocation failure for
> driver debugfs feature during probe. The reason is that we allow the
> driver probe can still continue (for this failure).
>
> Thanks,
> John

yea, I see, probe didn't deal with ENOMEM error. will drop this change.
thanks.

Bean