On 4/14/21 11:58 AM, Asutosh Das wrote:
> [ ... ]
Patches sent to the SCSI mailing list should not have a "scsi: " prefix
in the subject. That prefix is inserted before any SCSI patches go into
Martin's tree.
> diff --git a/drivers/scsi/ufs/cdns-pltfrm.c b/drivers/scsi/ufs/cdns-pltfrm.c
> index 13d9204..b9105e4 100644
> --- a/drivers/scsi/ufs/cdns-pltfrm.c
> +++ b/drivers/scsi/ufs/cdns-pltfrm.c
> @@ -323,6 +323,8 @@ static const struct dev_pm_ops cdns_ufs_dev_pm_ops = {
> .runtime_suspend = ufshcd_pltfrm_runtime_suspend,
> .runtime_resume = ufshcd_pltfrm_runtime_resume,
> .runtime_idle = ufshcd_pltfrm_runtime_idle,
> + .prepare = ufshcd_suspend_prepare,
> + .complete = ufshcd_resume_complete,
> };
>
> static struct platform_driver cdns_ufs_pltfrm_driver = {
> diff --git a/drivers/scsi/ufs/tc-dwc-g210-pci.c
> b/drivers/scsi/ufs/tc-dwc-g210-pci.c
> index 67a6a61..b01db12 100644
> --- a/drivers/scsi/ufs/tc-dwc-g210-pci.c
> +++ b/drivers/scsi/ufs/tc-dwc-g210-pci.c
> @@ -148,6 +148,8 @@ static const struct dev_pm_ops tc_dwc_g210_pci_pm_ops = {
> .runtime_suspend = tc_dwc_g210_pci_runtime_suspend,
> .runtime_resume = tc_dwc_g210_pci_runtime_resume,
> .runtime_idle = tc_dwc_g210_pci_runtime_idle,
> + .prepare = ufshcd_suspend_prepare,
> + .complete = ufshcd_resume_complete,
> };
[ ... ]
> --- a/drivers/scsi/ufs/ufs-exynos.c
> +++ b/drivers/scsi/ufs/ufs-exynos.c
> @@ -1267,6 +1267,8 @@ static const struct dev_pm_ops exynos_ufs_pm_ops = {
> .runtime_suspend = ufshcd_pltfrm_runtime_suspend,
> .runtime_resume = ufshcd_pltfrm_runtime_resume,
> .runtime_idle = ufshcd_pltfrm_runtime_idle,
> + .prepare = ufshcd_suspend_prepare,
> + .complete = ufshcd_resume_complete,
> };
>
> static struct platform_driver exynos_ufs_pltform = {
> diff --git a/drivers/scsi/ufs/ufs-hisi.c b/drivers/scsi/ufs/ufs-hisi.c
> index 0aa5813..d463b44 100644
> --- a/drivers/scsi/ufs/ufs-hisi.c
> +++ b/drivers/scsi/ufs/ufs-hisi.c
> @@ -574,6 +574,8 @@ static const struct dev_pm_ops ufs_hisi_pm_ops = {
> .runtime_suspend = ufshcd_pltfrm_runtime_suspend,
> .runtime_resume = ufshcd_pltfrm_runtime_resume,
> .runtime_idle = ufshcd_pltfrm_runtime_idle,
> + .prepare = ufshcd_suspend_prepare,
> + .complete = ufshcd_resume_complete,
> };
A minor comment about source code formatting: please make sure that the
equality signs are aligned in struct dev_pm_ops definitions.
> +static inline bool is_rpmb_wlun(struct scsi_device *sdev)
> +{
> + return (sdev->lun == ufshcd_upiu_wlun_to_scsi_wlun(UFS_UPIU_RPMB_WLUN));
> +}
> +
> +static inline bool is_device_wlun(struct scsi_device *sdev)
> +{
> + return (sdev->lun ==
> + ufshcd_upiu_wlun_to_scsi_wlun(UFS_UPIU_UFS_DEVICE_WLUN));
> +}
The Linux kernel coding style requires not to surround expressions with
parentheses in return statements.
> /**
> + * ufshcd_setup_links - associate link b/w device wlun and other luns
> + * @sdev: pointer to SCSI device
> + * @hba: pointer to ufs hba
> + */
> +static void ufshcd_setup_links(struct ufs_hba *hba, struct scsi_device *sdev)
> +{
> + struct device_link *link;
> +
> + /*
> + * device wlun is the supplier & rest of the luns are consumers
> + * This ensures that device wlun suspends after all other luns.
> + */
> + if (hba->sdev_ufs_device) {
> + link = device_link_add(&sdev->sdev_gendev,
> + &hba->sdev_ufs_device->sdev_gendev,
> + DL_FLAG_PM_RUNTIME|DL_FLAG_RPM_ACTIVE);
> + if (!link) {
> + dev_err(&sdev->sdev_gendev, "Failed establishing link -
> %s\n",
> + dev_name(&hba->sdev_ufs_device->sdev_gendev));
> + return;
> + }
> + hba->luns_avail--;
> + /* Ignore REPORT_LUN wlun probing */
> + if (hba->luns_avail == 1) {
> + ufshcd_rpm_put(hba);
> + return;
> + }
> + } else {
> + /* device wlun is probed */
> + hba->luns_avail--;
> + }
> +}
Please add a comment that explains that it is assumed that the WLUNs are
scanned before the other LUNs.
> @@ -4862,8 +4913,13 @@ static int ufshcd_slave_configure(struct scsi_device
> *sdev)
> blk_queue_update_dma_pad(q, PRDT_DATA_BYTE_COUNT_PAD - 1);
> if (hba->quirks & UFSHCD_QUIRK_ALIGN_SG_WITH_PAGE_SIZE)
> blk_queue_update_dma_alignment(q, PAGE_SIZE - 1);
> -
> - if (ufshcd_is_rpm_autosuspend_allowed(hba))
> + /*
> + * Block runtime-pm until all consumers are added.
> + * Refer ufshcd_setup_links().
> + */
> + if (is_device_wlun(sdev))
> + pm_runtime_get_noresume(&sdev->sdev_gendev);
> + else if (ufshcd_is_rpm_autosuspend_allowed(hba))
> sdev->rpm_autosuspend = 1;
>
> ufshcd_crypto_setup_rq_keyslot_manager(hba, q);
The following code is executed before ufshcd_async_scan() is called:
dev = hba->dev;
[ ... ]
/* Hold auto suspend until async scan completes */
pm_runtime_get_sync(dev);
and the following code occurs in ufshcd_add_lus():
pm_runtime_put_sync(hba->dev);
Isn't that sufficient to postpone enabling of runtime PM until LUN
scanning has finished? Or in other words, is adding a
pm_runtime_get_noresume() call in ufshcd_slave_configure() really necessary?
> @@ -4979,15 +5035,9 @@ ufshcd_transfer_rsp_status(struct ufs_hba *hba, struct
> ufshcd_lrb *lrbp)
> */
> if (!hba->pm_op_in_progress &&
> !ufshcd_eh_in_progress(hba) &&
> - ufshcd_is_exception_event(lrbp->ucd_rsp_ptr) &&
> - schedule_work(&hba->eeh_work)) {
> - /*
> - * Prevent suspend once eeh_work is scheduled
> - * to avoid deadlock between ufshcd_suspend
> - * and exception event handler.
> - */
> - pm_runtime_get_noresume(hba->dev);
> - }
> + ufshcd_is_exception_event(lrbp->ucd_rsp_ptr))
> + /* Flushed in suspend */
> + schedule_work(&hba->eeh_work);
What makes it safe to leave out the above pm_runtime_get_noresume() call?
Thanks,
Bart.
