Re: [RFC PATCH v3 1/1] scsi: ufs: Enable power management for wlun

[Asutosh Das]

On Tue, Feb 16 2021 at 09:44 -0800, Asutosh Das wrote:
> On Sat, Feb 13 2021 at 13:37 -0800, Avri Altman wrote:
> > > +       } else {
> > Is it possible to get here?
> > Scsi_scan_host is called only after successful add_wluns
>
> It looks so.
> scsi 0:0:0:49488: Link setup for lun - ufshcd_setup_links
> [...]
> Call trace:
> dump_backtrace+0x0/0x1d4
> show_stack+0x18/0x24
> dump_stack+0xc4/0x144
> ufshcd_setup_links+0xd8/0x100
> ufshcd_slave_alloc+0x134/0x1a0
> scsi_alloc_sdev+0x1c0/0x230
> scsi_probe_and_add_lun+0xc0/0xd48
> __scsi_add_device+0xc0/0x138
> ufshcd_scsi_add_wlus+0x30/0x1c0
> ufshcd_async_scan+0x58/0x240
> async_run_entry_fn+0x48/0x128
> process_one_work+0x1f0/0x470
> worker_thread+0x26c/0x4c8
> kthread+0x13c/0x320
> ret_from_fork+0x10/0x18
>
> > > +               /* device wlun is probed */
> > > +               hba->luns_avail--;
> > > +       }
> > > +}
> > > +
> >
> >
> > > /**
> > > @@ -7254,6 +7312,14 @@ static int ufshcd_scsi_add_wlus(struct ufs_hba
> > > *hba)
> > >                goto out;
> > >        }
> > >        ufshcd_blk_pm_runtime_init(hba->sdev_ufs_device);
> > > +       /*
> > > +        * A pm_runtime_put_sync is invoked when this device enables
> > > blk_pm_runtime
> > > +        * & would suspend the device-wlun upon timer expiry.
> > > +        * But suspending device wlun _may_ put the ufs device in the pre-defined
> > > +        * low power mode (SSU <rpm_lvl>). Probing of other luns may fail then.
> > > +        * Don't allow this suspend until all the luns have been probed.
> > Maybe add one more sentence: see pm_runtime_mark_last_busy in ufshcd_setup_links
> Done.
>
> > > -       ufshcd_clear_ua_wluns(hba);
> > Are there any callers left to ufshcd_clear_ua_wluns?
> > Can it be removed?
> Let me check.
I don't think this can be removed.
The reasoning behind this call as per the commit message indicates that if
there's a reset this request_sense is needed to clear uac.

In pm level 5, the reset would still happen. So I guess this is needed.
Please let me know if I'm missing something here.
The commit message didn't have much details otherwise.



> > > +       if (hba->wlun_dev_clr_ua)
> > > +               ufshcd_clear_ua_wlun(hba, UFS_UPIU_UFS_DEVICE_WLUN);
> > >
> > >        cmd[4] = pwr_mode << 4;