Re: [PATCH v3 0/2] scsi: libsas: Support spinup notification for SAS devices

Hi John

On 2026/8/4 15:35, John Garry wrote:
> On 03/08/2026 03:05, Xingui Yang wrote:
>> When a SAS device is in the Active_Wait or Idle_Wait power state, it
>> returns NOT_READY with ASC/ASCQ = 0x04/0x11 (notify (enable spinup)
>> required), indicating that a NOTIFY(ENABLE SPINUP) primitive is needed to
>> trigger media spinup.
>>
>> Without handling this condition, the SCSI mid-layer will indefinitely 
>> retry
>> the command with ACTION_DELAYED_RETRY, resulting in the disk never 
>> spinning
>> up and becoming unusable. A typical manifestation is:
>>
>>    sd 4:0:9:0: [sde] Spinning up disk...
>>    ...not responding...
>>    sd 4:0:9:0: [sde] Sense Key : Not Ready
>>    sd 4:0:9:0: [sde] Add. Sense: Logical unit not ready, notify 
>> (enable spinup) required
>>
>> To resolve this, the SAS controller needs to send a NOTIFY(ENABLE SPINUP)
>> primitive to the target phy, which transitions the device out of the
>> waiting state and allows normal spinup to proceed.
>>
> 
> How would other SAS HBAs which use libsas handle this scenario? Since 
> they have FW, would the FW automatically issue this NOTIFY(ENABLE SPINUP)?
> 
> I just wonder why hisi_sas seems to be only driver which would need this.

Thanks for the question. This affects only SAS HDDs. I checked pm8001, 
isci, and aic94xx — all handle NOTIFY(ENABLE SPINUP) at the driver level:

- pm8001: sends once at phy-up, then mdelay(200) — comment: "delay a 
moment to wait disk to spinup" — before notifying libsas.

- isci: enables hardware periodic insertion during link idle (ENABLE bit 
in notify_enable_spinup_control), cleared at phy stop.

- aic94xx: enables microcode periodic insertion during link idle 
(NOTIFY_TIMER_TIMEOUT = 500ms interval), stops at phy down.

hisi_sas also calls sl_notify_ssp() at phy-up, but the NOTIFY_EN bit is 
held for only 1ms — msleep(1) between setting and clearing — before 
immediately notifying libsas.

The root cause appears to be that 1ms is insufficient compared to 
pm8001's 200ms. An alternative to hisi_sas would be to simply increase 
the hold time in sl_notify_ssp() to match pm8001's approach, keeping the 
fix within hisi_sas.

Would you prefer this simpler approach, or do you still see value in the 
libsas-level sense detection and callback from hisi_sas? ^-^


Thanks,
Xingui Yang