Re: Regression on linux-next (next-20250708)

[Damien Le Moal <dlemoal@kernel.org>]

On 7/29/25 01:24, Borah, Chaitanya Kumar wrote:
> 
> 
> On 7/28/2025 9:41 AM, Damien Le Moal wrote:
>> On 7/25/25 3:43 PM, Borah, Chaitanya Kumar wrote:
>>> For some context in our kms_pm_rpm tests, we enable min_power policy for SATA
>>> so that we can reach deep runtime power states and restore the original policy
>>> after finishing. [5][6]
>>>
>>> IIUC, the above change is based on spec and not something which can be
>>> reverted. So as I see it, we have to drop this code path for external ports.
>>> However I am not sure if we can achieve deep power states without enforcing it
>>> through the sysfs entry.
>>>
>>> Atleast for the basic-rte subtest, the test passes if we comment out the
>>> functions controlling the SATA ports. We will need more testing to determine if
>>> this approach work. Any thoughts on it?
>>>
>>> Also, are there other ways to detect a port is external other than receiving
>>> EOPNOTSUPP on the sysfs write?
>>
>> The attached patch adds the "link_power_management_supported" sysfs device
>> attribute for drives connected to AHCI. Would that work for you ?
>>
> 
> Yes this could work. I quickly hacked the test to ignore writing policy 
> if this file returns 0.
> 
> Here is the state of the machine I am testing on.
> 
> /sys/class/scsi_host/host0/link_power_management_supported: 0
> /sys/class/scsi_host/host1/link_power_management_supported: 0
> /sys/class/scsi_host/host2/link_power_management_supported: 0
> /sys/class/scsi_host/host3/link_power_management_supported: 0
> /sys/class/scsi_host/host4/link_power_management_supported: 1
> /sys/class/scsi_host/host5/link_power_management_supported: 1
> /sys/class/scsi_host/host6/link_power_management_supported: 1
> /sys/class/scsi_host/host7/link_power_management_supported: 1

Looks good. My test machine looks exactly like this too.
I will send out this patch as it is useful anyway regardless of external/hotplug
ports since not all adapters/drives support LPM.

When posting this, can I tag it as a solution to the regression ?

And in addition to this patch, I will work on a flexible way of ignoring hotplug.


-- 
Damien Le Moal
Western Digital Research