Re: [RFC 2/2] ata: ahci: Protect users from setting policies their drives don't support

[Hans de Goede]

Hi,

On 2/25/22 22:24, Limonciello, Mario wrote:
> [AMD Official Use Only]
> 
> 
> 
>> -----Original Message-----
>> From: Hans de Goede <hdegoede@xxxxxxxxxx>
>> Sent: Friday, February 25, 2022 15:20
>> To: Limonciello, Mario <Mario.Limonciello@xxxxxxx>; Damien Le Moal
>> <damien.lemoal@xxxxxxxxxxxxxxxxxx>
>> Cc: open list:LIBATA SUBSYSTEM (Serial and Parallel ATA drivers) <linux-
>> ide@xxxxxxxxxxxxxxx>; open list <linux-kernel@xxxxxxxxxxxxxxx>
>> Subject: Re: [RFC 2/2] ata: ahci: Protect users from setting policies their
>> drives don't support
>>
>> Hi,
>>
>> On 2/25/22 19:10, Mario Limonciello wrote:
>>> As the default low power policy applies to more chipsets and drives, it's
>>> important to make sure that drives actually support the policy that a user
>>> selected in their kernel configuration.
>>>
>>> If the drive doesn't support slumber, don't let the default policy for the
>>> ATA port be `min_power` or `min_power_with_partial`.
>>>
>>> Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx>
>>> ---
>>>  drivers/ata/ahci.c | 8 ++++++++
>>>  1 file changed, 8 insertions(+)
>>>
>>> diff --git a/drivers/ata/ahci.c b/drivers/ata/ahci.c
>>> index 17d757ad7111..af8999453084 100644
>>> --- a/drivers/ata/ahci.c
>>> +++ b/drivers/ata/ahci.c
>>> @@ -1584,8 +1584,16 @@ static int ahci_init_msi(struct pci_dev *pdev,
>> unsigned int n_ports,
>>>  static void ahci_update_initial_lpm_policy(struct ata_port *ap,
>>>  					   struct ahci_host_priv *hpriv)
>>>  {
>>> +	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
>>>  	int policy = CONFIG_SATA_LPM_POLICY;
>>>
>>> +	if (policy >= ATA_LPM_MIN_POWER_WITH_PARTIAL &&
>>> +	   !(hpriv->cap & HOST_CAP_SSC)) {
>>> +		dev_warn(&pdev->dev,
>>> +			 "This drive doesn't support slumber; ignoring default
>> SATA policy\n");
>>> +		return;
>>> +	}
>>> +
>>
>> Don't the capabilties get checked later when the policy gets applied ?
>>
>> At least I think they do get checked later, but I have not looked
>> at this code for years  ...  ?
> 
> There's a bunch of layers of indirection so I might have missed something,
> but I didn't see anything in sata_link_scr_lpm or anywhere else for that
> matter that actually checked HOST_CAP_SSC.

Hmm, ok. Note that the user can still change the policy with an echo
to sysfs. So I think it would be better to do a fix where HOST_CAP_SSC
gets checked when the features are actually being enabled. Or at least
also at a HOST_CAP_SSC check to the sysfs write functions.

Regards,

Hans


>>>  	/* user modified policy via module param */
>>>  	if (mobile_lpm_policy != -1) {
>>>  		policy = mobile_lpm_policy;