Re: [PATCH] ata: ata_id_is_ssd() bugfix

[Jeff Garzik <jgarzik@pobox.com>]

Jens Axboe wrote:
> On Sun, Oct 19 2008, Bartlomiej Zolnierkiewicz wrote:
>> On Sunday 19 October 2008, Jens Axboe wrote:
>>> On Sat, Oct 18 2008, Bartlomiej Zolnierkiewicz wrote:
>>>> We need to explicitly check for major and minor version
>>>> of supported ATA spec as earlier revisions used word 217
>>>> for different purposes.
>>> What did they use to put in that word? Just curious if it does any harm,
>>>
>>> because as it stands, this patch will prevent ANY ssd from being
>>> correctly flagged as such. So I'm inclined to file this as too much spec
>>> fiddling, it'll do more harm than good.
>> IIRC it was marked as Reserved for < ATA8 (however I didn't go through
>> all previous specs) and some ATA8 minor versions (i.e. 3f) put "NV Cache
>> Read Transfer Speed in MB/s" there.
>>
>> Well, we may also drop minor versions checking assuming that no NV Cache
>> actually will have 1MB/s speed and major version checking assuming that
>> no vendor put anything special there but this would need an ACK from Alan
>> and maybe a comment on why we do it...
> 
> OK, so it's just NV cache. I'd consider that a non-issue.
> 
>>>> [ The issue was originally spotted by Alan Cox. ]
>>>>
>>>> This patch fixes regression introduced by:
>>>> commit 8bff7c6b0f63c7ee9c5e3a076338d74125b8debb
>>>> ("libata: set queue SSD flag for SSD devices").
>>>>
>>>> Cc: Jeff Garzik <jgarzik@pobox.com>
>>>> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
>>>> Cc: Jens Axboe <jens.axboe@oracle.com>
>>>> Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
>>>> ---
>>>> somebody owe me one for going through all these spec drafts... ;)
>>>>
>>>>  include/linux/ata.h |    8 +++++++-
>>>>  1 file changed, 7 insertions(+), 1 deletion(-)
>>>>
>>>> Index: b/include/linux/ata.h
>>>> ===================================================================
>>>> --- a/include/linux/ata.h
>>>> +++ b/include/linux/ata.h
>>>> @@ -75,6 +75,7 @@ enum {
>>>>  	ATA_ID_EIDE_PIO_IORDY	= 68,
>>>>  	ATA_ID_QUEUE_DEPTH	= 75,
>>>>  	ATA_ID_MAJOR_VER	= 80,
>>>> +	ATA_ID_MINOR_VER	= 81,
>>>>  	ATA_ID_COMMAND_SET_1	= 82,
>>>>  	ATA_ID_COMMAND_SET_2	= 83,
>>>>  	ATA_ID_CFSSE		= 84,
>>>> @@ -743,7 +744,12 @@ static inline int ata_id_is_cfa(const u1
>>>>  
>>>>  static inline int ata_id_is_ssd(const u16 *id)
>>>>  {
>>>> -	return id[ATA_ID_ROT_SPEED] == 0x01;
>>>> +	/* ATA8-ACS version 4c or higher (=> 4c or 6 at the moment) */
>>>> +	if (ata_id_major_version(id) >= 8 &&
>>>> +	    (id[ATA_ID_MINOR_VER] == 0x39 || id[ATA_ID_MINOR_VER] == 0x28) &&
>>>> +	    id[ATA_ID_ROT_SPEED] == 0x01)
>>>> +		return 1;
>>>> +	return 0;
>>> Is the check even correct? It'll match version 8 AND the currently
>>> listed minor version, not newer.
>> Checking minor versions is a non-trivial bussiness and improvements
>> are welcomed (though ATA8-ACS version 6 is the newest revision ATM).
>>
>> It could also be that the we should be checking something else than
>> the ATA_ID_ROT_SPEED to detect SSDs reliably but I don't know and don't
>> have time currently to look into it.  In the future please post ATA
> 
> Rotation speed == 1 is THE way to check for an SSD. The problem is just
> that lots of drives are out there and don't claim ATA8 compliance, since
> it was finalized until last month. Most of them SHOULD use word 217
> though, the ones I looked at sure do.

It seems highly unlikely that any drive that claims ATA8 would repurpose 
word 217, therefore upstream logic IMO should look like

	if (id[ATA_ID_ROT_SPEED] == 0x01 &&
	    ((ata_id_major_version(id) >= 8) ||
	     (a list of ATA6/7 exceptions that Jens wishes to supply)))

Regards,

	Jeff