On 9/20/18 11:04 AM, Sergei Shtylyov wrote:
> On 09/20/2018 06:15 PM, Jens Axboe wrote:
>
>>>> When we're comparing the hardware completion mask passed in from the
>>>> driver with the internal tag pending mask, we need to account for the
>>>> fact that the internal tag is different from the hardware tag. If not,
>>>> then we can end up either prematurely completing the internal tag (since
>>>> it's not set in the hw mask), or simply flag an error:
>>>>
>>>> ata2: illegal qc_active transition (100000000->00000001)
>>>>
>>>> If the internal tag is set, then swap that with the hardware tag in this
>>>> case before comparing with what the hardware reports.
>>>>
>>>> Fixes: 28361c403683 ("libata: add extra internal command")
>>>> Buglink: https://bugzilla.kernel.org/show_bug.cgi?id=201151
>>>> Cc: stable@xxxxxxxxxxxxxxx
>>>> Reported-by: Paul Sbarra <sbarra.paul@xxxxxxxxx>
>>>> Tested-by: Paul Sbarra <sbarra.paul@xxxxxxxxx>
>>>> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
>>>>
>>>> ---
>>>>
>>>> diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
>>>> index 599e01bcdef2..a9dd4ea7467d 100644
>>>> --- a/drivers/ata/libata-core.c
>>>> +++ b/drivers/ata/libata-core.c
>>>> @@ -5359,10 +5359,20 @@ void ata_qc_complete(struct ata_queued_cmd *qc)
>>>> */
>>>> int ata_qc_complete_multiple(struct ata_port *ap, u64 qc_active)
>>>> {
>>>> + u64 done_mask, ap_qc_active = ap->qc_active;
>>>> int nr_done = 0;
>>>> - u64 done_mask;
>>>>
>>>> - done_mask = ap->qc_active ^ qc_active;
>>>> + /*
>>>> + * If the internal tag is set on ap->qc_active, then we care about
>>>> + * bit0 on the passed in qc_active mask. Move that bit up to match
>>>> + * the internal tag.
>>>> + */
>>>> + if (ap_qc_active & (1ULL << ATA_TAG_INTERNAL)) {
>>>
>>> Maybe BIT_ULL(ATA_TAG_INTERNAL)?
>>
>> Honestly, I had defines like that, since they hide what's going on, whereas
>> the above is immediately readable to anyone. If it isn't, they probably
>> shouldn't be fiddling with that code :-)
>>
>> Besides, we don't use that anywhere in libata currently. If someone has
>
> To be precise, BIT() does get used in <linux/ata.h> and once in libata-core.c.
> Also some libata drivers do use that macro.
I actually think it's fine for things like BIT(9) use, since it's easy to
verify with the spec where as 0x40 may not be. But it sounds like we
basically agree here :-)
--
Jens Axboe
