Hello,

Le lundi 20 mai 2013 20:59:29, Tejun Heo a écrit :
> Ugh... so, this is inherently racy between the probing code and admin.
> Maybe we should just implement a new libata.force param and forget
> about dynamic configuration?

Something like this ? (2/2 - untested)
Is "horkage" just another way to say "quirks" in this context ? Google 
translate doesn't help, and urbandictionary has too many entries for "hork" to 
make me confident.
Alternatively, I would add a "dflags" field to struct ata_force_param, and 
reuse ATA_DFLAG_DMADIR instead of defining a new enum item.

As this completely supersedes the atapi_dmadir module argument, is there a way 
to deprecate it (if at all a good practice) ?

> One more thing.  In the ata_exec_internal_sg(), DMADIR should be set
> iff DMA is being used, right?  So, it should also check tf->protocol.
> It prolly should test tf->protocol == ATAPI_PROT_DMA instead of cdb.

Sounds logical. I lack ATA[PI] background a lot, so I'm guessing a lot.
I updated the patch (1/2 - untested).

I should find the time to test both patches tomorrow.

Regards,
-- 
Vincent Pelletier