Re: RFC: ATA to CAM integration patch

Alexander Motin wrote:
> Kamigishi Rei wrote:
>> Alexander Motin wrote:
>>> Sorry, my fault, it is the different bug, due to disabled invariants 
>>> I have missed two locking issues. Here is regenerated patch, 
>>> including this fix and all previous fixes and improvements:
>>> http://people.freebsd.org/~mav/cam-ata.20090629.patch
>>
>> OK, a question about the module: sometimes I'm getting loads (20+) of 
>> messages like
>>
>> ahcich1: ALL SLOTS BUSY!
>>
>> Is it purely informational or does it indicate a problem? (I'm using 
>> Intel Q35+ICH9 in AHCI mode.)
> 
> It is bad. It must never happen. It means that driver got more commands 
> then it has empty slots available. Results can be unpredictable.
> 
> How have you manage it? Were there any other messages around?

I think I have found possible bug triggering condition.
Apply this patch please:

--- ahci.c.prev 2009-06-29 12:48:45.000000000 +0300
+++ ahci.c      2009-06-29 17:25:29.000000000 +0300
_at__at_ -986,7 +986,7 _at__at_ ahci_begin_transaction(device_t dev, uni
                 if (ch->slot[tag].state == AHCI_SLOT_EMPTY)
                         break;
         } while (tag != ch->lastslot);
-       if (tag == ch->lastslot)
+       if (ch->slot[tag].state != AHCI_SLOT_EMPTY)
                 device_printf(ch->dev, "ALL SLOTS BUSY!\n");
         ch->lastslot = tag;
         /* Occupy chosen slot. */

-- 
Alexander Motin