On Mon, Dec 5, 2011 at 1:46 PM, Devin Heitmueller <dheitmueller@xxxxxxxxxxxxxx> wrote: > On Mon, Dec 5, 2011 at 1:35 PM, Mauro Carvalho Chehab > <mchehab@xxxxxxxxxx> wrote: >>> What's up with this change? Is this a bugfix for some race condition? >>> Why is it jammed into a patch for some particular product? >>> >>> It seems like a change such as this could significantly change the >>> timing of tuner initialization if you have multiple xc5000 based >>> products that might have a slow i2c bus. Was that intentional? >>> >>> This patch should be NACK'd and resubmitted as it's own bugfix where >>> it's implications can be fully understood in the context of all the >>> other products that use xc5000. >> >> >> It is too late for nacking the patch, as there are several other patches >> were already applied on the top of it, and we don't rebase the >> linux-media.git tree. >> >> Assuming that this is due to some bug that Eddi picked during xc5000 >> init, what it can be done now is to write a patch that would replace >> this xc5000-global mutex lock into a some other per-device locking >> schema. > > At this point we have zero idea why it's there *at all*. Eddi, can > you comment on what prompted this change? > > This patch should not have been accepted in the first place. It's an > undocumented change on a different driver than is advertised in the > subject line. Did you review the patch prior to merging? > > This change can result in a performance regression for all other > devices using xc5000, and it's not yet clear why it's there in the > first place. If its use cannot be explained then it should be rolled > back. If this breaks 930c, then the whole device support series > should be rolled back until somebody can figure out what is going on. > > It's crap like this that is the reason that every other week I get > complaints from some user that one of the drivers I wrote support for > worked fine for months/years until they upgraded to the latest kernel. Speaking of which, Mark Lord just tried out this change (he has an 800i and 950q - both xc5000 based), and now his DVB stack fails to load. And yes, he already has the fix to the mutex_unlock() regression which Dan Carpenter found six days ago and which this patch introduced. Devin -- Devin J. Heitmueller - Kernel Labs http://www.kernellabs.com -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
