Messages in this thread | | | Subject | Re: [GIT PULL] block updates for 2.6.31-rc1 | From | "Martin K. Petersen" <> | Date | Wed, 17 Jun 2009 18:49:15 -0400 |
| |
>>>>> "Andrew" == Andrew Morton <akpm@linux-foundation.org> writes:
Andrew> i386 allnoconfig:
Andrew> block/blk-settings.c: In function 'blk_set_default_limits': Andrew> block/blk-settings.c:115: warning: large integer implicitly Andrew> truncated to unsigned type
I sent Jens a fix earlier today.
Andrew> It doesn't look like it'll cause any runtime problems, but that Andrew> just means we got lucky.
I didn't do a compile check on 32-bit, I admit that. But the patch was explicitly done in a way that wouldn't affect the existing I/O path.
Andrew> Or is the code just buggy? What are the units of Andrew> BLK_BOUNCE_HIGH, BLK_BOUNCE_ANY and BLK_BOUNCE_ISA? Seems that Andrew> they are physical addresses. So why are we copying one of these Andrew> onto a variable which records pfns?
BLK_BOUNCE_* is the DMA mask, yes. It needs to be converted to pfns. That's what I messed up when I ripped the call out of DM to put it in the generic block layer function.
I'll defer to Jens wrt. the choice of -1ULL.
-- Martin K. Petersen Oracle Linux Engineering
| |