Skip to contentSkip to conversation

swiotlb_sync_sg() issue?

DISCUSSION Jan Beulich 1 participant 2 messages Dec 19, 2006

lore
Share & tools
lore
#0 lore raw

Due to its apparent obviousness, I'm afraid I'm missing something, but isn't it incorrect that lib/swiotlb.c:swiotlb_sync_sg() calls sync_single() with sg->dma_address simply cast to void* rather than going through phys_to_virt()?

Also, for ia64, isn't mark_clean() violating its own comment, i.e. to mark only *complete* pages? And then, for completeness, isn't swiotlb_sync_sg() missing the call to mark_clean()?

Thanks, Jan

#1 lore raw
Quoted 8 lines
Quoted 1 line
Quoted 1 line

"Jan Beulich" <[email protected]> 19.12.06 12:10 >>>

Due to its apparent obviousness, I'm afraid I'm missing something, but isn't it incorrect that lib/swiotlb.c:swiotlb_sync_sg() calls sync_single() with sg->dma_address simply cast to void* rather than going through phys_to_virt()?

Also, for ia64, isn't mark_clean() violating its own comment, i.e. to mark only *complete* pages? And then, for completeness, isn't swiotlb_sync_sg() missing the call to mark_clean()?

Okay, I see I mis-read mark_clean(), it's not violating what it claims to do. But the other two questions remain.

Thanks, Jan