Fix swap pager readahead after r292373.
ClosedPublic
Actions

Authored by markj on Aug 28 2016, 8:32 PM.

Details

Reviewers

kib
alc
glebius

Commits

rS305056: Restore swap pager readahead after r292373.

Summary

The removal of vm_fault_additional_pages() means that a hard fault on a
OBJT_SWAP-backed page will cause exactly one page to be read in. This
change restores readahead and readbehind for the swap pager in
swap_pager_getpages(). haspages() is modified to return the largest
contiguous non-resident range of pages in the swap block containing the
requested range.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

markj updated this revision to Diff 19764.Aug 28 2016, 8:32 PM

markj retitled this revision from to Fix swap pager readahead after r292373..

markj edited the test plan for this revision. (Show Details)

markj updated this object.

Herald added a subscriber: imp. · View Herald TranscriptAug 28 2016, 8:32 PM

markj added reviewers: alc, glebius, kib.Aug 28 2016, 8:35 PM

alc added inline comments.Aug 28 2016, 8:46 PM

sys/vm/swap_pager.c
1020 ↗	(On Diff #19764)	Since you're already editing this line for functional reasons, I suggest changing the "++i" to the more idiomatic "i++".
1034 ↗	(On Diff #19764)	Same here.

This change looks correct to me. Thanks for doing it.

This revision is now accepted and ready to land.Aug 28 2016, 9:03 PM

Style fix.

This revision now requires review to proceed.Aug 28 2016, 9:59 PM

kib accepted this revision.Aug 29 2016, 4:22 AM

kib edited edge metadata.

kib added inline comments.

sys/vm/swap_pager.c
1093 ↗	(On Diff #19766)	May be, assert that the object paging count > 0 there ? I wonder should we increment it locally instead of relying on the caller.

This revision is now accepted and ready to land.Aug 29 2016, 4:22 AM

markj added inline comments.Aug 29 2016, 4:42 AM

sys/vm/swap_pager.c
1093 ↗	(On Diff #19766)	Hm, I deleted the comment that references the pip reference because I thought it wasn't true in all cases. For instance, a tmpfs read may page in from swap via uiomove_object_page(), but it looks like the paging count will be 0 in this case, and the stability of the backing store is instead provided by the vnode lock. Is this incorrect?