diff options
author | Konstantin Belousov <kib@FreeBSD.org> | 2021-09-14 12:25:37 +0000 |
---|---|---|
committer | Konstantin Belousov <kib@FreeBSD.org> | 2021-09-16 17:01:19 +0000 |
commit | 181bfb42fd01bfa9f4636e803ccb3eeed8ac8ba4 (patch) | |
tree | 5b92f62ca49aa49d4d516d2624d84e66982ed47d /sys | |
parent | e8a872536042970b4dbf14dc75755a352fb14488 (diff) | |
download | src-181bfb42fd01bfa9f4636e803ccb3eeed8ac8ba4.tar.gz src-181bfb42fd01bfa9f4636e803ccb3eeed8ac8ba4.zip |
vm_phys: do not ignore phys_avail[] segments that do not fit completely into vm_phys segments
If phys_avail[] segment only intersect with some vm_phys segment, add
pages from it to the free list that belong to the given vm_phys_seg,
instead of dropping them.
The vm_phys segments are generally result of subdivision of phys_avail
segments, for instance DMA32 or LOWMEM boundaries split them. On
amd64, after UEFI in-place kernel activation (copy_staging disable)
was enabled, we typically have a large phys_avail[] segment below 4G
which crosses LOWMEM (1M) boundary. With the current way of requiring
phys_avail[] fully fit into vm_phys_seg, this memory was ignored.
Reported by: madpilot
Reviewed by: markj
Discussed with: alc
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D31958
Diffstat (limited to 'sys')
-rw-r--r-- | sys/vm/vm_page.c | 25 |
1 files changed, 12 insertions, 13 deletions
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c index 4ecea30e9219..d2e94ced6766 100644 --- a/sys/vm/vm_page.c +++ b/sys/vm/vm_page.c @@ -552,11 +552,12 @@ vm_offset_t vm_page_startup(vm_offset_t vaddr) { struct vm_phys_seg *seg; + struct vm_domain *vmd; vm_page_t m; char *list, *listend; vm_paddr_t end, high_avail, low_avail, new_end, size; vm_paddr_t page_range __unused; - vm_paddr_t last_pa, pa; + vm_paddr_t last_pa, pa, startp, endp; u_long pagecount; #if MINIDUMP_PAGE_TRACKING u_long vm_page_dump_size; @@ -770,21 +771,20 @@ vm_page_startup(vm_offset_t vaddr) vm_page_init_page(m, pa, segind); /* - * Add the segment to the free lists only if it is covered by - * one of the ranges in phys_avail. Because we've added the - * ranges to the vm_phys_segs array, we can assume that each - * segment is either entirely contained in one of the ranges, - * or doesn't overlap any of them. + * Add the segment's pages that are covered by one of + * phys_avail's ranges to the free lists. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) { - struct vm_domain *vmd; - - if (seg->start < phys_avail[i] || - seg->end > phys_avail[i + 1]) + if (seg->end < phys_avail[i] || + seg->start >= phys_avail[i + 1]) continue; - m = seg->first_page; - pagecount = (u_long)atop(seg->end - seg->start); + startp = MAX(seg->start, phys_avail[i]); + m = seg->first_page + atop(seg->start - startp); + endp = MIN(seg->end, phys_avail[i + 1]); + pagecount = (u_long)atop(endp - startp); + if (pagecount == 0) + continue; vmd = VM_DOMAIN(seg->domain); vm_domain_free_lock(vmd); @@ -796,7 +796,6 @@ vm_page_startup(vm_offset_t vaddr) vmd = VM_DOMAIN(seg->domain); vmd->vmd_page_count += (u_int)pagecount; vmd->vmd_segs |= 1UL << m->segind; - break; } } |