git: 181bfb42fd01 - main - vm_phys: do not ignore phys_avail[] segments that do not fit completely into vm_phys segments

[Konstantin Belousov]

The branch main has been updated by kib:

URL: https://cgit.FreeBSD.org/src/commit/?id=181bfb42fd01bfa9f4636e803ccb3eeed8ac8ba4

commit 181bfb42fd01bfa9f4636e803ccb3eeed8ac8ba4
Author:     Konstantin Belousov <kib at FreeBSD.org>
AuthorDate: 2021-09-14 12:25:37 +0000
Commit:     Konstantin Belousov <kib at FreeBSD.org>
CommitDate: 2021-09-16 17:01:19 +0000

    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
---
 sys/vm/vm_page.c | 25 ++++++++++++-------------
 1 file 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;
 		}
 	}