Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
x86/xen: Fix Xen PV guest boot
A recent patch moving the call of sparse_init() to common mm code
broke booting as a Xen PV guest.
Reason is that the Xen PV specific boot code relied on struct page area
being accessible rather early, but this changed by the move of the call
of sparse_init().
Fortunately the fix is rather easy: there is a static branch available
indicating whether struct page contents are usable by Xen. This static
branch just needs to be tested in some places for avoiding the access
of struct page.
Fixes: 4267739cabb8 ("arch, mm: consolidate initialization of SPARSE memory model")
Signed-off-by: Juergen Gross <jgross@suse.com>
Message-ID: <20260214135035.119357-1-jgross@suse.com>
+5
-1
+5
-1
arch/x86/xen/mmu_pv.c
+5
-1
arch/x86/xen/mmu_pv.c
···
509
509
unsigned offset = pgd - pgd_page;
510
510
pgd_t *user_ptr = NULL;
511
511
512
+
if (!static_branch_likely(&xen_struct_pages_ready))
513
+
return NULL;
514
+
512
515
if (offset < pgd_index(USER_LIMIT)) {
513
516
struct page *page = virt_to_page(pgd_page);
514
517
user_ptr = (pgd_t *)page->private;
···
1101
1098
1102
1099
if (unpin)
1103
1100
pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1104
-
ClearPagePinned(virt_to_page(__va(pa)));
1101
+
if (static_branch_likely(&xen_struct_pages_ready))
1102
+
ClearPagePinned(virt_to_page(__va(pa)));
1105
1103
xen_free_ro_pages(pa, PAGE_SIZE);
1106
1104
}
1107
1105