Re: [PATCH v4] x86/power/64: Fix kernel text mapping corruption during image restoration

From
Andy Lutomirski <luto@amacapital.net>
Date
2016-06-30 17:23:36
Also in
lkml
Reply to
Rafael J. Wysocki, 2016-06-30
On Thu, Jun 30, 2016 at 8:29 AM, Rafael J. Wysocki [off-list ref] wrote:
On Thu, Jun 30, 2016 at 5:24 PM, Andy Lutomirski [off-list ref] wrote:
quoted
On Thu, Jun 30, 2016 at 8:17 AM, Rafael J. Wysocki [off-list ref] wrote:
quoted
On Thu, Jun 30, 2016 at 5:05 PM, Borislav Petkov [off-list ref] wrote:
quoted
On Thu, Jun 30, 2016 at 03:17:20PM +0200, Rafael J. Wysocki wrote:
quoted
From: Rafael J. Wysocki <redacted>

Logan Gunthorpe reports that hibernation stopped working reliably for
him after commit ab76f7b4ab23 (x86/mm: Set NX on gap between __ex_table
and rodata).
...
quoted
+static int relocate_restore_code(void)
+{
+     pgd_t *pgd;
+     pud_t *pud;
+
+     relocated_restore_code = get_safe_page(GFP_ATOMIC);
+     if (!relocated_restore_code)
+             return -ENOMEM;
+
+     memcpy((void *)relocated_restore_code, &core_restore_code, PAGE_SIZE);
+
+     /* Make the page containing the relocated code executable */
+     pgd = (pgd_t *)__va(read_cr3()) + pgd_index(relocated_restore_code);
+     pud = pud_offset(pgd, relocated_restore_code);
+     if (pud_large(*pud)) {
+             set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
+     } else {
+             pmd_t *pmd = pmd_offset(pud, relocated_restore_code);
+
+             if (pmd_large(*pmd)) {
+                     set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
+             } else {
+                     pte_t *pte = pte_offset_kernel(pmd, relocated_restore_code);
+
+                     set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
+             }
+     }
+     flush_tlb_all();
I know you want to flush TLBs but this causes the splat below on the
resume kernel.

Most likely because:

resume_target_kernel() does local_irq_disable() and then

swsusp_arch_resume() -> relocate_restore_code() -> flush_tlb_all()

and smp_call_function_many() doesn't like it when IRQs are disabled.
Right.

Can I invoke __flush_tlb_all() from here to flush the TLB on the local
CPU?  That should be sufficient IMO.
Yes, unless there's another CPU already up at the time, but that seems unlikely.
They are all offline at that time.

OK, let me do that then.
quoted
FWIW, the CPU can and will cache the NX bit and the CPU will *not*
refresh the TLB before sending a page fault.  You definitely need to
flush.
OK
quoted
 I wonder if this is also why you're seeing a certain amount of
random corruption.
Well, the corruption we saw was not really random.  I can give you all
of the details if you're interested, maybe you'll be able to figure
out what might be happening. :-)
Sure, or at least in abbreviated form.  No guarantee that I'll have
any clue, though.

--Andy
Thread (36 messages) 36 messages, 8 authors, last reply 2016-06-30
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help