Re: EFI tree kernel panic in phys_efi_set_virtual_address_map()

[Borislav Petkov <bp-Gina5bIWoIWzQB+pC5nmwQ@public.gmane.org>]

On Wed, Dec 11, 2013 at 05:01:03PM -0700, Toshi Kani wrote:
> Hi Boris,
> 
> An EFI tree kernel panic'd during boot on one of my systems.  It boots
> fine when efi=old_map option is specified.  So, I think it is caused by
> your EFI virtual mapping changes.
> 
> The panic message is as follows.  I added some printk's to log the
> arguments of phys_efi_set_virtual_address_map().  The fault address is
> __pa(new_memmap) + 0x20 (too high for the map?).
> 
> Thanks,
> -Toshi
> 
> 
> 
> efi: >> Call phys_efi_set_virtual_address_map()
> efi:    count 29
> efi:    desc_size 0x30
> efi:    new_memmap 0xffff8a03fec16800
> efi:    __pa(new_memmap) 0x203fec16800
> 
> BUG: unable to handle kernel paging request at 00000203fec16820
> IP: [<0000000072dcda76>] 0x72dcda75
> PGD 0 

It looks like the page hierarchy which contains __pa(new_memmap) is not
mapped in the EFI page table. Nice.

> Oops: 0000 [#1] SMP 
> Modules linked in:
> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 3.12.0+ #52
> Hardware name: HP CB920s x1, BIOS Bundle: 005.028.018 SFW: 012.124.000
> 10/28/2013
> task: ffffffff81a10480 ti: ffffffff81a00000 task.ti: ffffffff81a00000
> RIP: 0010:[<0000000072dcda76>]  [<0000000072dcda76>] 0x72dcda75
> RSP: 0000:ffffffff81a01e08  EFLAGS: 00010206
> RAX: 00000000725fee18 RBX: 00000000725feda0 RCX: 00000203fec16800
> RDX: 0000000072dfe070 RSI: 0000000060000202 RDI: 0000000072dcdac8
> RBP: 0000000072dd0560 R08: 0000000000000000 R09: 000000000000001d
> R10: 0000000000000030 R11: 8000000000000000 R12: ffff8a03fec16800
> R13: 0000000000000001 R14: 000000000000001d R15: 000000000009c000
> FS:  0000000000000000(0000) GS:ffff88087fa00000(0000)
> knlGS:0000000000000000
> CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00000203fec16820 CR3: 000000000009c000 CR4: 00000000000406b0
> Stack:
>  0000000072dfb88f 0000000000000001 0000000000000d08 0000000000000000
>  ffffffff81d79a98 0000000072dcdac8 0000000072dcdb93 0000000000000001
>  ffffffff81a01f80 0000000000000570 0000000000000000 0000000000000000
> Call Trace:
>  [<ffffffff81045cec>] ? efi_call4+0x6c/0xf0
>  [<ffffffff81b02f65>] ? efi_enter_virtual_mode+0x229/0x3f1
>  [<ffffffff81aebdf2>] ? start_kernel+0x36c/0x407
>  [<ffffffff81aeb88f>] ? repair_env_string+0x5c/0x5c
>  [<ffffffff81aeb5a3>] ? x86_64_start_reservations+0x2a/0x2c
>  [<ffffffff81aeb696>] ? x86_64_start_kernel+0xf1/0xf4
> Code:  Bad RIP value.
> RIP  [<0000000072dcda76>] 0x72dcda75
>  RSP <ffffffff81a01e08>
> CR2: 00000203fec16820
> ---[ end trace e50b25032c120443 ]---

Ok, it is late here and I'm almost blocked but you could try the dirty
patch below - more fiddling tomorrow.

In the meantime, can you send me full dmesg, the exact tree you're using
and your .config?

Thanks.

--
diff --git a/arch/x86/platform/efi/efi.c b/arch/x86/platform/efi/efi.c
index f8ec4dafc74e..254d3f2eff9f 100644
--- a/arch/x86/platform/efi/efi.c
+++ b/arch/x86/platform/efi/efi.c
@@ -245,8 +245,13 @@ static efi_status_t __init phys_efi_set_virtual_address_map(
 	u32 descriptor_version,
 	efi_memory_desc_t *virtual_map)
 {
+	pgd_t *pgd = (pgd_t *)__va(real_mode_header->trampoline_pgd);
 	efi_status_t status;
 
+	memcpy(pgd + pgd_index((unsigned long)virtual_map),
+		init_mm.pgd + pgd_index((unsigned long)virtual_map),
+		sizeof(pgd_t));
+
 	efi_call_phys_prelog();
 	status = efi_call_phys4(efi_phys.set_virtual_address_map,
 				memory_map_size, descriptor_size,


-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--