quoted

Thanks for this v2.
Introducing the 'efi_switch_mm() ' helper instead of manually
twiddling with %cr3 seems more cleaner.

I have tested this patchset on a x86_64 machine with the following
configurations:

1. Primary kernel boot with efi=old_map 2. Primary kernel boot with
new efi map

While it seems that efi=old_map passes, the new efi map boot fails for
me on both the two x86 machine (Dell 3050MT and a SGI - UV300 machine.

It seems we are hitting a NULL pointer deference in
'efi_call_phys_prolog' while accessing '&efi_mm'.

Please see the log below for details:

[    0.020006] BUG: unable to handle kernel NULL pointer dereference
at           (null)
[    0.021000] IP: switch_mm_irqs_off+0x44/0x270
[    0.021000] Call Trace:
[    0.021000]  switch_mm+0x20/0x30
[    0.021000]  efi_switch_mm+0x49/0x60
[    0.021000]  efi_call_phys_prolog+0x56/0x1b3
[    0.021000]  efi_enter_virtual_mode+0x3a9/0x520
[    0.021000]  start_kernel+0x424/0x4c8
[    0.021000]  ? set_init_arg+0x5a/0x5a
[    0.021000]  ? early_idt_handler_array+0x120/0x120
[    0.021000]  x86_64_start_reservations+0x29/0x2b
[    0.021000]  x86_64_start_kernel+0x151/0x174
[    0.021000]  secondary_startup_64+0x9f/0x9f
[    0.021000] Code: 2d 82 51 d9 4f 65 c7 05 0f 65 da 4f 01 00 00 00
48 39 f7 0f 84 14 01 00 00 65 48 89 35 f6 64 da 4f 48 8b 86 e8 02 00
00 45 89 ed <f0> 4c 0f ab 28 bf 00 00 00 80 48 03 7e 50 48 8b 05 27 b0
b9 00
[    0.021000] RIP: switch_mm_irqs_off+0x44/0x270 RSP: ffffffffb0e035d0
[    0.021000] CR2: 0000000000000000
[    0.021000] ---[ end trace fb94349305e1cb8b ]---
[    0.021000] Kernel panic - not syncing: Fatal exception
[    0.021000] ---[ end Kernel panic - not syncing: Fatal exception

And I forgot to mention that I tried the patchset both with the efi/next and
linus's trees and saw the same result.

I would be happy to help in case you need further details of the test environment
or need help in testing this crash further.

Regards,
Bhupesh

Hi Bhupesh,

Thanks for trying the patches and raising the concern.
Could you also please also give a try on qemu? (if reproducible, we will be having a common platform to start debugging)
I have tested this patch set on qemu and real machines (different from one's you tried) in our lab and didn’t notice this issue.

Regards,
Sai