When PTI is enabled on x86-32 the kernel uses the GDT mapped in the fixmap
for the simple reason that this address is also mapped for user-space.
The efi_call_phys_prolog()/efi_call_phys_epilog() wrappers change the GDT
to call EFI runtime services and switch back to the kernel GDT when they
return. But the switch-back uses the writable GDT, not the fixmap GDT.
When that happened and and the CPU returns to user-space it switches to the
user %cr3 and tries to restore user segment registers. This fails because
the writable GDT is not mapped in the user page-table, and without a GDT
the fault handlers also can't be launched. The result is a triple fault and
reboot of the machine.
Fix that by restoring the GDT back to the fixmap GDT which is also mapped
in the user page-table.
Fixes: 7757d607c6b3 x86/pti: ('Allow CONFIG_PAGE_TABLE_ISOLATION for x86_32')
Reported-by: Guenter Roeck <[email protected]>
Signed-off-by: Joerg Roedel <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Tested-by: Guenter Roeck <[email protected]>
Cc: Ard Biesheuvel <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Pavel Machek <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: https://lkml.kernel.org/r/[email protected]
void __init efi_call_phys_epilog(pgd_t *save_pgd)
{
- struct desc_ptr gdt_descr;
-
- gdt_descr.address = (unsigned long)get_cpu_gdt_rw(0);
- gdt_descr.size = GDT_SIZE - 1;
- load_gdt(&gdt_descr);
-
load_cr3(save_pgd);
__flush_tlb_all();
+
+ load_fixmap_gdt(0);
}
void __init efi_runtime_update_mappings(void)