The UEFI stub executes in the context of the firmware, which identity
maps the available system RAM, which implies that only memory below
4 GB can be used for allocations on 32-bit architectures, even on [L]PAE
capable hardware.
So ignore any reported memory above 4 GB in efi_random_alloc(). This
also fixes a reported build problem on ARM under -Os, where the 64-bit
logical shift relies on a software routine that the ARM decompressor does
not provide.
A second [minor] issue is also fixed, where the '+ 1' is moved out of
the shift, where it belongs: the reason for its presence is that a
memory region where start == end should count as a single slot, given
that 'end' takes the desired size and alignment of the allocation into
account.
To clarify the code in this regard, rename start/end to 'first_slot' and
'last_slot', respectively, and introduce 'region_end' to describe the
last usable address of the current region.
Reported-by: Arnd Bergmann <[email protected]>
Signed-off-by: Ard Biesheuvel <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>
unsigned long align_shift)
{
unsigned long align = 1UL << align_shift;
- u64 start, end;
+ u64 first_slot, last_slot, region_end;
if (md->type != EFI_CONVENTIONAL_MEMORY)
return 0;
- start = round_up(md->phys_addr, align);
- end = round_down(md->phys_addr + md->num_pages * EFI_PAGE_SIZE - size,
- align);
+ region_end = min((u64)ULONG_MAX, md->phys_addr + md->num_pages*EFI_PAGE_SIZE - 1);
- if (start > end)
+ first_slot = round_up(md->phys_addr, align);
+ last_slot = round_down(region_end - size + 1, align);
+
+ if (first_slot > last_slot)
return 0;
- return (end - start + 1) >> align_shift;
+ return ((unsigned long)(last_slot - first_slot) >> align_shift) + 1;
}
/*