slaunch/txt: fix missing 64-bit arithmetic in MTRR calculation

[accek-itl]

Make the range sizes 64-bit: the 32-bit ~({mtrr_s,pages_in_range} - 1) dropped PHYSMASK bits above bit 43 on parts with MAXPHYADDR >= 45.

The broken calculations (bits above 31 truncated to 0) are here:

• https://github.com/TrenchBoot/grub/compare/tb-2.12-284-v4...accek-itl:grub:ptl-senter-fixes?expand=1#diff-beefc3492ec6895088ab2da67a2b0f5888410590ef4f0c02e321904463224a55L427
• https://github.com/TrenchBoot/grub/compare/tb-2.12-284-v4...accek-itl:grub:ptl-senter-fixes?expand=1#diff-beefc3492ec6895088ab2da67a2b0f5888410590ef4f0c02e321904463224a55L456

Possible alternative version if you prefer:

diff --git a/grub-core/loader/slaunch/txt.c b/grub-core/loader/slaunch/txt.c
index 43f82d5030c5..ece0902c8e6e 100644
--- a/grub-core/loader/slaunch/txt.c
+++ b/grub-core/loader/slaunch/txt.c
@@ -424,7 +424,7 @@ set_mtrr_mem_type (struct grub_txt_acm_header *sinit, grub_uint32_t mem_type)
       grub_wrmsr (GRUB_MSR_X86_MTRR_PHYSBASE0 + ndx*2, mtrr_physbase.raw);
 
       mtrr_physmask.raw = grub_rdmsr (GRUB_MSR_X86_MTRR_PHYSMASK0 + ndx*2);
-      mtrr_physmask.mask = ~(mtrr_s - 1) & mtrr_shifted_mask;
+      mtrr_physmask.mask = ~((grub_uint64_t) mtrr_s - 1) & mtrr_shifted_mask;
       mtrr_physmask.v = 1;
       grub_wrmsr (GRUB_MSR_X86_MTRR_PHYSMASK0 + ndx*2, mtrr_physmask.raw);
 
@@ -453,7 +453,7 @@ set_mtrr_mem_type (struct grub_txt_acm_header *sinit, grub_uint32_t mem_type)
       pages_in_range = 1 << (fls (num_pages) - 1);
 
       mtrr_physmask.raw = grub_rdmsr (GRUB_MSR_X86_MTRR_PHYSMASK0 + ndx*2);
-      mtrr_physmask.mask = ~(pages_in_range - 1) & mtrr_shifted_mask;
+      mtrr_physmask.mask = ~((grub_uint64_t) pages_in_range - 1) & mtrr_shifted_mask;
       mtrr_physmask.v = 1;
       grub_wrmsr (GRUB_MSR_X86_MTRR_PHYSMASK0 + ndx*2, mtrr_physmask.raw);

[SergiiDmytruk]

Made tb-upstream-next branch at the same commit as tb-2.12-284-v4 and updated the target branch here. v4 patchset has been published, so its branch shouldn't really change. I remember suggesting to use any branch, but a new one is actually better.

Possible alternative version if you prefer:

I think this version is better. The code makes a reasonable assumption that the number of pages within ranges fits into 32-bit integers. Changing types of those variables would have to also update how their values are computed to be consistent. Things go wrong only in those two expressions computing the masks and it makes sense to indicate their 64-bit nature explicitly.

[accek-itl]

Ok, I agree, adjusted commit to the alternative version.