x86/tsc: Fix native_sched_clock()
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest2 patches 4 repliesIn review
Patches in v1 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/2 | x86/tsc: Force inlining of cyc2ns bitsarch/ | +3 | 0 replies | |
| 2/2 | x86/percpu: Fix this_cpu_read()arch/ | +4 | 4 replies |
Reviews trailers per patch
1 of 2 patches has a Reviewed-by or Acked-by.
| Patch | Eric Dumazet |
|---|---|
| 1/2 | |
| 2/2 | A-b |
Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.
Cover letter [PATCH 0/2] x86/tsc: Fix native_sched_clock()
Eric reported that native_sched_clock() gets miscompiled; we loose a seqcount loop. These patches cure that.
After these patches:
native_sched_clock:
pushq %rbp # movq %rsp, %rbp #, ... jump label ... rdtsc salq $32, %rdx #, tmp110 orq %rax, %rdx # low, tmp110 movq %rdx, %r10 # tmp110, _23 movq $cyc2ns, %r9 #, tmp136 .L235: movl %gs:cyc2ns+32(%rip),%eax # cyc2ns.seq.sequence, pfo_ret__ movl %eax, %ecx # pfo_ret__, idx andl $1, %ecx #, idx salq $4, %rcx #, tmp116 addq %r9, %rcx # tmp136, tmp117 movq %gs:8(%rcx),%rdi # cyc2ns.data[idx_14].cyc2ns_offset, pfo_ret__ movl %gs:(%rcx),%esi # cyc2ns.data[idx_14].cyc2ns_mul, pfo_ret__ movl %gs:4(%rcx),%ecx # cyc2ns.data[idx_14].cyc2ns_shift, pfo_ret__ movl %gs:cyc2ns+32(%rip),%r8d # cyc2ns.seq.sequence, pfo_ret__ cmpl %r8d, %eax # pfo_ret__, pfo_ret__ jne .L235 #, movl %esi, %esi # pfo_ret__, pfo_ret__ movq %rsi, %rax # pfo_ret__, tmp133 mulq %r10 # _23 shrdq %rdx, %rax # pfo_ret__,, tmp134 shrq %cl, %rdx # pfo_ret__, testb $64, %cl #, pfo_ret__ cmovne %rdx, %rax #,, tmp134 addq %rdi, %rax # pfo_ret__, <retval> popq %rbp # ret