Messages in this thread | | | Date | Fri, 24 Jul 2015 12:36:09 +0200 | From | Peter Zijlstra <> | Subject | Re: Kernel broken on processors without performance counters |
| |
On Fri, Jul 24, 2015 at 07:29:05AM +0200, Borislav Petkov wrote: > On Thu, Jul 23, 2015 at 09:02:14PM +0200, Peter Zijlstra wrote: > > On Thu, Jul 23, 2015 at 07:54:36PM +0200, Borislav Petkov wrote: > > > On Thu, Jul 23, 2015 at 07:08:11PM +0200, Peter Zijlstra wrote: > > > > That would be bad, how can we force it to emit 5 bytes? > > > > > > .byte 0xe9 like we used to do in static_cpu_has_safe(). > > > > Like so then? > > > > static __always_inline bool arch_static_branch_jump(struct static_key *key, bool inv) > > { > > unsigned long kval = (unsigned long)key + inv; > > > > asm_volatile_goto("1:" > > ".byte 0xe9\n\t .long %l[l_yes]\n\t" > > ".pushsection __jump_table, \"aw\" \n\t" > > _ASM_ALIGN "\n\t" > > _ASM_PTR "1b, %l[l_yes], %c0 \n\t" > > ".popsection \n\t" > > : : "i" (kval) : : l_yes); > > > > return false; > > l_yes: > > return true; > > } > > Yap. > > But, we can do even better and note down what kind of JMP the compiler > generated and teach __jump_label_transform() to generate the right one. > Maybe this struct jump_entry would get a flags member or so. This way > we're optimal. > > Methinks...
Yes, Jason and Steve already have patches to go do that, but I'd really like to keep that separate for now, this thing is big enough as is.
| |