Re: [RFC PATCH 6/8] preempt/dynamic: Provide preempt_schedule[_notrace]() static calls

[Peter Zijlstra]

On Wed, Jan 27, 2021 at 05:18:37PM -0600, Josh Poimboeuf wrote:
> On Wed, Jan 27, 2021 at 01:02:18PM -0600, Josh Poimboeuf wrote:
> > On Wed, Jan 27, 2021 at 01:00:07PM -0600, Josh Poimboeuf wrote:
> > > On Wed, Jan 27, 2021 at 07:44:01PM +0100, Peter Zijlstra wrote:
> > > > On Wed, Jan 27, 2021 at 10:33:08AM -0600, Josh Poimboeuf wrote:
> > > > 
> > > > > What did you think about .static_call_tramp_key?  I could whip up a
> > > > > patch later unless you beat me to it.
> > > > 
> > > > Yeah, I'm not sure.. why duplicate information already present in
> > > > kallsyms?
> > > 
> > > Well, but it's not exactly duplicating kallsyms.  No need to store
> > > symbol names, just the pointer relationships.  And kallsyms is
> > > presumably slow.
> > > 
> > > > There's a fair number of features that already require KALLSYMS, I can't
> > > > really be bothered about adding one more (kprobes, function_tracer,
> > > > stack_tracer, ftrace_syscalls).
> 
> Here ya go.  It builds...  And the tramp_key section is nice and small.
> 
> Relocation section [1497] '.rela.static_call_tramp_key' for section [1496] '.static_call_tramp_key' at offset 0x179ab818 contains 8 entries:
>   Offset              Type            Value               Addend Name
>   000000000000000000  X86_64_PC32     0x00000000000004c0      +0 __SCT__preempt_schedule
>   0x0000000000000004  X86_64_PC32     0x000000000005ee10      +0 __SCK__preempt_schedule
>   0x0000000000000008  X86_64_PC32     0x00000000000004c8      +0 __SCT__preempt_schedule_notrace
>   0x000000000000000c  X86_64_PC32     0x000000000005ee00      +0 __SCK__preempt_schedule_notrace
>   0x0000000000000010  X86_64_PC32     0x00000000000004d0      +0 __SCT__cond_resched
>   0x0000000000000014  X86_64_PC32     0x000000000005dd20      +0 __SCK__cond_resched
>   0x0000000000000018  X86_64_PC32     0x00000000000004d8      +0 __SCT__might_resched
>   0x000000000000001c  X86_64_PC32     0x000000000005dd10      +0 __SCK__might_resched
> 

Fair enough I suppose. I'll slap a changelog and your SoB on it and I
suppose I'll got commit the whole lot. Then we can forget about it
again.