lkml.org 
[lkml]   [2021]   [Nov]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] sched/core: Export pelt_thermal_tp
Hi Peter

On 10/28/21 18:50, Peter Zijlstra wrote:
> On Thu, Oct 28, 2021 at 09:22:05AM -0700, Christoph Hellwig wrote:
> > On Thu, Oct 28, 2021 at 06:18:55PM +0200, Peter Zijlstra wrote:
> > > > > @@ -36,6 +36,7 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_rt_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_dl_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_irq_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_se_tp);
> > > > > +EXPORT_TRACEPOINT_SYMBOL_GPL(pelt_thermal_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(sched_cpu_capacity_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(sched_overutilized_tp);
> > > > > EXPORT_TRACEPOINT_SYMBOL_GPL(sched_util_est_cfs_tp);
> > > >
> > > > ... and while we're at it, all these exports are unused and should
> > > > be deleted as well.
> > >
> > > This is my concession wrt tracepoints. Actual tracepoints are ABI,
> > > exports are in-kernel interfaces and are explicitly not ABI.
> > >
> > > This way people can use an external module to get at the tracepoint data
> > > without having in-tree tracepoints.
> >
> > All of this makes no sense at all. These are entirely dead exports.
> > If you remove them nothing else changes. Note taht the tracepoints
> > do have in-kernel callers, so if people thing of them as an ABI you've
> > got your ABI already with or without the exports.
>
> These are not normal traceevents, these are tracepoints, the distinction
> is that these things do not show up in tracefs and there is no userspace
> visible representation of them. No userspace gives no ABI.
>
> All they provide is the in-code hook and in-kernel registration
> interface. These EXPORTS export that registration interface, such that
> an out-of-tree module can make use of them.
>
> And yes, unused exports are iffy, out-of-tree modules are iffy, but in
> this case I made an exception since ABI contraints are worse. We very
> clearly state there is no such thing is kabi, so breaking any user of
> these exports if fair game.
>
> Breaking users of userspace trace-events gets kernel patches reverted
> (been there, done that, never want to ever be there again).
>
> People want to trace this stuff, but I *REALLY* do not want to commit to
> ABI, this is the middle-ground that sucks least :/

AFAICS this wasn't picked up. Should I tweak the commit message to make things
clearer?

Thanks!

--
Qais Yousef

\
 
 \ /
  Last update: 2021-11-25 17:58    [W:0.038 / U:0.160 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site