Re: [PATCH 09/11] sysctl: Remove the end element in sysctl table arrays

[Joel Granados <j.granados@samsung.com>]

[-- Attachment #1: Type: text/plain, Size: 4805 bytes --]

On Wed, Jun 21, 2023 at 04:15:46PM +0300, Jani Nikula wrote:
> On Wed, 21 Jun 2023, Joel Granados <j.granados@samsung.com> wrote:
> > On Wed, Jun 21, 2023 at 02:16:55PM +0300, Jani Nikula wrote:
> >> On Wed, 21 Jun 2023, Joel Granados <j.granados@samsung.com> wrote:
> >> > Remove the empty end element from all the arrays that are passed to the
> >> > register sysctl calls. In some files this means reducing the explicit
> >> > array size by one. Also make sure that we are using the size in
> >> > ctl_table_header instead of evaluating the .procname element.
> >> 
> >> Where's the harm in removing the end elements driver by driver? This is
> >> an unwieldy patch to handle.
> >
> > I totally agree. Its a big one!!! but I'm concerned of breaking bisectibility:
> > * I could for example separate all the removes into separate commits and
> >   then have a final commit that removes the check for the empty element.
> >   But this will leave the tree in a state where the for loop will have
> >   undefined behavior when it looks for the empty end element. It might
> >   or might not work (probably not :) until the final commit where I fix
> >   that.
> >
> > * I could also change the logic that looks for the final element,
> >   commit that first and then remove the empty element one commit per
> >   driver after that. But then for all the arrays that still have an
> >   empty element, there would again be undefined behavior as it would
> >   think that the last element is valid (when it is really the sentinel).
> >
> > Any ideas on how to get around these?
> 
> First add size to the register calls, and allow the last one to be
> sentinel but do not require the sentinel.
> 
> Start removing sentinels, adjusting the size passed in.
This is a great idea! and I think I don't even have to adjust the size
because if I change the logic to stop on the sentinel or the size; so when
the sentinel is there, it will stop before the size. And when the
sentinel is not there, it will stop on the correct size.

There might be issues with indirection calls. And there might also be
lots of places where I need to adjust a for loop (as dan has pointed
out) but its worth a try for V2.

Best
> 
> Once enough sentinels have been removed, add warning if the final entry
> is a sentinel.
> 
> Never really remove the check? (But surely you can rework the logic to
> not count the number of elements up front, only while iterating.)
> 
> 
> BR,
> Jani.
> 
> >> 
> >> > diff --git a/drivers/gpu/drm/i915/i915_perf.c b/drivers/gpu/drm/i915/i915_perf.c
> >> > index f43950219ffc..e4d7372afb10 100644
> >> > --- a/drivers/gpu/drm/i915/i915_perf.c
> >> > +++ b/drivers/gpu/drm/i915/i915_perf.c
> >> > @@ -4884,24 +4884,23 @@ int i915_perf_remove_config_ioctl(struct drm_device *dev, void *data,
> >> >  
> >> >  static struct ctl_table oa_table[] = {
> >> >  	{
> >> > -	 .procname = "perf_stream_paranoid",
> >> > -	 .data = &i915_perf_stream_paranoid,
> >> > -	 .maxlen = sizeof(i915_perf_stream_paranoid),
> >> > -	 .mode = 0644,
> >> > -	 .proc_handler = proc_dointvec_minmax,
> >> > -	 .extra1 = SYSCTL_ZERO,
> >> > -	 .extra2 = SYSCTL_ONE,
> >> > -	 },
> >> > +		.procname = "perf_stream_paranoid",
> >> > +		.data = &i915_perf_stream_paranoid,
> >> > +		.maxlen = sizeof(i915_perf_stream_paranoid),
> >> > +		.mode = 0644,
> >> > +		.proc_handler = proc_dointvec_minmax,
> >> > +		.extra1 = SYSCTL_ZERO,
> >> > +		.extra2 = SYSCTL_ONE,
> >> > +	},
> >> >  	{
> >> > -	 .procname = "oa_max_sample_rate",
> >> > -	 .data = &i915_oa_max_sample_rate,
> >> > -	 .maxlen = sizeof(i915_oa_max_sample_rate),
> >> > -	 .mode = 0644,
> >> > -	 .proc_handler = proc_dointvec_minmax,
> >> > -	 .extra1 = SYSCTL_ZERO,
> >> > -	 .extra2 = &oa_sample_rate_hard_limit,
> >> > -	 },
> >> > -	{}
> >> > +		.procname = "oa_max_sample_rate",
> >> > +		.data = &i915_oa_max_sample_rate,
> >> > +		.maxlen = sizeof(i915_oa_max_sample_rate),
> >> > +		.mode = 0644,
> >> > +		.proc_handler = proc_dointvec_minmax,
> >> > +		.extra1 = SYSCTL_ZERO,
> >> > +		.extra2 = &oa_sample_rate_hard_limit,
> >> > +	}
> >> >  };
> >> 
> >> The existing indentation is off, but fixing it doesn't really belong in
> >> this patch.
> >
> > Agreed. But I actually was trying to fix something that checkpatch
> > flagged. I'll change these back (which will cause this patch to be
> > flagged).
> >
> > An alternative solution would be to fix the indentation as part of the
> > preparation patches. Tell me what you think.
> >
> > Thx
> >
> >> 
> >> BR,
> >> Jani.
> >> 
> >> 
> >> -- 
> >> Jani Nikula, Intel Open Source Graphics Center
> 
> -- 
> Jani Nikula, Intel Open Source Graphics Center

-- 

Joel Granados

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]