Re: [patch] kcsan: Make it RT aware

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2021-08-26 17:42:21 [+0200], Mike Galbraith wrote:
> On Thu, 2021-08-26 at 17:18 +0200, Sebastian Andrzej Siewior wrote:
> > On 2021-08-22 19:29:00 [+0200], Vlastimil Babka wrote:
> > > On 8/22/21 6:35 PM, Mike Galbraith wrote:
> > > > On second thought, burn both patches.  Making K[AC]SAN work with RT
> > > > ain't worth the warts.
> > >
> > > The stackdepot part of kasan patch will be useful at some point.
> > > slub_debug should eventually switch to stackdepot - it almost did in
> > > mainline already. I'll try to remember and dig that part up when
> > > needed.
> > > Thanks.
> >
> > If you want them dropped, sure.
> 
> If you think they're worth having, feel free.  The kcsan patch I found
> especially icky, and changed it like so, avoiding icky config dependent
> branches while giving myself an ever so tiny excuse for looping ;-)
> 
> kcsan: Make it RT aware
> 
> Converting report_filterlist_lock to raw_spinlock_t lets RT report
> problem free, but makes allocations in insert_report_filterlist()
> problematic.  Solve that by taking the lock after allocation.  That
> means all configs must check that the situation didn't change while
> acquiring if we're to avoid ugly config specific branches, but it
> also removes the need for GFP_ATOMIC allocations.
> 
> Signed-off-by: Mike Galbraith <efault@gmx.de>

I would have to look at this raw lock change now more closely…

Sebastian