Re: [PATCH v2] panic, kexec: Make __crash_kexec() NMI safe

[Petr Mladek <pmladek@suse.com>]

On Tue 2022-06-28 18:33:08, Valentin Schneider wrote:
> On 27/06/22 13:42, Valentin Schneider wrote:
> > On 25/06/22 12:04, Eric W. Biederman wrote:
> >> At this point I recommend going back to being ``unconventional'' with
> >> the kexec locking and effectively reverting commit 8c5a1cf0ad3a ("kexec:
> >> use a mutex for locking rather than xchg()").
> >>
> >> That would also mean that we don't have to worry about the lockdep code
> >> doing something weird in the future and breaking kexec.
> >>
> >> Your change starting to is atomic_cmpxchng is most halfway to a revert
> >> of commit 8c5a1cf0ad3a ("kexec: use a mutex for locking rather than
> >> xchg()").  So we might as well go the whole way and just document that
> >> the kexec on panic code can not use conventional kernel locking
> >> primitives and has to dig deep and build it's own.  At which point it
> >> makes no sense for the rest of the kexec code to use anything different.
> >>
> >
> > Hm, I'm a bit torn about that one, ideally I'd prefer to keep "homegrown"
> > locking primitives to just where they are needed (loading & kexec'ing), but
> > I'm also not immensely fond of the "hybrid" mutex+cmpxchg approach.
> >
> 
> 8c5a1cf0ad3a ("kexec: use a mutex for locking rather than xchg()") was
> straightforward enough because it turned
> 
>         if (xchg(&lock, 1))
>                 return -EBUSY;
> 
> into
> 
>         if (!mutex_trylock(&lock))
>                 return -EBUSY;
> 
> Now, most of the kexec_mutex uses are trylocks, except for:
> - crash_get_memory_size()
> - crash_shrink_memory()
> 
> I really don't want to go down the route of turning those into cmpxchg
> try-loops, would it be acceptable to make those use trylocks (i.e. return
> -EBUSY if the cmpxchg fails)?

IMHO, -EBUSY is acceptable for both crash_get_memory_size()
and crash_shrink_memory(). They are used in the sysfs interface.

> Otherwise, we keep the mutexes for functions like those which go nowhere
> near an NMI.

If we go this way then I would hide the locking into some wrappers,
like crash_kexec_trylock()/unlock() that would do both mutex
and xchg. The xchg part might be hidden in a separate wrapper
__crash_kexec_trylock()/unlock() or
crash_kexec_atomic_trylock()/unlock().

Best Regards,
Petr