Re: [PATCH] pstore: change mutex locking to spin_locks

From: Don Zickus <dzickus@redhat.com>
To: Tony Luck <tony.luck@intel.com>
Cc: "Huang, Ying" <ying.huang@intel.com>,
	LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] pstore: change mutex locking to spin_locks
Date: Mon, 25 Jul 2011 08:47:41 -0400	[thread overview]
Message-ID: <20110725124741.GB2866@redhat.com> (raw)
In-Reply-To: <CA+8MBb+E462CVOe+5FQb62Tdi5R6cBYgFZUW7sxWDDh2QTaxxA@mail.gmail.com>

On Fri, Jul 22, 2011 at 04:21:14PM -0700, Tony Luck wrote:
> On Fri, Jul 22, 2011 at 3:39 PM, Luck, Tony <tony.luck@intel.com> wrote:
> > Yup - oops is the only one where the current kernel continues - all the
> > others (halt, reboot, kexec, panic) lead to the termination of the
> > current kernel - and thus no user processes that would be able to
> > see the shiny new entry in the pstore file system.
> 
> Doh - I went to look at what it would take to fix this - and found that
> I'd already implemented it:
> 
> 
>                 if (reason == KMSG_DUMP_OOPS && pstore_is_mounted())
>                         pstore_mkfile(PSTORE_TYPE_DMESG, psinfo->name, id,
>                                       psinfo->buf, hsize + l1_cpy + l2_cpy,
>                                       CURRENT_TIME, psinfo);

Right, but it still has to be created with locks surrounding it.  And the
question is what type of locks should we use.  Mutexes don't work when
called from the NMI case and spinlocks don't seem to work when called from
the OOPs (pstore_mkfile) case.

Cheers,
Don