Re: [PATCH v10 10/10] mtd: maps: gpio-addr-flash: Add support for device-tree devices

[Ricardo Ribalda Delgado]

Hi Boris
On Fri, Oct 5, 2018 at 6:29 PM Boris Brezillon
<boris.brezillon@xxxxxxxxxxx> wrote:
>
> On Fri, 5 Oct 2018 17:40:44 +0200
> Ricardo Ribalda Delgado <ricardo.ribalda@xxxxxxxxx> wrote:
>
> > Hi Boris
> > On Fri, Oct 5, 2018 at 4:52 PM Boris Brezillon
> > <boris.brezillon@xxxxxxxxxxx> wrote:
> > >
> > > On Fri, 5 Oct 2018 16:06:57 +0200
> > > Ricardo Ribalda Delgado <ricardo.ribalda@xxxxxxxxx> wrote:
> > >
> > > > Hi again Boris
> > > >
> > > >
> > > > On Fri, Oct 5, 2018 at 2:10 PM Boris Brezillon
> > > > <boris.brezillon@xxxxxxxxxxx> wrote:
> > > > >
> > > > > On Fri, 5 Oct 2018 14:04:52 +0200
> > > > > Ricardo Ribalda Delgado <ricardo.ribalda@xxxxxxxxx> wrote:
> > > > >
> > > > > > Hi Boris
> > > > > > On Fri, Oct 5, 2018 at 12:12 PM Boris Brezillon
> > > > > > <boris.brezillon@xxxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > On Fri, 5 Oct 2018 11:54:18 +0200
> > > > > > > Ricardo Ribalda Delgado <ricardo.ribalda@xxxxxxxxx> wrote:
> > > > > > >
> > > > > > > > Hi Boris
> > > > > > > >
> > > > > > > > Just seen that  you already did the rebase at
> > > > > > > > https://github.com/bbrezillon/linux-0day/commits/mtd/physmap-cleanup
> > > > > > > >
> > > > > > > > Thanks for that.
> > > > > > > >
> > > > > > > > I am about to test it in real hw (unless you want me wait)
> > > > > > >
> > > > > > > Sure, go ahead and test it.
> > > > > > >
> > > > > > > Thanks,
> > > > > > >
> > > > > > > Boris
> > > > > > I had to change this on your patchset to have it working on hw:
> > > > > > https://pastebin.com/78A7yhJ9
> > > > > >
> > > > > > If you send the patchset to the mailing list I can review it patch by patch.
> > > > > >
> > > > > > Also
> > > > > > mtd: maps: Prepare merging of physmap and physmap_of
> > > > > >
> > > > > > I do not think that can be bisected. (Not sure, I have to test it)
> > > > >
> > > > > Okay, I'll have a look.
> > > > >
> > > > > >
> > > > > > I add the diff to the mail, but gmail will probably scramble the
> > > > > > lines(yes I know I have to use other mail client)
> > > > >
> > > > > The diff looks good, I'll fix that an send a push a new version.
> > > >
> > > > Also fix on physmap_flash_remove
> > > >
> > > > physmap_data->exit(dev); must be called BEFORE
> > > > map_destroy(info->mtds[i]);
> > >
> > > Hm, that's weird. That shouldn't happen. Do you have a non-NULL
> > > ->exit()? Can you detail why you think ->exit() call is the cause of
> > > this OOPS?
> > >
> >
> > No idea.  It was crashing at:
> > https://github.com/bbrezillon/linux-0day/blob/mtd/physmap-cleanup/drivers/mtd/chips/cfi_cmdset_0002.c#L2839
> > cfi_cmdset_0002.c seesm to play with cfi->chips on its reset callback
> >
> > I added some printfs:
> >
> > if (!cfi),
> > if (!chip)
> > if (!cfi->chips)
> >
> > sometimes it crashed on one place, sometimes in another :S. Reading
> > back our patch it seemed more logical (semantically :P) to
> > destroy after exit and not the other way around.
>
> Actually no, it makes more sense to call ->exit() after destroying the
> maps, because the platform-specific ->exit() implem might release
> resources that are used during the destroy_map() operation. Another
> reason to keep it in this order is that operations in the remove path
> should be in the reverse order of those done in the probe path, and
> ->init() is definitely called before map_probe().
>

I think I know what might be the issue. on cfi_cmdset_002.c
cfi_amdstd_reset can be called in parrallel to cfi_amdstd_destroy.

maybe we should call
unregister_reboot_notifier(&mtd->reboot_notifier);
before
cfi_amdstd_reset(mtd)

need to try that on real hw with some strategical printfs :P. But that
will ahve to wait til Monday

Cheers


> >
> > I made that change and it stopped OOPsing at reboot.
>
> Maybe it's just a side effect, especially since I'm pretty sure your
> physmap_data->exit() is NULL (assuming you do use DT to declare your
> flash).
>
> >
> > Havent had the time to dig deeper. But if it does not break anything
> > on your side to invert destroy and exit please do so. It was not
> > oopsing with my patchset.
>
> I'd like to understand what's happening before doing this change. As I
> said, I fear this reordering just hides something else.



-- 
Ricardo Ribalda