Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20201125193626.2266995-1-jim.cromie@gmail.com>
 <20201125193626.2266995-8-jim.cromie@gmail.com> <d57d4f39-b9d3-99f0-6fd0-b3f072895edd@akamai.com>
In-Reply-To: <d57d4f39-b9d3-99f0-6fd0-b3f072895edd@akamai.com>
From:   jim.cromie@gmail.com
Date:   Wed, 25 Nov 2020 14:23:00 -0700
Message-ID: <CAJfuBxyE2De7gjoiv9Dnh4XK7+WjLMA+_x9h9-wOiN7NL6bWiA@mail.gmail.com>
Subject: Re: [PATCH 7/7] dyndbg: enable 'cache' of active pr_debug callsites
To:     Jason Baron <jbaron@akamai.com>
Cc:     linux-mm@kvack.org, Greg KH <gregkh@linuxfoundation.org>,
        Rasmus Villemoes <linux@rasmusvillemoes.dk>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Nov 25, 2020 at 1:54 PM Jason Baron <jbaron@akamai.com> wrote:
>
>
>
> On 11/25/20 2:36 PM, Jim Cromie wrote:
> > In ddebug_putsite(), dont zs_unmap the callsite if it is enabled for
> > printing.  This means that the next time this pr_debug callsite is
> > executed, the _getsite() will succeed quickly without remapping the
> > zrec.
> >
> > Once the callsite is disabled via >control, a following _putsite()
> > will see the flag cleared, and zs_unmap it.
> >
> > This changes the lifetime of our zs_mappings from brief (only for the
> > single printk) to as long as uptime (if a prdebug is enabled til
> > poweroff).  This appears to be triggering the ensuing mayhem.
> >
> > I am able to get through init, to root console, by disabling all
> > dynamic-debugs, including the ones that are enabled at compile.
> >
> > $ kruna --kopt \*.dyndbg=-p --kopt debug_locks_proceed=1
> >
> > but enabling any pr-debug crashes.
> > Plain old boot also panics, as pasted below.
> >
>
>
> Hi Jim,
>
> Looks like you've made a lot of progress with this series...but if I
> understand correctly there are still some unresolved issues (panic).
> So I think this series is more 'RFC' at this point?
>

yes, RFC.  WIP suggests I know whats needed next.

certainly Ive seen a variety of sleeping while atomic ... BUGs,


> I was also expecting to see updates to the actual printing functions
> to now use dp->site. But perhaps I missed those bits?

those bits were changed in patch 5
basically changes are ~ s/dp/dc/g, and in ddebug_getsite / _putsite

>
> Thanks,
>
> -Jason
>
> > $ kruna --kopt debug_locks_proceed=1
> >


 [    1.915768]  ? rest_init+0x24d/0x24d
> > [    1.916417]  kernel_init+0xaf/0x103
> > [    1.917038]  ret_from_fork+0x22/0x30
> > [    1.917726] Kernel Offset: disabled
> > [    1.918324] ---[ end Kernel panic - not syncing: Requested init /bin/sh failed (error -14). ]---
> > QEMU 5.1.0 monitor - type 'help' for more information
> >
> > Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
> > ---
> >  lib/dynamic_debug.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
> > index 534d73e45844..0c69aa52395d 100644
> > --- a/lib/dynamic_debug.c
> > +++ b/lib/dynamic_debug.c
> > @@ -165,7 +165,9 @@ static void ddebug_putsite(struct _ddebug *dp)
> >               /* need to keep site until zmap backing exists */
> >               return;
> >
> > -     /* always unmap for now. if !pr-debug has locking issues */
> > +     if (dp->flags & _DPRINTK_FLAGS_PRINT)
> > +             return; /* keep maps of enabled pr_debugs */
> > +
> >       zs_unmap_object(dd_callsite_zpool, dp->zhandle);
> >       dp->site = NULL;
> >  }
> >