Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20191120054728.0979695C0FE4@us180.sjc.aristanetworks.com> <CAJwJo6ZTSw93zKdN3i-YqVtEEwh4qWc22Hk-emHvJth+HsaQbQ@mail.gmail.com>
In-Reply-To: <CAJwJo6ZTSw93zKdN3i-YqVtEEwh4qWc22Hk-emHvJth+HsaQbQ@mail.gmail.com>
From:   Francesco Ruggeri <fruggeri@arista.com>
Date:   Thu, 21 Nov 2019 14:49:34 -0800
Message-ID: <CA+HUmGh+tEnP7n+_FcJuwZOF4HBgfatLwUa8a=K_BdqnKETNCw@mail.gmail.com>
Subject: Re: [PATCH] ACPI: only free map once in osl.c
To:     Dmitry Safonov <0x7f454c46@gmail.com>
Cc:     lenb@kernel.org, "Rafael J. Wysocki" <rjw@rjwysocki.net>,
        open list <linux-kernel@vger.kernel.org>,
        linux-acpi@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Nov 21, 2019 at 1:19 PM Dmitry Safonov <0x7f454c46@gmail.com> wrote:
>
> Hi Francesco,
>
> I believe, there's still an issue with your patch.
>
> On Wed, 20 Nov 2019 at 05:50, Francesco Ruggeri <fruggeri@arista.com> wrote:
> > @@ -472,10 +477,11 @@ void acpi_os_unmap_generic_address(struct acpi_generic_address *gas)
> >                 mutex_unlock(&acpi_ioremap_lock);
> >                 return;
> >         }
> > -       acpi_os_drop_map_ref(map);
> > +       refcount = acpi_os_drop_map_ref(map);
> >         mutex_unlock(&acpi_ioremap_lock);
>
> Here comes acpi_os_get_iomem() increasing the refcount again.

Thanks Dmitry.
I think that any code that increments the refcount does so after
looking for map in acpi_ioremap under acpi_ioremap_lock,
and the process that drops the last reference removes map
from the list, also under acpi_ioremap_lock, so I am not sure
this could happen.
The synchronize_rcu_expedited in acpi_os_map_cleanup should
then take care of any other references to map (which it is my
understanding require acpi_ioremap_lock or rcu read lock).

Thanks,
Francesco