Re: linux-next: dock_link_device is oopsy

[Hillf Danton]

On Sat, Feb 18, 2012 at 7:14 PM, Holger Macht <holger@xxxxxxxx> wrote:
> On Fr 17. Feb - 15:49:02, Hugh Dickins wrote:
>> On Sat, 18 Feb 2012, Holger Macht wrote:
>> > On Fr 17. Feb - 14:42:31, Hugh Dickins wrote:
>> > > On Fri, 17 Feb 2012, Holger Macht wrote:
>> > > > On Fr 17. Feb - 13:46:04, Hugh Dickins wrote:
>> > > > > Matthew,
>> > > > >
>> > > > > A linux-next oops at bootup in dock_link_device() tells me that you
>> > > > > were not feeling well when you wrote that and dock_unlink_device():
>> > > > > I hope you're feeling better now and can rewrite them soon.
>> > > >
>> > > > Andrew Morton experienced a similar problem. What system are you using?
>> > > > I didn't encounter this problem with the systems I tested with.
>> > >
>> > > The two systems I got that on were both 4-year-old Core2 Duo systems,
>> > > one an HP quad desktop, one a Fujitsu-Siemens laptop.
>> >
>> > Thanks for the information I think this is really independent from the
>> > fact if a laptop, or more precicely if a system with dock station/bay is
>> > used.
>> >
>> > >
>> > > >
>> > > > Do you actually have a /sys/devices/platform/dock.?/ directory with a
>> > > > file 'type' that contains 'dock_station'?
>> > >
>> > > I'll have to report back on that this evening, I'm away from them now.
>> >
>> > I actually guess that those systems don't have a
>> > /sys/devices/platform/dock.? directory at all, which is fine.
>> >
>> > I also think this will fix it, would be great if you could confirm this:
>> >
>> > acpi: Bail out when linking devices and there are no dock stations
>> >
>> > If dock_station_count is zero, we allocate zero memory and don't check
>> > this at future references. So bail out if there are actually no dock
>> > stations.
>> >
>> > Signed-off-by: Holger Macht <holger@xxxxxxxx>
>>
>> Certainly won't fix it as is (well, it shifts the crash over into kfree).
>> This function is expected to return a pointer, not an error or success
>> code.
>
> Oh well, too late for me yesterday...
>
>>
>> I've little doubt that returning NULL rather than -ENODEV there would fix
>> the boot crash; and if you're in a hurry to fix up booting (understandable)
>> then I suppose that would do for the moment.
>
> I really think this will basically do the trick. dock_(un)link_device()
> is called by ata_acpi_(un)bind_dock(), which in turn is called by
> ata_scsi_scan_host() unconditionally, thus not depending on the presence
> of a dock device. And dock_(un)link_device() simply misses the check if
> there is a dock/bay device.
>
> So how about that?
>
> acpi: Bail out when linking devices and there are no dock stations
>
> If dock_station_count is zero, we allocate zero memory and don't check
> this at future references. So bail out if there are actually no dock
> stations.
>
> Signed-off-by: Holger Macht <holger@xxxxxxxx>
> ---
> Âdrivers/acpi/dock.c | Â 19 ++++++++++++++-----
> Â1 files changed, 14 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/acpi/dock.c b/drivers/acpi/dock.c
> index b5e4142..0b3072c 100644
> --- a/drivers/acpi/dock.c
> +++ b/drivers/acpi/dock.c
> @@ -281,11 +281,15 @@ EXPORT_SYMBOL_GPL(is_dock_device);
> Â*/
> Âstruct device **dock_link_device(acpi_handle handle)
> Â{
> - Â Â Â struct device *dev = acpi_get_physical_device(handle);
> + Â Â Â struct device *dev;
> Â Â Â Âstruct dock_station *dock_station;
> Â Â Â Âint ret, dock = 0;
> Â Â Â Âstruct device **devices;
>
> + Â Â Â if (!dock_station_count)
> + Â Â Â Â Â Â Â return NULL;
> +
> + Â Â Â dev = acpi_get_physical_device(handle);
> Â Â Â Âdevices = kmalloc(dock_station_count * sizeof(struct device *),
> Â Â Â Â Â Â Â Â Â Â Â Â ÂGFP_KERNEL);
>
> @@ -320,12 +324,17 @@ EXPORT_SYMBOL_GPL(dock_link_device);
> Â*/
> Âstruct device **dock_unlink_device(acpi_handle handle)
> Â{
> - Â Â Â struct device *dev = acpi_get_physical_device(handle);
> + Â Â Â struct device *dev;
> Â Â Â Âstruct dock_station *dock_station;
> Â Â Â Âint dock = 0;
> - Â Â Â struct device **devices =
> - Â Â Â Â Â Â Â kmalloc(dock_station_count * sizeof(struct device *),
> - Â Â Â Â Â Â Â Â Â Â Â GFP_KERNEL);
> + Â Â Â struct device **devices;
> +
> + Â Â Â if (!dock_station_count)
> + Â Â Â Â Â Â Â return NULL;
> +
> + Â Â Â dev = acpi_get_physical_device(handle);
> + Â Â Â devices = kmalloc(dock_station_count * sizeof(struct device *),
> + Â Â Â Â Â Â Â Â Â Â Â Â GFP_KERNEL);
>

If bail out in this way, another patch looks needed to fix up
mem leakage :-(

> Â Â Â Âif (!dev)
> Â Â Â Â Â Â Â Âreturn NULL;
> --
> 1.7.7
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/