Re: [PATCH] toshiba_acpi: Fix regression caused by backlight extra check code

From: Azael Avalos
Date: Tue Nov 11 2014 - 11:01:59 EST


2014-11-10 22:49 GMT-07:00 Darren Hart <dvhart@xxxxxxxxxxxxx>:
> On Mon, Nov 10, 2014 at 09:58:29AM -0700, Azael Avalos wrote:
>> Hi Darren,
>>
>> Sorry for the way late reply, I had to go out of town in a hurry.
>>
>> 2014-11-03 23:21 GMT-07:00 Darren Hart <dvhart@xxxxxxxxxxxxx>:
>> > On Mon, Nov 03, 2014 at 08:58:34PM -0700, Azael Avalos wrote:
>> >> Bug 86521 uncovered that some TOS6208 devices also return
>> >> non zero values on a write call to the backlight method,
>> >> thus getting caught and bailed out by the extra check code.
>> >>
>> >> This patch makes sure that the extra check is being done
>> >> on a TOS1900 device and then make the check for the broken
>> >> backlight code.
>> >>
>> >> Signed-off-by: Azael Avalos <coproscefalo@xxxxxxxxx>
>> >> ---
>> >> drivers/platform/x86/toshiba_acpi.c | 8 ++++++--
>> >> 1 file changed, 6 insertions(+), 2 deletions(-)
>> >>
>> >> diff --git a/drivers/platform/x86/toshiba_acpi.c b/drivers/platform/x86/toshiba_acpi.c
>> >> index ef3a190..e3fed12 100644
>> >> --- a/drivers/platform/x86/toshiba_acpi.c
>> >> +++ b/drivers/platform/x86/toshiba_acpi.c
>> >> @@ -944,9 +944,13 @@ static int set_lcd_brightness(struct toshiba_acpi_dev *dev, int value)
>> >> /* Extra check for "incomplete" backlight method, where the AML code
>> >> * doesn't check for HCI_SET or HCI_GET and returns TOS_SUCCESS,
>> >> * the actual brightness, and in some cases the max brightness.
>> >> + * Use the SPFC method as an indicator that we're on a TOS1900 device,
>> >> + * otherwise some TOS6208 devices might get bailed out, see bug 86521
>> >
>> > This needs a clearer description here in this comment, rather than redirecting
>> > the reader to a bug report (which may or may not exist when needed).
>>
>> Alright, will do whenever we reach an agreement below.
>>
>> >
>> >> */
>> >> - if (out[2] > 0 || out[3] == 0xE000)
>> >> - return -ENODEV;
>> >> + if (acpi_has_method(dev->acpi_dev->handle, "SPFC")) {
>> >
>> > Hrm, this checking for the existence of a specific method seems suspect to me.
>> > We would know if we are on a TOS1900 as we matches the acpi id already. Is the
>> > SPFC significant here, or is it just a "we only see SPFC on TOS1900 so it's a
>> > convenient test"? If the latter, it seems rather fragile and prone to other
>> > breakage to me.
>>
>> Yeah, its the latter, the "SPFC" method is specific to TOS1900 devices.
>>
>> All of the TOS1900 support the Toshiba specific backlight read-only,
>> and that test is just to get those implementations where the AML
>> code doesn't check for read/write registers, so far I've identified three
>> series of laptops with this issue (all Qosmios), X500, X505 and X75-A
>> (and there might be more around).
>>
>> We could dissable backlight on all TOS1900 or add those three models
>> to the (growing) DMI list on video.c, but of course, I would like to keep
>> the code in-house, but that's just me :-)
>>
>
> I'm having some trouble grokking the whole picture I think.
>
> So, let's try and clear it up. We have a function to set brightness. On model
> citizen laptops, after the ACPI call, out[2] is 0 (TOS_SUCCESS) and we return 0.

Yeah, at least on TOS1900 devices, all write calls set the out[n] to zero,
and then assign the specified return status to out[0] (either success or error).

>
> The extra check was added by f6aac65 to avoid registering certain badly behaved
> machines (Qosmios) which always return HCI_SUCCESS, even the read/write commands
> are missing. This was determined by checking for observed values in out[2]
> greater than 0 and out[3] equal to 0xE000 (which presumably map to actual or max
> brightness. I take it in a well behaved ACPI implementation, out[2] must be 0
> and out[3] is .... what? I'll call this out[2] || out[3] state the "signature"
> of a bad implementation.

Yeah, since TOS1900 set all out[n] values to zero, I was expecting that same
behaviour from a TOS62XX device, however, bug 86521 proved me wrong.

>
> Now, with bug 86521, we learn that these signatures are not necessarily bad,
> some working laptops also populate these fields in this way.

For TOS1900 devices, they are, but for TOS62XX, that's another story,
the methods
are hidden and I can't see what they're doing... :-(

>
> Finally, while it is only TOS1900 devices that are known to be bad, all TOS1900
> devices are not necessarily bad. So you don't want to block all TOS1900 devices
> at probe time.

We're doing it already, toshiba_acpi_setup_backlight checks for read and write
support in order to register the backlight, and since all* TOS1900
support it read-only
they all get bailed out by these checks, except those badly behaved models,
they simply return SUCCESS and thus pass that check, creating a broken backlight
device with just read capabilities.

* (Actually there are two TOS1900 devices, the ones that support it
read only, and the
ones that don't support it at all)

>
> Do I have this right?
>
> If so, this is looking more and more fragile to me. I'm inclined to push for a
> blacklist of the known bad models and strip out both this and the precious extra
> check, since they are based on circumstantial evidence of failure.

Yeah, I thought that from the beginning, but I found that a bit
cumbersome to add
those models to the growing list, if we can do it in-house. But I'm
fine with whatever
decision is made here.

>
>> >
>> > Rafael, any recommendations here?
>
>
> Rafael, what's your take on this? Does the above influence your position one way
> or the other?
>
>
>
>> >
>> >> + if (out[2] > 0 || out[3] == 0xE000)
>> >> + return -ENODEV;
>> >> + }
>> >>
>> >> return out[0] == TOS_SUCCESS ? 0 : -EIO;
>> >> }
>
> --
> Darren Hart
> Intel Open Source Technology Center

Cheers
Azael



--
-- El mundo apesta y vosotros apestais tambien --
--
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/