Re: i8042 error at booting an Intel Cherry Trail-based device

[Dmitry Torokhov]

On December 5, 2016 4:56:05 PM PST, Marcos Paulo de Souza <marcos.souza.org@xxxxxxxxx> wrote:
>Hi Takashi,
>
>On Fri, Dec 02, 2016 at 11:55:07AM +0100, Takashi Iwai wrote:
>> On Thu, 01 Dec 2016 08:19:46 +0100,
>> Takashi Iwai wrote:
>> > 
>> > On Thu, 01 Dec 2016 03:29:23 +0100,
>> > Dmitry Torokhov wrote:
>> > > 
>> > > Hi Takashi,
>> > > 
>> > > On Mon, Nov 28, 2016 at 02:56:36PM +0100, Takashi Iwai wrote:
>> > > > Hi Dmitry,
>> > > > 
>> > > > I've been testing a small machine with Intel Cherry Trail
>chipset, and
>> > > > noticed that the kernel spews errors always like:
>> > > > 
>> > > >  i8042: PNP: No PS/2 controller found. Probing ports directly.
>> > > >  i8042: Can't read CTR while initializing i8042
>> > > >  i8042: probe of i8042 failed with error -5
>> > > > 
>> > > > Especially the second one ("Can't read CTR...") is annoying
>since it's
>> > > > in KERN_ERR level and thus appears even booted with quiet boot
>> > > > option.  Actually this is the only error message appearing at
>boot, so
>> > > > I'd love to get rid of it.
>> > > > 
>> > > > What is the preferred way to reduce this?  For example, is a
>patch
>> > > > like below OK to simply change the log level and the error
>code?
>> > > 
>> > > No, because if controller is actually present this is a hard
>failure and
>> > > we should be reporting it, not suppressing it.
>> > > 
>> > > The issue is that we did not believe PNP data and in this case we
>should
>> > > have. Unfortunately in old days there was a lot of crap in
>PNP/ACPI
>> > > tables, but it could be better now. We can try, in addition to
>PNP
>> > > matching, checking 8042 flag in "Fixed ACPI Description Table
>Boot
>> > > Architecture Flags" in FADT and if it also shows there is no 8042
>then
>> > > bail.
>> > 
>> > That sounds promising.  Indeed FACL.dsl shows like:
>> > 
>> > [000h 0000   4]                    Signature : "FACP"    [Fixed
>ACPI Description Table (FADT)]
>> > [004h 0004   4]                 Table Length : 0000010C
>> > ....
>> >                Legacy Devices Supported (V2) : 0
>> >             8042 Present on ports 60/64 (V2) : 0
>> > 
>> > If a test patch gets ready, let me know, I'll give it a try.
>> 
>> FYI, a hack like below seems working.
>> 
>> 
>> Takashi
>> 
>> ---
>> diff --git a/drivers/input/serio/i8042-x86ia64io.h
>b/drivers/input/serio/i8042-x86ia64io.h
>> index 073246c7d163..ed6ab702e4b7 100644
>> --- a/drivers/input/serio/i8042-x86ia64io.h
>> +++ b/drivers/input/serio/i8042-x86ia64io.h
>> @@ -9,6 +9,7 @@
>>  
>>  #ifdef CONFIG_X86
>>  #include <asm/x86_init.h>
>> +#include <linux/acpi.h>
>>  #endif
>>  
>>  /*
>> @@ -1055,6 +1056,13 @@ static int __init i8042_pnp_init(void)
>>  #if defined(__ia64__)
>>  		return -ENODEV;
>>  #else
>> +#ifdef CONFIG_ACPI
>> +		if (acpi_gbl_FADT.header.revision >= 3 &&
>> +		    !(acpi_gbl_FADT.boot_flags & ACPI_FADT_8042)) {
>> +			pr_info("PNP: No PS/2 controller found and disabled in ACPI\n");
>> +			return -ENODEV;
>> +		}
>> +#endif
>>  		pr_info("PNP: No PS/2 controller found. Probing ports
>directly.\n");
>>  		return 0;
>>  #endif
>
>I'm not an expert in any subsystem but, maybe this "hack" could be
>added
>to default_i8042_detect in arch/x86/kernel/x86_init.c? Currently it is
>enabled by default, but different Intel platform like ce4100 and
>intel-mid disables it explicit.
>
>I mentioned "hack" because following osdev.org[1] using ACPI is the
>correct way to detect if i8042 exists. Pardon me if this not applies in
>this situation, or if I missed something.

That is the proper way of detecting i8042 if you trust firmware; historically we do not, and so we want to make sure that PNP data agrees with fadt data.


Thanks.

-- 
Dmitry