| Message ID | 20160211120551.4d8224a4@endymion |
|---|---|
| State | New |
| Headers | show |
On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: > The pinctrl-baytrail driver builds just fine as a module so give > users this option. IIRC the reason why this is built-in is that there are all kind of ACPI GPIO magic happening behind the scenes on Baytrail-T based machines (such as Asus T100) and it does not work without the GPIO driver. Some of the stuff is done pretty early too. I'm adding Mathias the original author of the driver if he remembers the details better. > Signed-off-by: Jean Delvare <jdelvare@suse.de> > Cc: Mika Westerberg <mika.westerberg@linux.intel.com> > Cc: Heikki Krogerus <heikki.krogerus@linux.intel.com> > Cc: Linus Walleij <linus.walleij@linaro.org> > --- > drivers/pinctrl/intel/Kconfig | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > --- linux-4.5-rc3.orig/drivers/pinctrl/intel/Kconfig 2016-01-11 00:01:32.000000000 +0100 > +++ linux-4.5-rc3/drivers/pinctrl/intel/Kconfig 2016-02-11 10:55:54.817711599 +0100 > @@ -3,7 +3,7 @@ > # > > config PINCTRL_BAYTRAIL > - bool "Intel Baytrail GPIO pin control" > + tristate "Intel Baytrail GPIO pin control" > depends on GPIOLIB && ACPI > select GPIOLIB_IRQCHIP > help > > > -- > Jean Delvare > SUSE L3 Support -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Feb 11, 2016 at 12:23 PM, Mika Westerberg <mika.westerberg@linux.intel.com> wrote: > On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: >> The pinctrl-baytrail driver builds just fine as a module so give >> users this option. > > IIRC the reason why this is built-in is that there are all kind of ACPI > GPIO magic happening behind the scenes on Baytrail-T based machines > (such as Asus T100) and it does not work without the GPIO driver. Some > of the stuff is done pretty early too. Hm I wonder if that could be the case for the AMD driver that just got tristated too... Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Feb 16, 2016 at 12:13:36AM +0100, Linus Walleij wrote: > On Thu, Feb 11, 2016 at 12:23 PM, Mika Westerberg > <mika.westerberg@linux.intel.com> wrote: > > On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: > >> The pinctrl-baytrail driver builds just fine as a module so give > >> users this option. > > > > IIRC the reason why this is built-in is that there are all kind of ACPI > > GPIO magic happening behind the scenes on Baytrail-T based machines > > (such as Asus T100) and it does not work without the GPIO driver. Some > > of the stuff is done pretty early too. > > Hm I wonder if that could be the case for the AMD driver that just > got tristated too... If they use ACPI GPIO events and operation regions early at boot then it can be the case. -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Mika, Linus, On Tue, 16 Feb 2016 10:53:46 +0200, Mika Westerberg wrote: > On Tue, Feb 16, 2016 at 12:13:36AM +0100, Linus Walleij wrote: > > On Thu, Feb 11, 2016 at 12:23 PM, Mika Westerberg > > <mika.westerberg@linux.intel.com> wrote: > > > On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: > > >> The pinctrl-baytrail driver builds just fine as a module so give > > >> users this option. > > > > > > IIRC the reason why this is built-in is that there are all kind of ACPI > > > GPIO magic happening behind the scenes on Baytrail-T based machines > > > (such as Asus T100) and it does not work without the GPIO driver. Some > > > of the stuff is done pretty early too. > > > > Hm I wonder if that could be the case for the AMD driver that just > > got tristated too... > > If they use ACPI GPIO events and operation regions early at boot then it > can be the case. TBH my patches were more RFCs or even wishful thinking than real patches. Sorry for not being clearer about this. Hardware-specific code which can only be built-in is a problem, because it doesn't scale in the long run, so the pinctrl-baytrail and pinctrl-amd drivers as they exist today do not please me. Same for acpi_lpss and acpi_apd, btw... Hardware-specific code that is always built-in. If we keep doing that then distribution kernels will just grow and grow beyond reason. So if there is any way to let this code be modularized, that would be great. At this point I still did not see the explanation as to why it can't be done. "All kind of ACPI GPIO magic happening behind the scenes" is not an explanation, that's something I would expect from the marketing department, not engineering ;-) I'd like to know what exactly is being done, how and why it technically can't be done after loading a module, it that's actually the case. Thanks,
On Tue, Feb 16, 2016 at 10:15:43AM +0100, Jean Delvare wrote: > Hi Mika, Linus, > > On Tue, 16 Feb 2016 10:53:46 +0200, Mika Westerberg wrote: > > On Tue, Feb 16, 2016 at 12:13:36AM +0100, Linus Walleij wrote: > > > On Thu, Feb 11, 2016 at 12:23 PM, Mika Westerberg > > > <mika.westerberg@linux.intel.com> wrote: > > > > On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: > > > >> The pinctrl-baytrail driver builds just fine as a module so give > > > >> users this option. > > > > > > > > IIRC the reason why this is built-in is that there are all kind of ACPI > > > > GPIO magic happening behind the scenes on Baytrail-T based machines > > > > (such as Asus T100) and it does not work without the GPIO driver. Some > > > > of the stuff is done pretty early too. > > > > > > Hm I wonder if that could be the case for the AMD driver that just > > > got tristated too... > > > > If they use ACPI GPIO events and operation regions early at boot then it > > can be the case. > > TBH my patches were more RFCs or even wishful thinking than real > patches. Sorry for not being clearer about this. > > Hardware-specific code which can only be built-in is a problem, because > it doesn't scale in the long run, so the pinctrl-baytrail and > pinctrl-amd drivers as they exist today do not please me. Same for > acpi_lpss and acpi_apd, btw... Hardware-specific code that is always > built-in. If we keep doing that then distribution kernels will just > grow and grow beyond reason. > > So if there is any way to let this code be modularized, that would be > great. At this point I still did not see the explanation as to why it > can't be done. "All kind of ACPI GPIO magic happening behind the > scenes" is not an explanation, that's something I would expect from the > marketing department, not engineering ;-) I'd like to know what exactly > is being done, how and why it technically can't be done after loading a > module, it that's actually the case. Like I said I remember there was some problems if the driver was not loaded early enough. By ACPI GPIO magic I mean that there are things like this for example in Asus T100: Device (GPO2) { ... Method (_AEI, 0, NotSerialized) // _AEI: ACPI Event Interrupts { Name (RBUF, ResourceTemplate () { GpioInt (Edge, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, "\\_SB.GPO2", 0x00, ResourceConsumer, , ) { // Pin list 0x0006 } GpioInt (Edge, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, "\\_SB.GPO2", 0x00, ResourceConsumer, , ) { // Pin list 0x0012 } }) Name (FBUF, ResourceTemplate () { GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, "\\_SB.GPO2", 0x00, ResourceConsumer, , ) { // Pin list 0x0006 } GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, "\\_SB.GPO2", 0x00, ResourceConsumer, , ) { // Pin list 0x0012 } }) If ((FSAS == One)) { Return (FBUF) /* \_SB_.GPO2._AEI.FBUF */ } Return (RBUF) /* \_SB_.GPO2._AEI.RBUF */ } Name (BFSA, Buffer (0x03) { 0x00, 0x01, 0x00 /* ... */ }) CreateByteField (BFSA, Zero, BYYY) CreateByteField (BFSA, 0x02, DATT) Name (BFSB, Buffer (0x03) { 0x00, 0x01, 0x00 /* ... */ }) CreateByteField (BFSB, Zero, YYYY) CreateByteField (BFSB, 0x02, DTTT) Method (_L06, 0, NotSerialized) // _Lxx: Level-Triggered GPE { } Name (BUFF, Buffer (0x07) { 0xFF /* . */ }) CreateByteField (BUFF, Zero, STAT) CreateByteField (BUFF, One, LEN) CreateByteField (BUFF, 0x02, TMP0) CreateByteField (BUFF, 0x03, AX00) CreateByteField (BUFF, 0x04, AX01) CreateByteField (BUFF, 0x05, AX10) CreateByteField (BUFF, 0x06, AX11) Method (_L12, 0, NotSerialized) // _Lxx: Level-Triggered GPE { Local0 = Zero If (CondRefOf (\_SB.I2C1.BATC, Local1)) { Local0 = ^^I2C1.BATC.INTR () If ((0xFF == Local0)) { ADBG ("INTR RD FAIL") Return (Zero) } If ((Zero == Local0)) { Return (Zero) } ADBG ("ULPMC INTR") ADBG (Local0) Notify (ADP1, 0x80) // Status Change Notify (^^I2C1.BATC, 0x80) // Status Change Notify (^^I2C1.BATC, 0x81) // Information Change } If ((Local0 == 0x0E)) { Notify (STR3, 0x90) // Device-Specific If ((^^I2C1.AVBL == One)) { BUFF = ^^I2C1.THRM /* \_SB_.I2C1.THRM */ If ((STAT == Zero)) { Local4 = AX01 /* \_SB_.GPO2.AX01 */ Local4 &= 0xEF Local5 = AX11 /* \_SB_.GPO2.AX11 */ Local5 &= 0xEF AX01 = Local4 AX11 = Local5 ^^I2C1.THRM = BUFF /* \_SB_.GPO2.BUFF */ } } } If (CondRefOf (\_SB.DPTF, Local3)) { If (((Local0 == One) || (Local0 == 0x02))) { Notify (DPTF, 0x86) // Device-Specific Notify (TCHG, 0x80) // Status Change } If (((Local0 == 0x07) || (Local0 == 0x08))) { Notify (DPTF, 0x86) // Device-Specific Notify (TCHG, 0x80) // Status Change } } } } The method _L12 is executed every time GPIO 12 is pulled low. Now, I'm not exactly sure what this method does but it at least mentions BATC so it might read battery status or something like that. If the GPIO triggers early enough the event is missing. Not sure if this is going to cause problems or not. I'm not opposing turning this driver to a module. If it breaks things we can always revert the patch ;-) -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, 16 Feb 2016 11:49:54 +0200, Mika Westerberg wrote: > On Tue, Feb 16, 2016 at 10:15:43AM +0100, Jean Delvare wrote: > > Hi Mika, Linus, > > > > On Tue, 16 Feb 2016 10:53:46 +0200, Mika Westerberg wrote: > > > On Tue, Feb 16, 2016 at 12:13:36AM +0100, Linus Walleij wrote: > > > > On Thu, Feb 11, 2016 at 12:23 PM, Mika Westerberg > > > > <mika.westerberg@linux.intel.com> wrote: > > > > > On Thu, Feb 11, 2016 at 12:05:51PM +0100, Jean Delvare wrote: > > > > >> The pinctrl-baytrail driver builds just fine as a module so give > > > > >> users this option. > > > > > > > > > > IIRC the reason why this is built-in is that there are all kind of ACPI > > > > > GPIO magic happening behind the scenes on Baytrail-T based machines > > > > > (such as Asus T100) and it does not work without the GPIO driver. Some > > > > > of the stuff is done pretty early too. > > > > > > > > Hm I wonder if that could be the case for the AMD driver that just > > > > got tristated too... > > > > > > If they use ACPI GPIO events and operation regions early at boot then it > > > can be the case. > > > > TBH my patches were more RFCs or even wishful thinking than real > > patches. Sorry for not being clearer about this. > > > > Hardware-specific code which can only be built-in is a problem, because > > it doesn't scale in the long run, so the pinctrl-baytrail and > > pinctrl-amd drivers as they exist today do not please me. Same for > > acpi_lpss and acpi_apd, btw... Hardware-specific code that is always > > built-in. If we keep doing that then distribution kernels will just > > grow and grow beyond reason. > > > > So if there is any way to let this code be modularized, that would be > > great. At this point I still did not see the explanation as to why it > > can't be done. "All kind of ACPI GPIO magic happening behind the > > scenes" is not an explanation, that's something I would expect from the > > marketing department, not engineering ;-) I'd like to know what exactly > > is being done, how and why it technically can't be done after loading a > > module, it that's actually the case. > > Like I said I remember there was some problems if the driver was not > loaded early enough. By ACPI GPIO magic I mean that there are things > like this for example in Asus T100: > > Device (GPO2) > { > ... > Method (_AEI, 0, NotSerialized) // _AEI: ACPI Event Interrupts > { > Name (RBUF, ResourceTemplate () > { > GpioInt (Edge, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, > "\\_SB.GPO2", 0x00, ResourceConsumer, , > ) > { // Pin list > 0x0006 > } > GpioInt (Edge, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, > "\\_SB.GPO2", 0x00, ResourceConsumer, , > ) > { // Pin list > 0x0012 > } > }) > Name (FBUF, ResourceTemplate () > { > GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, > "\\_SB.GPO2", 0x00, ResourceConsumer, , > ) > { // Pin list > 0x0006 > } > GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000, > "\\_SB.GPO2", 0x00, ResourceConsumer, , > ) > { // Pin list > 0x0012 > } > }) > If ((FSAS == One)) > { > Return (FBUF) /* \_SB_.GPO2._AEI.FBUF */ > } > > Return (RBUF) /* \_SB_.GPO2._AEI.RBUF */ > } > > Name (BFSA, Buffer (0x03) > { > 0x00, 0x01, 0x00 /* ... */ > }) > CreateByteField (BFSA, Zero, BYYY) > CreateByteField (BFSA, 0x02, DATT) > Name (BFSB, Buffer (0x03) > { > 0x00, 0x01, 0x00 /* ... */ > }) > CreateByteField (BFSB, Zero, YYYY) > CreateByteField (BFSB, 0x02, DTTT) > Method (_L06, 0, NotSerialized) // _Lxx: Level-Triggered GPE > { > } > > Name (BUFF, Buffer (0x07) > { > 0xFF /* . */ > }) > CreateByteField (BUFF, Zero, STAT) > CreateByteField (BUFF, One, LEN) > CreateByteField (BUFF, 0x02, TMP0) > CreateByteField (BUFF, 0x03, AX00) > CreateByteField (BUFF, 0x04, AX01) > CreateByteField (BUFF, 0x05, AX10) > CreateByteField (BUFF, 0x06, AX11) > Method (_L12, 0, NotSerialized) // _Lxx: Level-Triggered GPE > { > Local0 = Zero > If (CondRefOf (\_SB.I2C1.BATC, Local1)) > { > Local0 = ^^I2C1.BATC.INTR () > If ((0xFF == Local0)) > { > ADBG ("INTR RD FAIL") > Return (Zero) > } > > If ((Zero == Local0)) > { > Return (Zero) > } > > ADBG ("ULPMC INTR") > ADBG (Local0) > Notify (ADP1, 0x80) // Status Change > Notify (^^I2C1.BATC, 0x80) // Status Change > Notify (^^I2C1.BATC, 0x81) // Information Change > } > > If ((Local0 == 0x0E)) > { > Notify (STR3, 0x90) // Device-Specific > If ((^^I2C1.AVBL == One)) > { > BUFF = ^^I2C1.THRM /* \_SB_.I2C1.THRM */ > If ((STAT == Zero)) > { > Local4 = AX01 /* \_SB_.GPO2.AX01 */ > Local4 &= 0xEF > Local5 = AX11 /* \_SB_.GPO2.AX11 */ > Local5 &= 0xEF > AX01 = Local4 > AX11 = Local5 > ^^I2C1.THRM = BUFF /* \_SB_.GPO2.BUFF */ > } > } > } > > If (CondRefOf (\_SB.DPTF, Local3)) > { > If (((Local0 == One) || (Local0 == 0x02))) > { > Notify (DPTF, 0x86) // Device-Specific > Notify (TCHG, 0x80) // Status Change > } > > If (((Local0 == 0x07) || (Local0 == 0x08))) > { > Notify (DPTF, 0x86) // Device-Specific > Notify (TCHG, 0x80) // Status Change > } > } > } > > } > > The method _L12 is executed every time GPIO 12 is pulled low. Now, I'm > not exactly sure what this method does but it at least mentions BATC so > it might read battery status or something like that. If the GPIO > triggers early enough the event is missing. Not sure if this is going to > cause problems or not. Can you describe the respective roles of acpi_lpss and pinctrl-baytrail in this process? Both reference ACPI devices "INT33B2" and "INT33FC". Who instantiates these devices? Who pulls GPIO 12 low? Who detects when this happens? Who triggers the event that leads to _L12 being executed? > I'm not opposing turning this driver to a module. If it breaks things we > can always revert the patch ;-) ... or try to solve the problem cleanly, if that is possible.
On Tue, Feb 16, 2016 at 11:25:55AM +0100, Jean Delvare wrote: > Can you describe the respective roles of acpi_lpss and pinctrl-baytrail > in this process? Both reference ACPI devices "INT33B2" and "INT33FC". > Who instantiates these devices? Who pulls GPIO 12 low? Who detects when > this happens? Who triggers the event that leads to _L12 being executed? acpi_lpss enumerates the ACPI devices and creates platform devices from them. pinctrl-baytrail binds to the platform device. Hardware pulls down the GPIO 12. For example when the power adapter is connected. This is detected by the GPIO hardware and the driver (pinctrl-baytrail) that then executes the method (based on the code in gpiolib-acpi.c::acpi_gpiochip_request_interrupt()). > > I'm not opposing turning this driver to a module. If it breaks things we > > can always revert the patch ;-) > > ... or try to solve the problem cleanly, if that is possible. Sure. -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
--- linux-4.5-rc3.orig/drivers/pinctrl/intel/Kconfig 2016-01-11 00:01:32.000000000 +0100 +++ linux-4.5-rc3/drivers/pinctrl/intel/Kconfig 2016-02-11 10:55:54.817711599 +0100 @@ -3,7 +3,7 @@ # config PINCTRL_BAYTRAIL - bool "Intel Baytrail GPIO pin control" + tristate "Intel Baytrail GPIO pin control" depends on GPIOLIB && ACPI select GPIOLIB_IRQCHIP help
The pinctrl-baytrail driver builds just fine as a module so give users this option. Signed-off-by: Jean Delvare <jdelvare@suse.de> Cc: Mika Westerberg <mika.westerberg@linux.intel.com> Cc: Heikki Krogerus <heikki.krogerus@linux.intel.com> Cc: Linus Walleij <linus.walleij@linaro.org> --- drivers/pinctrl/intel/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)