Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
debouncing with WinBlue debounce behavior means that the GPIO doesn't
clear until a separate GPIO is used (such as touchpad).
Prefer to use legacy debouncing to avoid problems.
Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
---
drivers/pinctrl/pinctrl-amd.c | 34 ++++++++++++++++++++++++++++++++--
1 file changed, 32 insertions(+), 2 deletions(-)
diff --git a/drivers/pinctrl/pinctrl-amd.c b/drivers/pinctrl/pinctrl-amd.c
index a2468a988be3..2e1721a9249a 100644
--- a/drivers/pinctrl/pinctrl-amd.c
+++ b/drivers/pinctrl/pinctrl-amd.c
@@ -8,6 +8,7 @@
*
*/
+#include <linux/dmi.h>
#include <linux/err.h>
#include <linux/bug.h>
#include <linux/kernel.h>
@@ -41,6 +42,27 @@ module_param(powerbtn, int, 0444);
MODULE_PARM_DESC(powerbtn,
"Power button debouncing: 0=traditional, 1=windows, -1=auto");
+struct pinctrl_amd_dmi_quirk {
+ int powerbtn;
+};
+
+static const struct dmi_system_id pinctrl_amd_dmi_quirks[] __initconst = {
+ {
+ /*
+ * Lenovo Ideapad 5
+ * Power button GPIO not cleared until touchpad movement
+ * https://bugzilla.kernel.org/show_bug.cgi?id=217833
+ */
+ .matches = {
+ DMI_MATCH(DMI_SYS_VENDOR, "LENOVO"),
+ DMI_MATCH(DMI_PRODUCT_NAME, "82LM"),
+ },
+ .driver_data = &(struct pinctrl_amd_dmi_quirk) {
+ .powerbtn = 0,
+ },
+ }
+};
+
static int amd_gpio_get_direction(struct gpio_chip *gc, unsigned offset)
{
unsigned long flags;
@@ -1084,8 +1106,16 @@ static void handle_powerbtn(struct amd_gpio *gpio_dev)
{
u32 pin_reg;
- if (powerbtn == -1)
- return;
+ if (powerbtn == -1) {
+ const struct pinctrl_amd_dmi_quirk *quirk = NULL;
+ const struct dmi_system_id *id;
+
+ id = dmi_first_match(pinctrl_amd_dmi_quirks);
+ if (!id)
+ return;
+ quirk = id->driver_data;
+ powerbtn = quirk->powerbtn;
+ }
pin_reg = readl(gpio_dev->base + WAKE_INT_MASTER_REG);
switch (powerbtn) {
--
2.34.1Hi Mario,
On 8/29/23 18:56, Mario Limonciello wrote:
> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
> debouncing with WinBlue debounce behavior means that the GPIO doesn't
> clear until a separate GPIO is used (such as touchpad).
>
> Prefer to use legacy debouncing to avoid problems.
>
> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
I'm not happy to see yet another DMI quirk solution here.
and I guess you're not happy with this either...
Are we sure there is no other way? Did you check an acpidump
for the laptop and specifically for its ACPI powerbutton handling?
I would expect the ACPI powerbutton handler to somehow clear
the bit, like how patch 1/3 clears it from the GPIO chip's
own IRQ handler.
I see that drivers/acpi/button.c does:
static u32 acpi_button_event(void *data)
{
acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
return ACPI_INTERRUPT_HANDLED;
}
So unless I'm misreading something here, there is some AML being
executed on power-button events. So maybe there is something wrong
with how Linux interprets that AML ?
Regards,
Hans
> ---
> drivers/pinctrl/pinctrl-amd.c | 34 ++++++++++++++++++++++++++++++++--
> 1 file changed, 32 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/pinctrl/pinctrl-amd.c b/drivers/pinctrl/pinctrl-amd.c
> index a2468a988be3..2e1721a9249a 100644
> --- a/drivers/pinctrl/pinctrl-amd.c
> +++ b/drivers/pinctrl/pinctrl-amd.c
> @@ -8,6 +8,7 @@
> *
> */
>
> +#include <linux/dmi.h>
> #include <linux/err.h>
> #include <linux/bug.h>
> #include <linux/kernel.h>
> @@ -41,6 +42,27 @@ module_param(powerbtn, int, 0444);
> MODULE_PARM_DESC(powerbtn,
> "Power button debouncing: 0=traditional, 1=windows, -1=auto");
>
> +struct pinctrl_amd_dmi_quirk {
> + int powerbtn;
> +};
> +
> +static const struct dmi_system_id pinctrl_amd_dmi_quirks[] __initconst = {
> + {
> + /*
> + * Lenovo Ideapad 5
> + * Power button GPIO not cleared until touchpad movement
> + * https://bugzilla.kernel.org/show_bug.cgi?id=217833
> + */
> + .matches = {
> + DMI_MATCH(DMI_SYS_VENDOR, "LENOVO"),
> + DMI_MATCH(DMI_PRODUCT_NAME, "82LM"),
> + },
> + .driver_data = &(struct pinctrl_amd_dmi_quirk) {
> + .powerbtn = 0,
> + },
> + }
> +};
> +
> static int amd_gpio_get_direction(struct gpio_chip *gc, unsigned offset)
> {
> unsigned long flags;
> @@ -1084,8 +1106,16 @@ static void handle_powerbtn(struct amd_gpio *gpio_dev)
> {
> u32 pin_reg;
>
> - if (powerbtn == -1)
> - return;
> + if (powerbtn == -1) {
> + const struct pinctrl_amd_dmi_quirk *quirk = NULL;
> + const struct dmi_system_id *id;
> +
> + id = dmi_first_match(pinctrl_amd_dmi_quirks);
> + if (!id)
> + return;
> + quirk = id->driver_data;
> + powerbtn = quirk->powerbtn;
> + }
>
> pin_reg = readl(gpio_dev->base + WAKE_INT_MASTER_REG);
> switch (powerbtn) {
On 8/29/2023 14:54, Hans de Goede wrote:
> Hi Mario,
>
> On 8/29/23 18:56, Mario Limonciello wrote:
>> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
>> debouncing with WinBlue debounce behavior means that the GPIO doesn't
>> clear until a separate GPIO is used (such as touchpad).
>>
>> Prefer to use legacy debouncing to avoid problems.
>>
>> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>
> I'm not happy to see yet another DMI quirk solution here.
>
> and I guess you're not happy with this either...
Yeah I was really hoping the first patch was enough for the issue.
If we can't come up with anything else we can potentially drop patches 2
and 3. Even patch 1 alone will "significantly" improve the situation.
The other option I considered is to hardcode WinBlue debounce behavior
"off" in Linux.
I don't think this is a good idea though though because we will likely
trade bugs because the debounce values in the AML for systems using _AEI
aren't actually used in Windows and might not have good values.
>
> Are we sure there is no other way? Did you check an acpidump
> for the laptop and specifically for its ACPI powerbutton handling?
I'm not sure there is another way or not, but yes there is an acpidump
attached to the bug in case you or anyone else has some ideas.
>
> I would expect the ACPI powerbutton handler to somehow clear
> the bit, like how patch 1/3 clears it from the GPIO chip's
> own IRQ handler.
>
> I see that drivers/acpi/button.c does:
>
> static u32 acpi_button_event(void *data)
> {
> acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
> return ACPI_INTERRUPT_HANDLED;
> }
>
> So unless I'm misreading something here, there is some AML being
> executed on power-button events. So maybe there is something wrong
> with how Linux interprets that AML ?
>
The relevant ACPI spec section is here:
https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/04_ACPI_Hardware_Specification/ACPI_Hardware_Specification.html#control-method-power-button
I did look at the acpidump. GPE 08 notifies \_SB.PWRB (a PNP0C0C
device) with 0x2. According to the spec this is specifically for
letting the system know the power button is waking up the system from G1.
I don't see any notifications with data 0x80 in the AML. So I'm
wondering if the AML designer made an assumption that pressing the power
button caused the system to suspend or turn off. That's the Windows
behavior, isn't it?
The spec even says:
"While the system is in the working state, a power button press is a
user request to transition the system into either the sleeping (G1) or
soft-off state (G2)."
I think it would be a really good data point whether this happens with
Windows if possible too. Then we can know if we're looking at a Linux
bug or a platform bug.
Hi,
On 8/29/23 23:37, Mario Limonciello wrote:
> On 8/29/2023 14:54, Hans de Goede wrote:
>> Hi Mario,
>>
>> On 8/29/23 18:56, Mario Limonciello wrote:
>>> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
>>> debouncing with WinBlue debounce behavior means that the GPIO doesn't
>>> clear until a separate GPIO is used (such as touchpad).
>>>
>>> Prefer to use legacy debouncing to avoid problems.
>>>
>>> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
>>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
>>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>>
>> I'm not happy to see yet another DMI quirk solution here.
>>
>> and I guess you're not happy with this either...
>
> Yeah I was really hoping the first patch was enough for the issue.
>
> If we can't come up with anything else we can potentially drop patches 2 and 3. Even patch 1 alone will "significantly" improve the situation.
>
> The other option I considered is to hardcode WinBlue debounce behavior "off" in Linux.
>
> I don't think this is a good idea though though because we will likely trade bugs because the debounce values in the AML for systems using _AEI aren't actually used in Windows and might not have good values.
What if we turn off the WinBlue debounce behavior for GPIO0 and then just hardcode some sane debounce values for it, overriding whatever the DSDT _AEI entries contain ?
>> Are we sure there is no other way? Did you check an acpidump
>> for the laptop and specifically for its ACPI powerbutton handling?
>
> I'm not sure there is another way or not, but yes there is an acpidump attached to the bug in case you or anyone else has some ideas.
>
>>
>> I would expect the ACPI powerbutton handler to somehow clear
>> the bit, like how patch 1/3 clears it from the GPIO chip's
>> own IRQ handler.
>>
>> I see that drivers/acpi/button.c does:
>>
>> static u32 acpi_button_event(void *data)
>> {
>> acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
>> return ACPI_INTERRUPT_HANDLED;
>> }
>>
>> So unless I'm misreading something here, there is some AML being
>> executed on power-button events. So maybe there is something wrong
>> with how Linux interprets that AML ?
>>
> The relevant ACPI spec section is here:
>
> https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/04_ACPI_Hardware_Specification/ACPI_Hardware_Specification.html#control-method-power-button
>
> I did look at the acpidump. GPE 08 notifies \_SB.PWRB (a PNP0C0C device) with 0x2. According to the spec this is specifically for letting the system know the power button is waking up the system from G1.
Sorry, the acpi_os_execute() function name gave me the impression that this would actually call some ACPI defined function, since normally in acpi speak execute refers to an ACPI table defined method.
But that is not the case here it is just a wrapper to deferred-exec the passed in function pointer.
To be clear I was hoping that there was an ACPI defined (AML code) function which would maybe clear the GPIO for us and that that was maybe not working due to e.g. some opregion not being implemented by Linux. But no AML code is being executed at all, so this is all a red herring.
Regards,
Hans
On 8/30/2023 10:37, Hans de Goede wrote:
> Hi,
>
> On 8/29/23 23:37, Mario Limonciello wrote:
>> On 8/29/2023 14:54, Hans de Goede wrote:
>>> Hi Mario,
>>>
>>> On 8/29/23 18:56, Mario Limonciello wrote:
>>>> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
>>>> debouncing with WinBlue debounce behavior means that the GPIO doesn't
>>>> clear until a separate GPIO is used (such as touchpad).
>>>>
>>>> Prefer to use legacy debouncing to avoid problems.
>>>>
>>>> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
>>>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
>>>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>>>
>>> I'm not happy to see yet another DMI quirk solution here.
>>>
>>> and I guess you're not happy with this either...
>>
>> Yeah I was really hoping the first patch was enough for the issue.
>>
>> If we can't come up with anything else we can potentially drop patches 2 and 3. Even patch 1 alone will "significantly" improve the situation.
>>
>> The other option I considered is to hardcode WinBlue debounce behavior "off" in Linux.
>>
>> I don't think this is a good idea though though because we will likely trade bugs because the debounce values in the AML for systems using _AEI aren't actually used in Windows and might not have good values.
>
> What if we turn off the WinBlue debounce behavior for GPIO0 and then just hardcode some sane debounce values for it, overriding whatever the DSDT _AEI entries contain ?
>
I don't think this is a good idea.
Some vendors GPIO0 doesn't connect to the power button but instead to
the EC. If it's connected to the EC, the EC might instead trigger GPIO0
for lid or power button or whatever they decided for a design specific way.
I'd worry that we're going to end up with inconsistent results if they
have their own debouncing put in place in the EC *because* they were
relying upon the Winblue debounce behavior.
After all - this was fixed because of
https://bugzilla.kernel.org/show_bug.cgi?id=217315
>>> Are we sure there is no other way? Did you check an acpidump
>>> for the laptop and specifically for its ACPI powerbutton handling?
>>
>> I'm not sure there is another way or not, but yes there is an acpidump attached to the bug in case you or anyone else has some ideas.
>>
>>>
>>> I would expect the ACPI powerbutton handler to somehow clear
>>> the bit, like how patch 1/3 clears it from the GPIO chip's
>>> own IRQ handler.
>>>
>>> I see that drivers/acpi/button.c does:
>>>
>>> static u32 acpi_button_event(void *data)
>>> {
>>> acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
>>> return ACPI_INTERRUPT_HANDLED;
>>> }
>>>
>>> So unless I'm misreading something here, there is some AML being
>>> executed on power-button events. So maybe there is something wrong
>>> with how Linux interprets that AML ?
>>>
>> The relevant ACPI spec section is here:
>>
>> https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/04_ACPI_Hardware_Specification/ACPI_Hardware_Specification.html#control-method-power-button
>>
>> I did look at the acpidump. GPE 08 notifies \_SB.PWRB (a PNP0C0C device) with 0x2. According to the spec this is specifically for letting the system know the power button is waking up the system from G1.
>
> Sorry, the acpi_os_execute() function name gave me the impression that this would actually call some ACPI defined function, since normally in acpi speak execute refers to an ACPI table defined method.
>
> But that is not the case here it is just a wrapper to deferred-exec the passed in function pointer.
>
> To be clear I was hoping that there was an ACPI defined (AML code) function which would maybe clear the GPIO for us and that that was maybe not working due to e.g. some opregion not being implemented by Linux. But no AML code is being executed at all, so this is all a red herring.
>
> Regards,
>
> Hans
>
>
Something we could do is add an extra callback for ACPI button driver to
call the GPIO controller IRQ handler. Worst case the IRQ handler does
nothing, best case it fixes this issue.
I'm not sure how we'd tie it to something spec compliant.
Hi,
On 8/30/23 17:47, Mario Limonciello wrote:
> On 8/30/2023 10:37, Hans de Goede wrote:
>> Hi,
>>
>> On 8/29/23 23:37, Mario Limonciello wrote:
>>> On 8/29/2023 14:54, Hans de Goede wrote:
>>>> Hi Mario,
>>>>
>>>> On 8/29/23 18:56, Mario Limonciello wrote:
>>>>> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
>>>>> debouncing with WinBlue debounce behavior means that the GPIO doesn't
>>>>> clear until a separate GPIO is used (such as touchpad).
>>>>>
>>>>> Prefer to use legacy debouncing to avoid problems.
>>>>>
>>>>> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
>>>>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
>>>>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>>>>
>>>> I'm not happy to see yet another DMI quirk solution here.
>>>>
>>>> and I guess you're not happy with this either...
>>>
>>> Yeah I was really hoping the first patch was enough for the issue.
>>>
>>> If we can't come up with anything else we can potentially drop patches 2 and 3. Even patch 1 alone will "significantly" improve the situation.
>>>
>>> The other option I considered is to hardcode WinBlue debounce behavior "off" in Linux.
>>>
>>> I don't think this is a good idea though though because we will likely trade bugs because the debounce values in the AML for systems using _AEI aren't actually used in Windows and might not have good values.
>>
>> What if we turn off the WinBlue debounce behavior for GPIO0 and then just hardcode some sane debounce values for it, overriding whatever the DSDT _AEI entries contain ?
>>
>
> I don't think this is a good idea.
>
> Some vendors GPIO0 doesn't connect to the power button but instead to the EC. If it's connected to the EC, the EC might instead trigger GPIO0 for lid or power button or whatever they decided for a design specific way.
>
> I'd worry that we're going to end up with inconsistent results if they have their own debouncing put in place in the EC *because* they were relying upon the Winblue debounce behavior.
>
> After all - this was fixed because of https://bugzilla.kernel.org/show_bug.cgi?id=217315
Ok, that is fair.
>>>> Are we sure there is no other way? Did you check an acpidump
>>>> for the laptop and specifically for its ACPI powerbutton handling?
>>>
>>> I'm not sure there is another way or not, but yes there is an acpidump attached to the bug in case you or anyone else has some ideas.
>>>
>>>>
>>>> I would expect the ACPI powerbutton handler to somehow clear
>>>> the bit, like how patch 1/3 clears it from the GPIO chip's
>>>> own IRQ handler.
>>>>
>>>> I see that drivers/acpi/button.c does:
>>>>
>>>> static u32 acpi_button_event(void *data)
>>>> {
>>>> acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
>>>> return ACPI_INTERRUPT_HANDLED;
>>>> }
>>>>
>>>> So unless I'm misreading something here, there is some AML being
>>>> executed on power-button events. So maybe there is something wrong
>>>> with how Linux interprets that AML ?
>>>>
>>> The relevant ACPI spec section is here:
>>>
>>> https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/04_ACPI_Hardware_Specification/ACPI_Hardware_Specification.html#control-method-power-button
>>>
>>> I did look at the acpidump. GPE 08 notifies \_SB.PWRB (a PNP0C0C device) with 0x2. According to the spec this is specifically for letting the system know the power button is waking up the system from G1.
>>
>> Sorry, the acpi_os_execute() function name gave me the impression that this would actually call some ACPI defined function, since normally in acpi speak execute refers to an ACPI table defined method.
>>
>> But that is not the case here it is just a wrapper to deferred-exec the passed in function pointer.
>>
>> To be clear I was hoping that there was an ACPI defined (AML code) function which would maybe clear the GPIO for us and that that was maybe not working due to e.g. some opregion not being implemented by Linux. But no AML code is being executed at all, so this is all a red herring.
>>
>> Regards,
>>
>> Hans
>>
>>
>
> Something we could do is add an extra callback for ACPI button driver to call the GPIO controller IRQ handler. Worst case the IRQ handler does nothing, best case it fixes this issue.
> I'm not sure how we'd tie it to something spec compliant.
I was actually thinking the same thing. I think we should discuss going this route
(ACPI button driver to call the GPIO controller IRQ handler) with Rafael.
We can use the existing acpi_notifier_call_chain() mechanism and:
1. Have the button code call acpi_notifier_call_chain() on the PNP0C0C
event on button presses.
2. Have the AMD pinctrl driver register a notifier_block with
register_acpi_notifier() and then check if the source is a PNP0C0C
device based on the device_class of the acpi_bus_event struct
passed to the notifier and if it is check if GPIO0 needs
clearing.
I think this is preferable over the DMI quirk route, because this should
fix the same issue also on other affected models which we don't know
about.
Regards,
Hans
On 8/30/2023 11:18, Hans de Goede wrote:
> Hi,
>
> On 8/30/23 17:47, Mario Limonciello wrote:
>> On 8/30/2023 10:37, Hans de Goede wrote:
>>> Hi,
>>>
>>> On 8/29/23 23:37, Mario Limonciello wrote:
>>>> On 8/29/2023 14:54, Hans de Goede wrote:
>>>>> Hi Mario,
>>>>>
>>>>> On 8/29/23 18:56, Mario Limonciello wrote:
>>>>>> Lenovo ideapad 5 doesn't use interrupts for GPIO 0, and so internally
>>>>>> debouncing with WinBlue debounce behavior means that the GPIO doesn't
>>>>>> clear until a separate GPIO is used (such as touchpad).
>>>>>>
>>>>>> Prefer to use legacy debouncing to avoid problems.
>>>>>>
>>>>>> Reported-by: Luca Pigliacampo <lucapgl2001@gmail.com>
>>>>>> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217833
>>>>>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>>>>>
>>>>> I'm not happy to see yet another DMI quirk solution here.
>>>>>
>>>>> and I guess you're not happy with this either...
>>>>
>>>> Yeah I was really hoping the first patch was enough for the issue.
>>>>
>>>> If we can't come up with anything else we can potentially drop patches 2 and 3. Even patch 1 alone will "significantly" improve the situation.
>>>>
>>>> The other option I considered is to hardcode WinBlue debounce behavior "off" in Linux.
>>>>
>>>> I don't think this is a good idea though though because we will likely trade bugs because the debounce values in the AML for systems using _AEI aren't actually used in Windows and might not have good values.
>>>
>>> What if we turn off the WinBlue debounce behavior for GPIO0 and then just hardcode some sane debounce values for it, overriding whatever the DSDT _AEI entries contain ?
>>>
>>
>> I don't think this is a good idea.
>>
>> Some vendors GPIO0 doesn't connect to the power button but instead to the EC. If it's connected to the EC, the EC might instead trigger GPIO0 for lid or power button or whatever they decided for a design specific way.
>>
>> I'd worry that we're going to end up with inconsistent results if they have their own debouncing put in place in the EC *because* they were relying upon the Winblue debounce behavior.
>>
>> After all - this was fixed because of https://bugzilla.kernel.org/show_bug.cgi?id=217315
>
> Ok, that is fair.
>
>
>>>>> Are we sure there is no other way? Did you check an acpidump
>>>>> for the laptop and specifically for its ACPI powerbutton handling?
>>>>
>>>> I'm not sure there is another way or not, but yes there is an acpidump attached to the bug in case you or anyone else has some ideas.
>>>>
>>>>>
>>>>> I would expect the ACPI powerbutton handler to somehow clear
>>>>> the bit, like how patch 1/3 clears it from the GPIO chip's
>>>>> own IRQ handler.
>>>>>
>>>>> I see that drivers/acpi/button.c does:
>>>>>
>>>>> static u32 acpi_button_event(void *data)
>>>>> {
>>>>> acpi_os_execute(OSL_NOTIFY_HANDLER, acpi_button_notify_run, data);
>>>>> return ACPI_INTERRUPT_HANDLED;
>>>>> }
>>>>>
>>>>> So unless I'm misreading something here, there is some AML being
>>>>> executed on power-button events. So maybe there is something wrong
>>>>> with how Linux interprets that AML ?
>>>>>
>>>> The relevant ACPI spec section is here:
>>>>
>>>> https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/04_ACPI_Hardware_Specification/ACPI_Hardware_Specification.html#control-method-power-button
>>>>
>>>> I did look at the acpidump. GPE 08 notifies \_SB.PWRB (a PNP0C0C device) with 0x2. According to the spec this is specifically for letting the system know the power button is waking up the system from G1.
>>>
>>> Sorry, the acpi_os_execute() function name gave me the impression that this would actually call some ACPI defined function, since normally in acpi speak execute refers to an ACPI table defined method.
>>>
>>> But that is not the case here it is just a wrapper to deferred-exec the passed in function pointer.
>>>
>>> To be clear I was hoping that there was an ACPI defined (AML code) function which would maybe clear the GPIO for us and that that was maybe not working due to e.g. some opregion not being implemented by Linux. But no AML code is being executed at all, so this is all a red herring.
>>>
>>> Regards,
>>>
>>> Hans
>>>
>>>
>>
>> Something we could do is add an extra callback for ACPI button driver to call the GPIO controller IRQ handler. Worst case the IRQ handler does nothing, best case it fixes this issue.
>> I'm not sure how we'd tie it to something spec compliant.
>
> I was actually thinking the same thing. I think we should discuss going this route
> (ACPI button driver to call the GPIO controller IRQ handler) with Rafael.
>
> We can use the existing acpi_notifier_call_chain() mechanism and:
>
> 1. Have the button code call acpi_notifier_call_chain() on the PNP0C0C
> event on button presses.
>
> 2. Have the AMD pinctrl driver register a notifier_block with
> register_acpi_notifier() and then check if the source is a PNP0C0C
> device based on the device_class of the acpi_bus_event struct
> passed to the notifier and if it is check if GPIO0 needs
> clearing.
>
> I think this is preferable over the DMI quirk route, because this should
> fix the same issue also on other affected models which we don't know
> about.
>
Linus - please disregard version 1.
I provided Luca a new series that implements this approach that Hans and
I discussed and they confirmed it works.
I have some minor modifications to it to narrow where it's applied so we
don't have needless notifications and will send it for review after the
new modifications are tested as well.
On Thu, Aug 31, 2023 at 7:53 PM Mario Limonciello <mario.limonciello@amd.com> wrote: > Linus - please disregard version 1. OK! > I provided Luca a new series that implements this approach that Hans and > I discussed and they confirmed it works. > > I have some minor modifications to it to narrow where it's applied so we > don't have needless notifications and will send it for review after the > new modifications are tested as well. OK standing by, I'll wait for Hans' ACK and then merge it for fixes. Yours, Linus Walleij
Hi, On 9/12/23 09:08, Linus Walleij wrote: > On Thu, Aug 31, 2023 at 7:53 PM Mario Limonciello > <mario.limonciello@amd.com> wrote: > >> Linus - please disregard version 1. > > OK! > >> I provided Luca a new series that implements this approach that Hans and >> I discussed and they confirmed it works. >> >> I have some minor modifications to it to narrow where it's applied so we >> don't have needless notifications and will send it for review after the >> new modifications are tested as well. > > OK standing by, I'll wait for Hans' ACK and then merge it for fixes. AFAICT Mario has not posted a new version (yet), so there is nothing for me to ack (yet). Regards, Hans
On 9/12/2023 03:58, Hans de Goede wrote: > Hi, > > On 9/12/23 09:08, Linus Walleij wrote: >> On Thu, Aug 31, 2023 at 7:53 PM Mario Limonciello >> <mario.limonciello@amd.com> wrote: >> >>> Linus - please disregard version 1. >> >> OK! >> >>> I provided Luca a new series that implements this approach that Hans and >>> I discussed and they confirmed it works. >>> >>> I have some minor modifications to it to narrow where it's applied so we >>> don't have needless notifications and will send it for review after the >>> new modifications are tested as well. >> >> OK standing by, I'll wait for Hans' ACK and then merge it for fixes. > > AFAICT Mario has not posted a new version (yet), > so there is nothing for me to ack (yet). > > Regards, > > Hans > > Yeah it looked like we were about to have a new version to post last week but there are some inconsistent results that we need to understand still, especially with comparing to Windows.
On 9/12/2023 13:21, Mario Limonciello wrote: > On 9/12/2023 03:58, Hans de Goede wrote: >> Hi, >> >> On 9/12/23 09:08, Linus Walleij wrote: >>> On Thu, Aug 31, 2023 at 7:53 PM Mario Limonciello >>> <mario.limonciello@amd.com> wrote: >>> >>>> Linus - please disregard version 1. >>> >>> OK! >>> >>>> I provided Luca a new series that implements this approach that Hans >>>> and >>>> I discussed and they confirmed it works. >>>> >>>> I have some minor modifications to it to narrow where it's applied >>>> so we >>>> don't have needless notifications and will send it for review after the >>>> new modifications are tested as well. >>> >>> OK standing by, I'll wait for Hans' ACK and then merge it for fixes. >> >> AFAICT Mario has not posted a new version (yet), >> so there is nothing for me to ack (yet). >> >> Regards, >> >> Hans >> >> > > Yeah it looked like we were about to have a new version to post last > week but there are some inconsistent results that we need to understand > still, especially with comparing to Windows. I've got two pieces of news to share on this issue. 1. In further testing Luca confirmed that the issue was behaving identically in Windows. We were bug compliant. 2. In better news updating the BIOS fixed the issue in both Linux and Windows, no kernel patches needed. So no further work will be done on this series.
On Wed, Sep 13, 2023 at 11:21 PM Mario Limonciello <mario.limonciello@amd.com> wrote: > 2. In better news updating the BIOS fixed the issue in both Linux and > Windows, no kernel patches needed. > > So no further work will be done on this series. Is it easy for users to update BIOS? I.e. does fwupdmgr update work? Or does it require flashing special USB drives with FAT filesystems...? Because I'm not sure all users will do that. Or even be aware that they should. In that case detecting the situation and emitting a dev_err() telling the user to update their BIOS would be desirable I think? Yours, Linus Walleij
On 9/14/23 10:43, Linus Walleij wrote: > On Wed, Sep 13, 2023 at 11:21 PM Mario Limonciello > <mario.limonciello@amd.com> wrote: > >> 2. In better news updating the BIOS fixed the issue in both Linux and >> Windows, no kernel patches needed. >> >> So no further work will be done on this series. > Is it easy for users to update BIOS? I.e. does > fwupdmgr update work? > > Or does it require flashing special USB drives with FAT filesystems...? > > Because I'm not sure all users will do that. Or even be aware that > they should. In that case detecting the situation and emitting > a dev_err() telling the user to update their BIOS would be > desirable I think? > > Yours, > Linus Walleij sadly it's not convenient, the only way lenovo offers to update the bios is an executable to run on windows. So a user should either have a dual boot or install windows on an external drive and boot from that, also the update process might wipe every boot entry beside windows. I read that some bios updaters also run on freedos, but i didn't try
On 9/14/2023 04:08, Luca Pigliacampo wrote: > On 9/14/23 10:43, Linus Walleij wrote: > >> On Wed, Sep 13, 2023 at 11:21 PM Mario Limonciello >> <mario.limonciello@amd.com> wrote: >> >>> 2. In better news updating the BIOS fixed the issue in both Linux and >>> Windows, no kernel patches needed. >>> >>> So no further work will be done on this series. >> Is it easy for users to update BIOS? I.e. does >> fwupdmgr update work? >> >> Or does it require flashing special USB drives with FAT filesystems...? >> >> Because I'm not sure all users will do that. Or even be aware that >> they should. In that case detecting the situation and emitting >> a dev_err() telling the user to update their BIOS would be >> desirable I think? >> I'm not sure how to detect it without giving false positives to users with no problems. >> Yours, >> Linus Walleij > > sadly it's not convenient, > > the only way lenovo offers to update the bios > > is an executable to run on windows. > > > So a user should either have a dual boot > > or install windows on an external drive and boot from that, > > also the update process might wipe every boot entry beside windows. > > > I read that some bios updaters also run on freedos, but i didn't try > On some systems Lenovo offers native updates for Linux, but I guess not this one.
© 2016 - 2025 Red Hat, Inc.