RO and RW of EC may have different EC protocol version. If EC transitions
between RO and RW, but AP does not reboot (this is true for fingerprint
microcontroller / cros_fp, but not true for main ec / cros_ec), the AP
still uses the protocol version queried before transition, which can
cause problems. In the case of fingerprint microcontroller, this causes
AP to send the wrong version of EC_CMD_GET_NEXT_EVENT to RO in the
interrupt handler, which in turn prevents RO to clear the interrupt
line to AP, in an infinite loop.
Once an EC_HOST_EVENT_INTERFACE_READY is received, we know that there
might have been a transition between RO and RW, so re-query the protocol.
Signed-off-by: Yicheng Li <[email protected]>
---
Hi Enric and Marek,
> This patch landed recently in linux-next as commit
> 241a69ae8ea8e2defec751fe55dac1287aa044b8. Sadly, it causes following
> kernel oops on any key press on Samsung Exynos-based Chromebooks (Snow,
> Peach-Pit and Peach-Pi):
> Many thanks for report the issue, we will take a look ASAP and revert
> this commit meanwhile.
> Simply removing the BUG_ON() from cros_ec_get_host_event() function
> fixes the issue, but I don't know the protocol details to judge if this
> is the correct way of fixing it.
The issue was those Samsung Chromebooks (Snow, Peach-Pit and Peach-Pi)
do not support mkbp events, yet we applied the same thing to them, which
we shouldn't. This v6 should fix it: I Just added a check
if (ec_dev->mkbp_event_supported)
in cros_ec_register().
drivers/platform/chrome/cros_ec.c | 29 +++++++++++++++++++++
include/linux/platform_data/cros_ec_proto.h | 3 +++
2 files changed, 32 insertions(+)
diff --git a/drivers/platform/chrome/cros_ec.c b/drivers/platform/chrome/cros_ec.c
index 9b2d07422e17..f16804db805b 100644
--- a/drivers/platform/chrome/cros_ec.c
+++ b/drivers/platform/chrome/cros_ec.c
@@ -104,6 +104,23 @@ static int cros_ec_sleep_event(struct cros_ec_device *ec_dev, u8 sleep_event)
return ret;
}
+static int cros_ec_ready_event(struct notifier_block *nb,
+ unsigned long queued_during_suspend, void *_notify)
+{
+ struct cros_ec_device *ec_dev = container_of(nb, struct cros_ec_device,
+ notifier_ready);
+ u32 host_event = cros_ec_get_host_event(ec_dev);
+
+ if (host_event & EC_HOST_EVENT_MASK(EC_HOST_EVENT_INTERFACE_READY)) {
+ mutex_lock(&ec_dev->lock);
+ cros_ec_query_all(ec_dev);
+ mutex_unlock(&ec_dev->lock);
+ return NOTIFY_OK;
+ }
+
+ return NOTIFY_DONE;
+}
+
/**
* cros_ec_register() - Register a new ChromeOS EC, using the provided info.
* @ec_dev: Device to register.
@@ -201,6 +218,18 @@ int cros_ec_register(struct cros_ec_device *ec_dev)
dev_dbg(ec_dev->dev, "Error %d clearing sleep event to ec",
err);
+ if (ec_dev->mkbp_event_supported) {
+ /*
+ * Register the notifier for EC_HOST_EVENT_INTERFACE_READY
+ * event.
+ */
+ ec_dev->notifier_ready.notifier_call = cros_ec_ready_event;
+ err = blocking_notifier_chain_register(
+ &ec_dev->event_notifier, &ec_dev->notifier_ready);
+ if (err)
+ return err;
+ }
+
dev_info(dev, "Chrome EC device registered\n");
return 0;
diff --git a/include/linux/platform_data/cros_ec_proto.h b/include/linux/platform_data/cros_ec_proto.h
index 0d4e4aaed37a..a1c545c464e7 100644
--- a/include/linux/platform_data/cros_ec_proto.h
+++ b/include/linux/platform_data/cros_ec_proto.h
@@ -121,6 +121,8 @@ struct cros_ec_command {
* @event_data: Raw payload transferred with the MKBP event.
* @event_size: Size in bytes of the event data.
* @host_event_wake_mask: Mask of host events that cause wake from suspend.
+ * @notifier_ready: The notifier_block to let the kernel re-query EC
+ * communication protocol when the EC sends EC_HOST_EVENT_INTERFACE_READY.
* @ec: The platform_device used by the mfd driver to interface with the
* main EC.
* @pd: The platform_device used by the mfd driver to interface with the
@@ -161,6 +163,7 @@ struct cros_ec_device {
int event_size;
u32 host_event_wake_mask;
u32 last_resume_result;
+ struct notifier_block notifier_ready;
/* The platform devices used by the mfd driver */
struct platform_device *ec;
--
2.24.1
Hi
On 03.02.2020 23:53, Yicheng Li wrote:
> RO and RW of EC may have different EC protocol version. If EC transitions
> between RO and RW, but AP does not reboot (this is true for fingerprint
> microcontroller / cros_fp, but not true for main ec / cros_ec), the AP
> still uses the protocol version queried before transition, which can
> cause problems. In the case of fingerprint microcontroller, this causes
> AP to send the wrong version of EC_CMD_GET_NEXT_EVENT to RO in the
> interrupt handler, which in turn prevents RO to clear the interrupt
> line to AP, in an infinite loop.
>
> Once an EC_HOST_EVENT_INTERFACE_READY is received, we know that there
> might have been a transition between RO and RW, so re-query the protocol.
>
> Signed-off-by: Yicheng Li <[email protected]>
Tested-by: Marek Szyprowski <[email protected]>
Works fine on Samsung Exynos-based Chromebooks: Snow, Peach-Pit and
Peach-Pi.
> ---
> Hi Enric and Marek,
>
>> This patch landed recently in linux-next as commit
>> 241a69ae8ea8e2defec751fe55dac1287aa044b8. Sadly, it causes following
>> kernel oops on any key press on Samsung Exynos-based Chromebooks (Snow,
>> Peach-Pit and Peach-Pi):
>
>> Many thanks for report the issue, we will take a look ASAP and revert
>> this commit meanwhile.
>
>> Simply removing the BUG_ON() from cros_ec_get_host_event() function
>> fixes the issue, but I don't know the protocol details to judge if this
>> is the correct way of fixing it.
> The issue was those Samsung Chromebooks (Snow, Peach-Pit and Peach-Pi)
> do not support mkbp events, yet we applied the same thing to them, which
> we shouldn't. This v6 should fix it: I Just added a check
>
> if (ec_dev->mkbp_event_supported)
>
> in cros_ec_register().
>
>
>
> drivers/platform/chrome/cros_ec.c | 29 +++++++++++++++++++++
> include/linux/platform_data/cros_ec_proto.h | 3 +++
> 2 files changed, 32 insertions(+)
>
> diff --git a/drivers/platform/chrome/cros_ec.c b/drivers/platform/chrome/cros_ec.c
> index 9b2d07422e17..f16804db805b 100644
> --- a/drivers/platform/chrome/cros_ec.c
> +++ b/drivers/platform/chrome/cros_ec.c
> @@ -104,6 +104,23 @@ static int cros_ec_sleep_event(struct cros_ec_device *ec_dev, u8 sleep_event)
> return ret;
> }
>
> +static int cros_ec_ready_event(struct notifier_block *nb,
> + unsigned long queued_during_suspend, void *_notify)
> +{
> + struct cros_ec_device *ec_dev = container_of(nb, struct cros_ec_device,
> + notifier_ready);
> + u32 host_event = cros_ec_get_host_event(ec_dev);
> +
> + if (host_event & EC_HOST_EVENT_MASK(EC_HOST_EVENT_INTERFACE_READY)) {
> + mutex_lock(&ec_dev->lock);
> + cros_ec_query_all(ec_dev);
> + mutex_unlock(&ec_dev->lock);
> + return NOTIFY_OK;
> + }
> +
> + return NOTIFY_DONE;
> +}
> +
> /**
> * cros_ec_register() - Register a new ChromeOS EC, using the provided info.
> * @ec_dev: Device to register.
> @@ -201,6 +218,18 @@ int cros_ec_register(struct cros_ec_device *ec_dev)
> dev_dbg(ec_dev->dev, "Error %d clearing sleep event to ec",
> err);
>
> + if (ec_dev->mkbp_event_supported) {
> + /*
> + * Register the notifier for EC_HOST_EVENT_INTERFACE_READY
> + * event.
> + */
> + ec_dev->notifier_ready.notifier_call = cros_ec_ready_event;
> + err = blocking_notifier_chain_register(
> + &ec_dev->event_notifier, &ec_dev->notifier_ready);
> + if (err)
> + return err;
> + }
> +
> dev_info(dev, "Chrome EC device registered\n");
>
> return 0;
> diff --git a/include/linux/platform_data/cros_ec_proto.h b/include/linux/platform_data/cros_ec_proto.h
> index 0d4e4aaed37a..a1c545c464e7 100644
> --- a/include/linux/platform_data/cros_ec_proto.h
> +++ b/include/linux/platform_data/cros_ec_proto.h
> @@ -121,6 +121,8 @@ struct cros_ec_command {
> * @event_data: Raw payload transferred with the MKBP event.
> * @event_size: Size in bytes of the event data.
> * @host_event_wake_mask: Mask of host events that cause wake from suspend.
> + * @notifier_ready: The notifier_block to let the kernel re-query EC
> + * communication protocol when the EC sends EC_HOST_EVENT_INTERFACE_READY.
> * @ec: The platform_device used by the mfd driver to interface with the
> * main EC.
> * @pd: The platform_device used by the mfd driver to interface with the
> @@ -161,6 +163,7 @@ struct cros_ec_device {
> int event_size;
> u32 host_event_wake_mask;
> u32 last_resume_result;
> + struct notifier_block notifier_ready;
>
> /* The platform devices used by the mfd driver */
> struct platform_device *ec;
Best regards
--
Marek Szyprowski, PhD
Samsung R&D Institute Poland
Reviewed-by: Gwendal Grignou <[email protected]>
On Mon, Feb 3, 2020 at 11:30 PM Marek Szyprowski
<[email protected]> wrote:
>
> Hi
>
> On 03.02.2020 23:53, Yicheng Li wrote:
> > RO and RW of EC may have different EC protocol version. If EC transitions
> > between RO and RW, but AP does not reboot (this is true for fingerprint
> > microcontroller / cros_fp, but not true for main ec / cros_ec), the AP
> > still uses the protocol version queried before transition, which can
> > cause problems. In the case of fingerprint microcontroller, this causes
> > AP to send the wrong version of EC_CMD_GET_NEXT_EVENT to RO in the
> > interrupt handler, which in turn prevents RO to clear the interrupt
> > line to AP, in an infinite loop.
> >
> > Once an EC_HOST_EVENT_INTERFACE_READY is received, we know that there
> > might have been a transition between RO and RW, so re-query the protocol.
> >
> > Signed-off-by: Yicheng Li <[email protected]>
>
> Tested-by: Marek Szyprowski <[email protected]>
>
> Works fine on Samsung Exynos-based Chromebooks: Snow, Peach-Pit and
> Peach-Pi.
>
> > ---
> > Hi Enric and Marek,
> >
> >> This patch landed recently in linux-next as commit
> >> 241a69ae8ea8e2defec751fe55dac1287aa044b8. Sadly, it causes following
> >> kernel oops on any key press on Samsung Exynos-based Chromebooks (Snow,
> >> Peach-Pit and Peach-Pi):
> >
> >> Many thanks for report the issue, we will take a look ASAP and revert
> >> this commit meanwhile.
> >
> >> Simply removing the BUG_ON() from cros_ec_get_host_event() function
> >> fixes the issue, but I don't know the protocol details to judge if this
> >> is the correct way of fixing it.
> > The issue was those Samsung Chromebooks (Snow, Peach-Pit and Peach-Pi)
> > do not support mkbp events, yet we applied the same thing to them, which
> > we shouldn't. This v6 should fix it: I Just added a check
> >
> > if (ec_dev->mkbp_event_supported)
> >
> > in cros_ec_register().
> >
> >
> >
> > drivers/platform/chrome/cros_ec.c | 29 +++++++++++++++++++++
> > include/linux/platform_data/cros_ec_proto.h | 3 +++
> > 2 files changed, 32 insertions(+)
> >
> > diff --git a/drivers/platform/chrome/cros_ec.c b/drivers/platform/chrome/cros_ec.c
> > index 9b2d07422e17..f16804db805b 100644
> > --- a/drivers/platform/chrome/cros_ec.c
> > +++ b/drivers/platform/chrome/cros_ec.c
> > @@ -104,6 +104,23 @@ static int cros_ec_sleep_event(struct cros_ec_device *ec_dev, u8 sleep_event)
> > return ret;
> > }
> >
> > +static int cros_ec_ready_event(struct notifier_block *nb,
> > + unsigned long queued_during_suspend, void *_notify)
> > +{
> > + struct cros_ec_device *ec_dev = container_of(nb, struct cros_ec_device,
> > + notifier_ready);
> > + u32 host_event = cros_ec_get_host_event(ec_dev);
> > +
> > + if (host_event & EC_HOST_EVENT_MASK(EC_HOST_EVENT_INTERFACE_READY)) {
> > + mutex_lock(&ec_dev->lock);
> > + cros_ec_query_all(ec_dev);
> > + mutex_unlock(&ec_dev->lock);
> > + return NOTIFY_OK;
> > + }
> > +
> > + return NOTIFY_DONE;
> > +}
> > +
> > /**
> > * cros_ec_register() - Register a new ChromeOS EC, using the provided info.
> > * @ec_dev: Device to register.
> > @@ -201,6 +218,18 @@ int cros_ec_register(struct cros_ec_device *ec_dev)
> > dev_dbg(ec_dev->dev, "Error %d clearing sleep event to ec",
> > err);
> >
> > + if (ec_dev->mkbp_event_supported) {
> > + /*
> > + * Register the notifier for EC_HOST_EVENT_INTERFACE_READY
> > + * event.
> > + */
> > + ec_dev->notifier_ready.notifier_call = cros_ec_ready_event;
> > + err = blocking_notifier_chain_register(
> > + &ec_dev->event_notifier, &ec_dev->notifier_ready);
> > + if (err)
> > + return err;
> > + }
> > +
> > dev_info(dev, "Chrome EC device registered\n");
> >
> > return 0;
> > diff --git a/include/linux/platform_data/cros_ec_proto.h b/include/linux/platform_data/cros_ec_proto.h
> > index 0d4e4aaed37a..a1c545c464e7 100644
> > --- a/include/linux/platform_data/cros_ec_proto.h
> > +++ b/include/linux/platform_data/cros_ec_proto.h
> > @@ -121,6 +121,8 @@ struct cros_ec_command {
> > * @event_data: Raw payload transferred with the MKBP event.
> > * @event_size: Size in bytes of the event data.
> > * @host_event_wake_mask: Mask of host events that cause wake from suspend.
> > + * @notifier_ready: The notifier_block to let the kernel re-query EC
> > + * communication protocol when the EC sends EC_HOST_EVENT_INTERFACE_READY.
> > * @ec: The platform_device used by the mfd driver to interface with the
> > * main EC.
> > * @pd: The platform_device used by the mfd driver to interface with the
> > @@ -161,6 +163,7 @@ struct cros_ec_device {
> > int event_size;
> > u32 host_event_wake_mask;
> > u32 last_resume_result;
> > + struct notifier_block notifier_ready;
> >
> > /* The platform devices used by the mfd driver */
> > struct platform_device *ec;
>
> Best regards
> --
> Marek Szyprowski, PhD
> Samsung R&D Institute Poland
>
Hi,
On 4/2/20 21:40, Gwendal Grignou wrote:
> Reviewed-by: Gwendal Grignou <[email protected]>
>
> On Mon, Feb 3, 2020 at 11:30 PM Marek Szyprowski
> <[email protected]> wrote:
>>
>> Hi
>>
>> On 03.02.2020 23:53, Yicheng Li wrote:
>>> RO and RW of EC may have different EC protocol version. If EC transitions
>>> between RO and RW, but AP does not reboot (this is true for fingerprint
>>> microcontroller / cros_fp, but not true for main ec / cros_ec), the AP
>>> still uses the protocol version queried before transition, which can
>>> cause problems. In the case of fingerprint microcontroller, this causes
>>> AP to send the wrong version of EC_CMD_GET_NEXT_EVENT to RO in the
>>> interrupt handler, which in turn prevents RO to clear the interrupt
>>> line to AP, in an infinite loop.
>>>
>>> Once an EC_HOST_EVENT_INTERFACE_READY is received, we know that there
>>> might have been a transition between RO and RW, so re-query the protocol.
>>>
>>> Signed-off-by: Yicheng Li <[email protected]>
>>
>> Tested-by: Marek Szyprowski <[email protected]>
>>
>> Works fine on Samsung Exynos-based Chromebooks: Snow, Peach-Pit and
>> Peach-Pi.
>>
Thanks, we will pick this patch once 5.6-rc1 is out. For my own reference:
Acked-for-chrome-platform: Enric Balletbo i Serra <[email protected]>
>>> ---
>>> Hi Enric and Marek,
>>>
>>>> This patch landed recently in linux-next as commit
>>>> 241a69ae8ea8e2defec751fe55dac1287aa044b8. Sadly, it causes following
>>>> kernel oops on any key press on Samsung Exynos-based Chromebooks (Snow,
>>>> Peach-Pit and Peach-Pi):
>>>
>>>> Many thanks for report the issue, we will take a look ASAP and revert
>>>> this commit meanwhile.
>>>
>>>> Simply removing the BUG_ON() from cros_ec_get_host_event() function
>>>> fixes the issue, but I don't know the protocol details to judge if this
>>>> is the correct way of fixing it.
>>> The issue was those Samsung Chromebooks (Snow, Peach-Pit and Peach-Pi)
>>> do not support mkbp events, yet we applied the same thing to them, which
>>> we shouldn't. This v6 should fix it: I Just added a check
>>>
>>> if (ec_dev->mkbp_event_supported)
>>>
>>> in cros_ec_register().
>>>
>>>
>>>
>>> drivers/platform/chrome/cros_ec.c | 29 +++++++++++++++++++++
>>> include/linux/platform_data/cros_ec_proto.h | 3 +++
>>> 2 files changed, 32 insertions(+)
>>>
>>> diff --git a/drivers/platform/chrome/cros_ec.c b/drivers/platform/chrome/cros_ec.c
>>> index 9b2d07422e17..f16804db805b 100644
>>> --- a/drivers/platform/chrome/cros_ec.c
>>> +++ b/drivers/platform/chrome/cros_ec.c
>>> @@ -104,6 +104,23 @@ static int cros_ec_sleep_event(struct cros_ec_device *ec_dev, u8 sleep_event)
>>> return ret;
>>> }
>>>
>>> +static int cros_ec_ready_event(struct notifier_block *nb,
>>> + unsigned long queued_during_suspend, void *_notify)
>>> +{
>>> + struct cros_ec_device *ec_dev = container_of(nb, struct cros_ec_device,
>>> + notifier_ready);
>>> + u32 host_event = cros_ec_get_host_event(ec_dev);
>>> +
>>> + if (host_event & EC_HOST_EVENT_MASK(EC_HOST_EVENT_INTERFACE_READY)) {
>>> + mutex_lock(&ec_dev->lock);
>>> + cros_ec_query_all(ec_dev);
>>> + mutex_unlock(&ec_dev->lock);
>>> + return NOTIFY_OK;
>>> + }
>>> +
>>> + return NOTIFY_DONE;
>>> +}
>>> +
>>> /**
>>> * cros_ec_register() - Register a new ChromeOS EC, using the provided info.
>>> * @ec_dev: Device to register.
>>> @@ -201,6 +218,18 @@ int cros_ec_register(struct cros_ec_device *ec_dev)
>>> dev_dbg(ec_dev->dev, "Error %d clearing sleep event to ec",
>>> err);
>>>
>>> + if (ec_dev->mkbp_event_supported) {
>>> + /*
>>> + * Register the notifier for EC_HOST_EVENT_INTERFACE_READY
>>> + * event.
>>> + */
>>> + ec_dev->notifier_ready.notifier_call = cros_ec_ready_event;
>>> + err = blocking_notifier_chain_register(
>>> + &ec_dev->event_notifier, &ec_dev->notifier_ready);
>>> + if (err)
>>> + return err;
>>> + }
>>> +
>>> dev_info(dev, "Chrome EC device registered\n");
>>>
>>> return 0;
>>> diff --git a/include/linux/platform_data/cros_ec_proto.h b/include/linux/platform_data/cros_ec_proto.h
>>> index 0d4e4aaed37a..a1c545c464e7 100644
>>> --- a/include/linux/platform_data/cros_ec_proto.h
>>> +++ b/include/linux/platform_data/cros_ec_proto.h
>>> @@ -121,6 +121,8 @@ struct cros_ec_command {
>>> * @event_data: Raw payload transferred with the MKBP event.
>>> * @event_size: Size in bytes of the event data.
>>> * @host_event_wake_mask: Mask of host events that cause wake from suspend.
>>> + * @notifier_ready: The notifier_block to let the kernel re-query EC
>>> + * communication protocol when the EC sends EC_HOST_EVENT_INTERFACE_READY.
>>> * @ec: The platform_device used by the mfd driver to interface with the
>>> * main EC.
>>> * @pd: The platform_device used by the mfd driver to interface with the
>>> @@ -161,6 +163,7 @@ struct cros_ec_device {
>>> int event_size;
>>> u32 host_event_wake_mask;
>>> u32 last_resume_result;
>>> + struct notifier_block notifier_ready;
>>>
>>> /* The platform devices used by the mfd driver */
>>> struct platform_device *ec;
>>
>> Best regards
>> --
>> Marek Szyprowski, PhD
>> Samsung R&D Institute Poland
>>
On 3/2/20 23:53, Yicheng Li wrote:
> RO and RW of EC may have different EC protocol version. If EC transitions
> between RO and RW, but AP does not reboot (this is true for fingerprint
> microcontroller / cros_fp, but not true for main ec / cros_ec), the AP
> still uses the protocol version queried before transition, which can
> cause problems. In the case of fingerprint microcontroller, this causes
> AP to send the wrong version of EC_CMD_GET_NEXT_EVENT to RO in the
> interrupt handler, which in turn prevents RO to clear the interrupt
> line to AP, in an infinite loop.
>
> Once an EC_HOST_EVENT_INTERFACE_READY is received, we know that there
> might have been a transition between RO and RW, so re-query the protocol.
>
> Signed-off-by: Yicheng Li <[email protected]>
> ---
Queued for 5.7 with all the collected tags, thanks.