Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
On Fri, 2020-12-04 at 14:51 -0700, Jerry Snitselaar wrote: > James Bottomley @ 2020-12-03 14:05 MST: > > > On Thu, 2020-12-03 at 13:14 -0700, Jerry Snitselaar wrote: > > > Jerry Snitselaar @ 2020-12-02 23:11 MST: > > [...] > > > > The interrupt storm detection code works on the T490s. I'm not > > > > sure what is going on with the L490. I will see if I can get > > > > access to one. > > > > > > > > Jerry > > > > > > Lenovo verified that the L490 hangs. > > > > Just to confirm, that's this system: > > > > https://www.lenovo.com/us/en/laptops/thinkpad/thinkpad-l/ThinkPad-L490/p/22TP2TBL490 > > > > We could ask if lenovo will give us one, but if not we could pull a > > Jens. [the backstory is that when Jens was doing queueing in the > > block layer, there were lots of SATA devices that didn't work quite > > right but you couldn't tell unless you actually tried them > > out. Getting manufacturers to send samples is rather arduous, so > > he took to ordering them online, testing them out, and then > > returning them for a full refund within the allowed window] > > > > It looks like Lenovo has a nice christmas returns policy: > > > > https://www.lenovo.com/us/en/shopping-faq/#returns > > > > James > > Yes, that is the one. I'm seeing if we have any located somewhere, or > if Lenovo will loan me one. OK, that would be best since it's your patch. We've got 12 days left to get the free returns policy, so if you haven't managed to find one by the end of next week, I'll order one and play with it ... you could always put your manager's credit card down for one as well ... especially as it should get refunded ... > I think for the time being the patch that disabled interrupts for > the T490s could be changed to it for the L490 instead. I'll post a v3 > of my current patchset. It would probably make sense for it to go in > with your patches when they land. Absolutely ... we can't turn interrupts on and then have a load of hung laptops we already knew about ... James
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
James Bottomley @ 2020-12-03 14:05 MST: > On Thu, 2020-12-03 at 13:14 -0700, Jerry Snitselaar wrote: >> Jerry Snitselaar @ 2020-12-02 23:11 MST: > [...] >> > The interrupt storm detection code works on the T490s. I'm not sure >> > what is going on with the L490. I will see if I can get access to >> > one. >> > >> > Jerry >> >> Lenovo verified that the L490 hangs. > > Just to confirm, that's this system: > > https://www.lenovo.com/us/en/laptops/thinkpad/thinkpad-l/ThinkPad-L490/p/22TP2TBL490 > > We could ask if lenovo will give us one, but if not we could pull a > Jens. [the backstory is that when Jens was doing queueing in the block > layer, there were lots of SATA devices that didn't work quite right but > you couldn't tell unless you actually tried them out. Getting > manufacturers to send samples is rather arduous, so he took to ordering > them online, testing them out, and then returning them for a full > refund within the allowed window] > > It looks like Lenovo has a nice christmas returns policy: > > https://www.lenovo.com/us/en/shopping-faq/#returns > > James Yes, that is the one. I'm seeing if we have any located somewhere, or if Lenovo will loan me one. I think for the time being the patch that disabled interrupts for the T490s could be changed to it for the L490 instead. I'll post a v3 of my current patchset. It would probably make sense for it to go in with your patches when they land.
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
On Wed, Dec 02, 2020 at 11:11:41PM -0700, Jerry Snitselaar wrote:
>
> Jerry Snitselaar @ 2020-12-02 17:02 MST:
>
> > Jarkko Sakkinen @ 2020-12-02 09:49 MST:
> >
> >> On Tue, Dec 01, 2020 at 12:59:23PM -0700, Jerry Snitselaar wrote:
> >>>
> >>> Jerry Snitselaar @ 2020-11-30 20:26 MST:
> >>>
> >>> > Jerry Snitselaar @ 2020-11-30 19:58 MST:
> >>> >
> >>> >> When enabling the interrupt code for the tpm_tis driver we have
> >>> >> noticed some systems have a bios issue causing an interrupt storm to
> >>> >> occur. The issue isn't limited to a single tpm or system manufacturer
> >>> >> so keeping a denylist of systems with the issue isn't optimal. Instead
> >>> >> try to detect the problem occurring, disable interrupts, and revert to
> >>> >> polling when it happens.
> >>> >>
> >>> >> Cc: Jarkko Sakkinen
> >>> >> Cc: Jason Gunthorpe
> >>> >> Cc: Peter Huewe
> >>> >> Cc: James Bottomley
> >>> >> Cc: Matthew Garrett
> >>> >> Cc: Hans de Goede
> >>> >> Signed-off-by: Jerry Snitselaar
> >>> >> ---
> >>> >> v2: drop tpm_tis specific workqueue and use just system_wq
> >>> >>
> >>> >> drivers/char/tpm/tpm_tis_core.c | 27 +++
> >>> >> drivers/char/tpm/tpm_tis_core.h | 2 ++
> >>> >> 2 files changed, 29 insertions(+)
> >>> >>
> >>> >> diff --git a/drivers/char/tpm/tpm_tis_core.c
> >>> >> b/drivers/char/tpm/tpm_tis_core.c
> >>> >> index 23b60583928b..72cc8a5a152c 100644
> >>> >> --- a/drivers/char/tpm/tpm_tis_core.c
> >>> >> +++ b/drivers/char/tpm/tpm_tis_core.c
> >>> >> @@ -24,6 +24,8 @@
> >>> >> #include
> >>> >> #include
> >>> >> #include
> >>> >> +#include
> >>> >> +#include
> >>> >> #include "tpm.h"
> >>> >> #include "tpm_tis_core.h"
> >>> >>
> >>> >> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy,
> >>> >> void *dev_id)
> >>> >> {
> >>> >>struct tpm_chip *chip = dev_id;
> >>> >>struct tpm_tis_data *priv = dev_get_drvdata(>dev);
> >>> >> + static bool check_storm = true;
> >>> >> + static unsigned int check_start;
> >>> >>u32 interrupt;
> >>> >>int i, rc;
> >>> >>
> >>> >> + if (unlikely(check_storm)) {
> >>> >> + if (!check_start) {
> >>> >> + check_start = jiffies_to_msecs(jiffies);
> >>> >> + } else if ((kstat_irqs(priv->irq) > 1000) &&
> >>> >> + (jiffies_to_msecs(jiffies) - check_start <
> >>> >> 500)) {
> >>> >> + check_storm = false;
> >>> >> + schedule_work(>storm_work);
> >>> >> + } else if (jiffies_to_msecs(jiffies) - check_start >=
> >>> >> 500) {
> >>> >> + check_storm = false;
> >>> >> + }
> >>> >> + }
> >>> >> +
> >>> >>rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality),
> >>> >> );
> >>> >>if (rc < 0)
> >>> >>return IRQ_NONE;
> >>> >> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
> >>> >>.clk_enable = tpm_tis_clkrun_enable,
> >>> >> };
> >>> >>
> >>> >> +static void tpm_tis_storm_work(struct work_struct *work)
> >>> >> +{
> >>> >> + struct tpm_tis_data *priv = container_of(work, struct
> >>> >> tpm_tis_data, storm_work);
> >>> >> +
> >>> >> + disable_interrupts(priv->chip);
> >>> >> + dev_warn(>chip->dev, "Interrupt storm detected, using
> >>> >> polling.\n");
> >>> >> +}
> >>> >> +
> >>> >> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv,
> >>> >> int irq,
> >>> >> const struct tpm_tis_phy_ops *phy_ops,
> >>> >> acpi_handle acpi_dev_handle)
> >>> >> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
> >>> >> tpm_tis_data *priv, int irq,
> >>> >>if (IS_ERR(chip))
> >>> >>return PTR_ERR(chip);
> >>> >>
> >>> >> + priv->chip = chip;
> >>> >> + INIT_WORK(>storm_work, tpm_tis_storm_work);
> >>> >> +
> >>> >> #ifdef CONFIG_ACPI
> >>> >>chip->acpi_dev_handle = acpi_dev_handle;
> >>> >> #endif
> >>> >> diff --git a/drivers/char/tpm/tpm_tis_core.h
> >>> >> b/drivers/char/tpm/tpm_tis_core.h
> >>> >> index edeb5dc61c95..5630f294dc0c 100644
> >>> >> --- a/drivers/char/tpm/tpm_tis_core.h
> >>> >> +++ b/drivers/char/tpm/tpm_tis_core.h
> >>> >> @@ -95,6 +95,8 @@ struct tpm_tis_data {
> >>> >>u16 clkrun_enabled;
> >>> >>wait_queue_head_t int_queue;
> >>> >>wait_queue_head_t read_queue;
> >>> >> + struct work_struct storm_work;
> >>> >> + struct tpm_chip *chip;
> >>> >>const struct tpm_tis_phy_ops *phy_ops;
> >>> >>unsigned short rng_quality;
> >>> >> };
> >>> >
> >>> > I've tested this with the Intel platform that has an Infineon chip that
> >>> > I found the other week. It works, but isn't the complete fix. With this
> >>> > on top of James' patchset I sometimes see the message "Lost Interrupt
> >>> > waiting for TPM stat", so I guess there needs to be a check in
>
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
On Thu, 2020-12-03 at 13:14 -0700, Jerry Snitselaar wrote: > Jerry Snitselaar @ 2020-12-02 23:11 MST: [...] > > The interrupt storm detection code works on the T490s. I'm not sure > > what is going on with the L490. I will see if I can get access to > > one. > > > > Jerry > > Lenovo verified that the L490 hangs. Just to confirm, that's this system: https://www.lenovo.com/us/en/laptops/thinkpad/thinkpad-l/ThinkPad-L490/p/22TP2TBL490 We could ask if lenovo will give us one, but if not we could pull a Jens. [the backstory is that when Jens was doing queueing in the block layer, there were lots of SATA devices that didn't work quite right but you couldn't tell unless you actually tried them out. Getting manufacturers to send samples is rather arduous, so he took to ordering them online, testing them out, and then returning them for a full refund within the allowed window] It looks like Lenovo has a nice christmas returns policy: https://www.lenovo.com/us/en/shopping-faq/#returns James
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
Jerry Snitselaar @ 2020-12-02 23:11 MST:
> Jerry Snitselaar @ 2020-12-02 17:02 MST:
>
>> Jarkko Sakkinen @ 2020-12-02 09:49 MST:
>>
>>> On Tue, Dec 01, 2020 at 12:59:23PM -0700, Jerry Snitselaar wrote:
Jerry Snitselaar @ 2020-11-30 20:26 MST:
> Jerry Snitselaar @ 2020-11-30 19:58 MST:
>
>> When enabling the interrupt code for the tpm_tis driver we have
>> noticed some systems have a bios issue causing an interrupt storm to
>> occur. The issue isn't limited to a single tpm or system manufacturer
>> so keeping a denylist of systems with the issue isn't optimal. Instead
>> try to detect the problem occurring, disable interrupts, and revert to
>> polling when it happens.
>>
>> Cc: Jarkko Sakkinen
>> Cc: Jason Gunthorpe
>> Cc: Peter Huewe
>> Cc: James Bottomley
>> Cc: Matthew Garrett
>> Cc: Hans de Goede
>> Signed-off-by: Jerry Snitselaar
>> ---
>> v2: drop tpm_tis specific workqueue and use just system_wq
>>
>> drivers/char/tpm/tpm_tis_core.c | 27 +++
>> drivers/char/tpm/tpm_tis_core.h | 2 ++
>> 2 files changed, 29 insertions(+)
>>
>> diff --git a/drivers/char/tpm/tpm_tis_core.c
>> b/drivers/char/tpm/tpm_tis_core.c
>> index 23b60583928b..72cc8a5a152c 100644
>> --- a/drivers/char/tpm/tpm_tis_core.c
>> +++ b/drivers/char/tpm/tpm_tis_core.c
>> @@ -24,6 +24,8 @@
>> #include
>> #include
>> #include
>> +#include
>> +#include
>> #include "tpm.h"
>> #include "tpm_tis_core.h"
>>
>> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
>> *dev_id)
>> {
>> struct tpm_chip *chip = dev_id;
>> struct tpm_tis_data *priv = dev_get_drvdata(>dev);
>> + static bool check_storm = true;
>> + static unsigned int check_start;
>> u32 interrupt;
>> int i, rc;
>>
>> + if (unlikely(check_storm)) {
>> + if (!check_start) {
>> + check_start = jiffies_to_msecs(jiffies);
>> + } else if ((kstat_irqs(priv->irq) > 1000) &&
>> + (jiffies_to_msecs(jiffies) - check_start <
>> 500)) {
>> + check_storm = false;
>> + schedule_work(>storm_work);
>> + } else if (jiffies_to_msecs(jiffies) - check_start >=
>> 500) {
>> + check_storm = false;
>> + }
>> + }
>> +
>> rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality),
>> );
>> if (rc < 0)
>> return IRQ_NONE;
>> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
>> .clk_enable = tpm_tis_clkrun_enable,
>> };
>>
>> +static void tpm_tis_storm_work(struct work_struct *work)
>> +{
>> + struct tpm_tis_data *priv = container_of(work, struct
>> tpm_tis_data, storm_work);
>> +
>> + disable_interrupts(priv->chip);
>> + dev_warn(>chip->dev, "Interrupt storm detected, using
>> polling.\n");
>> +}
>> +
>> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv,
>> int irq,
>> const struct tpm_tis_phy_ops *phy_ops,
>> acpi_handle acpi_dev_handle)
>> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
>> tpm_tis_data *priv, int irq,
>> if (IS_ERR(chip))
>> return PTR_ERR(chip);
>>
>> + priv->chip = chip;
>> + INIT_WORK(>storm_work, tpm_tis_storm_work);
>> +
>> #ifdef CONFIG_ACPI
>> chip->acpi_dev_handle = acpi_dev_handle;
>> #endif
>> diff --git a/drivers/char/tpm/tpm_tis_core.h
>> b/drivers/char/tpm/tpm_tis_core.h
>> index edeb5dc61c95..5630f294dc0c 100644
>> --- a/drivers/char/tpm/tpm_tis_core.h
>> +++ b/drivers/char/tpm/tpm_tis_core.h
>> @@ -95,6 +95,8 @@ struct tpm_tis_data {
>> u16 clkrun_enabled;
>> wait_queue_head_t int_queue;
>> wait_queue_head_t read_queue;
>> + struct work_struct storm_work;
>> + struct tpm_chip *chip;
>> const struct tpm_tis_phy_ops *phy_ops;
>> unsigned short rng_quality;
>> };
>
> I've tested this with the Intel platform that has an Infineon chip that
> I found the other week. It works, but isn't the complete fix. With this
> on top of James' patchset I sometimes see the message "Lost Interrupt
> waiting for TPM stat", so I guess there needs to be a check in
> wait_for_tpm_stat and request_locality to see if interrupts were
> disabled when the
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
Jerry Snitselaar @ 2020-12-02 17:02 MST:
> Jarkko Sakkinen @ 2020-12-02 09:49 MST:
>
>> On Tue, Dec 01, 2020 at 12:59:23PM -0700, Jerry Snitselaar wrote:
>>>
>>> Jerry Snitselaar @ 2020-11-30 20:26 MST:
>>>
>>> > Jerry Snitselaar @ 2020-11-30 19:58 MST:
>>> >
>>> >> When enabling the interrupt code for the tpm_tis driver we have
>>> >> noticed some systems have a bios issue causing an interrupt storm to
>>> >> occur. The issue isn't limited to a single tpm or system manufacturer
>>> >> so keeping a denylist of systems with the issue isn't optimal. Instead
>>> >> try to detect the problem occurring, disable interrupts, and revert to
>>> >> polling when it happens.
>>> >>
>>> >> Cc: Jarkko Sakkinen
>>> >> Cc: Jason Gunthorpe
>>> >> Cc: Peter Huewe
>>> >> Cc: James Bottomley
>>> >> Cc: Matthew Garrett
>>> >> Cc: Hans de Goede
>>> >> Signed-off-by: Jerry Snitselaar
>>> >> ---
>>> >> v2: drop tpm_tis specific workqueue and use just system_wq
>>> >>
>>> >> drivers/char/tpm/tpm_tis_core.c | 27 +++
>>> >> drivers/char/tpm/tpm_tis_core.h | 2 ++
>>> >> 2 files changed, 29 insertions(+)
>>> >>
>>> >> diff --git a/drivers/char/tpm/tpm_tis_core.c
>>> >> b/drivers/char/tpm/tpm_tis_core.c
>>> >> index 23b60583928b..72cc8a5a152c 100644
>>> >> --- a/drivers/char/tpm/tpm_tis_core.c
>>> >> +++ b/drivers/char/tpm/tpm_tis_core.c
>>> >> @@ -24,6 +24,8 @@
>>> >> #include
>>> >> #include
>>> >> #include
>>> >> +#include
>>> >> +#include
>>> >> #include "tpm.h"
>>> >> #include "tpm_tis_core.h"
>>> >>
>>> >> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
>>> >> *dev_id)
>>> >> {
>>> >> struct tpm_chip *chip = dev_id;
>>> >> struct tpm_tis_data *priv = dev_get_drvdata(>dev);
>>> >> +static bool check_storm = true;
>>> >> +static unsigned int check_start;
>>> >> u32 interrupt;
>>> >> int i, rc;
>>> >>
>>> >> +if (unlikely(check_storm)) {
>>> >> +if (!check_start) {
>>> >> +check_start = jiffies_to_msecs(jiffies);
>>> >> +} else if ((kstat_irqs(priv->irq) > 1000) &&
>>> >> + (jiffies_to_msecs(jiffies) - check_start <
>>> >> 500)) {
>>> >> +check_storm = false;
>>> >> +schedule_work(>storm_work);
>>> >> +} else if (jiffies_to_msecs(jiffies) - check_start >=
>>> >> 500) {
>>> >> +check_storm = false;
>>> >> +}
>>> >> +}
>>> >> +
>>> >> rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality),
>>> >> );
>>> >> if (rc < 0)
>>> >> return IRQ_NONE;
>>> >> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
>>> >> .clk_enable = tpm_tis_clkrun_enable,
>>> >> };
>>> >>
>>> >> +static void tpm_tis_storm_work(struct work_struct *work)
>>> >> +{
>>> >> +struct tpm_tis_data *priv = container_of(work, struct
>>> >> tpm_tis_data, storm_work);
>>> >> +
>>> >> +disable_interrupts(priv->chip);
>>> >> +dev_warn(>chip->dev, "Interrupt storm detected, using
>>> >> polling.\n");
>>> >> +}
>>> >> +
>>> >> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv,
>>> >> int irq,
>>> >>const struct tpm_tis_phy_ops *phy_ops,
>>> >>acpi_handle acpi_dev_handle)
>>> >> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
>>> >> tpm_tis_data *priv, int irq,
>>> >> if (IS_ERR(chip))
>>> >> return PTR_ERR(chip);
>>> >>
>>> >> +priv->chip = chip;
>>> >> +INIT_WORK(>storm_work, tpm_tis_storm_work);
>>> >> +
>>> >> #ifdef CONFIG_ACPI
>>> >> chip->acpi_dev_handle = acpi_dev_handle;
>>> >> #endif
>>> >> diff --git a/drivers/char/tpm/tpm_tis_core.h
>>> >> b/drivers/char/tpm/tpm_tis_core.h
>>> >> index edeb5dc61c95..5630f294dc0c 100644
>>> >> --- a/drivers/char/tpm/tpm_tis_core.h
>>> >> +++ b/drivers/char/tpm/tpm_tis_core.h
>>> >> @@ -95,6 +95,8 @@ struct tpm_tis_data {
>>> >> u16 clkrun_enabled;
>>> >> wait_queue_head_t int_queue;
>>> >> wait_queue_head_t read_queue;
>>> >> +struct work_struct storm_work;
>>> >> +struct tpm_chip *chip;
>>> >> const struct tpm_tis_phy_ops *phy_ops;
>>> >> unsigned short rng_quality;
>>> >> };
>>> >
>>> > I've tested this with the Intel platform that has an Infineon chip that
>>> > I found the other week. It works, but isn't the complete fix. With this
>>> > on top of James' patchset I sometimes see the message "Lost Interrupt
>>> > waiting for TPM stat", so I guess there needs to be a check in
>>> > wait_for_tpm_stat and request_locality to see if interrupts were
>>> > disabled when the wait_event_interruptible_timeout call times out.
>>>
>>> As kernel test robot pointed out. kstat_irqs isn't visible when tpm_tis
>>> builds
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
Jarkko Sakkinen @ 2020-12-02 09:49 MST:
> On Tue, Dec 01, 2020 at 12:59:23PM -0700, Jerry Snitselaar wrote:
>>
>> Jerry Snitselaar @ 2020-11-30 20:26 MST:
>>
>> > Jerry Snitselaar @ 2020-11-30 19:58 MST:
>> >
>> >> When enabling the interrupt code for the tpm_tis driver we have
>> >> noticed some systems have a bios issue causing an interrupt storm to
>> >> occur. The issue isn't limited to a single tpm or system manufacturer
>> >> so keeping a denylist of systems with the issue isn't optimal. Instead
>> >> try to detect the problem occurring, disable interrupts, and revert to
>> >> polling when it happens.
>> >>
>> >> Cc: Jarkko Sakkinen
>> >> Cc: Jason Gunthorpe
>> >> Cc: Peter Huewe
>> >> Cc: James Bottomley
>> >> Cc: Matthew Garrett
>> >> Cc: Hans de Goede
>> >> Signed-off-by: Jerry Snitselaar
>> >> ---
>> >> v2: drop tpm_tis specific workqueue and use just system_wq
>> >>
>> >> drivers/char/tpm/tpm_tis_core.c | 27 +++
>> >> drivers/char/tpm/tpm_tis_core.h | 2 ++
>> >> 2 files changed, 29 insertions(+)
>> >>
>> >> diff --git a/drivers/char/tpm/tpm_tis_core.c
>> >> b/drivers/char/tpm/tpm_tis_core.c
>> >> index 23b60583928b..72cc8a5a152c 100644
>> >> --- a/drivers/char/tpm/tpm_tis_core.c
>> >> +++ b/drivers/char/tpm/tpm_tis_core.c
>> >> @@ -24,6 +24,8 @@
>> >> #include
>> >> #include
>> >> #include
>> >> +#include
>> >> +#include
>> >> #include "tpm.h"
>> >> #include "tpm_tis_core.h"
>> >>
>> >> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
>> >> *dev_id)
>> >> {
>> >> struct tpm_chip *chip = dev_id;
>> >> struct tpm_tis_data *priv = dev_get_drvdata(>dev);
>> >> + static bool check_storm = true;
>> >> + static unsigned int check_start;
>> >> u32 interrupt;
>> >> int i, rc;
>> >>
>> >> + if (unlikely(check_storm)) {
>> >> + if (!check_start) {
>> >> + check_start = jiffies_to_msecs(jiffies);
>> >> + } else if ((kstat_irqs(priv->irq) > 1000) &&
>> >> +(jiffies_to_msecs(jiffies) - check_start < 500)) {
>> >> + check_storm = false;
>> >> + schedule_work(>storm_work);
>> >> + } else if (jiffies_to_msecs(jiffies) - check_start >= 500) {
>> >> + check_storm = false;
>> >> + }
>> >> + }
>> >> +
>> >> rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), );
>> >> if (rc < 0)
>> >> return IRQ_NONE;
>> >> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
>> >> .clk_enable = tpm_tis_clkrun_enable,
>> >> };
>> >>
>> >> +static void tpm_tis_storm_work(struct work_struct *work)
>> >> +{
>> >> + struct tpm_tis_data *priv = container_of(work, struct tpm_tis_data,
>> >> storm_work);
>> >> +
>> >> + disable_interrupts(priv->chip);
>> >> + dev_warn(>chip->dev, "Interrupt storm detected, using
>> >> polling.\n");
>> >> +}
>> >> +
>> >> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int
>> >> irq,
>> >> const struct tpm_tis_phy_ops *phy_ops,
>> >> acpi_handle acpi_dev_handle)
>> >> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
>> >> tpm_tis_data *priv, int irq,
>> >> if (IS_ERR(chip))
>> >> return PTR_ERR(chip);
>> >>
>> >> + priv->chip = chip;
>> >> + INIT_WORK(>storm_work, tpm_tis_storm_work);
>> >> +
>> >> #ifdef CONFIG_ACPI
>> >> chip->acpi_dev_handle = acpi_dev_handle;
>> >> #endif
>> >> diff --git a/drivers/char/tpm/tpm_tis_core.h
>> >> b/drivers/char/tpm/tpm_tis_core.h
>> >> index edeb5dc61c95..5630f294dc0c 100644
>> >> --- a/drivers/char/tpm/tpm_tis_core.h
>> >> +++ b/drivers/char/tpm/tpm_tis_core.h
>> >> @@ -95,6 +95,8 @@ struct tpm_tis_data {
>> >> u16 clkrun_enabled;
>> >> wait_queue_head_t int_queue;
>> >> wait_queue_head_t read_queue;
>> >> + struct work_struct storm_work;
>> >> + struct tpm_chip *chip;
>> >> const struct tpm_tis_phy_ops *phy_ops;
>> >> unsigned short rng_quality;
>> >> };
>> >
>> > I've tested this with the Intel platform that has an Infineon chip that
>> > I found the other week. It works, but isn't the complete fix. With this
>> > on top of James' patchset I sometimes see the message "Lost Interrupt
>> > waiting for TPM stat", so I guess there needs to be a check in
>> > wait_for_tpm_stat and request_locality to see if interrupts were
>> > disabled when the wait_event_interruptible_timeout call times out.
>>
>> As kernel test robot pointed out. kstat_irqs isn't visible when tpm_tis
>> builds as a module. It looks like it is only called by kstat_irq_usrs,
>> and that is only by the fs/proc code. I have a patch to export it, but
>> the i915 driver open codes their own version instead of using it. Is
>> there any reason not to export it?
>
> If you add a patch that exports it, then for coherency it'd be better to
> also patch i915 driver. Jani?
>
> /Jarkko
It looks like this might not solve all cases. I'm having Lenovo test
another build
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
On Tue, Dec 01, 2020 at 12:59:23PM -0700, Jerry Snitselaar wrote:
>
> Jerry Snitselaar @ 2020-11-30 20:26 MST:
>
> > Jerry Snitselaar @ 2020-11-30 19:58 MST:
> >
> >> When enabling the interrupt code for the tpm_tis driver we have
> >> noticed some systems have a bios issue causing an interrupt storm to
> >> occur. The issue isn't limited to a single tpm or system manufacturer
> >> so keeping a denylist of systems with the issue isn't optimal. Instead
> >> try to detect the problem occurring, disable interrupts, and revert to
> >> polling when it happens.
> >>
> >> Cc: Jarkko Sakkinen
> >> Cc: Jason Gunthorpe
> >> Cc: Peter Huewe
> >> Cc: James Bottomley
> >> Cc: Matthew Garrett
> >> Cc: Hans de Goede
> >> Signed-off-by: Jerry Snitselaar
> >> ---
> >> v2: drop tpm_tis specific workqueue and use just system_wq
> >>
> >> drivers/char/tpm/tpm_tis_core.c | 27 +++
> >> drivers/char/tpm/tpm_tis_core.h | 2 ++
> >> 2 files changed, 29 insertions(+)
> >>
> >> diff --git a/drivers/char/tpm/tpm_tis_core.c
> >> b/drivers/char/tpm/tpm_tis_core.c
> >> index 23b60583928b..72cc8a5a152c 100644
> >> --- a/drivers/char/tpm/tpm_tis_core.c
> >> +++ b/drivers/char/tpm/tpm_tis_core.c
> >> @@ -24,6 +24,8 @@
> >> #include
> >> #include
> >> #include
> >> +#include
> >> +#include
> >> #include "tpm.h"
> >> #include "tpm_tis_core.h"
> >>
> >> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
> >> *dev_id)
> >> {
> >>struct tpm_chip *chip = dev_id;
> >>struct tpm_tis_data *priv = dev_get_drvdata(>dev);
> >> + static bool check_storm = true;
> >> + static unsigned int check_start;
> >>u32 interrupt;
> >>int i, rc;
> >>
> >> + if (unlikely(check_storm)) {
> >> + if (!check_start) {
> >> + check_start = jiffies_to_msecs(jiffies);
> >> + } else if ((kstat_irqs(priv->irq) > 1000) &&
> >> + (jiffies_to_msecs(jiffies) - check_start < 500)) {
> >> + check_storm = false;
> >> + schedule_work(>storm_work);
> >> + } else if (jiffies_to_msecs(jiffies) - check_start >= 500) {
> >> + check_storm = false;
> >> + }
> >> + }
> >> +
> >>rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), );
> >>if (rc < 0)
> >>return IRQ_NONE;
> >> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
> >>.clk_enable = tpm_tis_clkrun_enable,
> >> };
> >>
> >> +static void tpm_tis_storm_work(struct work_struct *work)
> >> +{
> >> + struct tpm_tis_data *priv = container_of(work, struct tpm_tis_data,
> >> storm_work);
> >> +
> >> + disable_interrupts(priv->chip);
> >> + dev_warn(>chip->dev, "Interrupt storm detected, using
> >> polling.\n");
> >> +}
> >> +
> >> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int
> >> irq,
> >> const struct tpm_tis_phy_ops *phy_ops,
> >> acpi_handle acpi_dev_handle)
> >> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
> >> tpm_tis_data *priv, int irq,
> >>if (IS_ERR(chip))
> >>return PTR_ERR(chip);
> >>
> >> + priv->chip = chip;
> >> + INIT_WORK(>storm_work, tpm_tis_storm_work);
> >> +
> >> #ifdef CONFIG_ACPI
> >>chip->acpi_dev_handle = acpi_dev_handle;
> >> #endif
> >> diff --git a/drivers/char/tpm/tpm_tis_core.h
> >> b/drivers/char/tpm/tpm_tis_core.h
> >> index edeb5dc61c95..5630f294dc0c 100644
> >> --- a/drivers/char/tpm/tpm_tis_core.h
> >> +++ b/drivers/char/tpm/tpm_tis_core.h
> >> @@ -95,6 +95,8 @@ struct tpm_tis_data {
> >>u16 clkrun_enabled;
> >>wait_queue_head_t int_queue;
> >>wait_queue_head_t read_queue;
> >> + struct work_struct storm_work;
> >> + struct tpm_chip *chip;
> >>const struct tpm_tis_phy_ops *phy_ops;
> >>unsigned short rng_quality;
> >> };
> >
> > I've tested this with the Intel platform that has an Infineon chip that
> > I found the other week. It works, but isn't the complete fix. With this
> > on top of James' patchset I sometimes see the message "Lost Interrupt
> > waiting for TPM stat", so I guess there needs to be a check in
> > wait_for_tpm_stat and request_locality to see if interrupts were
> > disabled when the wait_event_interruptible_timeout call times out.
>
> As kernel test robot pointed out. kstat_irqs isn't visible when tpm_tis
> builds as a module. It looks like it is only called by kstat_irq_usrs,
> and that is only by the fs/proc code. I have a patch to export it, but
> the i915 driver open codes their own version instead of using it. Is
> there any reason not to export it?
If you add a patch that exports it, then for coherency it'd be better to
also patch i915 driver. Jani?
/Jarkko
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
Jerry Snitselaar @ 2020-11-30 20:26 MST:
> Jerry Snitselaar @ 2020-11-30 19:58 MST:
>
>> When enabling the interrupt code for the tpm_tis driver we have
>> noticed some systems have a bios issue causing an interrupt storm to
>> occur. The issue isn't limited to a single tpm or system manufacturer
>> so keeping a denylist of systems with the issue isn't optimal. Instead
>> try to detect the problem occurring, disable interrupts, and revert to
>> polling when it happens.
>>
>> Cc: Jarkko Sakkinen
>> Cc: Jason Gunthorpe
>> Cc: Peter Huewe
>> Cc: James Bottomley
>> Cc: Matthew Garrett
>> Cc: Hans de Goede
>> Signed-off-by: Jerry Snitselaar
>> ---
>> v2: drop tpm_tis specific workqueue and use just system_wq
>>
>> drivers/char/tpm/tpm_tis_core.c | 27 +++
>> drivers/char/tpm/tpm_tis_core.h | 2 ++
>> 2 files changed, 29 insertions(+)
>>
>> diff --git a/drivers/char/tpm/tpm_tis_core.c
>> b/drivers/char/tpm/tpm_tis_core.c
>> index 23b60583928b..72cc8a5a152c 100644
>> --- a/drivers/char/tpm/tpm_tis_core.c
>> +++ b/drivers/char/tpm/tpm_tis_core.c
>> @@ -24,6 +24,8 @@
>> #include
>> #include
>> #include
>> +#include
>> +#include
>> #include "tpm.h"
>> #include "tpm_tis_core.h"
>>
>> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
>> *dev_id)
>> {
>> struct tpm_chip *chip = dev_id;
>> struct tpm_tis_data *priv = dev_get_drvdata(>dev);
>> +static bool check_storm = true;
>> +static unsigned int check_start;
>> u32 interrupt;
>> int i, rc;
>>
>> +if (unlikely(check_storm)) {
>> +if (!check_start) {
>> +check_start = jiffies_to_msecs(jiffies);
>> +} else if ((kstat_irqs(priv->irq) > 1000) &&
>> + (jiffies_to_msecs(jiffies) - check_start < 500)) {
>> +check_storm = false;
>> +schedule_work(>storm_work);
>> +} else if (jiffies_to_msecs(jiffies) - check_start >= 500) {
>> +check_storm = false;
>> +}
>> +}
>> +
>> rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), );
>> if (rc < 0)
>> return IRQ_NONE;
>> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
>> .clk_enable = tpm_tis_clkrun_enable,
>> };
>>
>> +static void tpm_tis_storm_work(struct work_struct *work)
>> +{
>> +struct tpm_tis_data *priv = container_of(work, struct tpm_tis_data,
>> storm_work);
>> +
>> +disable_interrupts(priv->chip);
>> +dev_warn(>chip->dev, "Interrupt storm detected, using
>> polling.\n");
>> +}
>> +
>> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int
>> irq,
>>const struct tpm_tis_phy_ops *phy_ops,
>>acpi_handle acpi_dev_handle)
>> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
>> tpm_tis_data *priv, int irq,
>> if (IS_ERR(chip))
>> return PTR_ERR(chip);
>>
>> +priv->chip = chip;
>> +INIT_WORK(>storm_work, tpm_tis_storm_work);
>> +
>> #ifdef CONFIG_ACPI
>> chip->acpi_dev_handle = acpi_dev_handle;
>> #endif
>> diff --git a/drivers/char/tpm/tpm_tis_core.h
>> b/drivers/char/tpm/tpm_tis_core.h
>> index edeb5dc61c95..5630f294dc0c 100644
>> --- a/drivers/char/tpm/tpm_tis_core.h
>> +++ b/drivers/char/tpm/tpm_tis_core.h
>> @@ -95,6 +95,8 @@ struct tpm_tis_data {
>> u16 clkrun_enabled;
>> wait_queue_head_t int_queue;
>> wait_queue_head_t read_queue;
>> +struct work_struct storm_work;
>> +struct tpm_chip *chip;
>> const struct tpm_tis_phy_ops *phy_ops;
>> unsigned short rng_quality;
>> };
>
> I've tested this with the Intel platform that has an Infineon chip that
> I found the other week. It works, but isn't the complete fix. With this
> on top of James' patchset I sometimes see the message "Lost Interrupt
> waiting for TPM stat", so I guess there needs to be a check in
> wait_for_tpm_stat and request_locality to see if interrupts were
> disabled when the wait_event_interruptible_timeout call times out.
As kernel test robot pointed out. kstat_irqs isn't visible when tpm_tis
builds as a module. It looks like it is only called by kstat_irq_usrs,
and that is only by the fs/proc code. I have a patch to export it, but
the i915 driver open codes their own version instead of using it. Is
there any reason not to export it?
Re: [PATCH v2] tpm_tis: Disable interrupts if interrupt storm detected
Jerry Snitselaar @ 2020-11-30 19:58 MST:
> When enabling the interrupt code for the tpm_tis driver we have
> noticed some systems have a bios issue causing an interrupt storm to
> occur. The issue isn't limited to a single tpm or system manufacturer
> so keeping a denylist of systems with the issue isn't optimal. Instead
> try to detect the problem occurring, disable interrupts, and revert to
> polling when it happens.
>
> Cc: Jarkko Sakkinen
> Cc: Jason Gunthorpe
> Cc: Peter Huewe
> Cc: James Bottomley
> Cc: Matthew Garrett
> Cc: Hans de Goede
> Signed-off-by: Jerry Snitselaar
> ---
> v2: drop tpm_tis specific workqueue and use just system_wq
>
> drivers/char/tpm/tpm_tis_core.c | 27 +++
> drivers/char/tpm/tpm_tis_core.h | 2 ++
> 2 files changed, 29 insertions(+)
>
> diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
> index 23b60583928b..72cc8a5a152c 100644
> --- a/drivers/char/tpm/tpm_tis_core.c
> +++ b/drivers/char/tpm/tpm_tis_core.c
> @@ -24,6 +24,8 @@
> #include
> #include
> #include
> +#include
> +#include
> #include "tpm.h"
> #include "tpm_tis_core.h"
>
> @@ -745,9 +747,23 @@ static irqreturn_t tis_int_handler(int dummy, void
> *dev_id)
> {
> struct tpm_chip *chip = dev_id;
> struct tpm_tis_data *priv = dev_get_drvdata(>dev);
> + static bool check_storm = true;
> + static unsigned int check_start;
> u32 interrupt;
> int i, rc;
>
> + if (unlikely(check_storm)) {
> + if (!check_start) {
> + check_start = jiffies_to_msecs(jiffies);
> + } else if ((kstat_irqs(priv->irq) > 1000) &&
> +(jiffies_to_msecs(jiffies) - check_start < 500)) {
> + check_storm = false;
> + schedule_work(>storm_work);
> + } else if (jiffies_to_msecs(jiffies) - check_start >= 500) {
> + check_storm = false;
> + }
> + }
> +
> rc = tpm_tis_read32(priv, TPM_INT_STATUS(priv->locality), );
> if (rc < 0)
> return IRQ_NONE;
> @@ -987,6 +1003,14 @@ static const struct tpm_class_ops tpm_tis = {
> .clk_enable = tpm_tis_clkrun_enable,
> };
>
> +static void tpm_tis_storm_work(struct work_struct *work)
> +{
> + struct tpm_tis_data *priv = container_of(work, struct tpm_tis_data,
> storm_work);
> +
> + disable_interrupts(priv->chip);
> + dev_warn(>chip->dev, "Interrupt storm detected, using
> polling.\n");
> +}
> +
> int tpm_tis_core_init(struct device *dev, struct tpm_tis_data *priv, int irq,
> const struct tpm_tis_phy_ops *phy_ops,
> acpi_handle acpi_dev_handle)
> @@ -1003,6 +1027,9 @@ int tpm_tis_core_init(struct device *dev, struct
> tpm_tis_data *priv, int irq,
> if (IS_ERR(chip))
> return PTR_ERR(chip);
>
> + priv->chip = chip;
> + INIT_WORK(>storm_work, tpm_tis_storm_work);
> +
> #ifdef CONFIG_ACPI
> chip->acpi_dev_handle = acpi_dev_handle;
> #endif
> diff --git a/drivers/char/tpm/tpm_tis_core.h b/drivers/char/tpm/tpm_tis_core.h
> index edeb5dc61c95..5630f294dc0c 100644
> --- a/drivers/char/tpm/tpm_tis_core.h
> +++ b/drivers/char/tpm/tpm_tis_core.h
> @@ -95,6 +95,8 @@ struct tpm_tis_data {
> u16 clkrun_enabled;
> wait_queue_head_t int_queue;
> wait_queue_head_t read_queue;
> + struct work_struct storm_work;
> + struct tpm_chip *chip;
> const struct tpm_tis_phy_ops *phy_ops;
> unsigned short rng_quality;
> };
I've tested this with the Intel platform that has an Infineon chip that
I found the other week. It works, but isn't the complete fix. With this
on top of James' patchset I sometimes see the message "Lost Interrupt
waiting for TPM stat", so I guess there needs to be a check in
wait_for_tpm_stat and request_locality to see if interrupts were
disabled when the wait_event_interruptible_timeout call times out.
