Re: [PATCH v4] platform/x86: intel_pmc_core: export platform global_reset via sysfs.

[Hans de Goede]

Hi,

On 4/7/21 11:31 AM, Winkler, Tomas wrote:
> 
> 
>> Hi,
>>
>> On 4/7/21 8:51 AM, Winkler, Tomas wrote:
>>>>>
>>>>> During PCH (platform/board) manufacturing process a global reset has
>>>>> to be induced in order for configuration changes take the effect
>>>>> upon following platform reset.
>>>>> This setting was commonly done by accessing PMC registers via
>>>>> /dev/mem but due to security concern /dev/mem access is much
>>>>> restricted, hence the reason for exposing this setting via dedicated sysfs
>> interface.
>>>>> To prevent post manufacturing abuse the register is protected by
>>>>> hardware locking.
>>>>
>>>> The purpose of this reset functionality is not entirely clear to me.
>>>>
>>>> Is this only used during production of a board? Or is this also
>>>> something which a user/reseller may use as part of a factory-reset
>> procedure?
>>>
>>> Board production and refurbishing of the board. I can try to rephrase but I
>> thought all the info is in the commit message.
>>> As a runtime feature a user can check that her/his platform is correctly
>> sealed.
>>
>> Manufacturing is clear, refurbishing is very much not clear, do you mean
>> actually desoldering the chip and replacing it with a new one ?
>>
>>>> If this is only used once during production, then I'm not sure if
>>>> introducing a sysfs file for this is desirable.
>>>
>>> What do you suggest, than?  I'm just guessing is where are you heading
>>> so the answer is that the manufacturing is often already run on the
>> production OS installation,  w/o going into details swapping or reconfiguring
>> the OS is not always an option.
>>> The manufacturer is also a user of ours.
>>
>> Ok, so lets compromise here, please make use of the visibility sysfs attribute
>> callback, which returns a umask and make the file read-only at the umask
>> level if it has been sealed, to make it clear to users that they cannot write to
>> it, the -EACCES error means 'Permission denied' so if the user is already root
>> they are going to get mightily confused if ls -l shows the file is writable.
> Okay, it seems a better solution if the file is the global reset,  
> but maybe this path should not be taken if we rename it to  extended_test_mode_register3, than it's better to get EACCESS on a specific bit.

Ack, I was thinking about this perhaps not being the best option if we
expose more bits myself too (when I wrote the rest of my email).

Still it might be an idea to do this if all bits which we allow setting
are locked ?  Note this is just a suggestion / something to consider,
I think that if we rename the file to extended_test_mode_register3
most of my concerns are solved.

I'm still not entirely happy with the -EACCESS though, can you perhaps do a
dev_err_ratelimited() in that case, logging why the EACCESS is happening?

As I said before, returning EACCESS when the file-mode bits say the file is
writable and the user is root will lead to some head-scratching I'm afraid.
Another option would be to pick another errno value. Although even with
another errno value a dev_err_ratelimited() logging the reason of the
failure is probably a good idea.

>> Also on set you are checking that the written value is bit 20, and on show you
>> are showing the contents of the "Extended Test Mode Register 3" in hex, or
>> at least those bits you are willing to show.
> 
> The intention was to left the user space behave same as with direct register access (/dev/mem)

Ack, that is fine.

>> So in essence what you are doing here is giving userspace (some) access to
>> the "Extended Test Mode Register 3", I would prefer to spell that out
>> explicitly. The global_reset sysfs file name to me too much hints at something
>> which the user can trigger / use while it is not intended for user usage.
> 
> Yeah, Global reset is maybe too ambiguous name in a general context, this is not the standard platform reset. 

Ack.

> I've left it in register form in order to keep the user space as it has access to the register (/dev/mem)
>>
>> Also the Documentation/ABI/testing/sysfs-platform-intel-pm file pretty much
>> describes this as direct register access rather then as some reset mechanism.
>>
>> So I think it would be better to call the new file
>> extended_test_mode_register3, this will also be useful if we need to provide
>> access to other bits in the same register later; and this will be a good
>> template to follow if we need to provide some access to other registers later
>> too.
> 
> Need to sync with David on that he pointed just ow, that he plans to expose some more bits. 

Ok, I assume I'll eventually see a new version appear.

Regards,

Hans





>>>> Can you please provide a new version where the purpsoe of the newly
>>>> introduced sysfs file is made more clear, both in the commit-msg as
>>>> well as in
>>>> the:
>>> Okay I can do that.
>>>>
>>>> Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>
>>>> File ?
>>>>
>>>> Regards,
>>>>
>>>> Hans
>>>>
>>>>
>>>>
>>>>>
>>>>> The register in MMIO space is defined for Cannon Lake and newer PCHs.
>>>>>
>>>>> Cc: David E Box <david.e.box@xxxxxxxxx>
>>>>> Reviewed-by: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
>>>>> Signed-off-by: Tamar Mashiah <tamar.mashiah@xxxxxxxxx>
>>>>> Signed-off-by: Tomas Winkler <tomas.winkler@xxxxxxxxx>
>>>>> ---
>>>>> 2:
>>>>> 1. Add locking for reading the ET3 register  (Andy) 2. Fix few style
>>>>> issues (Andy)
>>>>> V3:
>>>>> 1. Resend
>>>>> v4:
>>>>> 1. Fix return statement (Andy)
>>>>> 2. Specify manufacturing process (Enrico)
>>>>>
>>>>>  .../ABI/testing/sysfs-platform-intel-pmc      | 11 +++
>>>>>  MAINTAINERS                                   |  1 +
>>>>>  drivers/platform/x86/intel_pmc_core.c         | 97 +++++++++++++++++++
>>>>>  drivers/platform/x86/intel_pmc_core.h         |  6 ++
>>>>>  4 files changed, 115 insertions(+)
>>>>>  create mode 100644
>>>>> Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>>
>>>>> diff --git a/Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>> b/Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>> new file mode 100644
>>>>> index 000000000000..7ce00e77fbcd
>>>>> --- /dev/null
>>>>> +++ b/Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>> @@ -0,0 +1,11 @@
>>>>> +What:		/sys/devices/platform/<platform>/global_reset
>>>>> +Date:		Apr 2021
>>>>> +KernelVersion:	5.13
>>>>> +Contact:	"Tomas Winkler" <tomas.winkler@xxxxxxxxx>
>>>>> +Description:
>>>>> +		Display global reset setting bits for PMC.
>>>>> +			* bit 31 - global reset is locked
>>>>> +			* bit 20 - global reset is set
>>>>> +		Writing bit 20 value to the global_reset will induce
>>>>> +		a platform global reset upon consequent platform reset.
>>>>> +		in case the register is not locked.
>>>>> diff --git a/MAINTAINERS b/MAINTAINERS index
>>>>> 04f68e0cda64..618676eba8c8 100644
>>>>> --- a/MAINTAINERS
>>>>> +++ b/MAINTAINERS
>>>>> @@ -9166,6 +9166,7 @@ M:	Rajneesh Bhardwaj
>>>> <irenic.rajneesh@xxxxxxxxx>
>>>>>  M:	David E Box <david.e.box@xxxxxxxxx>
>>>>>  L:	platform-driver-x86@xxxxxxxxxxxxxxx
>>>>>  S:	Maintained
>>>>> +F:	Documentation/ABI/testing/sysfs-platform-intel-pmc
>>>>>  F:	drivers/platform/x86/intel_pmc_core*
>>>>>
>>>>>  INTEL PMIC GPIO DRIVERS
>>>>> diff --git a/drivers/platform/x86/intel_pmc_core.c
>>>>> b/drivers/platform/x86/intel_pmc_core.c
>>>>> index ee2f757515b0..8afc198550a4 100644
>>>>> --- a/drivers/platform/x86/intel_pmc_core.c
>>>>> +++ b/drivers/platform/x86/intel_pmc_core.c
>>>>> @@ -401,6 +401,7 @@ static const struct pmc_reg_map cnp_reg_map =
>> {
>>>>>  	.pm_cfg_offset = CNP_PMC_PM_CFG_OFFSET,
>>>>>  	.pm_read_disable_bit = CNP_PMC_READ_DISABLE_BIT,
>>>>>  	.ltr_ignore_max = CNP_NUM_IP_IGN_ALLOWED,
>>>>> +	.etr3_offset = ETR3_OFFSET,
>>>>>  };
>>>>>
>>>>>  static const struct pmc_reg_map icl_reg_map = { @@ -418,6 +419,7 @@
>>>>> static const struct pmc_reg_map icl_reg_map = {
>>>>>  	.pm_cfg_offset = CNP_PMC_PM_CFG_OFFSET,
>>>>>  	.pm_read_disable_bit = CNP_PMC_READ_DISABLE_BIT,
>>>>>  	.ltr_ignore_max = ICL_NUM_IP_IGN_ALLOWED,
>>>>> +	.etr3_offset = ETR3_OFFSET,
>>>>>  };
>>>>>
>>>>>  static const struct pmc_bit_map tgl_clocksource_status_map[] = { @@
>>>>> -585,6 +587,7 @@ static const struct pmc_reg_map tgl_reg_map = {
>>>>>  	.lpm_sts = tgl_lpm_maps,
>>>>>  	.lpm_status_offset = TGL_LPM_STATUS_OFFSET,
>>>>>  	.lpm_live_status_offset = TGL_LPM_LIVE_STATUS_OFFSET,
>>>>> +	.etr3_offset = ETR3_OFFSET,
>>>>>  };
>>>>>
>>>>>  static inline u32 pmc_core_reg_read(struct pmc_dev *pmcdev, int
>>>>> reg_offset) @@ -603,6 +606,99 @@ static inline u64
>>>> pmc_core_adjust_slp_s0_step(struct pmc_dev *pmcdev, u32 value)
>>>>>  	return (u64)value * pmcdev->map->slp_s0_res_counter_step;
>>>>>  }
>>>>>
>>>>> +static int set_global_reset(struct pmc_dev *pmcdev) {
>>>>> +	const struct pmc_reg_map *map = pmcdev->map;
>>>>> +	u32 reg;
>>>>> +	int err;
>>>>> +
>>>>> +	if (!map->etr3_offset)
>>>>> +		return -EOPNOTSUPP;
>>>>> +
>>>>> +	mutex_lock(&pmcdev->lock);
>>>>> +
>>>>> +	/* check if CF9 is locked */
>>>>> +	reg = pmc_core_reg_read(pmcdev, map->etr3_offset);
>>>>> +	if (reg & ETR3_CF9LOCK) {
>>>>> +		err = -EACCES;
>>>>> +		goto out_unlock;
>>>>> +	}
>>>>> +
>>>>> +	/* write CF9 global reset bit */
>>>>> +	reg |= ETR3_CF9GR;
>>>>> +	pmc_core_reg_write(pmcdev, map->etr3_offset, reg);
>>>>> +
>>>>> +	reg = pmc_core_reg_read(pmcdev, map->etr3_offset);
>>>>> +	if (!(reg & ETR3_CF9GR)) {
>>>>> +		err = -EIO;
>>>>> +		goto out_unlock;
>>>>> +	}
>>>>> +
>>>>> +	err = 0;
>>>>> +
>>>>> +out_unlock:
>>>>> +	mutex_unlock(&pmcdev->lock);
>>>>> +	return err;
>>>>> +}
>>>>> +
>>>>> +static ssize_t global_reset_show(struct device *dev,
>>>>> +				 struct device_attribute *attr, char *buf) {
>>>>> +	struct pmc_dev *pmcdev = dev_get_drvdata(dev);
>>>>> +	const struct pmc_reg_map *map = pmcdev->map;
>>>>> +	u32 reg;
>>>>> +
>>>>> +	if (!map->etr3_offset)
>>>>> +		return -EOPNOTSUPP;
>>>>> +
>>>>> +	mutex_lock(&pmcdev->lock);
>>>>> +
>>>>> +	reg = pmc_core_reg_read(pmcdev, map->etr3_offset);
>>>>> +	reg &= ETR3_CF9GR | ETR3_CF9LOCK;
>>>>> +
>>>>> +	mutex_unlock(&pmcdev->lock);
>>>>> +
>>>>> +	return sysfs_emit(buf, "0x%08x", reg); }
>>>>> +
>>>>> +static ssize_t global_reset_store(struct device *dev,
>>>>> +				  struct device_attribute *attr,
>>>>> +				  const char *buf, size_t len)
>>>>> +{
>>>>> +	struct pmc_dev *pmcdev = dev_get_drvdata(dev);
>>>>> +	int err;
>>>>> +	u32 reg;
>>>>> +
>>>>> +	err = kstrtouint(buf, 16, &reg);
>>>>> +	if (err)
>>>>> +		return err;
>>>>> +
>>>>> +	/* allow only CF9 writes */
>>>>> +	if (reg != ETR3_CF9GR)
>>>>> +		return -EINVAL;
>>>>> +
>>>>> +	err = set_global_reset(pmcdev);
>>>>> +	if (err)
>>>>> +		return err;
>>>>> +
>>>>> +	return len;
>>>>> +}
>>>>> +static DEVICE_ATTR_RW(global_reset);
>>>>> +
>>>>> +static struct attribute *pmc_attrs[] = {
>>>>> +	&dev_attr_global_reset.attr,
>>>>> +	NULL
>>>>> +};
>>>>> +
>>>>> +static const struct attribute_group pmc_attr_group = {
>>>>> +	.attrs = pmc_attrs,
>>>>> +};
>>>>> +
>>>>> +static const struct attribute_group *pmc_dev_groups[] = {
>>>>> +	&pmc_attr_group,
>>>>> +	NULL
>>>>> +};
>>>>> +
>>>>>  static int pmc_core_dev_state_get(void *data, u64 *val)  {
>>>>>  	struct pmc_dev *pmcdev = data;
>>>>> @@ -1364,6 +1460,7 @@ static struct platform_driver pmc_core_driver
>> = {
>>>>>  		.name = "intel_pmc_core",
>>>>>  		.acpi_match_table = ACPI_PTR(pmc_core_acpi_ids),
>>>>>  		.pm = &pmc_core_pm_ops,
>>>>> +		.dev_groups = pmc_dev_groups,
>>>>>  	},
>>>>>  	.probe = pmc_core_probe,
>>>>>  	.remove = pmc_core_remove,
>>>>> diff --git a/drivers/platform/x86/intel_pmc_core.h
>>>>> b/drivers/platform/x86/intel_pmc_core.h
>>>>> index f33cd2c34835..98ebdfe57138 100644
>>>>> --- a/drivers/platform/x86/intel_pmc_core.h
>>>>> +++ b/drivers/platform/x86/intel_pmc_core.h
>>>>> @@ -200,6 +200,11 @@ enum ppfear_regs {
>>>>>  #define TGL_LPM_STATUS_OFFSET			0x1C3C
>>>>>  #define TGL_LPM_LIVE_STATUS_OFFSET		0x1C5C
>>>>>
>>>>> +/* Extended Test Mode Register 3 (CNL and later) */
>>>>> +#define ETR3_OFFSET				0x1048
>>>>> +#define ETR3_CF9GR				BIT(20)
>>>>> +#define ETR3_CF9LOCK				BIT(31)
>>>>> +
>>>>>  const char *tgl_lpm_modes[] = {
>>>>>  	"S0i2.0",
>>>>>  	"S0i2.1",
>>>>> @@ -263,6 +268,7 @@ struct pmc_reg_map {
>>>>>  	const u32 lpm_residency_offset;
>>>>>  	const u32 lpm_status_offset;
>>>>>  	const u32 lpm_live_status_offset;
>>>>> +	const u32 etr3_offset;
>>>>>  };
>>>>>
>>>>>  /**
>>>>>
>>>
>