Re: [PATCH v3 8/8] drm/xe/gt_throttle: Avoid TOCTOU when monitoring reasons

["Vivi, Rodrigo" <rodrigo.vivi@intel.com>]

On Thu, 2025-10-30 at 13:43 -0500, Lucas De Marchi wrote:
> On Thu, Oct 30, 2025 at 05:06:46PM +0100, Raag Jadav wrote:
> > On Thu, Oct 30, 2025 at 11:47:03AM -0400, Rodrigo Vivi wrote:
> > > On Thu, Oct 30, 2025 at 09:55:30AM -0500, Lucas De Marchi wrote:
> > > > On Thu, Oct 30, 2025 at 10:53:36AM +0100, Raag Jadav wrote:
> > > > > On Wed, Oct 29, 2025 at 04:45:10PM -0700, Lucas De Marchi
> > > > > wrote:
> > > > > > It's currently not possible to safely monitor if there's
> > > > > > throttling
> > > > > > happening and what are the reasons. The approach of reading
> > > > > > the status
> > > > > > and then reading the reasons is not reliable as by the time
> > > > > > sysadmin
> > > > > > reads the reason, the throttling could not be happening
> > > > > > anymore.
> > > > > > 
> > > > > > Previous tentative to fix that[1] was breaking the ABI and
> > > > > > potentially
> > > > > > sysadmin's scripts. This takes a different approach of
> > > > > > adding and
> > > > > > documenting the additional attribute. It's still valuable,
> > > > > > though
> > > > > > redundant, to provide the simpler 0/1 interface.
> > > > > > 
> > > > > > In order to avoid userspace knowledge on the bitmask
> > > > > > meaning and to be
> > > > > > able to maintain the kernel side in sync with possible
> > > > > > changes in
> > > > > > future, just walk the attribute group and check what are
> > > > > > the masks that
> > > > > > match the value read.
> > > > > > 
> > > > > > [1]
> > > > > > https://lore.kernel.org/intel-xe/20241025092238.167042-1-raag.jadav@intel.com/
> > > > > 
> > > > > ...
> > > > > 
> > > > > > +static const struct attribute_group
> > > > > > *get_platform_throttle_group(struct xe_device *xe);
> > > > > > +
> > > > > > +static ssize_t status_reasons_show(struct kobject *kobj,
> > 
> > Semantically there's much of a 'status' here, so this could simply
> > be
> > 'reasons' (and same for the attribute name).
> > 
> > > > > > +				   struct kobj_attribute
> > > > > > *attr, char *buff)
> > > > > > +{
> > > > > > +	struct xe_gt *gt = throttle_to_gt(kobj);
> > > > > > +	struct xe_device *xe = gt_to_xe(gt);
> > > > > > +	const struct attribute_group *group;
> > > > > > +	struct attribute **pother;
> > > > > > +	ssize_t ret = 0;
> > > > > > +	u32 reasons;
> > > > > > +
> > > > > > +	reasons = xe_gt_throttle_get_limit_reasons(gt);
> > > > > > +	group = get_platform_throttle_group(xe);
> > > > > > +
> > > > > > +	for (pother = group->attrs; *pother; pother++) {
> > > > > > +		struct kobj_attribute *kattr =
> > > > > > container_of(*pother, struct kobj_attribute, attr);
> > > > > > +		struct throttle_attribute *other_ta =
> > > > > > kobj_attribute_to_throttle(kattr);
> > > > > > +
> > > > > > +		if (other_ta->mask != U32_MAX && reasons &
> > > > > > other_ta->mask)
> > > > > > +			ret += sysfs_emit_at(buff, ret,
> > > > > > "%s ", (*pother)->name);
> > > > > 
> > > > > Much better.
> > > > > 
> > > > > > +	}
> > > > > > +
> > > > > > +	/* Drop extra space from last iteration above */
> > > > > > +	if (ret)
> > > > > > +		ret--;
> > > > > > +
> > > > > > +	ret += sysfs_emit_at(buff, ret, "\n");
> > > > > 
> > > > > I went through the documentation again and I couldn't find
> > > > > any rules
> > > > > related to empty files or whether it is allowed (just
> > > > > thinking out
> > > > > loud about no throttling cases).
> > > > 
> > > > do you mean if "empty" files are allowed in sysfs? I don't
> > > > think there's
> > > > any problem with that. It's also not empty, it has a newline
> > > > there ;)
> > > 
> > > alternatively we could print the entire reg in hex format?
> > > But I prefer the text line in this patch.
> > > 
> > > Nothing against the 'empty' file with or without the new-line,
> > > but perhaps we could consider to track that in the loop
> > > and if none is add we print
> > > 
> > > if (ret)
> > > 	ret--;
> > > else
> > > 	sysfs_emit_at(buff, ret, "none");
> > > 
> > > and document that above...
> > 
> > +1.
> 
> collecting all the review commends from you 2, that would be with
> this
> diff squashed:
> 
> diff --git a/drivers/gpu/drm/xe/xe_gt_throttle.c
> b/drivers/gpu/drm/xe/xe_gt_throttle.c
> index ca45aea8c17a6..f3476fda7f4f6 100644
> --- a/drivers/gpu/drm/xe/xe_gt_throttle.c
> +++ b/drivers/gpu/drm/xe/xe_gt_throttle.c
> @@ -22,15 +22,15 @@
>    * Their availability depend on the platform and some may not be
> visible if that
>    * reason is not available.
>    *
> - * The ``status_reasons`` attribute can be used by sysadmin
> monitoring all
> - * possible reasons for throttling and reporting them. It's
> preferred over
> - * monitoring ``status`` and then reading the reason both for
> simplicity and to
> - * avoid TOCTOU (time-of-check to time-of-use).
> + * The ``reasons`` attribute can be used by sysadmin to monitor all
> possible
> + * reasons for throttling and report them. It's preferred over
> monitoring
> + * ``status`` and then reading the reason from individual attributes
> since that
> + * is racy. If there's no throttling happening, "none" is returned.
>    *
>    * The following attributes are available on Crescent Island
> platform:
>    *
>    * - ``status``: Overall throttle status (0: no throttling, 1:
> throttling)
> - * - ``status_reasons``: All reasons causing throttling separated by
> newline.
> + * - ``reasons``: All reasons causing throttling separated by space
>    * - ``reason_pl1``: package PL1
>    * - ``reason_pl2``: package PL2
>    * - ``reason_pl4``: package PL4
> @@ -50,7 +50,7 @@
>    * Other platforms support the following reasons:
>    *
>    * - ``status``: Overall throttle status (0: no throttling, 1:
> throttling)
> - * - ``status_reasons``: All reasons causing throttling separated by
> newline.
> + * - ``reasons``: All reasons causing throttling separated by space
>    * - ``reason_pl1``: package PL1
>    * - ``reason_pl2``: package PL2
>    * - ``reason_pl4``: package PL4, Iccmax etc.
> @@ -120,8 +120,8 @@ static ssize_t reason_show(struct kobject *kobj,
>   
>   static const struct attribute_group
> *get_platform_throttle_group(struct xe_device *xe);
>   
> -static ssize_t status_reasons_show(struct kobject *kobj,
> -				   struct kobj_attribute *attr, char
> *buff)
> +static ssize_t reasons_show(struct kobject *kobj,
> +			    struct kobj_attribute *attr, char *buff)
>   {
>   	struct xe_gt *gt = throttle_to_gt(kobj);
>   	struct xe_device *xe = gt_to_xe(gt);
> @@ -141,9 +141,11 @@ static ssize_t status_reasons_show(struct
> kobject *kobj,
>   			ret += sysfs_emit_at(buff, ret, "%s ",
> (*pother)->name);
>   	}
>   
> -	/* Drop extra space from last iteration above */
>   	if (ret)
> +		/* Drop extra space from last iteration above */
>   		ret--;
> +	else
> +		ret += sysfs_emit_at(buff, ret, "none");
>   
>   	ret += sysfs_emit_at(buff, ret, "\n");
>   
> @@ -162,7 +164,7 @@ static ssize_t status_reasons_show(struct kobject
> *kobj,
>   		.mask = _mask,					\
>   	}
>   
> -static THROTTLE_ATTR_RO_FUNC(status_reasons, 0,
> status_reasons_show);
> +static THROTTLE_ATTR_RO_FUNC(reasons, 0, reasons_show);
>   static THROTTLE_ATTR_RO(status, U32_MAX);
>   static THROTTLE_ATTR_RO(reason_pl1, POWER_LIMIT_1_MASK);
>   static THROTTLE_ATTR_RO(reason_pl2, POWER_LIMIT_2_MASK);
> @@ -174,7 +176,7 @@ static THROTTLE_ATTR_RO(reason_vr_thermalert,
> VR_THERMALERT_MASK);
>   static THROTTLE_ATTR_RO(reason_vr_tdc, VR_TDC_MASK);
>   
>   static struct attribute *throttle_attrs[] = {
> -	&attr_status_reasons.attr.attr,
> +	&attr_reasons.attr.attr,
>   	&attr_status.attr.attr,
>   	&attr_reason_pl1.attr.attr,
>   	&attr_reason_pl2.attr.attr,
> @@ -200,7 +202,7 @@ static THROTTLE_ATTR_RO(reason_psys_crit,
> PSYS_CRIT_MASK);
>   
>   static struct attribute *cri_throttle_attrs[] = {
>   	/* Common */
> -	&attr_status_reasons.attr.attr,
> +	&attr_reasons.attr.attr,
>   	&attr_status.attr.attr,
>   	&attr_reason_pl1.attr.attr,
>   	&attr_reason_pl2.attr.attr,

works for me. count on my rv-b.

unless Raag has any further concern...

> 
> Lucas De Marchi
> 
> > 
> > PS: A good read[1] if anyone's interested.
> > 
> > [1] https://lwn.net/Articles/378884/
> > 
> > Raag