The ARM Architecture Reference Manual allows for caches to be "invisible" and
thus not specified in the system registers under some scenarios such as if the
cache cannot be managed by set/way operations.
However, such caches may be specified in the ACPI PPTT table for workload
performance/scheduling optimizations.
Currently such caches can cause an error in lscpu -
lscpu: cannot open /sys/devices/system/cpu/cpu0/cache/index3/type: No such
file or directory
and result in no output, providing a poor user experience. lstopo is also
affected as such caches are not included in the output.
Address these issues by attempting to be a little more discerning about when
cache information is provided to userspace, and also utilize all sources for
cache information when possible.
[v2]
-Updated cacheinfo per Sudeep's suggestion
-Integrated the PPTT fix into existing PPTT code per Sudeep's suggestion
Jeffrey Hugo (2):
drivers: base: cacheinfo: Do not populate sysfs for unknown cache
types
ACPI/PPTT: Handle architecturally unknown cache types
drivers/acpi/pptt.c | 15 +++++++++++----
drivers/base/cacheinfo.c | 2 ++
2 files changed, 13 insertions(+), 4 deletions(-)
--
Qualcomm Datacenter Technologies as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.
If a cache has an unknown type because neither the hardware nor the
firmware told us, an entry in the sysfs tree will be made, but the type
file will not be present. lscpu depends on the type file being present
for every entry, and will error out without printing system information
if lscpu cannot open the type file.
Presenting information about a cache without indicating its type is not
useful, therefore if we hit a cache with an unknown type, stop populating
sysfs so that userspace has the maximum amount of useful information.
This addresses the following lscpu error, which prevents any output.
lscpu: cannot open /sys/devices/system/cpu/cpu0/cache/index3/type: No such
file or directory
Suggested-by: Sudeep Holla <[email protected]>
Signed-off-by: Jeffrey Hugo <[email protected]>
---
drivers/base/cacheinfo.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
index 5d5b598..cf78fa6 100644
--- a/drivers/base/cacheinfo.c
+++ b/drivers/base/cacheinfo.c
@@ -615,6 +615,8 @@ static int cache_add_dev(unsigned int cpu)
this_leaf = this_cpu_ci->info_list + i;
if (this_leaf->disable_sysfs)
continue;
+ if (this_leaf->type == CACHE_TYPE_NOCACHE)
+ break;
cache_groups = cache_get_attribute_groups(this_leaf);
ci_dev = cpu_device_create(parent, this_leaf, cache_groups,
"index%1u", i);
--
Qualcomm Datacenter Technologies as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.
The type of a cache might not be specified by architectural mechanisms (ie
system registers), but its type might be specified in the PPTT. In this
case, we should populate the type of the cache, rather than leave it
undefined.
This fixes the issue where the cacheinfo driver will not populate sysfs
for such caches, resulting in the information missing from utilities like
lstopo and lscpu, thus degrading the user experience.
Fixes: 2bd00bcd73e5 (ACPI/PPTT: Add Processor Properties Topology Table parsing)
Reported-by: Vijaya Kumar K <[email protected]>
Signed-off-by: Jeffrey Hugo <[email protected]>
---
drivers/acpi/pptt.c | 15 +++++++++++----
1 file changed, 11 insertions(+), 4 deletions(-)
diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c
index d1e26cb..bb00ed9 100644
--- a/drivers/acpi/pptt.c
+++ b/drivers/acpi/pptt.c
@@ -402,11 +402,18 @@ static void update_cache_properties(struct cacheinfo *this_leaf,
}
}
/*
- * If the above flags are valid, and the cache type is NOCACHE
- * update the cache type as well.
+ * If cache type is NOCACHE, then the cache hasn't been specified
+ * via other mechanisms. Update the type if either the cache has
+ * been fully specified in PPTT, or a cache type has been provided.
+ *
+ * Note, we assume such caches are unified based on conventional system
+ * design and known examples. Significant work is required elsewhere to
+ * fully support data/instruction only type caches which are only
+ * specified in PPTT.
*/
- if (this_leaf->type == CACHE_TYPE_NOCACHE &&
- valid_flags == PPTT_CHECKED_ATTRIBUTES)
+ if ((this_leaf->type == CACHE_TYPE_NOCACHE) &&
+ (valid_flags == PPTT_CHECKED_ATTRIBUTES ||
+ found_cache->flags & ACPI_PPTT_CACHE_TYPE_VALID))
this_leaf->type = CACHE_TYPE_UNIFIED;
}
--
Qualcomm Datacenter Technologies as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.
Hi,
On 09/14/2018 11:28 AM, Jeffrey Hugo wrote:
> If a cache has an unknown type because neither the hardware nor the
> firmware told us, an entry in the sysfs tree will be made, but the type
> file will not be present. lscpu depends on the type file being present
> for every entry, and will error out without printing system information
> if lscpu cannot open the type file.
>
> Presenting information about a cache without indicating its type is not
> useful, therefore if we hit a cache with an unknown type, stop populating
> sysfs so that userspace has the maximum amount of useful information.
>
> This addresses the following lscpu error, which prevents any output.
> lscpu: cannot open /sys/devices/system/cpu/cpu0/cache/index3/type: No such
> file or directory
>
> Suggested-by: Sudeep Holla <[email protected]>
> Signed-off-by: Jeffrey Hugo <[email protected]>
> ---
> drivers/base/cacheinfo.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
> index 5d5b598..cf78fa6 100644
> --- a/drivers/base/cacheinfo.c
> +++ b/drivers/base/cacheinfo.c
> @@ -615,6 +615,8 @@ static int cache_add_dev(unsigned int cpu)
> this_leaf = this_cpu_ci->info_list + i;
> if (this_leaf->disable_sysfs)
> continue;
> + if (this_leaf->type == CACHE_TYPE_NOCACHE)
> + break;
> cache_groups = cache_get_attribute_groups(this_leaf);
> ci_dev = cpu_device_create(parent, this_leaf, cache_groups,
> "index%1u", i);
>
Looks fine to me:
Reviewed-by: Jeremy Linton <[email protected]>
Thanks,
On 14/09/18 17:28, Jeffrey Hugo wrote:
> The type of a cache might not be specified by architectural mechanisms (ie
> system registers), but its type might be specified in the PPTT. In this
> case, we should populate the type of the cache, rather than leave it
> undefined.
>
> This fixes the issue where the cacheinfo driver will not populate sysfs
> for such caches, resulting in the information missing from utilities like
> lstopo and lscpu, thus degrading the user experience.
>
> Fixes: 2bd00bcd73e5 (ACPI/PPTT: Add Processor Properties Topology Table parsing)
> Reported-by: Vijaya Kumar K <[email protected]>
> Signed-off-by: Jeffrey Hugo <[email protected]>
> ---
> drivers/acpi/pptt.c | 15 +++++++++++----
> 1 file changed, 11 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c
> index d1e26cb..bb00ed9 100644
> --- a/drivers/acpi/pptt.c
> +++ b/drivers/acpi/pptt.c
> @@ -402,11 +402,18 @@ static void update_cache_properties(struct cacheinfo *this_leaf,
> }
> }
> /*
> - * If the above flags are valid, and the cache type is NOCACHE
> - * update the cache type as well.
> + * If cache type is NOCACHE, then the cache hasn't been specified
> + * via other mechanisms. Update the type if either the cache has
> + * been fully specified in PPTT, or a cache type has been provided.
> + *
> + * Note, we assume such caches are unified based on conventional system
> + * design and known examples. Significant work is required elsewhere to
> + * fully support data/instruction only type caches which are only
> + * specified in PPTT.
> */
> - if (this_leaf->type == CACHE_TYPE_NOCACHE &&
> - valid_flags == PPTT_CHECKED_ATTRIBUTES)
> + if ((this_leaf->type == CACHE_TYPE_NOCACHE) &&
> + (valid_flags == PPTT_CHECKED_ATTRIBUTES ||
> + found_cache->flags & ACPI_PPTT_CACHE_TYPE_VALID))
> this_leaf->type = CACHE_TYPE_UNIFIED;
I thought I did mention that we can drop the valid_flags altogether
unless Jeremy has reasons not to.
--
Regards,
Sudeep
On 9/17/2018 10:17 AM, Sudeep Holla wrote:
>
>
> On 14/09/18 17:28, Jeffrey Hugo wrote:
>> The type of a cache might not be specified by architectural mechanisms (ie
>> system registers), but its type might be specified in the PPTT. In this
>> case, we should populate the type of the cache, rather than leave it
>> undefined.
>>
>> This fixes the issue where the cacheinfo driver will not populate sysfs
>> for such caches, resulting in the information missing from utilities like
>> lstopo and lscpu, thus degrading the user experience.
>>
>> Fixes: 2bd00bcd73e5 (ACPI/PPTT: Add Processor Properties Topology Table parsing)
>> Reported-by: Vijaya Kumar K <[email protected]>
>> Signed-off-by: Jeffrey Hugo <[email protected]>
>> ---
>> drivers/acpi/pptt.c | 15 +++++++++++----
>> 1 file changed, 11 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c
>> index d1e26cb..bb00ed9 100644
>> --- a/drivers/acpi/pptt.c
>> +++ b/drivers/acpi/pptt.c
>> @@ -402,11 +402,18 @@ static void update_cache_properties(struct cacheinfo *this_leaf,
>> }
>> }
>> /*
>> - * If the above flags are valid, and the cache type is NOCACHE
>> - * update the cache type as well.
>> + * If cache type is NOCACHE, then the cache hasn't been specified
>> + * via other mechanisms. Update the type if either the cache has
>> + * been fully specified in PPTT, or a cache type has been provided.
>> + *
>> + * Note, we assume such caches are unified based on conventional system
>> + * design and known examples. Significant work is required elsewhere to
>> + * fully support data/instruction only type caches which are only
>> + * specified in PPTT.
>> */
>> - if (this_leaf->type == CACHE_TYPE_NOCACHE &&
>> - valid_flags == PPTT_CHECKED_ATTRIBUTES)
>> + if ((this_leaf->type == CACHE_TYPE_NOCACHE) &&
>> + (valid_flags == PPTT_CHECKED_ATTRIBUTES ||
>> + found_cache->flags & ACPI_PPTT_CACHE_TYPE_VALID))
>> this_leaf->type = CACHE_TYPE_UNIFIED;
>
> I thought I did mention that we can drop the valid_flags altogether
> unless Jeremy has reasons not to.
>
You suggested that perhaps that could be the case. It seemed like an
open question to me. I'm at Linaro Connect without access to the device
this week, so I guess someone has roughly a week to chime in that the
valid flags should be kept, otherwise I'll try a v3 with them removed.
--
Jeffrey Hugo
Qualcomm Datacenter Technologies as an affiliate of Qualcomm
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the
Code Aurora Forum, a Linux Foundation Collaborative Project.
Hi,
On 09/17/2018 05:46 PM, Jeffrey Hugo wrote:
> On 9/17/2018 10:17 AM, Sudeep Holla wrote:
>>
>>
>> On 14/09/18 17:28, Jeffrey Hugo wrote:
>>> The type of a cache might not be specified by architectural
>>> mechanisms (ie
>>> system registers), but its type might be specified in the PPTT. In this
>>> case, we should populate the type of the cache, rather than leave it
>>> undefined.
>>>
>>> This fixes the issue where the cacheinfo driver will not populate sysfs
>>> for such caches, resulting in the information missing from utilities
>>> like
>>> lstopo and lscpu, thus degrading the user experience.
>>>
>>> Fixes: 2bd00bcd73e5 (ACPI/PPTT: Add Processor Properties Topology
>>> Table parsing)
>>> Reported-by: Vijaya Kumar K <[email protected]>
>>> Signed-off-by: Jeffrey Hugo <[email protected]>
>>> ---
>>> drivers/acpi/pptt.c | 15 +++++++++++----
>>> 1 file changed, 11 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/acpi/pptt.c b/drivers/acpi/pptt.c
>>> index d1e26cb..bb00ed9 100644
>>> --- a/drivers/acpi/pptt.c
>>> +++ b/drivers/acpi/pptt.c
>>> @@ -402,11 +402,18 @@ static void update_cache_properties(struct
>>> cacheinfo *this_leaf,
>>> }
>>> }
>>> /*
>>> - * If the above flags are valid, and the cache type is NOCACHE
>>> - * update the cache type as well.
>>> + * If cache type is NOCACHE, then the cache hasn't been specified
>>> + * via other mechanisms. Update the type if either the cache has
>>> + * been fully specified in PPTT, or a cache type has been provided.
>>> + *
>>> + * Note, we assume such caches are unified based on conventional
>>> system
>>> + * design and known examples. Significant work is required
>>> elsewhere to
>>> + * fully support data/instruction only type caches which are only
>>> + * specified in PPTT.
>>> */
>>> - if (this_leaf->type == CACHE_TYPE_NOCACHE &&
>>> - valid_flags == PPTT_CHECKED_ATTRIBUTES)
>>> + if ((this_leaf->type == CACHE_TYPE_NOCACHE) &&
>>> + (valid_flags == PPTT_CHECKED_ATTRIBUTES ||
>>> + found_cache->flags & ACPI_PPTT_CACHE_TYPE_VALID))
>>> this_leaf->type = CACHE_TYPE_UNIFIED;
>>
>> I thought I did mention that we can drop the valid_flags altogether
>> unless Jeremy has reasons not to.
>>
>
> You suggested that perhaps that could be the case. It seemed like an
> open question to me. I'm at Linaro Connect without access to the device
> this week, so I guess someone has roughly a week to chime in that the
> valid flags should be kept, otherwise I'll try a v3 with them removed.
>
The point of the valid_flags/CHECKED_ATTRIBUTE was to help assure that a
minimum set of attributes were being provided by the firmware. If we are
going to reset the CACHE_TYPE, then we might as well remove the
valid_flag/CHECKED_ATTRIBUTE counts as it can be easily bypassed.
So, yes please, remove the valid_flags with this change.
Thanks,