lkml.org 
[lkml]   [2017]   [Nov]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] drivers: base: cacheinfo: support DT overrides for cache type
From
Date
On 2017/11/17 10:13, Tan Xiaojun wrote:
> On 2017/11/16 23:23, Sudeep Holla wrote:
>>
>>
>> On 16/11/17 12:58, Tan Xiaojun wrote:
>>> Since commit dfea747d2aba ("drivers: base: cacheinfo: support DT
>>> overrides for cache properties"), we can set the correct cacheinfo
>>> via DT. But the cache type can't be set in the same way.
>>>
>>> I found this may be a problem in recent tests. I tested L3 cache
>>> node setting in DT in Hisilicon D03/D05 board. And I got cacheinfo
>>> via sysfs below:
>>
>> I assume L3 is outer non-architected system cache.
>>
>
> Yes.
>
>>>
>>> $ cat /sys/devices/system/cpu/cpu*/cache/index3/
>>> allocation_policy level power/
>>> shared_cpu_map uevent write_policy
>>> coherency_line_size number_of_sets shared_cpu_list
>>> size ways_of_associativity
>>>
>>> This is incomplete, no type file to display type info. Because L3
>>> cache is uncore, we can't get correct type info from system
>>> register, and will get a default type "CACHE_TYPE_NOCACHE". Then
>>> use "lscpu" will print an error like below:
>>>
>>> $ lscpu
>>> lscpu: cannot open /sys/devices/system/cpu/cpu0/cache/index3/type:
>>> No such file or directory
>>>
>>> So I think maybe we can set correct cache type via DT too.
>>>
>>> Signed-off-by: Tan Xiaojun <tanxiaojun@huawei.com>
>>> ---
>>> drivers/base/cacheinfo.c | 10 ++++++++++
>>> 1 file changed, 10 insertions(+)
>>>
>>> diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
>>> index eb3af27..3e650dc 100644
>>> --- a/drivers/base/cacheinfo.c
>>> +++ b/drivers/base/cacheinfo.c
>>> @@ -122,6 +122,15 @@ static inline int get_cacheinfo_idx(enum cache_type type)
>>> return type;
>>> }
>>>
>>> +static void cache_type(struct cacheinfo *this_leaf)
>>> +{
>>> + const __be32 *cache_type;
>>> +
>>> + cache_type = of_get_property(this_leaf->of_node, "type", NULL);
>>
>> NACK for this:
>>
>> 1. We don't have any DT binding for the "type"
>> 2. Even if we had, we will never have such a binding that magical
>> returns the enum used in Linux implementation. That's not how DT
>> bindings are designed.
>>
>> Please refer ePAPR or Devicetree Specification Release 0.1 from
>> devicetree.org, we have cache-unified boolean for type.
>>
>> Let me know if the below patch works for you, I will submit it
>> preferably with your tested-by.
>>
>> Regards,
>> Sudeep
>>
>
> OK. That's fine. I will test this.
> By the way, Arm64 tend to use acpi way to boot now. Is there some
> suitable solution for acpi?
>
> Thanks.
> Xiaojun.
>

Test passed, this is indeed a better solution.

Thanks.
Xiaojun.

>> -->8
>>
>> diff --git i/drivers/base/cacheinfo.c w/drivers/base/cacheinfo.c
>> index eb3af2739537..07532d83be0b 100644
>> --- i/drivers/base/cacheinfo.c
>> +++ w/drivers/base/cacheinfo.c
>> @@ -186,6 +186,11 @@ static void cache_associativity(struct cacheinfo
>> *this_leaf)
>> this_leaf->ways_of_associativity = (size / nr_sets) /
>> line_size;
>> }
>>
>> +static bool cache_node_is_unified(struct cacheinfo *this_leaf)
>> +{
>> + return of_property_read_bool(this_leaf->of_node, "cache-unified");
>> +}
>> +
>> static void cache_of_override_properties(unsigned int cpu)
>> {
>> int index;
>> @@ -194,6 +199,14 @@ static void cache_of_override_properties(unsigned
>> int cpu)
>>
>> for (index = 0; index < cache_leaves(cpu); index++) {
>> this_leaf = this_cpu_ci->info_list + index;
>> + /*
>> + * init_cache_level must setup the cache level correctly
>> + * overriding the architecturally specified levels, so
>> + * if type is NONE at this stage, it should be unified
>> + */
>> + if (this_leaf->type == CACHE_TYPE_NOCACHE &&
>> + cache_node_is_unified(this_leaf))
>> + this_leaf->type = CACHE_TYPE_UNIFIED;
>> cache_size(this_leaf);
>> cache_get_line_size(this_leaf);
>> cache_nr_sets(this_leaf);
>>
>>
>> .
>>
>
>
>
> .
>


\
 
 \ /
  Last update: 2017-11-17 17:18    [W:0.110 / U:0.832 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site