Re: [PATCH] drivers: base: cacheinfo: support DT overrides for cache type

[Sudeep Holla]

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.

> 
> $ 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@xxxxxxxxxx>
> ---
>  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

-->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);