Re: [PATCH] arch_topology: support parsing cache topology from DT

[Dietmar Eggemann]

On 06/04/2022 11:18, Qing Wang wrote:
> From: wangqing <11112896@xxxxxxxxxx>

[...]

> +void init_cpu_cache_topology(void)
> +{
> +	struct device_node *node_cpu, *node_cache;
> +	int cpu;
> +	int level = 0;
> +
> +	for_each_possible_cpu(cpu) {
> +		node_cpu = of_get_cpu_node(cpu, NULL);
> +		if (!node_cpu)
> +			continue;
> +
> +		level = 0;
> +		node_cache = node_cpu;
> +		while (level < MAX_CACHE_LEVEL) {
> +			node_cache = of_parse_phandle(node_cache, "next-level-cache", 0);
> +			if (!node_cache)
> +				break;
> +
> +			cache_topology[cpu][level++] = node_cache;
> +		}
> +		of_node_put(node_cpu);
> +	}
> +}

>From where is init_cpu_cache_topology() called?

> +bool cpu_share_llc(int cpu1, int cpu2)
> +{
> +	int cache_level;
> +
> +	for (cache_level = MAX_CACHE_LEVEL - 1; cache_level > 0; cache_level--) {
> +		if (!cache_topology[cpu1][cache_level])
> +			continue;
> +
> +		if (cache_topology[cpu1][cache_level] == cache_topology[cpu2][cache_level])
> +			return true;
> +
> +		return false;
> +	}
> +
> +	return false;
> +}

Like I mentioned in:

https://lkml.kernel.org/r/73b491fe-b5e8-ebca-081e-fa339cc903e1@xxxxxxx

the correct setting in DT's cpu-map node (only core nodes in your case
(One DynamIQ cluster) will give you the correct LLC (highest
SD_SHARE_PKG_RESOURCES) setting.

https://www.kernel.org/doc/Documentation/devicetree/bindings/arm/topology.txt

> +
> +bool cpu_share_l2c(int cpu1, int cpu2)
> +{
> +	if (!cache_topology[cpu1][0])
> +		return false;
> +
> +	if (cache_topology[cpu1][0] == cache_topology[cpu2][0])
> +		return true;
> +
> +	return false;
> +}
> +
>  /*
>   * cpu topology table
>   */
> @@ -662,7 +720,7 @@ const struct cpumask *cpu_coregroup_mask(int cpu)
>  		/* not numa in package, lets use the package siblings */
>  		core_mask = &cpu_topology[cpu].core_sibling;
>  	}
> -	if (cpu_topology[cpu].llc_id != -1) {
> +	if (cpu_topology[cpu].llc_id != -1 || cache_topology[cpu][0]) {
>  		if (cpumask_subset(&cpu_topology[cpu].llc_sibling, core_mask))
>  			core_mask = &cpu_topology[cpu].llc_sibling;
>  	}
> @@ -684,7 +742,8 @@ void update_siblings_masks(unsigned int cpuid)
>  	for_each_online_cpu(cpu) {
>  		cpu_topo = &cpu_topology[cpu];
>  
> -		if (cpuid_topo->llc_id == cpu_topo->llc_id) {
> +		if ((cpuid_topo->llc_id != -1 && cpuid_topo->llc_id == cpu_topo->llc_id)
> +			|| (cpuid_topo->llc_id == -1 && cpu_share_llc(cpu, cpuid))) {

Assuming a:

      .---------------.
CPU   |0 1 2 3 4 5 6 7|
      +---------------+
uarch |l l l l m m m b| (so called tri-gear: little, medium, big)
      +---------------+
  L2  |   |   | | | | |
      +---------------+
  L3  |<--         -->|
      +---------------+
      |<-- cluster -->|
      +---------------+
      |<--   DSU   -->|
      '---------------'

system, I guess you would get (w/ Phantom SD and L2/L3 cache info in DT):

CPU0 .. 3:

MC	   SD_SHARE_PKG_RESOURCES
DIE     no SD_SHARE_PKG_RESOURCES

CPU 4...7:

DIE     no SD_SHARE_PKG_RESOURCES

I can't see how this would make any sense ...

Reason is cpu_share_llc(). You don't check cache_level=0 and w/

CPU0 .. 3:
cache_topology[CPUX][0] == L2
cache_topology[CPUX][1] == L3

CPU4...7:
cache_topology[CPUX][0] == L3

there is, except for CPU0-1 and CPU2-3, no LLC match.

[...]