Re: [PATCH 01/01][retry 1] x86: L3 cache index disable for 2.6.26

[Mark Langsdorf <mark.langsdorf@amd.com>]

On Friday 08 August 2008, Pavel Machek wrote:

> I think there's one-value-per-file rule in sysfs...
> 
> I guess it is better to return -EOPNOTSUP (or something) instead of
> english text explaining that...
> 
> No, really, what you created is impossible to parse -- /proc like
> nightmare.

Okay, this is a simpler version that includes most of Ingo's
clean-ups and style changes.  It only displays the two
cache index values.  Is this acceptable?



New versions of AMD processors have support to disable parts
of their L3 caches if too many MCEs are generated by the
L3 cache.  

This patch provides a /sysfs interface under the cache
hierarchy to display which caches indices are disabled
(if any) and to ALLOW monitoring applications to disable a
cache index.

This patch does not set an automatic policy to disable
the L3 cache.  Policy decisions would need to be made
by a RAS handler.  This patch merely makes it easier to
see what indices are currently disabled.

Signed-off-by: Mark Langsdorf <mark.langsdorf@amd.com>

diff -r f3f819497a68 arch/x86/kernel/cpu/intel_cacheinfo.c
--- a/arch/x86/kernel/cpu/intel_cacheinfo.c	Thu Aug 07 04:24:53 2008 -0500
+++ b/arch/x86/kernel/cpu/intel_cacheinfo.c	Tue Aug 12 05:29:37 2008 -0500
@@ -130,6 +130,7 @@ struct _cpuid4_info {
 	union _cpuid4_leaf_ebx ebx;
 	union _cpuid4_leaf_ecx ecx;
 	unsigned long size;
+	unsigned long can_disable;
 	cpumask_t shared_cpu_map;	/* future?: only cpus/node is needed */
 };
 
@@ -251,6 +252,14 @@ static void __cpuinit amd_cpuid4(int lea
 		(ebx->split.ways_of_associativity + 1) - 1;
 }
 
+static void __cpuinit 
+amd_check_l3_disable(int index, struct _cpuid4_info *this_leaf)
+{
+	if (index < 3)
+		return;
+	this_leaf->can_disable = 1;     
+}
+
 static int __cpuinit cpuid4_cache_lookup(int index, struct _cpuid4_info *this_leaf)
 {
 	union _cpuid4_leaf_eax 	eax;
@@ -258,10 +267,13 @@ static int __cpuinit cpuid4_cache_lookup
 	union _cpuid4_leaf_ecx 	ecx;
 	unsigned		edx;
 
-	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
 		amd_cpuid4(index, &eax, &ebx, &ecx);
-	else
-		cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full,  &edx);
+		if (boot_cpu_data.x86 >= 0x10)
+			amd_check_l3_disable(index, this_leaf);
+	} else {
+		cpuid_count(4, index, &eax.full, &ebx.full, &ecx.full, &edx);
+	}
 	if (eax.split.type == CACHE_TYPE_NULL)
 		return -EIO; /* better error ? */
 
@@ -269,9 +281,9 @@ static int __cpuinit cpuid4_cache_lookup
 	this_leaf->ebx = ebx;
 	this_leaf->ecx = ecx;
 	this_leaf->size = (ecx.split.number_of_sets + 1) *
-		(ebx.split.coherency_line_size + 1) *
-		(ebx.split.physical_line_partition + 1) *
-		(ebx.split.ways_of_associativity + 1);
+			(ebx.split.coherency_line_size + 1) *
+			(ebx.split.physical_line_partition + 1) *
+			(ebx.split.ways_of_associativity + 1);
 	return 0;
 }
 
@@ -574,6 +586,9 @@ static DEFINE_PER_CPU(struct _index_kobj
 static DEFINE_PER_CPU(struct _index_kobject *, index_kobject);
 #define INDEX_KOBJECT_PTR(x, y)    (&((per_cpu(index_kobject, x))[y]))
 
+#define to_object(k)    container_of(k, struct _index_kobject, kobj)
+#define to_attr(a)      container_of(a, struct _cache_attr, attr)
+
 #define show_one_plus(file_name, object, val)				\
 static ssize_t show_##file_name						\
 			(struct _cpuid4_info *this_leaf, char *buf)	\
@@ -618,6 +633,83 @@ static inline ssize_t show_shared_cpu_li
 static inline ssize_t show_shared_cpu_list(struct _cpuid4_info *leaf, char *buf)
 {
 	return show_shared_cpu_map_func(leaf, 1, buf);
+}
+
+#if defined(CONFIG_PCI) && defined(CONFIG_K8_NB)
+#include <linux/pci.h>
+#include <asm/k8.h>
+static struct pci_dev *get_k8_northbridge(int node)
+{
+        return k8_northbridges[node];
+}
+#else
+static inline int pci_write_config_dword(struct pci_dev *dev, int where,
+                                         u32 val)
+{
+	return 0;
+}
+
+static inline int pci_read_config_dword(struct pci_dev *dev, int where,
+                                        u32 *val)
+{
+	return 0;
+}
+
+static struct pci_dev *get_k8_northbridge(int node)
+{
+        return NULL;
+}
+#endif
+
+static ssize_t show_cache_disable(struct _cpuid4_info *this_leaf, char *buf)
+{
+        int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
+        struct pci_dev *dev = get_k8_northbridge(node);
+        ssize_t ret = 0;
+        int i;
+
+        if (!this_leaf->can_disable) 
+                return sprintf(buf, "-1");
+
+        for (i = 0; i < 2; i++) {
+                unsigned int reg = 0;
+
+                pci_read_config_dword(dev, 0x1BC + i * 4, &reg);
+
+                ret += sprintf(buf, "%s %x\t", buf, reg);
+        }
+	ret += sprintf(buf,"%s\n", buf);
+
+	return ret;
+}
+
+static ssize_t 
+store_cache_disable(struct _cpuid4_info *this_leaf, const char *buf,
+                size_t count)
+{
+        int node = cpu_to_node(first_cpu(this_leaf->shared_cpu_map));
+        struct pci_dev *dev = get_k8_northbridge(node);
+        ssize_t ret = 0;
+        unsigned int index, val;
+
+        if (!this_leaf->can_disable)
+                return 0;
+
+        if (strlen(buf) > 15)
+                return -EINVAL;
+
+        ret = sscanf(buf, "%x %x\n", &index, &val);
+        if (ret != 2)
+                return -EINVAL;
+        if (index > 1) 
+                return -EINVAL;
+
+        val |= 0xc0000000;
+        pci_write_config_dword(dev, 0x1BC + index * 4, val & ~0x40000000);
+        wbinvd();
+        pci_write_config_dword(dev, 0x1BC + index * 4, val);
+
+        return strlen(buf);
 }
 
 static ssize_t show_type(struct _cpuid4_info *this_leaf, char *buf) {
@@ -657,6 +749,8 @@ define_one_ro(shared_cpu_map);
 define_one_ro(shared_cpu_map);
 define_one_ro(shared_cpu_list);
 
+static struct _cache_attr cache_disable = __ATTR(cache_disable, 0644, show_cache_disable, store_cache_disable);
+
 static struct attribute * default_attrs[] = {
 	&type.attr,
 	&level.attr,
@@ -667,11 +761,9 @@ static struct attribute * default_attrs[
 	&size.attr,
 	&shared_cpu_map.attr,
 	&shared_cpu_list.attr,
+	&cache_disable.attr,
 	NULL
 };
-
-#define to_object(k) container_of(k, struct _index_kobject, kobj)
-#define to_attr(a) container_of(a, struct _cache_attr, attr)
 
 static ssize_t show(struct kobject * kobj, struct attribute * attr, char * buf)
 {
@@ -689,7 +781,15 @@ static ssize_t store(struct kobject * ko
 static ssize_t store(struct kobject * kobj, struct attribute * attr,
 		     const char * buf, size_t count)
 {
-	return 0;
+	struct _cache_attr *fattr = to_attr(attr);
+	struct _index_kobject *this_leaf = to_object(kobj);
+	ssize_t ret;
+
+	ret = fattr->store ?
+		fattr->store(CPUID4_INFO_IDX(this_leaf->cpu, this_leaf->index),
+			 buf, count) :
+		0;
+	return ret;
 }
 
 static struct sysfs_ops sysfs_ops = {