* [PATCH 1/6] kill useless SMT code in prom_hold_cpus
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-27 5:24 ` [PATCH 2/6] register_cpu_online should be __cpuinit Nathan Lynch
` (5 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
I think this code that counts SMT threads and compares against NR_CPUS
is an artifact of pre-powerpc-merge ppc64. We care about starting
only primary threads in the OF client code.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/prom_init.c | 39 +++------------------------------------
1 files changed, 3 insertions(+), 36 deletions(-)
diff --git a/arch/powerpc/kernel/prom_init.c b/arch/powerpc/kernel/prom_init.c
index c4ab219..b72849a 100644
--- a/arch/powerpc/kernel/prom_init.c
+++ b/arch/powerpc/kernel/prom_init.c
@@ -205,8 +205,6 @@ static int __initdata mem_reserve_cnt;
static cell_t __initdata regbuf[1024];
-#define MAX_CPU_THREADS 2
-
/*
* Error results ... some OF calls will return "-1" on error, some
* will return 0, some will return either. To simplify, here are
@@ -1339,10 +1337,6 @@ static void __init prom_hold_cpus(void)
unsigned int reg;
phandle node;
char type[64];
- int cpuid = 0;
- unsigned int interrupt_server[MAX_CPU_THREADS];
- unsigned int cpu_threads, hw_cpu_num;
- int propsize;
struct prom_t *_prom = &RELOC(prom);
unsigned long *spinloop
= (void *) LOW_ADDR(__secondary_hold_spinloop);
@@ -1386,7 +1380,6 @@ static void __init prom_hold_cpus(void)
reg = -1;
prom_getprop(node, "reg", ®, sizeof(reg));
- prom_debug("\ncpuid = 0x%x\n", cpuid);
prom_debug("cpu hw idx = 0x%x\n", reg);
/* Init the acknowledge var which will be reset by
@@ -1395,28 +1388,9 @@ static void __init prom_hold_cpus(void)
*/
*acknowledge = (unsigned long)-1;
- propsize = prom_getprop(node, "ibm,ppc-interrupt-server#s",
- &interrupt_server,
- sizeof(interrupt_server));
- if (propsize < 0) {
- /* no property. old hardware has no SMT */
- cpu_threads = 1;
- interrupt_server[0] = reg; /* fake it with phys id */
- } else {
- /* We have a threaded processor */
- cpu_threads = propsize / sizeof(u32);
- if (cpu_threads > MAX_CPU_THREADS) {
- prom_printf("SMT: too many threads!\n"
- "SMT: found %x, max is %x\n",
- cpu_threads, MAX_CPU_THREADS);
- cpu_threads = 1; /* ToDo: panic? */
- }
- }
-
- hw_cpu_num = interrupt_server[0];
- if (hw_cpu_num != _prom->cpu) {
+ if (reg != _prom->cpu) {
/* Primary Thread of non-boot cpu */
- prom_printf("%x : starting cpu hw idx %x... ", cpuid, reg);
+ prom_printf("starting cpu hw idx %x... ", reg);
call_prom("start-cpu", 3, 0, node,
secondary_hold, reg);
@@ -1431,17 +1405,10 @@ static void __init prom_hold_cpus(void)
}
#ifdef CONFIG_SMP
else
- prom_printf("%x : boot cpu %x\n", cpuid, reg);
+ prom_printf("boot cpu hw idx %x\n", reg);
#endif /* CONFIG_SMP */
-
- /* Reserve cpu #s for secondary threads. They start later. */
- cpuid += cpu_threads;
}
- if (cpuid > NR_CPUS)
- prom_printf("WARNING: maximum CPUs (" __stringify(NR_CPUS)
- ") exceeded: ignoring extras\n");
-
prom_debug("prom_hold_cpus: end...\n");
}
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH 2/6] register_cpu_online should be __cpuinit
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
2008-07-27 5:24 ` [PATCH 1/6] kill useless SMT code in prom_hold_cpus Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-27 5:24 ` [PATCH 3/6] Update cpu_sibling_maps dynamically Nathan Lynch
` (4 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
It is called only in cpu online paths.
(caught by CONFIG_DEBUG_SECTION_MISMATCH=y)
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/sysfs.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/arch/powerpc/kernel/sysfs.c b/arch/powerpc/kernel/sysfs.c
index 800e5e9..1568080 100644
--- a/arch/powerpc/kernel/sysfs.c
+++ b/arch/powerpc/kernel/sysfs.c
@@ -298,7 +298,7 @@ static struct sysdev_attribute pa6t_attrs[] = {
};
-static void register_cpu_online(unsigned int cpu)
+static void __cpuinit register_cpu_online(unsigned int cpu)
{
struct cpu *c = &per_cpu(cpu_devices, cpu);
struct sys_device *s = &c->sysdev;
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH 3/6] Update cpu_sibling_maps dynamically
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
2008-07-27 5:24 ` [PATCH 1/6] kill useless SMT code in prom_hold_cpus Nathan Lynch
2008-07-27 5:24 ` [PATCH 2/6] register_cpu_online should be __cpuinit Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-27 5:24 ` [PATCH 4/6] make core sibling information available to userspace Nathan Lynch
` (3 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
Rather doing one initialization pass over all the per-cpu
cpu_sibling_maps at boot, update the maps at cpu online/offline time.
This is a behavior change -- the thread_siblings attribute now
reflects only online siblings, whereas it would display offline
siblings before. The new behavior matches that of x86, and is
arguably more useful.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/setup-common.c | 24 ------------------------
arch/powerpc/kernel/setup_64.c | 3 ---
arch/powerpc/kernel/smp.c | 32 +++++++++++++++++++++++++++++---
3 files changed, 29 insertions(+), 30 deletions(-)
diff --git a/arch/powerpc/kernel/setup-common.c b/arch/powerpc/kernel/setup-common.c
index 61a3f41..9cc5a52 100644
--- a/arch/powerpc/kernel/setup-common.c
+++ b/arch/powerpc/kernel/setup-common.c
@@ -367,7 +367,6 @@ static void __init cpu_init_thread_core_maps(int tpc)
* setup_cpu_maps - initialize the following cpu maps:
* cpu_possible_map
* cpu_present_map
- * cpu_sibling_map
*
* Having the possible map set up early allows us to restrict allocations
* of things like irqstacks to num_possible_cpus() rather than NR_CPUS.
@@ -475,29 +474,6 @@ void __init smp_setup_cpu_maps(void)
*/
cpu_init_thread_core_maps(nthreads);
}
-
-/*
- * Being that cpu_sibling_map is now a per_cpu array, then it cannot
- * be initialized until the per_cpu areas have been created. This
- * function is now called from setup_per_cpu_areas().
- */
-void __init smp_setup_cpu_sibling_map(void)
-{
-#ifdef CONFIG_PPC64
- int i, cpu, base;
-
- for_each_possible_cpu(cpu) {
- DBG("Sibling map for CPU %d:", cpu);
- base = cpu_first_thread_in_core(cpu);
- for (i = 0; i < threads_per_core; i++) {
- cpu_set(base + i, per_cpu(cpu_sibling_map, cpu));
- DBG(" %d", base + i);
- }
- DBG("\n");
- }
-
-#endif /* CONFIG_PPC64 */
-}
#endif /* CONFIG_SMP */
#ifdef CONFIG_PCSPKR_PLATFORM
diff --git a/arch/powerpc/kernel/setup_64.c b/arch/powerpc/kernel/setup_64.c
index 04d8de9..8b25f51 100644
--- a/arch/powerpc/kernel/setup_64.c
+++ b/arch/powerpc/kernel/setup_64.c
@@ -611,9 +611,6 @@ void __init setup_per_cpu_areas(void)
paca[i].data_offset = ptr - __per_cpu_start;
memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
}
-
- /* Now that per_cpu is setup, initialize cpu_sibling_map */
- smp_setup_cpu_sibling_map();
}
#endif
diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index f5ae9fa..3c4d07e 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -41,6 +41,7 @@
#include <asm/smp.h>
#include <asm/time.h>
#include <asm/machdep.h>
+#include <asm/cputhreads.h>
#include <asm/cputable.h>
#include <asm/system.h>
#include <asm/mpic.h>
@@ -228,6 +229,7 @@ void __devinit smp_prepare_boot_cpu(void)
BUG_ON(smp_processor_id() != boot_cpuid);
cpu_set(boot_cpuid, cpu_online_map);
+ cpu_set(boot_cpuid, per_cpu(cpu_sibling_map, boot_cpuid));
#ifdef CONFIG_PPC64
paca[boot_cpuid].__current = current;
#endif
@@ -380,6 +382,7 @@ int __cpuinit __cpu_up(unsigned int cpu)
int __devinit start_secondary(void *unused)
{
unsigned int cpu = smp_processor_id();
+ int i, base;
atomic_inc(&init_mm.mm_count);
current->active_mm = &init_mm;
@@ -400,6 +403,14 @@ int __devinit start_secondary(void *unused)
ipi_call_lock();
cpu_set(cpu, cpu_online_map);
+ /* Update sibling maps */
+ base = cpu_first_thread_in_core(cpu);
+ for (i = 0; i < threads_per_core; i++) {
+ if (cpu_is_offline(base + i))
+ continue;
+ cpu_set(cpu, per_cpu(cpu_sibling_map, base + i));
+ cpu_set(base + i, per_cpu(cpu_sibling_map, cpu));
+ }
ipi_call_unlock();
local_irq_enable();
@@ -437,10 +448,25 @@ void __init smp_cpus_done(unsigned int max_cpus)
#ifdef CONFIG_HOTPLUG_CPU
int __cpu_disable(void)
{
- if (smp_ops->cpu_disable)
- return smp_ops->cpu_disable();
+ int cpu = smp_processor_id();
+ int base, i;
+ int err;
- return -ENOSYS;
+ if (!smp_ops->cpu_disable)
+ return -ENOSYS;
+
+ err = smp_ops->cpu_disable();
+ if (err)
+ return err;
+
+ /* Update sibling maps */
+ base = cpu_first_thread_in_core(cpu);
+ for (i = 0; i < threads_per_core; i++) {
+ cpu_clear(cpu, per_cpu(cpu_sibling_map, base + i));
+ cpu_clear(base + i, per_cpu(cpu_sibling_map, cpu));
+ }
+
+ return 0;
}
void __cpu_die(unsigned int cpu)
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH 4/6] make core sibling information available to userspace
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
` (2 preceding siblings ...)
2008-07-27 5:24 ` [PATCH 3/6] Update cpu_sibling_maps dynamically Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-27 5:24 ` [PATCH 5/6] make core id " Nathan Lynch
` (2 subsequent siblings)
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
Implement the notion of "core siblings" for powerpc. This makes
/sys/devices/system/cpu/cpu*/topology/core_siblings present sensible
values, indicating online CPUs which share an L2 cache.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/smp.c | 71 ++++++++++++++++++++++++++++++++++++++++
include/asm-powerpc/smp.h | 1 +
include/asm-powerpc/topology.h | 1 +
3 files changed, 73 insertions(+), 0 deletions(-)
diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index 3c4d07e..e8bc6a0 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -63,10 +63,12 @@ struct thread_info *secondary_ti;
cpumask_t cpu_possible_map = CPU_MASK_NONE;
cpumask_t cpu_online_map = CPU_MASK_NONE;
DEFINE_PER_CPU(cpumask_t, cpu_sibling_map) = CPU_MASK_NONE;
+DEFINE_PER_CPU(cpumask_t, cpu_core_map) = CPU_MASK_NONE;
EXPORT_SYMBOL(cpu_online_map);
EXPORT_SYMBOL(cpu_possible_map);
EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
+EXPORT_PER_CPU_SYMBOL(cpu_core_map);
/* SMP operations for this machine */
struct smp_ops_t *smp_ops;
@@ -230,6 +232,7 @@ void __devinit smp_prepare_boot_cpu(void)
cpu_set(boot_cpuid, cpu_online_map);
cpu_set(boot_cpuid, per_cpu(cpu_sibling_map, boot_cpuid));
+ cpu_set(boot_cpuid, per_cpu(cpu_core_map, boot_cpuid));
#ifdef CONFIG_PPC64
paca[boot_cpuid].__current = current;
#endif
@@ -377,11 +380,43 @@ int __cpuinit __cpu_up(unsigned int cpu)
return 0;
}
+/* Must be called when no change can occur to cpu_present_map,
+ * i.e. during cpu online or offline.
+ */
+static struct device_node *cpu_to_l2cache(int cpu)
+{
+ struct device_node *np;
+ const u32 *l2cache_phandle;
+
+ if (!cpu_present(cpu))
+ return NULL;
+
+ np = of_get_cpu_node(cpu, NULL);
+
+ l2cache_phandle = of_get_property(np, "l2-cache", NULL);
+
+ of_node_put(np);
+
+ if (!l2cache_phandle)
+ return NULL;
+
+ np = NULL;
+ for_each_node_by_type(np, "cache") {
+ const u32 *phandle = of_get_property(np, "ibm,phandle", NULL);
+ if (!phandle)
+ continue;
+ if (*phandle == *l2cache_phandle)
+ return np;
+ }
+
+ return NULL;
+}
/* Activate a secondary processor. */
int __devinit start_secondary(void *unused)
{
unsigned int cpu = smp_processor_id();
+ struct device_node *l2_cache;
int i, base;
atomic_inc(&init_mm.mm_count);
@@ -410,7 +445,26 @@ int __devinit start_secondary(void *unused)
continue;
cpu_set(cpu, per_cpu(cpu_sibling_map, base + i));
cpu_set(base + i, per_cpu(cpu_sibling_map, cpu));
+
+ /* cpu_core_map should be a superset of
+ * cpu_sibling_map even if we don't have cache
+ * information, so update the former here, too.
+ */
+ cpu_set(cpu, per_cpu(cpu_core_map, base +i));
+ cpu_set(base + i, per_cpu(cpu_core_map, cpu));
+ }
+ l2_cache = cpu_to_l2cache(cpu);
+ for_each_online_cpu(i) {
+ struct device_node *np = cpu_to_l2cache(i);
+ if (!np)
+ continue;
+ if (np == l2_cache) {
+ cpu_set(cpu, per_cpu(cpu_core_map, i));
+ cpu_set(i, per_cpu(cpu_core_map, cpu));
+ }
+ of_node_put(np);
}
+ of_node_put(l2_cache);
ipi_call_unlock();
local_irq_enable();
@@ -448,6 +502,7 @@ void __init smp_cpus_done(unsigned int max_cpus)
#ifdef CONFIG_HOTPLUG_CPU
int __cpu_disable(void)
{
+ struct device_node *l2_cache;
int cpu = smp_processor_id();
int base, i;
int err;
@@ -464,8 +519,24 @@ int __cpu_disable(void)
for (i = 0; i < threads_per_core; i++) {
cpu_clear(cpu, per_cpu(cpu_sibling_map, base + i));
cpu_clear(base + i, per_cpu(cpu_sibling_map, cpu));
+ cpu_clear(cpu, per_cpu(cpu_core_map, base +i));
+ cpu_clear(base + i, per_cpu(cpu_core_map, cpu));
}
+ l2_cache = cpu_to_l2cache(cpu);
+ for_each_present_cpu(i) {
+ struct device_node *np = cpu_to_l2cache(i);
+ if (!np)
+ continue;
+ if (np == l2_cache) {
+ cpu_clear(cpu, per_cpu(cpu_core_map, i));
+ cpu_clear(i, per_cpu(cpu_core_map, cpu));
+ }
+ of_node_put(np);
+ }
+ of_node_put(l2_cache);
+
+
return 0;
}
diff --git a/include/asm-powerpc/smp.h b/include/asm-powerpc/smp.h
index 416d4c2..32e9100 100644
--- a/include/asm-powerpc/smp.h
+++ b/include/asm-powerpc/smp.h
@@ -62,6 +62,7 @@ extern int smp_hw_index[];
#endif
DECLARE_PER_CPU(cpumask_t, cpu_sibling_map);
+DECLARE_PER_CPU(cpumask_t, cpu_core_map);
/* Since OpenPIC has only 4 IPIs, we use slightly different message numbers.
*
diff --git a/include/asm-powerpc/topology.h b/include/asm-powerpc/topology.h
index 100c6fb..f00e8e5 100644
--- a/include/asm-powerpc/topology.h
+++ b/include/asm-powerpc/topology.h
@@ -108,6 +108,7 @@ static inline void sysfs_remove_device_from_node(struct sys_device *dev,
#include <asm/smp.h>
#define topology_thread_siblings(cpu) (per_cpu(cpu_sibling_map, cpu))
+#define topology_core_siblings(cpu) (per_cpu(cpu_core_map, cpu))
#endif
#endif
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH 5/6] make core id information available to userspace
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
` (3 preceding siblings ...)
2008-07-27 5:24 ` [PATCH 4/6] make core sibling information available to userspace Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-27 5:24 ` [PATCH 6/6] show processor cache information in sysfs Nathan Lynch
2008-07-28 6:34 ` [PATCH 0/6] powerpc topology updates Benjamin Herrenschmidt
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
Existing Open Firmware practice is to report each processor core as a
separate node in the device tree. Report the value of the "reg" OF
property corresponding to a logical CPU's device node as the core_id
attribute in /sys/devices/system/cpu/cpu*/topology/core_id.
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/smp.c | 23 +++++++++++++++++++++++
include/asm-powerpc/smp.h | 1 +
include/asm-powerpc/topology.h | 1 +
3 files changed, 25 insertions(+), 0 deletions(-)
diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index e8bc6a0..f3a2360 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -380,6 +380,29 @@ int __cpuinit __cpu_up(unsigned int cpu)
return 0;
}
+/* Return the value of the reg property corresponding to the given
+ * logical cpu.
+ */
+int cpu_to_core_id(int cpu)
+{
+ struct device_node *np;
+ const int *reg;
+ int id = -1;
+
+ np = of_get_cpu_node(cpu, NULL);
+ if (!np)
+ goto out;
+
+ reg = of_get_property(np, "reg", NULL);
+ if (!reg)
+ goto out;
+
+ id = *reg;
+out:
+ of_node_put(np);
+ return id;
+}
+
/* Must be called when no change can occur to cpu_present_map,
* i.e. during cpu online or offline.
*/
diff --git a/include/asm-powerpc/smp.h b/include/asm-powerpc/smp.h
index 32e9100..4d28e1e 100644
--- a/include/asm-powerpc/smp.h
+++ b/include/asm-powerpc/smp.h
@@ -63,6 +63,7 @@ extern int smp_hw_index[];
DECLARE_PER_CPU(cpumask_t, cpu_sibling_map);
DECLARE_PER_CPU(cpumask_t, cpu_core_map);
+extern int cpu_to_core_id(int cpu);
/* Since OpenPIC has only 4 IPIs, we use slightly different message numbers.
*
diff --git a/include/asm-powerpc/topology.h b/include/asm-powerpc/topology.h
index f00e8e5..c32da6f 100644
--- a/include/asm-powerpc/topology.h
+++ b/include/asm-powerpc/topology.h
@@ -109,6 +109,7 @@ static inline void sysfs_remove_device_from_node(struct sys_device *dev,
#define topology_thread_siblings(cpu) (per_cpu(cpu_sibling_map, cpu))
#define topology_core_siblings(cpu) (per_cpu(cpu_core_map, cpu))
+#define topology_core_id(cpu) (cpu_to_core_id(cpu))
#endif
#endif
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* [PATCH 6/6] show processor cache information in sysfs
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
` (4 preceding siblings ...)
2008-07-27 5:24 ` [PATCH 5/6] make core id " Nathan Lynch
@ 2008-07-27 5:24 ` Nathan Lynch
2008-07-28 6:34 ` [PATCH 0/6] powerpc topology updates Benjamin Herrenschmidt
6 siblings, 0 replies; 8+ messages in thread
From: Nathan Lynch @ 2008-07-27 5:24 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev, Paul Mackerras
Collect cache information from the OF device tree and display it in
the cpu hierarchy in sysfs. This is intended to be compatible at the
userspace level with x86's implementation[1], hence some of the funny
attribute names. The arrangement of cache info is not immediately
intuitive, but (again) it's for compatibility's sake.
The cache attributes exposed are:
type (Data, Instruction, or Unified)
level (1, 2, 3...)
size
coherency_line_size
number_of_sets
ways_of_associativity
All of these can be derived on platforms that follow the OF PowerPC
Processor binding. The code "publishes" only those attributes for
which it is able to determine values; attributes for values which
cannot be determined are not created at all.
[1] arch/x86/kernel/cpu/intel_cacheinfo.c
Signed-off-by: Nathan Lynch <ntl@pobox.com>
---
arch/powerpc/kernel/sysfs.c | 308 +++++++++++++++++++++++++++++++++++++++++++
1 files changed, 308 insertions(+), 0 deletions(-)
diff --git a/arch/powerpc/kernel/sysfs.c b/arch/powerpc/kernel/sysfs.c
index 1568080..7f56cc8 100644
--- a/arch/powerpc/kernel/sysfs.c
+++ b/arch/powerpc/kernel/sysfs.c
@@ -22,6 +22,8 @@
static DEFINE_PER_CPU(struct cpu, cpu_devices);
+static DEFINE_PER_CPU(struct kobject *, cache_toplevel);
+
/* SMT stuff */
#ifdef CONFIG_PPC_MULTIPLATFORM
@@ -297,6 +299,286 @@ static struct sysdev_attribute pa6t_attrs[] = {
#endif /* CONFIG_DEBUG_KERNEL */
};
+struct cache_desc {
+ struct kobject kobj;
+ struct cache_desc *next;
+ const char *type; /* Instruction, Data, or Unified */
+ u32 size; /* total cache size in KB */
+ u32 line_size; /* in bytes */
+ u32 nr_sets; /* number of sets */
+ u32 level; /* e.g. 1, 2, 3... */
+ u32 associativity; /* e.g. 8-way... 0 is fully associative */
+};
+
+DEFINE_PER_CPU(struct cache_desc *, cache_desc);
+
+static struct cache_desc *kobj_to_cache_desc(struct kobject *k)
+{
+ return container_of(k, struct cache_desc, kobj);
+}
+
+static void cache_desc_release(struct kobject *k)
+{
+ struct cache_desc *desc = kobj_to_cache_desc(k);
+
+ printk("releasing %s\n", kobject_name(k));
+
+ if (desc->next)
+ kobject_put(&desc->next->kobj);
+
+ kfree(kobj_to_cache_desc(k));
+}
+
+static ssize_t cache_desc_show(struct kobject *k, struct attribute *attr, char *buf)
+{
+ struct kobj_attribute *kobj_attr;
+
+ kobj_attr = container_of(attr, struct kobj_attribute, attr);
+
+ return kobj_attr->show(k, kobj_attr, buf);
+}
+
+static struct sysfs_ops cache_desc_sysfs_ops = {
+ .show = cache_desc_show,
+};
+
+static struct kobj_type cache_desc_type = {
+ .release = cache_desc_release,
+ .sysfs_ops = &cache_desc_sysfs_ops,
+};
+
+static ssize_t cache_size_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%uK\n", cache->size);
+}
+
+static struct kobj_attribute cache_size_attr =
+ __ATTR(size, 0444, cache_size_show, NULL);
+
+static ssize_t cache_line_size_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%u\n", cache->line_size);
+}
+
+static struct kobj_attribute cache_line_size_attr =
+ __ATTR(coherency_line_size, 0444, cache_line_size_show, NULL);
+
+static ssize_t cache_nr_sets_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%u\n", cache->nr_sets);
+}
+
+static struct kobj_attribute cache_nr_sets_attr =
+ __ATTR(number_of_sets, 0444, cache_nr_sets_show, NULL);
+
+static ssize_t cache_type_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%s\n", cache->type);
+}
+
+static struct kobj_attribute cache_type_attr =
+ __ATTR(type, 0444, cache_type_show, NULL);
+
+static ssize_t cache_level_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%u\n", cache->level);
+}
+
+static struct kobj_attribute cache_level_attr =
+ __ATTR(level, 0444, cache_level_show, NULL);
+
+static ssize_t cache_assoc_show(struct kobject *k, struct kobj_attribute *attr, char *buf)
+{
+ struct cache_desc *cache = kobj_to_cache_desc(k);
+
+ return sprintf(buf, "%u\n", cache->associativity);
+}
+
+static struct kobj_attribute cache_assoc_attr =
+ __ATTR(ways_of_associativity, 0444, cache_assoc_show, NULL);
+
+struct cache_desc_info {
+ const char *type;
+ const char *size_prop;
+ const char *line_size_prop;
+ const char *nr_sets_prop;
+};
+
+/* PowerPC Processor binding says the [di]-cache-* must be equal on
+ * unified caches, so just use d-cache properties. */
+static struct cache_desc_info ucache_info = {
+ .type = "Unified",
+ .size_prop = "d-cache-size",
+ .line_size_prop = "d-cache-line-size",
+ .nr_sets_prop = "d-cache-sets",
+};
+
+static struct cache_desc_info dcache_info = {
+ .type = "Data",
+ .size_prop = "d-cache-size",
+ .line_size_prop = "d-cache-line-size",
+ .nr_sets_prop = "d-cache-sets",
+};
+
+static struct cache_desc_info icache_info = {
+ .type = "Instruction",
+ .size_prop = "i-cache-size",
+ .line_size_prop = "i-cache-line-size",
+ .nr_sets_prop = "i-cache-sets",
+};
+
+static struct cache_desc * __cpuinit create_cache_desc(struct device_node *np, struct kobject *parent, int index, int level, struct cache_desc_info *info)
+{
+ const u32 *cache_line_size;
+ struct cache_desc *new;
+ const u32 *cache_size;
+ const u32 *nr_sets;
+ int rc;
+
+ new = kzalloc(sizeof(*new), GFP_KERNEL);
+ if (!new)
+ return NULL;
+
+ rc = kobject_init_and_add(&new->kobj, &cache_desc_type, parent,
+ "index%d", index);
+ if (rc)
+ goto err;
+
+ /* type */
+ new->type = info->type;
+ rc = sysfs_create_file(&new->kobj, &cache_type_attr.attr);
+ WARN_ON(rc);
+
+ /* level */
+ new->level = level;
+ rc = sysfs_create_file(&new->kobj, &cache_level_attr.attr);
+ WARN_ON(rc);
+
+ /* size */
+ cache_size = of_get_property(np, info->size_prop, NULL);
+ if (cache_size) {
+ new->size = *cache_size / 1024;
+ rc = sysfs_create_file(&new->kobj,
+ &cache_size_attr.attr);
+ WARN_ON(rc);
+ }
+
+ /* coherency_line_size */
+ cache_line_size = of_get_property(np, info->line_size_prop, NULL);
+ if (cache_line_size) {
+ new->line_size = *cache_line_size;
+ rc = sysfs_create_file(&new->kobj,
+ &cache_line_size_attr.attr);
+ WARN_ON(rc);
+ }
+
+ /* number_of_sets */
+ nr_sets = of_get_property(np, info->nr_sets_prop, NULL);
+ if (nr_sets) {
+ new->nr_sets = *nr_sets;
+ rc = sysfs_create_file(&new->kobj,
+ &cache_nr_sets_attr.attr);
+ WARN_ON(rc);
+ }
+
+ /* ways_of_associativity */
+ if (new->nr_sets == 1) {
+ /* fully associative */
+ new->associativity = 0;
+ goto create_assoc;
+ }
+
+ if (new->nr_sets && new->size && new->line_size) {
+ /* If we have values for all of these we can derive
+ * the associativity. */
+ new->associativity =
+ ((new->size * 1024) / new->nr_sets) / new->line_size;
+create_assoc:
+ rc = sysfs_create_file(&new->kobj,
+ &cache_assoc_attr.attr);
+ WARN_ON(rc);
+ }
+
+ return new;
+err:
+ kfree(new);
+ return NULL;
+}
+
+static bool cache_is_unified(struct device_node *np)
+{
+ return of_get_property(np, "cache-unified", NULL);
+}
+
+static struct cache_desc * __cpuinit create_cache_index_info(struct device_node *np, struct kobject *parent, int index, int level)
+{
+ const phandle *next_cache_phandle;
+ struct device_node *next_cache;
+ struct cache_desc *new, **end;
+
+ printk("%s(np = %s, index = %d)\n", __func__, np->name, index);
+ if (cache_is_unified(np)) {
+ new = create_cache_desc(np, parent, index, level,
+ &ucache_info);
+ } else {
+ new = create_cache_desc(np, parent, index, level,
+ &dcache_info);
+ if (new) {
+ index++;
+ new->next = create_cache_desc(np, parent, index, level,
+ &icache_info);
+ }
+ }
+ if (!new)
+ return NULL;
+
+ end = &new->next;
+ while (*end)
+ end = &(*end)->next;
+
+ next_cache_phandle = of_get_property(np, "l2-cache", NULL);
+ if (!next_cache_phandle)
+ goto out;
+
+ next_cache = of_find_node_by_phandle(*next_cache_phandle);
+ if (!next_cache)
+ goto out;
+
+ *end = create_cache_index_info(next_cache, parent, ++index, ++level);
+
+ of_node_put(next_cache);
+out:
+ return new;
+}
+
+static void __cpuinit create_cache_info(struct sys_device *sysdev)
+{
+ struct kobject *cache_toplevel;
+ struct device_node *np = NULL;
+ int cpu = sysdev->id;
+
+ cache_toplevel = kobject_create_and_add("cache", &sysdev->kobj);
+ if (!cache_toplevel)
+ return;
+ per_cpu(cache_toplevel, cpu) = cache_toplevel;
+
+ np = of_get_cpu_node(cpu, NULL);
+ per_cpu(cache_desc, cpu) =
+ create_cache_index_info(np, cache_toplevel, 0, 1);
+ of_node_put(np);
+
+ return;
+}
static void __cpuinit register_cpu_online(unsigned int cpu)
{
@@ -346,9 +628,33 @@ static void __cpuinit register_cpu_online(unsigned int cpu)
if (cpu_has_feature(CPU_FTR_DSCR))
sysdev_create_file(s, &attr_dscr);
+
+ create_cache_info(s);
}
#ifdef CONFIG_HOTPLUG_CPU
+static void remove_cache_info(struct sys_device *sysdev)
+{
+ struct kobject *cache_toplevel;
+ struct cache_desc *cache_desc;
+ int cpu = sysdev->id;
+
+ cache_desc = per_cpu(cache_desc, cpu);
+
+ sysfs_remove_file(&cache_desc->kobj, &cache_size_attr.attr);
+ sysfs_remove_file(&cache_desc->kobj, &cache_line_size_attr.attr);
+ sysfs_remove_file(&cache_desc->kobj, &cache_type_attr.attr);
+ sysfs_remove_file(&cache_desc->kobj, &cache_level_attr.attr);
+ sysfs_remove_file(&cache_desc->kobj, &cache_nr_sets_attr.attr);
+ sysfs_remove_file(&cache_desc->kobj, &cache_assoc_attr.attr);
+
+ kobject_put(&cache_desc->kobj);
+
+ cache_toplevel = per_cpu(cache_toplevel, cpu);
+
+ kobject_put(cache_toplevel);
+}
+
static void unregister_cpu_online(unsigned int cpu)
{
struct cpu *c = &per_cpu(cpu_devices, cpu);
@@ -399,6 +705,8 @@ static void unregister_cpu_online(unsigned int cpu)
if (cpu_has_feature(CPU_FTR_DSCR))
sysdev_remove_file(s, &attr_dscr);
+
+ remove_cache_info(s);
}
#endif /* CONFIG_HOTPLUG_CPU */
--
1.5.6.2
^ permalink raw reply related [flat|nested] 8+ messages in thread* Re: [PATCH 0/6] powerpc topology updates
2008-07-27 5:24 [PATCH 0/6] powerpc topology updates Nathan Lynch
` (5 preceding siblings ...)
2008-07-27 5:24 ` [PATCH 6/6] show processor cache information in sysfs Nathan Lynch
@ 2008-07-28 6:34 ` Benjamin Herrenschmidt
6 siblings, 0 replies; 8+ messages in thread
From: Benjamin Herrenschmidt @ 2008-07-28 6:34 UTC (permalink / raw)
To: Nathan Lynch; +Cc: linuxppc-dev, Paul Mackerras
On Sun, 2008-07-27 at 00:24 -0500, Nathan Lynch wrote:
> This series consists of a couple of cleanups and a few feature
> additions, all of which are more or less related to system topology
> (threads, cores, caches, sysfs...) These are independent of
> each other except for 4 and 5 (core sibling and core id info).
>
> I'd say the highlights are the last three patches, which add core and
> cache information to sysfs. Here is some example output with the
> patches applied:
Applied with minor changes.
I made cpu_to_l2cache() use of_find_device_by_phandle() which should
work as well on pseries and should have the added advantage of working
on others too (ie. it will use device_node->linux_phandle to do the
compare, which is setup with the base phandle and overriden by the
ibm,phandle if there is one).
I also did a few cleanups of printk's and error checking.
Cheers,
Ben.
^ permalink raw reply [flat|nested] 8+ messages in thread