Subject: [PATCH 0/2] oprofile, x86: updates for AMD family 15h systems

(resending due to mail corruption)

This patch set adds support for AMD family 15h systems. It adds cpu
detection code for family 15h and implements support for 6 counters.

Suravee already sent updates for the oprofile userland to the oprofile
mailing list.

Please review. If the patches are fine I will apply them to the
oprofile git repostitory on kernel.org.

Thanks,

-Robert



Subject: [PATCH 2/2] oprofile, x86: Add support for 6 counters (AMD family 15h)

This patch adds support for up to 6 hardware counters for AMD family
15h cpus. There is a new MSR range for hardware counters beginning at
MSRC001_0200 Performance Event Select (PERF_CTL0).

Signed-off-by: Robert Richter <[email protected]>
---
arch/x86/include/asm/msr-index.h | 4 +++
arch/x86/oprofile/op_model_amd.c | 54 +++++++++++++++++++++++++------------
2 files changed, 40 insertions(+), 18 deletions(-)

diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 3ea3dc4..e1636d6 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -123,6 +123,10 @@
#define MSR_AMD64_IBSCTL 0xc001103a
#define MSR_AMD64_IBSBRTARGET 0xc001103b

+/* Fam 15h MSRs */
+#define MSR_F15H_PERF_CTL 0xc0010200
+#define MSR_F15H_PERF_CTR 0xc0010201
+
/* Fam 10h MSRs */
#define MSR_FAM10H_MMIO_CONF_BASE 0xc0010058
#define FAM10H_MMIO_CONF_ENABLE (1<<0)
diff --git a/arch/x86/oprofile/op_model_amd.c b/arch/x86/oprofile/op_model_amd.c
index a011bcc..f2984d4 100644
--- a/arch/x86/oprofile/op_model_amd.c
+++ b/arch/x86/oprofile/op_model_amd.c
@@ -29,11 +29,12 @@
#include "op_x86_model.h"
#include "op_counter.h"

-#define NUM_COUNTERS 4
+#define NUM_COUNTERS 4
+#define NUM_COUNTERS_F15H 6
#ifdef CONFIG_OPROFILE_EVENT_MULTIPLEX
-#define NUM_VIRT_COUNTERS 32
+#define NUM_VIRT_COUNTERS 32
#else
-#define NUM_VIRT_COUNTERS NUM_COUNTERS
+#define NUM_VIRT_COUNTERS 0
#endif

#define OP_EVENT_MASK 0x0FFF
@@ -41,7 +42,8 @@

#define MSR_AMD_EVENTSEL_RESERVED ((0xFFFFFCF0ULL<<32)|(1ULL<<21))

-static unsigned long reset_value[NUM_VIRT_COUNTERS];
+static int num_counters;
+static unsigned long reset_value[OP_MAX_COUNTER];

#define IBS_FETCH_SIZE 6
#define IBS_OP_SIZE 12
@@ -387,7 +389,7 @@ static void op_mux_switch_ctrl(struct op_x86_model_spec const *model,
int i;

/* enable active counters */
- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
int virt = op_x86_phys_to_virt(i);
if (!reset_value[virt])
continue;
@@ -406,7 +408,7 @@ static void op_amd_shutdown(struct op_msrs const * const msrs)
{
int i;

- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
if (!msrs->counters[i].addr)
continue;
release_perfctr_nmi(MSR_K7_PERFCTR0 + i);
@@ -418,7 +420,7 @@ static int op_amd_fill_in_addresses(struct op_msrs * const msrs)
{
int i;

- for (i = 0; i < NUM_COUNTERS; i++) {
+ for (i = 0; i < num_counters; i++) {
if (!reserve_perfctr_nmi(MSR_K7_PERFCTR0 + i))
goto fail;
if (!reserve_evntsel_nmi(MSR_K7_EVNTSEL0 + i)) {
@@ -426,8 +428,13 @@ static int op_amd_fill_in_addresses(struct op_msrs * const msrs)
goto fail;
}
/* both registers must be reserved */
- msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
- msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
+ if (num_counters == NUM_COUNTERS_F15H) {
+ msrs->counters[i].addr = MSR_F15H_PERF_CTR + (i << 1);
+ msrs->controls[i].addr = MSR_F15H_PERF_CTL + (i << 1);
+ } else {
+ msrs->controls[i].addr = MSR_K7_EVNTSEL0 + i;
+ msrs->counters[i].addr = MSR_K7_PERFCTR0 + i;
+ }
continue;
fail:
if (!counter_config[i].enabled)
@@ -447,7 +454,7 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
int i;

/* setup reset_value */
- for (i = 0; i < NUM_VIRT_COUNTERS; ++i) {
+ for (i = 0; i < OP_MAX_COUNTER; ++i) {
if (counter_config[i].enabled
&& msrs->counters[op_x86_virt_to_phys(i)].addr)
reset_value[i] = counter_config[i].count;
@@ -456,7 +463,7 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
}

/* clear all counters */
- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
if (!msrs->controls[i].addr)
continue;
rdmsrl(msrs->controls[i].addr, val);
@@ -472,7 +479,7 @@ static void op_amd_setup_ctrs(struct op_x86_model_spec const *model,
}

/* enable active counters */
- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
int virt = op_x86_phys_to_virt(i);
if (!reset_value[virt])
continue;
@@ -503,7 +510,7 @@ static int op_amd_check_ctrs(struct pt_regs * const regs,
u64 val;
int i;

- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
int virt = op_x86_phys_to_virt(i);
if (!reset_value[virt])
continue;
@@ -526,7 +533,7 @@ static void op_amd_start(struct op_msrs const * const msrs)
u64 val;
int i;

- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
if (!reset_value[op_x86_phys_to_virt(i)])
continue;
rdmsrl(msrs->controls[i].addr, val);
@@ -546,7 +553,7 @@ static void op_amd_stop(struct op_msrs const * const msrs)
* Subtle: stop on all counters to avoid race with setting our
* pm callback
*/
- for (i = 0; i < NUM_COUNTERS; ++i) {
+ for (i = 0; i < num_counters; ++i) {
if (!reset_value[op_x86_phys_to_virt(i)])
continue;
rdmsrl(msrs->controls[i].addr, val);
@@ -698,18 +705,29 @@ static int setup_ibs_files(struct super_block *sb, struct dentry *root)
return 0;
}

+struct op_x86_model_spec op_amd_spec;
+
static int op_amd_init(struct oprofile_operations *ops)
{
init_ibs();
create_arch_files = ops->create_files;
ops->create_files = setup_ibs_files;
+
+ if (boot_cpu_data.x86 == 0x15) {
+ num_counters = NUM_COUNTERS_F15H;
+ } else {
+ num_counters = NUM_COUNTERS;
+ }
+
+ op_amd_spec.num_counters = num_counters;
+ op_amd_spec.num_controls = num_counters;
+ op_amd_spec.num_virt_counters = max(num_counters, NUM_VIRT_COUNTERS);
+
return 0;
}

struct op_x86_model_spec op_amd_spec = {
- .num_counters = NUM_COUNTERS,
- .num_controls = NUM_COUNTERS,
- .num_virt_counters = NUM_VIRT_COUNTERS,
+ /* num_counters/num_controls filled in at runtime */
.reserved = MSR_AMD_EVENTSEL_RESERVED,
.event_mask = OP_EVENT_MASK,
.init = op_amd_init,
--
1.7.3.2

Subject: [PATCH 1/2] oprofile, x86: Add support for AMD family 15h

This patch adds support for AMD family 15h (Interlagos/Valencia/
Zambezi) cpus.

Signed-off-by: Robert Richter <[email protected]>
---
arch/x86/oprofile/nmi_int.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/arch/x86/oprofile/nmi_int.c b/arch/x86/oprofile/nmi_int.c
index 4e8baad..358c8b9 100644
--- a/arch/x86/oprofile/nmi_int.c
+++ b/arch/x86/oprofile/nmi_int.c
@@ -732,6 +732,9 @@ int __init op_nmi_init(struct oprofile_operations *ops)
case 0x14:
cpu_type = "x86-64/family14h";
break;
+ case 0x15:
+ cpu_type = "x86-64/family15h";
+ break;
default:
return -ENODEV;
}
--
1.7.3.2

Subject: Re: [PATCH 0/2] oprofile, x86: updates for AMD family 15h systems

On 10.12.10 06:49:04, Robert Richter wrote:
> (resending due to mail corruption)
>
> This patch set adds support for AMD family 15h systems. It adds cpu
> detection code for family 15h and implements support for 6 counters.
>
> Suravee already sent updates for the oprofile userland to the oprofile
> mailing list.
>
> Please review. If the patches are fine I will apply them to the
> oprofile git repostitory on kernel.org.

I applied the patches to

git://git.kernel.org/pub/scm/linux/kernel/git/rric/oprofile.git core

-Robert

--
Advanced Micro Devices, Inc.
Operating System Research Center