* xenpm: provide core/package cstate residencies
@ 2010-07-12 11:03 Wei, Gang
2010-07-12 11:26 ` Jan Beulich
2010-07-12 15:22 ` Wei, Gang
0 siblings, 2 replies; 11+ messages in thread
From: Wei, Gang @ 2010-07-12 11:03 UTC (permalink / raw)
To: xen-devel@lists.xensource.com; +Cc: Keir Fraser, Wei, Gang
[-- Attachment #1: Type: text/plain, Size: 12055 bytes --]
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend sysctl_get_pmstat interface to pass the core/package cstate residencies, and modify xenpm to output those information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 17:47:25 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 17:33:17 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 17:38:39 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
diff -r 1af6303f103c xen/arch/x86/acpi/cpu_idle.c
--- a/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 17:40:52 2010 +0800
@@ -55,6 +55,14 @@
/*#define DEBUG_PM_CX*/
+#define GET_HW_RES_IN_NS(msr, val) \
+ do { rdmsrl(msr, val); val = tsc_ticks2ns(val); } while( 0 )
+#define GET_PC3_RES(val) GET_HW_RES_IN_NS(0x3F8, val)
+#define GET_PC6_RES(val) GET_HW_RES_IN_NS(0x3F9, val)
+#define GET_PC7_RES(val) GET_HW_RES_IN_NS(0x3FA, val)
+#define GET_CC3_RES(val) GET_HW_RES_IN_NS(0x3FC, val)
+#define GET_CC6_RES(val) GET_HW_RES_IN_NS(0x3FD, val)
+
static void lapic_timer_nop(void) { }
static void (*lapic_timer_off)(void);
static void (*lapic_timer_on)(void);
@@ -75,6 +83,47 @@ boolean_param("lapic_timer_c2_ok", local
boolean_param("lapic_timer_c2_ok", local_apic_timer_c2_ok);
static struct acpi_processor_power *__read_mostly processor_powers[NR_CPUS];
+
+struct hw_residencies
+{
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
+};
+
+static void do_get_hw_residencies(void *arg)
+{
+ struct hw_residencies *hw_res = (struct hw_residencies *)arg;
+
+ GET_PC3_RES(hw_res->pc3);
+ GET_PC6_RES(hw_res->pc6);
+ GET_PC7_RES(hw_res->pc7);
+ GET_CC3_RES(hw_res->cc3);
+ GET_CC6_RES(hw_res->cc6);
+}
+
+static void get_hw_residencies(uint32_t cpu, struct hw_residencies *hw_res)
+{
+ if ( smp_processor_id() == cpu )
+ do_get_hw_residencies((void *)hw_res);
+ else
+ on_selected_cpus(cpumask_of(cpu),
+ do_get_hw_residencies, (void *)hw_res, 1);
+}
+
+static void print_hw_residencies(uint32_t cpu)
+{
+ struct hw_residencies hw_res;
+
+ get_hw_residencies(cpu, &hw_res);
+
+ printk("PC3[%"PRId64"] PC6[%"PRId64"] PC7[%"PRId64"]\n",
+ hw_res.pc3, hw_res.pc6, hw_res.pc7);
+ printk("CC3[%"PRId64"] CC6[%"PRId64"]\n",
+ hw_res.cc3, hw_res.cc6);
+}
static char* acpi_cstate_method_name[] =
{
@@ -113,6 +162,7 @@ static void print_acpi_power(uint32_t cp
printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
idle_usage, NOW() - idle_res);
+ print_hw_residencies(cpu);
}
static void dump_cx(unsigned char key)
@@ -933,6 +983,7 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
const struct acpi_processor_power *power = processor_powers[cpuid];
uint64_t usage, res, idle_usage = 0, idle_res = 0;
int i;
+ struct hw_residencies hw_res = { 0 };
if ( power == NULL )
{
@@ -965,6 +1016,26 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
return -EFAULT;
}
+ switch ( cpuid_eax(1) & 0x0FFF0FF0 )
+ {
+ /* Nehalem */
+ case 0x106A0:
+ case 0x106E0:
+ case 0x106F0:
+ case 0x206E0:
+ /* Westmere */
+ case 0x20650:
+ case 0x206C0:
+ get_hw_residencies(cpuid, &hw_res);
+ break;
+ }
+
+ stat->pc3 = hw_res.pc3;
+ stat->pc6 = hw_res.pc6;
+ stat->pc7 = hw_res.pc7;
+ stat->cc3 = hw_res.cc3;
+ stat->cc6 = hw_res.cc6;
+
return 0;
}
diff -r 1af6303f103c xen/arch/x86/time.c
--- a/xen/arch/x86/time.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/time.c Mon Jul 12 14:13:03 2010 +0800
@@ -785,6 +785,13 @@ s_time_t get_s_time(void)
now = t->stime_local_stamp + scale_delta(delta, &t->tsc_scale);
return now;
+}
+
+uint64_t tsc_ticks2ns(uint64_t ticks)
+{
+ struct cpu_time *t = &this_cpu(cpu_time);
+
+ return scale_delta(ticks, &t->tsc_scale);
}
/* Explicitly OR with 1 just in case version number gets out of sync. */
diff -r 1af6303f103c xen/include/asm-x86/time.h
--- a/xen/include/asm-x86/time.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/asm-x86/time.h Mon Jul 12 14:28:57 2010 +0800
@@ -56,6 +56,8 @@ uint64_t acpi_pm_tick_to_ns(uint64_t tic
uint64_t acpi_pm_tick_to_ns(uint64_t ticks);
uint64_t ns_to_acpi_pm_tick(uint64_t ns);
+uint64_t tsc_ticks2ns(uint64_t ticks);
+
void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp);
u64 gtime_to_gtsc(struct domain *d, u64 tsc);
diff -r 1af6303f103c xen/include/public/sysctl.h
--- a/xen/include/public/sysctl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/public/sysctl.h Mon Jul 12 17:34:58 2010 +0800
@@ -223,6 +223,11 @@ struct pm_cx_stat {
uint64_aligned_t idle_time; /* idle time from boot */
XEN_GUEST_HANDLE_64(uint64) triggers; /* Cx trigger counts */
XEN_GUEST_HANDLE_64(uint64) residencies; /* Cx residencies */
+ uint64_aligned_t pc3;
+ uint64_aligned_t pc6;
+ uint64_aligned_t pc7;
+ uint64_aligned_t cc3;
+ uint64_aligned_t cc6;
};
struct xen_sysctl_get_pmstat {
[-- Attachment #2: xenpm-package-cstate.patch --]
[-- Type: application/octet-stream, Size: 11725 bytes --]
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend sysctl_get_pmstat interface to pass the core/package cstate residencies, and modify xenpm to output those information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 17:47:25 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 17:33:17 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 17:38:39 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
diff -r 1af6303f103c xen/arch/x86/acpi/cpu_idle.c
--- a/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 17:40:52 2010 +0800
@@ -55,6 +55,14 @@
/*#define DEBUG_PM_CX*/
+#define GET_HW_RES_IN_NS(msr, val) \
+ do { rdmsrl(msr, val); val = tsc_ticks2ns(val); } while( 0 )
+#define GET_PC3_RES(val) GET_HW_RES_IN_NS(0x3F8, val)
+#define GET_PC6_RES(val) GET_HW_RES_IN_NS(0x3F9, val)
+#define GET_PC7_RES(val) GET_HW_RES_IN_NS(0x3FA, val)
+#define GET_CC3_RES(val) GET_HW_RES_IN_NS(0x3FC, val)
+#define GET_CC6_RES(val) GET_HW_RES_IN_NS(0x3FD, val)
+
static void lapic_timer_nop(void) { }
static void (*lapic_timer_off)(void);
static void (*lapic_timer_on)(void);
@@ -75,6 +83,47 @@ boolean_param("lapic_timer_c2_ok", local
boolean_param("lapic_timer_c2_ok", local_apic_timer_c2_ok);
static struct acpi_processor_power *__read_mostly processor_powers[NR_CPUS];
+
+struct hw_residencies
+{
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
+};
+
+static void do_get_hw_residencies(void *arg)
+{
+ struct hw_residencies *hw_res = (struct hw_residencies *)arg;
+
+ GET_PC3_RES(hw_res->pc3);
+ GET_PC6_RES(hw_res->pc6);
+ GET_PC7_RES(hw_res->pc7);
+ GET_CC3_RES(hw_res->cc3);
+ GET_CC6_RES(hw_res->cc6);
+}
+
+static void get_hw_residencies(uint32_t cpu, struct hw_residencies *hw_res)
+{
+ if ( smp_processor_id() == cpu )
+ do_get_hw_residencies((void *)hw_res);
+ else
+ on_selected_cpus(cpumask_of(cpu),
+ do_get_hw_residencies, (void *)hw_res, 1);
+}
+
+static void print_hw_residencies(uint32_t cpu)
+{
+ struct hw_residencies hw_res;
+
+ get_hw_residencies(cpu, &hw_res);
+
+ printk("PC3[%"PRId64"] PC6[%"PRId64"] PC7[%"PRId64"]\n",
+ hw_res.pc3, hw_res.pc6, hw_res.pc7);
+ printk("CC3[%"PRId64"] CC6[%"PRId64"]\n",
+ hw_res.cc3, hw_res.cc6);
+}
static char* acpi_cstate_method_name[] =
{
@@ -113,6 +162,7 @@ static void print_acpi_power(uint32_t cp
printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
idle_usage, NOW() - idle_res);
+ print_hw_residencies(cpu);
}
static void dump_cx(unsigned char key)
@@ -933,6 +983,7 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
const struct acpi_processor_power *power = processor_powers[cpuid];
uint64_t usage, res, idle_usage = 0, idle_res = 0;
int i;
+ struct hw_residencies hw_res = { 0 };
if ( power == NULL )
{
@@ -965,6 +1016,26 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
return -EFAULT;
}
+ switch ( cpuid_eax(1) & 0x0FFF0FF0 )
+ {
+ /* Nehalem */
+ case 0x106A0:
+ case 0x106E0:
+ case 0x106F0:
+ case 0x206E0:
+ /* Westmere */
+ case 0x20650:
+ case 0x206C0:
+ get_hw_residencies(cpuid, &hw_res);
+ break;
+ }
+
+ stat->pc3 = hw_res.pc3;
+ stat->pc6 = hw_res.pc6;
+ stat->pc7 = hw_res.pc7;
+ stat->cc3 = hw_res.cc3;
+ stat->cc6 = hw_res.cc6;
+
return 0;
}
diff -r 1af6303f103c xen/arch/x86/time.c
--- a/xen/arch/x86/time.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/time.c Mon Jul 12 14:13:03 2010 +0800
@@ -785,6 +785,13 @@ s_time_t get_s_time(void)
now = t->stime_local_stamp + scale_delta(delta, &t->tsc_scale);
return now;
+}
+
+uint64_t tsc_ticks2ns(uint64_t ticks)
+{
+ struct cpu_time *t = &this_cpu(cpu_time);
+
+ return scale_delta(ticks, &t->tsc_scale);
}
/* Explicitly OR with 1 just in case version number gets out of sync. */
diff -r 1af6303f103c xen/include/asm-x86/time.h
--- a/xen/include/asm-x86/time.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/asm-x86/time.h Mon Jul 12 14:28:57 2010 +0800
@@ -56,6 +56,8 @@ uint64_t acpi_pm_tick_to_ns(uint64_t tic
uint64_t acpi_pm_tick_to_ns(uint64_t ticks);
uint64_t ns_to_acpi_pm_tick(uint64_t ns);
+uint64_t tsc_ticks2ns(uint64_t ticks);
+
void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp);
u64 gtime_to_gtsc(struct domain *d, u64 tsc);
diff -r 1af6303f103c xen/include/public/sysctl.h
--- a/xen/include/public/sysctl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/public/sysctl.h Mon Jul 12 17:34:58 2010 +0800
@@ -223,6 +223,11 @@ struct pm_cx_stat {
uint64_aligned_t idle_time; /* idle time from boot */
XEN_GUEST_HANDLE_64(uint64) triggers; /* Cx trigger counts */
XEN_GUEST_HANDLE_64(uint64) residencies; /* Cx residencies */
+ uint64_aligned_t pc3;
+ uint64_aligned_t pc6;
+ uint64_aligned_t pc7;
+ uint64_aligned_t cc3;
+ uint64_aligned_t cc6;
};
struct xen_sysctl_get_pmstat {
[-- Attachment #3: Type: text/plain, Size: 138 bytes --]
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: xenpm: provide core/package cstate residencies
2010-07-12 11:03 xenpm: provide core/package cstate residencies Wei, Gang
@ 2010-07-12 11:26 ` Jan Beulich
2010-07-12 15:11 ` Wei, Gang
2010-07-12 15:22 ` Wei, Gang
1 sibling, 1 reply; 11+ messages in thread
From: Jan Beulich @ 2010-07-12 11:26 UTC (permalink / raw)
To: Gang Wei; +Cc: xen-devel@lists.xensource.com, Keir Fraser
>>> On 12.07.10 at 13:03, "Wei, Gang" <gang.wei@intel.com> wrote:
> @@ -113,6 +162,7 @@ static void print_acpi_power(uint32_t cp
> printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
> idle_usage, NOW() - idle_res);
>
> + print_hw_residencies(cpu);
Isn't this resulting in a set of unconditional rdmsr-s on possibly
unimplemented MSRs?
Jan
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: xenpm: provide core/package cstate residencies
2010-07-12 11:26 ` Jan Beulich
@ 2010-07-12 15:11 ` Wei, Gang
0 siblings, 0 replies; 11+ messages in thread
From: Wei, Gang @ 2010-07-12 15:11 UTC (permalink / raw)
To: Jan Beulich; +Cc: xen-devel@lists.xensource.com, Keir Fraser, Wei, Gang
On 2010 07 12 19:27, Jan Beulich wrote:
>>>> On 12.07.10 at 13:03, "Wei, Gang" <gang.wei@intel.com> wrote:
>> @@ -113,6 +162,7 @@ static void print_acpi_power(uint32_t cp
>> printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
>> idle_usage, NOW() - idle_res);
>>
>> + print_hw_residencies(cpu);
>
> Isn't this resulting in a set of unconditional rdmsr-s on possibly
> unimplemented MSRs?
Oh, yes, my oversight. Thanks for pointing it out. I will resend a modified version.
Jimmy
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: xenpm: provide core/package cstate residencies
2010-07-12 11:03 xenpm: provide core/package cstate residencies Wei, Gang
2010-07-12 11:26 ` Jan Beulich
@ 2010-07-12 15:22 ` Wei, Gang
2010-07-12 17:34 ` Keir Fraser
1 sibling, 1 reply; 11+ messages in thread
From: Wei, Gang @ 2010-07-12 15:22 UTC (permalink / raw)
To: xen-devel@lists.xensource.com; +Cc: Keir Fraser, Wei, Gang
[-- Attachment #1: Type: text/plain, Size: 12199 bytes --]
Resend it.
=======
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend sysctl_get_pmstat interface to pass the core/package cstate residencies, and modify xenpm to output those information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 22:16:04 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 22:16:04 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 22:16:04 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
diff -r 1af6303f103c xen/arch/x86/acpi/cpu_idle.c
--- a/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 22:58:25 2010 +0800
@@ -55,6 +55,14 @@
/*#define DEBUG_PM_CX*/
+#define GET_HW_RES_IN_NS(msr, val) \
+ do { rdmsrl(msr, val); val = tsc_ticks2ns(val); } while( 0 )
+#define GET_PC3_RES(val) GET_HW_RES_IN_NS(0x3F8, val)
+#define GET_PC6_RES(val) GET_HW_RES_IN_NS(0x3F9, val)
+#define GET_PC7_RES(val) GET_HW_RES_IN_NS(0x3FA, val)
+#define GET_CC3_RES(val) GET_HW_RES_IN_NS(0x3FC, val)
+#define GET_CC6_RES(val) GET_HW_RES_IN_NS(0x3FD, val)
+
static void lapic_timer_nop(void) { }
static void (*lapic_timer_off)(void);
static void (*lapic_timer_on)(void);
@@ -75,6 +83,63 @@ boolean_param("lapic_timer_c2_ok", local
boolean_param("lapic_timer_c2_ok", local_apic_timer_c2_ok);
static struct acpi_processor_power *__read_mostly processor_powers[NR_CPUS];
+
+struct hw_residencies
+{
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
+};
+
+static void do_get_hw_residencies(void *arg)
+{
+ struct cpuinfo_x86 *c = ¤t_cpu_data;
+ struct hw_residencies *hw_res = (struct hw_residencies *)arg;
+
+ if ( c->x86_vendor != X86_VENDOR_INTEL || c->x86 != 6 )
+ return;
+
+ switch ( c->x86_model )
+ {
+ /* Nehalem */
+ case 0x1A:
+ case 0x1E:
+ case 0x1F:
+ case 0x2E:
+ /* Westmere */
+ case 0x25:
+ case 0x2C:
+ GET_PC3_RES(hw_res->pc3);
+ GET_PC6_RES(hw_res->pc6);
+ GET_PC7_RES(hw_res->pc7);
+ GET_CC3_RES(hw_res->cc3);
+ GET_CC6_RES(hw_res->cc6);
+ break;
+ }
+}
+
+static void get_hw_residencies(uint32_t cpu, struct hw_residencies *hw_res)
+{
+ if ( smp_processor_id() == cpu )
+ do_get_hw_residencies((void *)hw_res);
+ else
+ on_selected_cpus(cpumask_of(cpu),
+ do_get_hw_residencies, (void *)hw_res, 1);
+}
+
+static void print_hw_residencies(uint32_t cpu)
+{
+ struct hw_residencies hw_res = {0};
+
+ get_hw_residencies(cpu, &hw_res);
+
+ printk("PC3[%"PRId64"] PC6[%"PRId64"] PC7[%"PRId64"]\n",
+ hw_res.pc3, hw_res.pc6, hw_res.pc7);
+ printk("CC3[%"PRId64"] CC6[%"PRId64"]\n",
+ hw_res.cc3, hw_res.cc6);
+}
static char* acpi_cstate_method_name[] =
{
@@ -113,6 +178,7 @@ static void print_acpi_power(uint32_t cp
printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
idle_usage, NOW() - idle_res);
+ print_hw_residencies(cpu);
}
static void dump_cx(unsigned char key)
@@ -933,6 +999,7 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
const struct acpi_processor_power *power = processor_powers[cpuid];
uint64_t usage, res, idle_usage = 0, idle_res = 0;
int i;
+ struct hw_residencies hw_res = {0};
if ( power == NULL )
{
@@ -965,6 +1032,14 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
return -EFAULT;
}
+ get_hw_residencies(cpuid, &hw_res);
+
+ stat->pc3 = hw_res.pc3;
+ stat->pc6 = hw_res.pc6;
+ stat->pc7 = hw_res.pc7;
+ stat->cc3 = hw_res.cc3;
+ stat->cc6 = hw_res.cc6;
+
return 0;
}
diff -r 1af6303f103c xen/arch/x86/time.c
--- a/xen/arch/x86/time.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/time.c Mon Jul 12 22:16:04 2010 +0800
@@ -785,6 +785,13 @@ s_time_t get_s_time(void)
now = t->stime_local_stamp + scale_delta(delta, &t->tsc_scale);
return now;
+}
+
+uint64_t tsc_ticks2ns(uint64_t ticks)
+{
+ struct cpu_time *t = &this_cpu(cpu_time);
+
+ return scale_delta(ticks, &t->tsc_scale);
}
/* Explicitly OR with 1 just in case version number gets out of sync. */
diff -r 1af6303f103c xen/include/asm-x86/time.h
--- a/xen/include/asm-x86/time.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/asm-x86/time.h Mon Jul 12 22:16:04 2010 +0800
@@ -56,6 +56,8 @@ uint64_t acpi_pm_tick_to_ns(uint64_t tic
uint64_t acpi_pm_tick_to_ns(uint64_t ticks);
uint64_t ns_to_acpi_pm_tick(uint64_t ns);
+uint64_t tsc_ticks2ns(uint64_t ticks);
+
void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp);
u64 gtime_to_gtsc(struct domain *d, u64 tsc);
diff -r 1af6303f103c xen/include/public/sysctl.h
--- a/xen/include/public/sysctl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/public/sysctl.h Mon Jul 12 22:16:04 2010 +0800
@@ -223,6 +223,11 @@ struct pm_cx_stat {
uint64_aligned_t idle_time; /* idle time from boot */
XEN_GUEST_HANDLE_64(uint64) triggers; /* Cx trigger counts */
XEN_GUEST_HANDLE_64(uint64) residencies; /* Cx residencies */
+ uint64_aligned_t pc3;
+ uint64_aligned_t pc6;
+ uint64_aligned_t pc7;
+ uint64_aligned_t cc3;
+ uint64_aligned_t cc6;
};
struct xen_sysctl_get_pmstat {
[-- Attachment #2: xenpm-package-cstate-v2.patch --]
[-- Type: application/octet-stream, Size: 11842 bytes --]
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend sysctl_get_pmstat interface to pass the core/package cstate residencies, and modify xenpm to output those information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 22:16:04 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 22:16:04 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 22:16:04 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
diff -r 1af6303f103c xen/arch/x86/acpi/cpu_idle.c
--- a/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 22:58:25 2010 +0800
@@ -55,6 +55,14 @@
/*#define DEBUG_PM_CX*/
+#define GET_HW_RES_IN_NS(msr, val) \
+ do { rdmsrl(msr, val); val = tsc_ticks2ns(val); } while( 0 )
+#define GET_PC3_RES(val) GET_HW_RES_IN_NS(0x3F8, val)
+#define GET_PC6_RES(val) GET_HW_RES_IN_NS(0x3F9, val)
+#define GET_PC7_RES(val) GET_HW_RES_IN_NS(0x3FA, val)
+#define GET_CC3_RES(val) GET_HW_RES_IN_NS(0x3FC, val)
+#define GET_CC6_RES(val) GET_HW_RES_IN_NS(0x3FD, val)
+
static void lapic_timer_nop(void) { }
static void (*lapic_timer_off)(void);
static void (*lapic_timer_on)(void);
@@ -75,6 +83,63 @@ boolean_param("lapic_timer_c2_ok", local
boolean_param("lapic_timer_c2_ok", local_apic_timer_c2_ok);
static struct acpi_processor_power *__read_mostly processor_powers[NR_CPUS];
+
+struct hw_residencies
+{
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
+};
+
+static void do_get_hw_residencies(void *arg)
+{
+ struct cpuinfo_x86 *c = ¤t_cpu_data;
+ struct hw_residencies *hw_res = (struct hw_residencies *)arg;
+
+ if ( c->x86_vendor != X86_VENDOR_INTEL || c->x86 != 6 )
+ return;
+
+ switch ( c->x86_model )
+ {
+ /* Nehalem */
+ case 0x1A:
+ case 0x1E:
+ case 0x1F:
+ case 0x2E:
+ /* Westmere */
+ case 0x25:
+ case 0x2C:
+ GET_PC3_RES(hw_res->pc3);
+ GET_PC6_RES(hw_res->pc6);
+ GET_PC7_RES(hw_res->pc7);
+ GET_CC3_RES(hw_res->cc3);
+ GET_CC6_RES(hw_res->cc6);
+ break;
+ }
+}
+
+static void get_hw_residencies(uint32_t cpu, struct hw_residencies *hw_res)
+{
+ if ( smp_processor_id() == cpu )
+ do_get_hw_residencies((void *)hw_res);
+ else
+ on_selected_cpus(cpumask_of(cpu),
+ do_get_hw_residencies, (void *)hw_res, 1);
+}
+
+static void print_hw_residencies(uint32_t cpu)
+{
+ struct hw_residencies hw_res = {0};
+
+ get_hw_residencies(cpu, &hw_res);
+
+ printk("PC3[%"PRId64"] PC6[%"PRId64"] PC7[%"PRId64"]\n",
+ hw_res.pc3, hw_res.pc6, hw_res.pc7);
+ printk("CC3[%"PRId64"] CC6[%"PRId64"]\n",
+ hw_res.cc3, hw_res.cc6);
+}
static char* acpi_cstate_method_name[] =
{
@@ -113,6 +178,7 @@ static void print_acpi_power(uint32_t cp
printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
idle_usage, NOW() - idle_res);
+ print_hw_residencies(cpu);
}
static void dump_cx(unsigned char key)
@@ -933,6 +999,7 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
const struct acpi_processor_power *power = processor_powers[cpuid];
uint64_t usage, res, idle_usage = 0, idle_res = 0;
int i;
+ struct hw_residencies hw_res = {0};
if ( power == NULL )
{
@@ -965,6 +1032,14 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
return -EFAULT;
}
+ get_hw_residencies(cpuid, &hw_res);
+
+ stat->pc3 = hw_res.pc3;
+ stat->pc6 = hw_res.pc6;
+ stat->pc7 = hw_res.pc7;
+ stat->cc3 = hw_res.cc3;
+ stat->cc6 = hw_res.cc6;
+
return 0;
}
diff -r 1af6303f103c xen/arch/x86/time.c
--- a/xen/arch/x86/time.c Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/arch/x86/time.c Mon Jul 12 22:16:04 2010 +0800
@@ -785,6 +785,13 @@ s_time_t get_s_time(void)
now = t->stime_local_stamp + scale_delta(delta, &t->tsc_scale);
return now;
+}
+
+uint64_t tsc_ticks2ns(uint64_t ticks)
+{
+ struct cpu_time *t = &this_cpu(cpu_time);
+
+ return scale_delta(ticks, &t->tsc_scale);
}
/* Explicitly OR with 1 just in case version number gets out of sync. */
diff -r 1af6303f103c xen/include/asm-x86/time.h
--- a/xen/include/asm-x86/time.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/asm-x86/time.h Mon Jul 12 22:16:04 2010 +0800
@@ -56,6 +56,8 @@ uint64_t acpi_pm_tick_to_ns(uint64_t tic
uint64_t acpi_pm_tick_to_ns(uint64_t ticks);
uint64_t ns_to_acpi_pm_tick(uint64_t ns);
+uint64_t tsc_ticks2ns(uint64_t ticks);
+
void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp);
u64 gtime_to_gtsc(struct domain *d, u64 tsc);
diff -r 1af6303f103c xen/include/public/sysctl.h
--- a/xen/include/public/sysctl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/xen/include/public/sysctl.h Mon Jul 12 22:16:04 2010 +0800
@@ -223,6 +223,11 @@ struct pm_cx_stat {
uint64_aligned_t idle_time; /* idle time from boot */
XEN_GUEST_HANDLE_64(uint64) triggers; /* Cx trigger counts */
XEN_GUEST_HANDLE_64(uint64) residencies; /* Cx residencies */
+ uint64_aligned_t pc3;
+ uint64_aligned_t pc6;
+ uint64_aligned_t pc7;
+ uint64_aligned_t cc3;
+ uint64_aligned_t cc6;
};
struct xen_sysctl_get_pmstat {
[-- Attachment #3: Type: text/plain, Size: 138 bytes --]
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: xenpm: provide core/package cstate residencies
2010-07-12 15:22 ` Wei, Gang
@ 2010-07-12 17:34 ` Keir Fraser
2010-07-13 0:36 ` Wei, Gang
` (2 more replies)
0 siblings, 3 replies; 11+ messages in thread
From: Keir Fraser @ 2010-07-12 17:34 UTC (permalink / raw)
To: Wei, Gang, xen-devel@lists.xensource.com; +Cc: Ian Jackson
On 12/07/2010 16:22, "Wei, Gang" <gang.wei@intel.com> wrote:
> Resend it.
>
> =======
> xenpm: provide core/package cstate residencies
>
> According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel
> Nehalem/Westmere processors provide h/w MSR to report the core/package cstate
> residencies.Extend sysctl_get_pmstat interface to pass the core/package cstate
> residencies, and modify xenpm to output those information.
I applied the hypervisor component of this patch. I leave it to Ian Jackson
to deal with the tools part. In future please split patches that touch both
hypervisor and tools into a patch series in which each component patch
touches only one or the other.
Thanks,
Keir
> Signed-off-by: Wei Gang <gang.wei@intel.com>
>
> diff -r 1af6303f103c tools/libxc/xc_pm.c
> --- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
> +++ b/tools/libxc/xc_pm.c Mon Jul 12 22:16:04 2010 +0800
> @@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
> cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
> cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
> cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
> + cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
> + cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
> + cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
> + cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
> + cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
>
> unlock_3:
> unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
> diff -r 1af6303f103c tools/libxc/xenctrl.h
> --- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
> +++ b/tools/libxc/xenctrl.h Mon Jul 12 22:16:04 2010 +0800
> @@ -1393,6 +1393,11 @@ struct xc_cx_stat {
> uint64_t idle_time; /* idle time from boot */
> uint64_t *triggers; /* Cx trigger counts */
> uint64_t *residencies; /* Cx residencies */
> + uint64_t pc3;
> + uint64_t pc6;
> + uint64_t pc7;
> + uint64_t cc3;
> + uint64_t cc6;
> };
> typedef struct xc_cx_stat xc_cx_stat_t;
>
> diff -r 1af6303f103c tools/misc/xenpm.c
> --- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
> +++ b/tools/misc/xenpm.c Mon Jul 12 22:16:04 2010 +0800
> @@ -15,6 +15,7 @@
> * this program; if not, write to the Free Software Foundation, Inc., 59
> Temple
> * Place - Suite 330, Boston, MA 02111-1307 USA.
> */
> +#define MAX_NR_CPU 512
>
> #include <stdio.h>
> #include <stdlib.h>
> @@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
> printf(" residency [%020"PRIu64" ms]\n",
> cxstat->residencies[i]/1000000UL);
> }
> + printf("pc3 : [%020"PRIu64" ms]\n"
> + "pc6 : [%020"PRIu64" ms]\n"
> + "pc7 : [%020"PRIu64" ms]\n",
> + cxstat->pc3/1000000UL, cxstat->pc6/1000000UL,
> cxstat->pc7/1000000UL);
> + printf("cc3 : [%020"PRIu64" ms]\n"
> + "cc6 : [%020"PRIu64" ms]\n",
> + cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
> printf("\n");
> }
>
> @@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
>
> static void signal_int_handler(int signo)
> {
> - int i, j;
> + int i, j, k, ret;
> struct timeval tv;
> int cx_cap = 0, px_cap = 0;
> + uint32_t cpu_to_core[MAX_NR_CPU];
> + uint32_t cpu_to_socket[MAX_NR_CPU];
> + uint32_t cpu_to_node[MAX_NR_CPU];
> + xc_topologyinfo_t info = { 0 };
>
> if ( gettimeofday(&tv, NULL) == -1 )
> {
> @@ -369,6 +381,93 @@ static void signal_int_handler(int signo
> pxstat_start[i].pt[j].residency;
> printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
> res / 1000000UL, 100UL * res / (double)sum_px[i]);
> + }
> + }
> + }
> +
> + set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
> + set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
> + set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
> + info.max_cpu_index = MAX_NR_CPU - 1;
> +
> + ret = xc_topologyinfo(xc_handle, &info);
> + if ( !ret )
> + {
> + uint32_t socket_ids[MAX_NR_CPU];
> + uint32_t core_ids[MAX_NR_CPU];
> + uint32_t socket_nr = 0;
> + uint32_t core_nr = 0;
> +
> + if ( info.max_cpu_index > MAX_NR_CPU - 1 )
> + info.max_cpu_index = MAX_NR_CPU - 1;
> + /* check validity */
> + for ( i = 0; i <= info.max_cpu_index; i++ )
> + {
> + if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
> + cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
> + break;
> + }
> + if ( i > info.max_cpu_index )
> + {
> + /* find socket nr & core nr per socket */
> + for ( i = 0; i <= info.max_cpu_index; i++ )
> + {
> + for ( j = 0; j < socket_nr; j++ )
> + if ( cpu_to_socket[i] == socket_ids[j] )
> + break;
> + if ( j == socket_nr )
> + {
> + socket_ids[j] = cpu_to_socket[i];
> + socket_nr++;
> + }
> +
> + for ( j = 0; j < core_nr; j++ )
> + if ( cpu_to_core[i] == core_ids[j] )
> + break;
> + if ( j == core_nr )
> + {
> + core_ids[j] = cpu_to_core[i];
> + core_nr++;
> + }
> + }
> +
> + /* print out CC? and PC? */
> + for ( i = 0; i < socket_nr; i++ )
> + {
> + uint64_t res;
> + for ( j = 0; j <= info.max_cpu_index; j++ )
> + {
> + if ( cpu_to_socket[j] == socket_ids[i] )
> + break;
> + }
> + printf("Socket %d\n", socket_ids[i]);
> + res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
> + printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
> + 100UL * res / (double)sum_cx[j]);
> + res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
> + printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
> + 100UL * res / (double)sum_cx[j]);
> + res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
> + printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
> + 100UL * res / (double)sum_cx[j]);
> + for ( k = 0; k < core_nr; k++ )
> + {
> + for ( j = 0; j <= info.max_cpu_index; j++ )
> + {
> + if ( cpu_to_socket[j] == socket_ids[i] &&
> + cpu_to_core[j] == core_ids[k] )
> + break;
> + }
> + printf("\t Core %d CPU %d\n", core_ids[k], j);
> + res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
> + printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res /
> 1000000UL,
> + 100UL * res / (double)sum_cx[j]);
> + res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
> + printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res /
> 1000000UL,
> + 100UL * res / (double)sum_cx[j]);
> + printf("\n");
> +
> + }
> }
> }
> printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
> @@ -833,8 +932,6 @@ out:
> fprintf(stderr, "failed to set governor name\n");
> }
>
> -#define MAX_NR_CPU 512
> -
> void cpu_topology_func(int argc, char *argv[])
> {
> uint32_t cpu_to_core[MAX_NR_CPU];
> diff -r 1af6303f103c xen/arch/x86/acpi/cpu_idle.c
> --- a/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 14:12:54 2010 +0800
> +++ b/xen/arch/x86/acpi/cpu_idle.c Mon Jul 12 22:58:25 2010 +0800
> @@ -55,6 +55,14 @@
>
> /*#define DEBUG_PM_CX*/
>
> +#define GET_HW_RES_IN_NS(msr, val) \
> + do { rdmsrl(msr, val); val = tsc_ticks2ns(val); } while( 0 )
> +#define GET_PC3_RES(val) GET_HW_RES_IN_NS(0x3F8, val)
> +#define GET_PC6_RES(val) GET_HW_RES_IN_NS(0x3F9, val)
> +#define GET_PC7_RES(val) GET_HW_RES_IN_NS(0x3FA, val)
> +#define GET_CC3_RES(val) GET_HW_RES_IN_NS(0x3FC, val)
> +#define GET_CC6_RES(val) GET_HW_RES_IN_NS(0x3FD, val)
> +
> static void lapic_timer_nop(void) { }
> static void (*lapic_timer_off)(void);
> static void (*lapic_timer_on)(void);
> @@ -75,6 +83,63 @@ boolean_param("lapic_timer_c2_ok", local
> boolean_param("lapic_timer_c2_ok", local_apic_timer_c2_ok);
>
> static struct acpi_processor_power *__read_mostly processor_powers[NR_CPUS];
> +
> +struct hw_residencies
> +{
> + uint64_t pc3;
> + uint64_t pc6;
> + uint64_t pc7;
> + uint64_t cc3;
> + uint64_t cc6;
> +};
> +
> +static void do_get_hw_residencies(void *arg)
> +{
> + struct cpuinfo_x86 *c = ¤t_cpu_data;
> + struct hw_residencies *hw_res = (struct hw_residencies *)arg;
> +
> + if ( c->x86_vendor != X86_VENDOR_INTEL || c->x86 != 6 )
> + return;
> +
> + switch ( c->x86_model )
> + {
> + /* Nehalem */
> + case 0x1A:
> + case 0x1E:
> + case 0x1F:
> + case 0x2E:
> + /* Westmere */
> + case 0x25:
> + case 0x2C:
> + GET_PC3_RES(hw_res->pc3);
> + GET_PC6_RES(hw_res->pc6);
> + GET_PC7_RES(hw_res->pc7);
> + GET_CC3_RES(hw_res->cc3);
> + GET_CC6_RES(hw_res->cc6);
> + break;
> + }
> +}
> +
> +static void get_hw_residencies(uint32_t cpu, struct hw_residencies *hw_res)
> +{
> + if ( smp_processor_id() == cpu )
> + do_get_hw_residencies((void *)hw_res);
> + else
> + on_selected_cpus(cpumask_of(cpu),
> + do_get_hw_residencies, (void *)hw_res, 1);
> +}
> +
> +static void print_hw_residencies(uint32_t cpu)
> +{
> + struct hw_residencies hw_res = {0};
> +
> + get_hw_residencies(cpu, &hw_res);
> +
> + printk("PC3[%"PRId64"] PC6[%"PRId64"] PC7[%"PRId64"]\n",
> + hw_res.pc3, hw_res.pc6, hw_res.pc7);
> + printk("CC3[%"PRId64"] CC6[%"PRId64"]\n",
> + hw_res.cc3, hw_res.cc6);
> +}
>
> static char* acpi_cstate_method_name[] =
> {
> @@ -113,6 +178,7 @@ static void print_acpi_power(uint32_t cp
> printk(" C0:\tusage[%08d] duration[%"PRId64"]\n",
> idle_usage, NOW() - idle_res);
>
> + print_hw_residencies(cpu);
> }
>
> static void dump_cx(unsigned char key)
> @@ -933,6 +999,7 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
> const struct acpi_processor_power *power = processor_powers[cpuid];
> uint64_t usage, res, idle_usage = 0, idle_res = 0;
> int i;
> + struct hw_residencies hw_res = {0};
>
> if ( power == NULL )
> {
> @@ -965,6 +1032,14 @@ int pmstat_get_cx_stat(uint32_t cpuid, s
> return -EFAULT;
> }
>
> + get_hw_residencies(cpuid, &hw_res);
> +
> + stat->pc3 = hw_res.pc3;
> + stat->pc6 = hw_res.pc6;
> + stat->pc7 = hw_res.pc7;
> + stat->cc3 = hw_res.cc3;
> + stat->cc6 = hw_res.cc6;
> +
> return 0;
> }
>
> diff -r 1af6303f103c xen/arch/x86/time.c
> --- a/xen/arch/x86/time.c Mon Jul 12 14:12:54 2010 +0800
> +++ b/xen/arch/x86/time.c Mon Jul 12 22:16:04 2010 +0800
> @@ -785,6 +785,13 @@ s_time_t get_s_time(void)
> now = t->stime_local_stamp + scale_delta(delta, &t->tsc_scale);
>
> return now;
> +}
> +
> +uint64_t tsc_ticks2ns(uint64_t ticks)
> +{
> + struct cpu_time *t = &this_cpu(cpu_time);
> +
> + return scale_delta(ticks, &t->tsc_scale);
> }
>
> /* Explicitly OR with 1 just in case version number gets out of sync. */
> diff -r 1af6303f103c xen/include/asm-x86/time.h
> --- a/xen/include/asm-x86/time.h Mon Jul 12 14:12:54 2010 +0800
> +++ b/xen/include/asm-x86/time.h Mon Jul 12 22:16:04 2010 +0800
> @@ -56,6 +56,8 @@ uint64_t acpi_pm_tick_to_ns(uint64_t tic
> uint64_t acpi_pm_tick_to_ns(uint64_t ticks);
> uint64_t ns_to_acpi_pm_tick(uint64_t ns);
>
> +uint64_t tsc_ticks2ns(uint64_t ticks);
> +
> void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp);
> u64 gtime_to_gtsc(struct domain *d, u64 tsc);
>
> diff -r 1af6303f103c xen/include/public/sysctl.h
> --- a/xen/include/public/sysctl.h Mon Jul 12 14:12:54 2010 +0800
> +++ b/xen/include/public/sysctl.h Mon Jul 12 22:16:04 2010 +0800
> @@ -223,6 +223,11 @@ struct pm_cx_stat {
> uint64_aligned_t idle_time; /* idle time from boot */
> XEN_GUEST_HANDLE_64(uint64) triggers; /* Cx trigger counts */
> XEN_GUEST_HANDLE_64(uint64) residencies; /* Cx residencies */
> + uint64_aligned_t pc3;
> + uint64_aligned_t pc6;
> + uint64_aligned_t pc7;
> + uint64_aligned_t cc3;
> + uint64_aligned_t cc6;
> };
>
> struct xen_sysctl_get_pmstat {
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: xenpm: provide core/package cstate residencies
2010-07-12 17:34 ` Keir Fraser
@ 2010-07-13 0:36 ` Wei, Gang
2010-07-13 1:09 ` Wei, Gang
2010-07-13 7:31 ` Jan Beulich
2 siblings, 0 replies; 11+ messages in thread
From: Wei, Gang @ 2010-07-13 0:36 UTC (permalink / raw)
To: Keir Fraser, xen-devel@lists.xensource.com; +Cc: Ian Jackson, Wei, Gang
On 2010 07 13 01:35, Keir Fraser wrote:
> I applied the hypervisor component of this patch. I leave it to Ian
> Jackson to deal with the tools part. In future please split patches
> that touch both hypervisor and tools into a patch series in which
> each component patch touches only one or the other.
Sure, thanks.
Jimmy
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: xenpm: provide core/package cstate residencies
2010-07-12 17:34 ` Keir Fraser
2010-07-13 0:36 ` Wei, Gang
@ 2010-07-13 1:09 ` Wei, Gang
2010-07-13 18:42 ` Ian Jackson
2010-07-13 7:31 ` Jan Beulich
2 siblings, 1 reply; 11+ messages in thread
From: Wei, Gang @ 2010-07-13 1:09 UTC (permalink / raw)
To: xen-devel@lists.xensource.com; +Cc: Ian Jackson, Keir Fraser, Wei, Gang
[-- Attachment #1: Type: text/plain, Size: 7586 bytes --]
On 2010 07 13 01:35, Keir Fraser wrote:
> I applied the hypervisor component of this patch. I leave it to Ian
> Jackson to deal with the tools part. In future please split patches
> that touch both hypervisor and tools into a patch series in which
> each component patch touches only one or the other.
Here is the tools part of the previous large patch.
============
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend libxc to support the extended sysctl_get_pmstat, and modify xenpm to output those extended information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 22:16:04 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 22:16:04 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 22:16:04 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
[-- Attachment #2: xenpm-package-cstate-v3.patch --]
[-- Type: application/octet-stream, Size: 7031 bytes --]
xenpm: provide core/package cstate residencies
According to Intel 64 and IA32 Architectures SDM 3B Appendix B, Intel Nehalem/Westmere processors provide h/w MSR to report the core/package cstate residencies.Extend libxc to support the extended sysctl_get_pmstat, and modify xenpm to output those extended information.
Signed-off-by: Wei Gang <gang.wei@intel.com>
diff -r 1af6303f103c tools/libxc/xc_pm.c
--- a/tools/libxc/xc_pm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xc_pm.c Mon Jul 12 22:16:04 2010 +0800
@@ -152,6 +152,11 @@ int xc_pm_get_cxstat(xc_interface *xch,
cxpt->nr = sysctl.u.get_pmstat.u.getcx.nr;
cxpt->last = sysctl.u.get_pmstat.u.getcx.last;
cxpt->idle_time = sysctl.u.get_pmstat.u.getcx.idle_time;
+ cxpt->pc3 = sysctl.u.get_pmstat.u.getcx.pc3;
+ cxpt->pc6 = sysctl.u.get_pmstat.u.getcx.pc6;
+ cxpt->pc7 = sysctl.u.get_pmstat.u.getcx.pc7;
+ cxpt->cc3 = sysctl.u.get_pmstat.u.getcx.cc3;
+ cxpt->cc6 = sysctl.u.get_pmstat.u.getcx.cc6;
unlock_3:
unlock_pages(cxpt->residencies, max_cx * sizeof(uint64_t));
diff -r 1af6303f103c tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/libxc/xenctrl.h Mon Jul 12 22:16:04 2010 +0800
@@ -1393,6 +1393,11 @@ struct xc_cx_stat {
uint64_t idle_time; /* idle time from boot */
uint64_t *triggers; /* Cx trigger counts */
uint64_t *residencies; /* Cx residencies */
+ uint64_t pc3;
+ uint64_t pc6;
+ uint64_t pc7;
+ uint64_t cc3;
+ uint64_t cc6;
};
typedef struct xc_cx_stat xc_cx_stat_t;
diff -r 1af6303f103c tools/misc/xenpm.c
--- a/tools/misc/xenpm.c Mon Jul 12 14:12:54 2010 +0800
+++ b/tools/misc/xenpm.c Mon Jul 12 22:16:04 2010 +0800
@@ -15,6 +15,7 @@
* this program; if not, write to the Free Software Foundation, Inc., 59 Temple
* Place - Suite 330, Boston, MA 02111-1307 USA.
*/
+#define MAX_NR_CPU 512
#include <stdio.h>
#include <stdlib.h>
@@ -91,6 +92,13 @@ static void print_cxstat(int cpuid, stru
printf(" residency [%020"PRIu64" ms]\n",
cxstat->residencies[i]/1000000UL);
}
+ printf("pc3 : [%020"PRIu64" ms]\n"
+ "pc6 : [%020"PRIu64" ms]\n"
+ "pc7 : [%020"PRIu64" ms]\n",
+ cxstat->pc3/1000000UL, cxstat->pc6/1000000UL, cxstat->pc7/1000000UL);
+ printf("cc3 : [%020"PRIu64" ms]\n"
+ "cc6 : [%020"PRIu64" ms]\n",
+ cxstat->cc3/1000000UL, cxstat->cc6/1000000UL);
printf("\n");
}
@@ -306,9 +314,13 @@ static uint64_t *sum, *sum_cx, *sum_px;
static void signal_int_handler(int signo)
{
- int i, j;
+ int i, j, k, ret;
struct timeval tv;
int cx_cap = 0, px_cap = 0;
+ uint32_t cpu_to_core[MAX_NR_CPU];
+ uint32_t cpu_to_socket[MAX_NR_CPU];
+ uint32_t cpu_to_node[MAX_NR_CPU];
+ xc_topologyinfo_t info = { 0 };
if ( gettimeofday(&tv, NULL) == -1 )
{
@@ -369,6 +381,93 @@ static void signal_int_handler(int signo
pxstat_start[i].pt[j].residency;
printf(" P%d\t%"PRIu64"\t(%5.2f%%)\n", j,
res / 1000000UL, 100UL * res / (double)sum_px[i]);
+ }
+ }
+ }
+
+ set_xen_guest_handle(info.cpu_to_core, cpu_to_core);
+ set_xen_guest_handle(info.cpu_to_socket, cpu_to_socket);
+ set_xen_guest_handle(info.cpu_to_node, cpu_to_node);
+ info.max_cpu_index = MAX_NR_CPU - 1;
+
+ ret = xc_topologyinfo(xc_handle, &info);
+ if ( !ret )
+ {
+ uint32_t socket_ids[MAX_NR_CPU];
+ uint32_t core_ids[MAX_NR_CPU];
+ uint32_t socket_nr = 0;
+ uint32_t core_nr = 0;
+
+ if ( info.max_cpu_index > MAX_NR_CPU - 1 )
+ info.max_cpu_index = MAX_NR_CPU - 1;
+ /* check validity */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ if ( cpu_to_core[i] == INVALID_TOPOLOGY_ID ||
+ cpu_to_socket[i] == INVALID_TOPOLOGY_ID )
+ break;
+ }
+ if ( i > info.max_cpu_index )
+ {
+ /* find socket nr & core nr per socket */
+ for ( i = 0; i <= info.max_cpu_index; i++ )
+ {
+ for ( j = 0; j < socket_nr; j++ )
+ if ( cpu_to_socket[i] == socket_ids[j] )
+ break;
+ if ( j == socket_nr )
+ {
+ socket_ids[j] = cpu_to_socket[i];
+ socket_nr++;
+ }
+
+ for ( j = 0; j < core_nr; j++ )
+ if ( cpu_to_core[i] == core_ids[j] )
+ break;
+ if ( j == core_nr )
+ {
+ core_ids[j] = cpu_to_core[i];
+ core_nr++;
+ }
+ }
+
+ /* print out CC? and PC? */
+ for ( i = 0; i < socket_nr; i++ )
+ {
+ uint64_t res;
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] )
+ break;
+ }
+ printf("Socket %d\n", socket_ids[i]);
+ res = cxstat_end[j].pc3 - cxstat_start[j].pc3;
+ printf("\tPC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc6 - cxstat_start[j].pc6;
+ printf("\tPC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].pc7 - cxstat_start[j].pc7;
+ printf("\tPC7\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ for ( k = 0; k < core_nr; k++ )
+ {
+ for ( j = 0; j <= info.max_cpu_index; j++ )
+ {
+ if ( cpu_to_socket[j] == socket_ids[i] &&
+ cpu_to_core[j] == core_ids[k] )
+ break;
+ }
+ printf("\t Core %d CPU %d\n", core_ids[k], j);
+ res = cxstat_end[j].cc3 - cxstat_start[j].cc3;
+ printf("\t\tCC3\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ res = cxstat_end[j].cc6 - cxstat_start[j].cc6;
+ printf("\t\tCC6\t%"PRIu64" ms\t%.2f%%\n", res / 1000000UL,
+ 100UL * res / (double)sum_cx[j]);
+ printf("\n");
+
+ }
}
}
printf(" Avg freq\t%d\tKHz\n", avgfreq[i]);
@@ -833,8 +932,6 @@ out:
fprintf(stderr, "failed to set governor name\n");
}
-#define MAX_NR_CPU 512
-
void cpu_topology_func(int argc, char *argv[])
{
uint32_t cpu_to_core[MAX_NR_CPU];
[-- Attachment #3: Type: text/plain, Size: 138 bytes --]
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: xenpm: provide core/package cstate residencies
2010-07-12 17:34 ` Keir Fraser
2010-07-13 0:36 ` Wei, Gang
2010-07-13 1:09 ` Wei, Gang
@ 2010-07-13 7:31 ` Jan Beulich
2010-07-13 7:55 ` Keir Fraser
2 siblings, 1 reply; 11+ messages in thread
From: Jan Beulich @ 2010-07-13 7:31 UTC (permalink / raw)
To: Keir Fraser; +Cc: xen-devel@lists.xensource.com, Ian Jackson
>>> On 12.07.10 at 19:34, Keir Fraser <keir.fraser@eu.citrix.com> wrote:
> I applied the hypervisor component of this patch. I leave it to Ian Jackson
> to deal with the tools part. In future please split patches that touch both
> hypervisor and tools into a patch series in which each component patch
> touches only one or the other.
Hmm, is that really a good mechanism, especially when a change
modifies the hypervisor <-> tools interface in an incompatible way?
The resulting inconsistency, besides being bad by itself, would likely
make bisecting more difficult.
Jan
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Re: xenpm: provide core/package cstate residencies
2010-07-13 7:31 ` Jan Beulich
@ 2010-07-13 7:55 ` Keir Fraser
0 siblings, 0 replies; 11+ messages in thread
From: Keir Fraser @ 2010-07-13 7:55 UTC (permalink / raw)
To: Jan Beulich; +Cc: xen-devel@lists.xensource.com, Ian Jackson
On 13/07/2010 08:31, "Jan Beulich" <JBeulich@novell.com> wrote:
>>>> On 12.07.10 at 19:34, Keir Fraser <keir.fraser@eu.citrix.com> wrote:
>> I applied the hypervisor component of this patch. I leave it to Ian Jackson
>> to deal with the tools part. In future please split patches that touch both
>> hypervisor and tools into a patch series in which each component patch
>> touches only one or the other.
>
> Hmm, is that really a good mechanism, especially when a change
> modifies the hypervisor <-> tools interface in an incompatible way?
> The resulting inconsistency, besides being bad by itself, would likely
> make bisecting more difficult.
In this case the change isn't an API, or even an ABI, breakage. A consistent
build of hypervisor and tools will work, and continue to work, across the
hypervisor-only changeset.
I think there will rarely be interface-breaking patches. If there are we
might have to consider them as one unit. We'll see. A patch series will
usually be the correct way to go however.
-- Keir
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: xenpm: provide core/package cstate residencies
2010-07-13 1:09 ` Wei, Gang
@ 2010-07-13 18:42 ` Ian Jackson
2010-07-14 0:17 ` Wei, Gang
0 siblings, 1 reply; 11+ messages in thread
From: Ian Jackson @ 2010-07-13 18:42 UTC (permalink / raw)
To: Wei, Gang; +Cc: Keir, xen-devel@lists.xensource.com, Fraser
Wei, Gang writes ("[Xen-devel] RE: xenpm: provide core/package cstate residencies"):
> On 2010 07 13 01:35, Keir Fraser wrote:
> > I applied the hypervisor component of this patch. I leave it to Ian
> > Jackson to deal with the tools part. In future please split patches
> > that touch both hypervisor and tools into a patch series in which
> > each component patch touches only one or the other.
>
> Here is the tools part of the previous large patch.
Thanks.
Can I ask why these new performance counters are just named with
numbers (pc3, pc6, pc7, cc3, cc6) ? Is this normal ? Should they be
given more expansive names in the output from xenpm ?
Also, I just went and looked at xenpm.c since you were adding a lot of
code to the SIGINT handler, and it seems to me that the code is signal
unsafe. It calls printf both from the main program and from the
signal handler, and if you hit ^C while it's starting up it will
probably crash in stdio. This is of course not your fault and not a
reason not to apply your patch.
Also, why loop again over the CPU topology, rather than doing the work
in cpu_topology_func ?
Sorry to ask lots of questions but I'm not very familiar with the
xenpm code and how people use it.
Thanks,
Ian.
^ permalink raw reply [flat|nested] 11+ messages in thread
* RE: RE: xenpm: provide core/package cstate residencies
2010-07-13 18:42 ` Ian Jackson
@ 2010-07-14 0:17 ` Wei, Gang
0 siblings, 0 replies; 11+ messages in thread
From: Wei, Gang @ 2010-07-14 0:17 UTC (permalink / raw)
To: Ian Jackson; +Cc: Keir, xen-devel@lists.xensource.com, Fraser, Wei, Gang
On 2010 07 14 02:43, Ian Jackson wrote:
> Can I ask why these new performance counters are just named with
> numbers (pc3, pc6, pc7, cc3, cc6) ? Is this normal ? Should they be
> given more expansive names in the output from xenpm ?
pc stands for package cstate, cc stands for core cstate, it should be succinct. Peopole caring about those numbers should understanding them correctly. More expansive names in the xenpm output are not really needed.
> Also, I just went and looked at xenpm.c since you were adding a lot of
> code to the SIGINT handler, and it seems to me that the code is signal
> unsafe. It calls printf both from the main program and from the
> signal handler, and if you hit ^C while it's starting up it will
> probably crash in stdio. This is of course not your fault and not a
> reason not to apply your patch.
Let's discuss how to improve it. Maybe we can move those printf & loops in SIGINT to main program and use a flag instead in SIGINT to control. But anyway, let's do another incremental patch for this.
> Also, why loop again over the CPU topology, rather than doing the work
> in cpu_topology_func ?
cpu_topology_func just outputs the topology got via hypercall. But here we want to calculate the cstate residencs for every core/package in a period, it is not suitable to add this into cpu_topology_func.
> Sorry to ask lots of questions but I'm not very familiar with the
> xenpm code and how people use it.
Thanks for review it carefully. :-)
Jimmy
^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2010-07-14 0:17 UTC | newest]
Thread overview: 11+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2010-07-12 11:03 xenpm: provide core/package cstate residencies Wei, Gang
2010-07-12 11:26 ` Jan Beulich
2010-07-12 15:11 ` Wei, Gang
2010-07-12 15:22 ` Wei, Gang
2010-07-12 17:34 ` Keir Fraser
2010-07-13 0:36 ` Wei, Gang
2010-07-13 1:09 ` Wei, Gang
2010-07-13 18:42 ` Ian Jackson
2010-07-14 0:17 ` Wei, Gang
2010-07-13 7:31 ` Jan Beulich
2010-07-13 7:55 ` Keir Fraser
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).