* [PATCH] [PATCH] kvm: powerpc: add exit timing statistics
@ 2008-11-04 12:33 Ehrhardt Christian
2008-11-04 12:42 ` Christian Ehrhardt
2008-11-07 4:58 ` Hollis Blanchard
0 siblings, 2 replies; 3+ messages in thread
From: Ehrhardt Christian @ 2008-11-04 12:33 UTC (permalink / raw)
To: kvm-ppc
From: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Other existing kvm stats are either just counters (kvm_stat) reported for kvm
generally or trace based aproaches like kvm_trace.
For kvm on powerpc we had the need to track the timings of the different exit
types. While this could be achieved parsing data created with a kvm_trace
extension this adds too muhc overhead (at least on embedded powerpc) slowing
down the workloads we wanted to measure.
Therefore this patch adds a in kernel exit timing statistic to the powerpc kvm
code. These statistic is available per vm&vcpu under the kvm debugfs directory.
As this statistic is low, but still some overhead it can be enabled via a
.config entry and should be off by default.
Since this patch touched all powerpc kvm_stat code anyway this code is now
merged and simpliefied together with the exit timing statistic code (still
working with exit timing disabled in .config).
Here is a sample output (after postprocessing with the awk script I'll post in reply to this patch) how the results look like.
sum of time 27504898
MMIO: count 824 min 51 max 555 sum 75825 sum_quad 9232871 avg 92.0206 stddev 52.318 % 0.28
DCR: count 140 min 44 max 92 sum 6746 sum_quad 327658 avg 48.1857 stddev 4.307 % 0.02
SIGNAL: count 2 min 309 max 993 sum 1302 sum_quad 1081530 avg 651.0000 stddev 342.000 % 0.00
ITLBREAL: count 293 min 11 max 14 sum 3515 sum_quad 42175 avg 11.9966 stddev 0.155 % 0.01
ITLBVIRT: count 113822 min 20 max 338 sum 2595412 sum_quad 60256824 avg 22.8024 stddev 3.074 % 9.44
DTLBREAL: count 242 min 11 max 14 sum 2908 sum_quad 34974 avg 12.0165 stddev 0.352 % 0.01
DTLBVIRT: count 66687 min 21 max 329 sum 1530048 sum_quad 35434926 avg 22.9437 stddev 2.224 % 5.56
SYSCALL: count 72 min 9 max 10 sum 649 sum_quad 5851 avg 9.0139 stddev 0.117 % 0.00
ISI: count 56 min 9 max 10 sum 506 sum_quad 4574 avg 9.0357 stddev 0.186 % 0.00
DSI: count 49 min 9 max 10 sum 448 sum_quad 4102 avg 9.1429 stddev 0.350 % 0.00
EMULINST: count 211220 min 7 max 7700 sum 3292984 sum_quad 5797023806 avg 15.5903 stddev 164.931 % 11.97
DEC: count 6582 min 49 max 322 sum 367567 sum_quad 22996737 avg 55.8443 stddev 19.373 % 1.34
EXTINT: count 4 min 79 max 513 sum 797 sum_quad 290423 avg 199.2500 stddev 181.398 % 0.00
TIMEINGUEST: count 399993 min 0 max 3952 sum 19626191 sum_quad 61148587807 avg 49.0663 stddev 387.900 % 71.36
Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
---
[diffstat]
arch/powerpc/include/asm/kvm_host.h | 49 ++++++++
arch/powerpc/include/asm/kvm_timing_stats.h | 142 ++++++++++++++++++++++++
arch/powerpc/include/asm/mmu-44x.h | 1
arch/powerpc/kernel/asm-offsets.c | 11 +
arch/powerpc/kvm/Kconfig | 9 +
arch/powerpc/kvm/booke.c | 35 +++---
arch/powerpc/kvm/booke.h | 7 +
arch/powerpc/kvm/booke_interrupts.S | 24 ++++
arch/powerpc/kvm/powerpc.c | 163 +++++++++++++++++++++++++++-
9 files changed, 425 insertions(+), 16 deletions(-)
[diff]
diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
--- a/arch/powerpc/include/asm/kvm_host.h
+++ b/arch/powerpc/include/asm/kvm_host.h
@@ -71,7 +71,43 @@
u32 word2;
};
+enum kvm_exit_types {
+ MMIO_EXITS,
+ DCR_EXITS,
+ SIGNAL_EXITS,
+ ITLB_REAL_MISS_EXITS,
+ ITLB_VIRT_MISS_EXITS,
+ DTLB_REAL_MISS_EXITS,
+ DTLB_VIRT_MISS_EXITS,
+ SYSCALL_EXITS,
+ ISI_EXITS,
+ DSI_EXITS,
+ EMULATED_INST_EXITS,
+ EMULATED_MTMSRWE_EXITS,
+ DEC_EXITS,
+ EXT_INTR_EXITS,
+ HALT_WAKEUP,
+ USR_PR_INST,
+ FP_UNAVAIL,
+ DEBUG_EXITS,
+ TIMEINGUEST,
+ __NUMBER_OF_KVM_EXIT_TYPES
+};
+
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+/* allow access to big endian 32bit upper/lower parts and 64bit var */
+typedef union {
+ u64 tv64;
+ struct {
+ u32 tbu, tbl;
+ } tv32;
+} exit_timing_t;
+#endif
+
struct kvm_arch {
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ unsigned int vm_id;
+#endif
};
struct kvm_vcpu_arch {
@@ -130,6 +166,19 @@
u32 dbcr0;
u32 dbcr1;
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ exit_timing_t timing_exit;
+ exit_timing_t timing_last_enter;
+ u32 last_exit_type;
+ u32 timing_count_type[__NUMBER_OF_KVM_EXIT_TYPES];
+ u64 timing_sum_duration[__NUMBER_OF_KVM_EXIT_TYPES];
+ u64 timing_sum_quad_duration[__NUMBER_OF_KVM_EXIT_TYPES];
+ u64 timing_min_duration[__NUMBER_OF_KVM_EXIT_TYPES];
+ u64 timing_max_duration[__NUMBER_OF_KVM_EXIT_TYPES];
+ u64 timing_last_exit;
+ struct dentry *debugfs_exit_timing;
+#endif
+
u32 last_inst;
ulong fault_dear;
ulong fault_esr;
diff --git a/arch/powerpc/include/asm/kvm_timing_stats.h b/arch/powerpc/include/asm/kvm_timing_stats.h
new file mode 100644
--- /dev/null
+++ b/arch/powerpc/include/asm/kvm_timing_stats.h
@@ -0,0 +1,168 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * Copyright IBM Corp. 2008
+ *
+ * Authors: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
+ */
+
+#ifndef __POWERPC_KVM_EXITTIMING_H__
+#define __POWERPC_KVM_EXITTIMING_H__
+
+#include <linux/kvm_host.h>
+#include <asm/time.h>
+#include <asm-generic/div64.h>
+#include <linux/sysrq.h>
+
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+static inline void init_timing_stats(struct kvm_vcpu *vcpu)
+{
+ int i;
+
+ /* pause guest execution to avoid concurrent updates */
+ local_irq_disable();
+ mutex_lock(&vcpu->mutex);
+
+ vcpu->arch.last_exit_type = 0xDEAD;
+ for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+ vcpu->arch.timing_count_type[i] = 0;
+ vcpu->arch.timing_max_duration[i] = 0;
+ vcpu->arch.timing_min_duration[i] = 0xFFFFFFFF;
+ vcpu->arch.timing_sum_duration[i] = 0;
+ vcpu->arch.timing_sum_quad_duration[i] = 0;
+ }
+ vcpu->arch.timing_last_exit = 0;
+ vcpu->arch.timing_exit.tv64 = 0;
+ vcpu->arch.timing_last_enter.tv64 = 0;
+
+ mutex_unlock(&vcpu->mutex);
+ local_irq_enable();
+}
+
+static inline void add_exit_timing(struct kvm_vcpu *vcpu,
+ u64 duration, int type)
+{
+ u64 old;
+
+ do_div(duration, tb_ticks_per_usec);
+ if (unlikely(duration > 0xFFFFFFFF)) {
+ printk(KERN_ERR"%s - duration too big -> overflow"
+ " duration %lld type %d exit #%d\n",
+ __func__, duration, type,
+ vcpu->arch.timing_count_type[type]);
+ return;
+ }
+
+ vcpu->arch.timing_count_type[type]++;
+
+ /* sum */
+ old = vcpu->arch.timing_sum_duration[type];
+ vcpu->arch.timing_sum_duration[type] += duration;
+ if (unlikely(old > vcpu->arch.timing_sum_duration[type])) {
+ printk(KERN_ERR"%s - wrap adding sum of durations"
+ " old %lld new %lld type %d exit # of type %d\n",
+ __func__, old, vcpu->arch.timing_sum_duration[type],
+ type, vcpu->arch.timing_count_type[type]);
+ }
+
+ /* square sum */
+ old = vcpu->arch.timing_sum_quad_duration[type];
+ vcpu->arch.timing_sum_quad_duration[type] += (duration*duration);
+ if (unlikely(old > vcpu->arch.timing_sum_quad_duration[type])) {
+ printk(KERN_ERR"%s - wrap adding sum of squared durations"
+ " old %lld new %lld type %d exit # of type %d\n",
+ __func__, old,
+ vcpu->arch.timing_sum_quad_duration[type],
+ type, vcpu->arch.timing_count_type[type]);
+ }
+
+ /* set min/max */
+ if (unlikely(duration < vcpu->arch.timing_min_duration[type]))
+ vcpu->arch.timing_min_duration[type] = duration;
+ if (unlikely(duration > vcpu->arch.timing_max_duration[type]))
+ vcpu->arch.timing_max_duration[type] = duration;
+}
+
+static inline void update_timing_stats(struct kvm_vcpu *vcpu)
+{
+ u64 exit = vcpu->arch.timing_last_exit;
+ u64 enter = vcpu->arch.timing_last_enter.tv64;
+
+ /* save exit time, used next exit when the reenter time is known */
+ vcpu->arch.timing_last_exit = vcpu->arch.timing_exit.tv64;
+
+ if (unlikely(vcpu->arch.last_exit_type = 0xDEAD || exit = 0))
+ return; /* skip incomplete cycle (e.g. after reset) */
+
+ /* update statistics for average and standard deviation */
+ add_exit_timing(vcpu, (enter - exit), vcpu->arch.last_exit_type);
+ /* enter -> timing_last_exit is time spent in guest - log this too */
+ add_exit_timing(vcpu, (vcpu->arch.timing_last_exit - enter),
+ TIMEINGUEST);
+}
+#else
+#define init_timing_stats(x) do { } while (0)
+#define update_timing_stats(x) do { } while (0)
+#endif /* CONFIG_KVM_BOOKE_EXIT_TIMING */
+
+static inline void account_exit(struct kvm_vcpu *vcpu, int type)
+{
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ vcpu->arch.last_exit_type = type;
+#endif
+ /* type is usually known at build time */
+ switch (type) {
+ case EXT_INTR_EXITS:
+ vcpu->stat.ext_intr_exits++;
+ break;
+ case DEC_EXITS:
+ vcpu->stat.dec_exits++;
+ break;
+ case EMULATED_INST_EXITS:
+ vcpu->stat.emulated_inst_exits++;
+ break;
+ case DCR_EXITS:
+ vcpu->stat.dcr_exits++;
+ break;
+ case DSI_EXITS:
+ vcpu->stat.dsi_exits++;
+ break;
+ case ISI_EXITS:
+ vcpu->stat.isi_exits++;
+ break;
+ case SYSCALL_EXITS:
+ vcpu->stat.syscall_exits++;
+ break;
+ case DTLB_REAL_MISS_EXITS:
+ vcpu->stat.dtlb_real_miss_exits++;
+ break;
+ case DTLB_VIRT_MISS_EXITS:
+ vcpu->stat.dtlb_virt_miss_exits++;
+ break;
+ case MMIO_EXITS:
+ vcpu->stat.mmio_exits++;
+ break;
+ case ITLB_REAL_MISS_EXITS:
+ vcpu->stat.itlb_real_miss_exits++;
+ break;
+ case ITLB_VIRT_MISS_EXITS:
+ vcpu->stat.itlb_virt_miss_exits++;
+ break;
+ case SIGNAL_EXITS:
+ vcpu->stat.signal_exits++;
+ break;
+ }
+}
+
+#endif /* __POWERPC_KVM_EXITTIMING_H__ */
diff --git a/arch/powerpc/include/asm/mmu-44x.h b/arch/powerpc/include/asm/mmu-44x.h
--- a/arch/powerpc/include/asm/mmu-44x.h
+++ b/arch/powerpc/include/asm/mmu-44x.h
@@ -56,6 +56,7 @@
#ifndef __ASSEMBLY__
extern unsigned int tlb_44x_hwater;
+extern unsigned int tlb_44x_index;
typedef struct {
unsigned long id;
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -387,5 +387,16 @@
DEFINE(PTE_SHIFT, PTE_SHIFT);
#endif
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ DEFINE(VCPU_TIMING_EXIT_TBU, offsetof(struct kvm_vcpu,
+ arch.timing_exit.tv32.tbu));
+ DEFINE(VCPU_TIMING_EXIT_TBL, offsetof(struct kvm_vcpu,
+ arch.timing_exit.tv32.tbl));
+ DEFINE(VCPU_TIMING_LAST_ENTER_TBU, offsetof(struct kvm_vcpu,
+ arch.timing_last_enter.tv32.tbu));
+ DEFINE(VCPU_TIMING_LAST_ENTER_TBL, offsetof(struct kvm_vcpu,
+ arch.timing_last_enter.tv32.tbl));
+#endif
+
return 0;
}
diff --git a/arch/powerpc/kvm/Kconfig b/arch/powerpc/kvm/Kconfig
--- a/arch/powerpc/kvm/Kconfig
+++ b/arch/powerpc/kvm/Kconfig
@@ -34,6 +34,15 @@
---help---
KVM can run unmodified 440 guest kernels on 440 host processors.
+config KVM_BOOKE_EXIT_TIMING
+ bool "Trace detailed exit Timing"
+ depends on KVM && 44x
+ ---help---
+ Inserts code to trace timestamps for every exit/enter cycle. A per vcpu
+ report is available in debugfs kvm/VM_###/VPCU_###_exit_timing.
+ The overhead is relatively small, however it is not recommended for
+ production environments.
+
config KVM_TRACE
bool "KVM trace support"
depends on KVM && MARKERS && SYSFS
diff --git a/arch/powerpc/kvm/booke.c b/arch/powerpc/kvm/booke.c
--- a/arch/powerpc/kvm/booke.c
+++ b/arch/powerpc/kvm/booke.c
@@ -27,6 +27,7 @@
#include <asm/cputable.h>
#include <asm/uaccess.h>
#include <asm/kvm_ppc.h>
+#include <asm/kvm_timing_stats.h>
#include <asm/cacheflush.h>
#include "booke.h"
@@ -183,6 +184,9 @@
enum emulation_result er;
int r = RESUME_HOST;
+ /* update before a new last_exit_type is rewritten */
+ update_timing_stats(vcpu);
+
local_irq_enable();
run->exit_reason = KVM_EXIT_UNKNOWN;
@@ -196,7 +200,7 @@
break;
case BOOKE_INTERRUPT_EXTERNAL:
- vcpu->stat.ext_intr_exits++;
+ account_exit(vcpu, EXT_INTR_EXITS);
if (need_resched())
cond_resched();
r = RESUME_GUEST;
@@ -211,7 +215,7 @@
* we do reschedule the host will fault over it. Perhaps we
* should politely restore the host's entries to minimize
* misses before ceding control. */
- vcpu->stat.dec_exits++;
+ account_exit(vcpu, DEC_EXITS);
if (need_resched())
cond_resched();
r = RESUME_GUEST;
@@ -224,6 +228,7 @@
vcpu->arch.esr = vcpu->arch.fault_esr;
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_PROGRAM);
r = RESUME_GUEST;
+ account_exit(vcpu, USR_PR_INST);
break;
}
@@ -232,12 +237,12 @@
case EMULATE_DONE:
/* Future optimization: only reload non-volatiles if
* they were actually modified by emulation. */
- vcpu->stat.emulated_inst_exits++;
+ account_exit(vcpu, EMULATED_INST_EXITS);
r = RESUME_GUEST_NV;
break;
case EMULATE_DO_DCR:
run->exit_reason = KVM_EXIT_DCR;
- vcpu->stat.dcr_exits++;
+ account_exit(vcpu, DCR_EXITS);
r = RESUME_HOST;
break;
case EMULATE_FAIL:
@@ -257,6 +262,7 @@
case BOOKE_INTERRUPT_FP_UNAVAIL:
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_FP_UNAVAIL);
+ account_exit(vcpu, FP_UNAVAIL);
r = RESUME_GUEST;
break;
@@ -264,20 +270,20 @@
vcpu->arch.dear = vcpu->arch.fault_dear;
vcpu->arch.esr = vcpu->arch.fault_esr;
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DATA_STORAGE);
- vcpu->stat.dsi_exits++;
+ account_exit(vcpu, DSI_EXITS);
r = RESUME_GUEST;
break;
case BOOKE_INTERRUPT_INST_STORAGE:
vcpu->arch.esr = vcpu->arch.fault_esr;
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_INST_STORAGE);
- vcpu->stat.isi_exits++;
+ account_exit(vcpu, ISI_EXITS);
r = RESUME_GUEST;
break;
case BOOKE_INTERRUPT_SYSCALL:
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_SYSCALL);
- vcpu->stat.syscall_exits++;
+ account_exit(vcpu, SYSCALL_EXITS);
r = RESUME_GUEST;
break;
@@ -293,7 +299,7 @@
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_DTLB_MISS);
vcpu->arch.dear = vcpu->arch.fault_dear;
vcpu->arch.esr = vcpu->arch.fault_esr;
- vcpu->stat.dtlb_real_miss_exits++;
+ account_exit(vcpu, DTLB_REAL_MISS_EXITS);
r = RESUME_GUEST;
break;
}
@@ -310,13 +316,13 @@
* invoking the guest. */
kvmppc_mmu_map(vcpu, eaddr, gfn, gtlbe->tid,
gtlbe->word2);
- vcpu->stat.dtlb_virt_miss_exits++;
+ account_exit(vcpu, DTLB_VIRT_MISS_EXITS);
r = RESUME_GUEST;
} else {
/* Guest has mapped and accessed a page which is not
* actually RAM. */
r = kvmppc_emulate_mmio(run, vcpu);
- vcpu->stat.mmio_exits++;
+ account_exit(vcpu, MMIO_EXITS);
}
break;
@@ -334,11 +340,11 @@
if (!gtlbe) {
/* The guest didn't have a mapping for it. */
kvmppc_booke_queue_irqprio(vcpu, BOOKE_IRQPRIO_ITLB_MISS);
- vcpu->stat.itlb_real_miss_exits++;
+ account_exit(vcpu, ITLB_REAL_MISS_EXITS);
break;
}
- vcpu->stat.itlb_virt_miss_exits++;
+ account_exit(vcpu, ITLB_VIRT_MISS_EXITS);
gfn = tlb_xlate(gtlbe, eaddr) >> PAGE_SHIFT;
@@ -370,6 +376,7 @@
mtspr(SPRN_DBSR, dbsr);
run->exit_reason = KVM_EXIT_DEBUG;
+ account_exit(vcpu, DEBUG_EXITS);
r = RESUME_HOST;
break;
}
@@ -390,7 +397,7 @@
if (signal_pending(current)) {
run->exit_reason = KVM_EXIT_INTR;
r = (-EINTR << 2) | RESUME_HOST | (r & RESUME_FLAG_NV);
- vcpu->stat.signal_exits++;
+ account_exit(vcpu, SIGNAL_EXITS);
}
}
@@ -409,6 +416,8 @@
/* Eye-catching number so we know if the guest takes an interrupt
* before it's programmed its own IVPR. */
vcpu->arch.ivpr = 0x55550000;
+
+ init_timing_stats(vcpu);
return kvmppc_core_vcpu_setup(vcpu);
}
diff --git a/arch/powerpc/kvm/booke.h b/arch/powerpc/kvm/booke.h
--- a/arch/powerpc/kvm/booke.h
+++ b/arch/powerpc/kvm/booke.h
@@ -22,6 +22,7 @@
#include <linux/types.h>
#include <linux/kvm_host.h>
+#include <asm/kvm_timing_stats.h>
/* interrupt priortity ordering */
#define BOOKE_IRQPRIO_DATA_STORAGE 0
@@ -50,8 +51,12 @@
vcpu->arch.msr = new_msr;
- if (vcpu->arch.msr & MSR_WE)
+ if (vcpu->arch.msr & MSR_WE) {
+ /* whatever exit it was before, this would clobber its timings
+ * count MSR_WE wait's in seperate exit category */
+ account_exit(vcpu, EMULATED_MTMSRWE_EXITS);
kvm_vcpu_block(vcpu);
+ }
}
#endif /* __KVM_BOOKE_H__ */
diff --git a/arch/powerpc/kvm/booke_interrupts.S b/arch/powerpc/kvm/booke_interrupts.S
--- a/arch/powerpc/kvm/booke_interrupts.S
+++ b/arch/powerpc/kvm/booke_interrupts.S
@@ -106,6 +106,18 @@
li r6, 1
slw r6, r6, r5
+
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ /* save exit time */
+..exit_tbu_overflow_loop:
+ mfspr r7, SPRN_TBRU
+ mfspr r8, SPRN_TBRL
+ mfspr r9, SPRN_TBRU
+ cmpw r9, r7
+ bne ..exit_tbu_overflow_loop
+ stw r8, VCPU_TIMING_EXIT_TBL(r4)
+ stw r9, VCPU_TIMING_EXIT_TBU(r4)
+#endif
/* Save the faulting instruction and all GPRs for emulation. */
andi. r7, r6, NEED_INST_MASK
@@ -423,6 +435,18 @@
lwz r3, VCPU_SPRG7(r4)
mtspr SPRN_SPRG7, r3
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ /* save enter time */
+..enter_tbu_overflow_loop:
+ mfspr r6, SPRN_TBRU
+ mfspr r7, SPRN_TBRL
+ mfspr r8, SPRN_TBRU
+ cmpw r8, r6
+ bne ..enter_tbu_overflow_loop
+ stw r7, VCPU_TIMING_LAST_ENTER_TBL(r4)
+ stw r8, VCPU_TIMING_LAST_ENTER_TBU(r4)
+#endif
+
/* Finish loading guest volatiles and jump to guest. */
lwz r3, VCPU_CTR(r4)
mtctr r3
diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
--- a/arch/powerpc/kvm/powerpc.c
+++ b/arch/powerpc/kvm/powerpc.c
@@ -28,7 +28,10 @@
#include <asm/uaccess.h>
#include <asm/kvm_ppc.h>
#include <asm/tlbflush.h>
-
+#include <asm/kvm_timing_stats.h>
+#include <asm/atomic.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
{
@@ -101,6 +104,153 @@
*(int *)rtn = kvmppc_core_check_processor_compat();
}
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+static const char *kvm_exit_names[__NUMBER_OF_KVM_EXIT_TYPES] = {
+ "MMIO",
+ "DCR",
+ "SIGNAL",
+ "ITLBREAL",
+ "ITLBVIRT",
+ "DTLBREAL",
+ "DTLBVIRT",
+ "SYSCALL",
+ "ISI",
+ "DSI",
+ "EMULINST",
+ "EMUL_MSR_WE",
+ "DEC",
+ "EXTINT",
+ "HALT",
+ "USR_PR_INST",
+ "FP_UNAVAIL",
+ "DEBUG",
+ "TIMEINGUEST"
+};
+
+/* assign a unique number to each vm created */
+static atomic_t vm_count = ATOMIC_INIT(0);
+
+static int kvmppc_exit_timing_show(struct seq_file *m, void *private)
+{
+ struct kvm_vcpu *vcpu = m->private;
+ int i;
+ u64 min, max;
+
+ for (i = 0; i < __NUMBER_OF_KVM_EXIT_TYPES; i++) {
+ if (vcpu->arch.timing_min_duration[i] = 0xFFFFFFFF)
+ min = 0;
+ else
+ min = vcpu->arch.timing_min_duration[i];
+ if (vcpu->arch.timing_max_duration[i] = 0)
+ max = 0;
+ else
+ max = vcpu->arch.timing_max_duration[i];
+
+ seq_printf(m, "%12s: count %10d min %10lld "
+ "max %10lld sum %20lld sum_quad %20lld\n",
+ kvm_exit_names[i], vcpu->arch.timing_count_type[i],
+ vcpu->arch.timing_min_duration[i],
+ vcpu->arch.timing_max_duration[i],
+ vcpu->arch.timing_sum_duration[i],
+ vcpu->arch.timing_sum_quad_duration[i]);
+ }
+ return 0;
+}
+
+static ssize_t kvmppc_exit_timing_write(struct file *file,
+ const char __user *user_buf,
+ size_t count, loff_t *ppos)
+{
+ size_t len;
+ int err;
+ const char __user *p;
+ char c;
+
+ len = 0;
+ p = user_buf;
+ while (len < count) {
+ if (get_user(c, p++))
+ err = -EFAULT;
+ if (c = 0 || c = '\n')
+ break;
+ len++;
+ }
+
+ if (len > 1) {
+ err = -EINVAL;
+ goto done;
+ }
+
+ if (copy_from_user(&c, user_buf, sizeof(c))) {
+ err = -EFAULT;
+ goto done;
+ }
+
+ if (c = 'c') {
+ struct seq_file *seqf = (struct seq_file *)file->private_data;
+ struct kvm_vcpu *vcpu = seqf->private;
+ /* write does not affect out buffers previsously generated with
+ * show. Seq file is locked here to prevent races of init with
+ * a show call */
+ mutex_lock(&seqf->lock);
+ init_timing_stats(vcpu);
+ mutex_unlock(&seqf->lock);
+ err = count;
+ } else {
+ err = -EINVAL;
+ goto done;
+ }
+
+done:
+ return err;
+}
+
+static int kvmppc_exit_timing_open(struct inode *inode, struct file *file)
+{
+ return single_open(file, kvmppc_exit_timing_show, inode->i_private);
+}
+
+static struct file_operations kvmppc_exit_timing_fops = {
+ .owner = THIS_MODULE,
+ .open = kvmppc_exit_timing_open,
+ .read = seq_read,
+ .write = kvmppc_exit_timing_write,
+ .llseek = seq_lseek,
+ .release = single_release,
+};
+
+static void kvmppc_create_vcpu_debugfs(struct kvm_vcpu *vcpu, unsigned int id)
+{
+ static char dbg_fname[28];
+ struct dentry *debugfs_file;
+
+ snprintf(dbg_fname, sizeof(dbg_fname), "VM_%03u_VCPU_%03u_exit_timing",
+ vcpu->kvm->arch.vm_id, id);
+ debugfs_file = debugfs_create_file(dbg_fname, 0666,
+ kvm_debugfs_dir, vcpu,
+ &kvmppc_exit_timing_fops);
+
+ if (!debugfs_file) {
+ printk(KERN_ERR"%s: error creating debugfs file %s\n",
+ __func__, dbg_fname);
+ return;
+ }
+
+ vcpu->arch.debugfs_exit_timing = debugfs_file;
+}
+
+static void kvmppc_remove_vcpu_debugfs(struct kvm_vcpu *vcpu)
+{
+ if (vcpu->arch.debugfs_exit_timing) {
+ debugfs_remove(vcpu->arch.debugfs_exit_timing);
+ vcpu->arch.debugfs_exit_timing = NULL;
+ }
+}
+#else
+#define kvmppc_create_vcpu_debugfs(x, y) do { } while (0)
+#define kvmppc_remove_vcpu_debugfs(x) do { } while (0)
+#endif /* CONFIG_KVM_BOOKE_EXIT_TIMING */
+
struct kvm *kvm_arch_create_vm(void)
{
struct kvm *kvm;
@@ -108,6 +258,10 @@
kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
if (!kvm)
return ERR_PTR(-ENOMEM);
+
+#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
+ kvm->arch.vm_id = atomic_inc_return(&vm_count);
+#endif
return kvm;
}
@@ -170,11 +324,15 @@
struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
{
- return kvmppc_core_vcpu_create(kvm, id);
+ struct kvm_vcpu *vcpu;
+ vcpu = kvmppc_core_vcpu_create(kvm, id);
+ kvmppc_create_vcpu_debugfs(vcpu, id);
+ return vcpu;
}
void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
{
+ kvmppc_remove_vcpu_debugfs(vcpu);
kvmppc_core_vcpu_free(vcpu);
}
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] [PATCH] kvm: powerpc: add exit timing statistics
2008-11-04 12:33 [PATCH] [PATCH] kvm: powerpc: add exit timing statistics Ehrhardt Christian
@ 2008-11-04 12:42 ` Christian Ehrhardt
2008-11-07 4:58 ` Hollis Blanchard
1 sibling, 0 replies; 3+ messages in thread
From: Christian Ehrhardt @ 2008-11-04 12:42 UTC (permalink / raw)
To: kvm-ppc
This patch reports the square sum in addition to the classic sum/count
(+min/max) to allow more details statistical post processing.
That way you can process average, % of whole duration and using the
square sum also the standard deviation.
A simple way to get these numbers with awk is:
#!/bin/bash
echo "processing file ${1}"
sumsum=`awk 'BEGIN { csum=0; sumsum=0; count=0}{ csum=csum+$3; count++;
sumsum = sumsum+$9;}END { printf("%s\n", sumsum)}' $1`
echo "sum of time ${sumsum}"
awk '!/count 0/ { printf("%s",$0); printf(" avg %16.4lf stddev
%12.3lf %% %5.2f\n",$9/$3, sqrt($11/$3-(($9/$3)*($9/$3))),
($9/('$sumsum'/100))); }' $1
This runs awk twice, because for % of overall time you need the sum the
durations and I personally don't like all those "rewind hacks" for awk.
Ehrhardt Christian wrote:
> From: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
>
> Other existing kvm stats are either just counters (kvm_stat) reported for kvm
> generally or trace based aproaches like kvm_trace.
> For kvm on powerpc we had the need to track the timings of the different exit
> types. While this could be achieved parsing data created with a kvm_trace
> extension this adds too muhc overhead (at least on embedded powerpc) slowing
> down the workloads we wanted to measure.
>
> Therefore this patch adds a in kernel exit timing statistic to the powerpc kvm
> code. These statistic is available per vm&vcpu under the kvm debugfs directory.
> As this statistic is low, but still some overhead it can be enabled via a
> .config entry and should be off by default.
>
> Since this patch touched all powerpc kvm_stat code anyway this code is now
> merged and simpliefied together with the exit timing statistic code (still
> working with exit timing disabled in .config).
>
> Here is a sample output (after postprocessing with the awk script I'll post in reply to this patch) how the results look like.
> sum of time 27504898
> MMIO: count 824 min 51 max 555 sum 75825 sum_quad 9232871 avg 92.0206 stddev 52.318 % 0.28
> DCR: count 140 min 44 max 92 sum 6746 sum_quad 327658 avg 48.1857 stddev 4.307 % 0.02
> SIGNAL: count 2 min 309 max 993 sum 1302 sum_quad 1081530 avg 651.0000 stddev 342.000 % 0.00
> ITLBREAL: count 293 min 11 max 14 sum 3515 sum_quad 42175 avg 11.9966 stddev 0.155 % 0.01
> ITLBVIRT: count 113822 min 20 max 338 sum 2595412 sum_quad 60256824 avg 22.8024 stddev 3.074 % 9.44
> DTLBREAL: count 242 min 11 max 14 sum 2908 sum_quad 34974 avg 12.0165 stddev 0.352 % 0.01
> DTLBVIRT: count 66687 min 21 max 329 sum 1530048 sum_quad 35434926 avg 22.9437 stddev 2.224 % 5.56
> SYSCALL: count 72 min 9 max 10 sum 649 sum_quad 5851 avg 9.0139 stddev 0.117 % 0.00
> ISI: count 56 min 9 max 10 sum 506 sum_quad 4574 avg 9.0357 stddev 0.186 % 0.00
> DSI: count 49 min 9 max 10 sum 448 sum_quad 4102 avg 9.1429 stddev 0.350 % 0.00
> EMULINST: count 211220 min 7 max 7700 sum 3292984 sum_quad 5797023806 avg 15.5903 stddev 164.931 % 11.97
> DEC: count 6582 min 49 max 322 sum 367567 sum_quad 22996737 avg 55.8443 stddev 19.373 % 1.34
> EXTINT: count 4 min 79 max 513 sum 797 sum_quad 290423 avg 199.2500 stddev 181.398 % 0.00
> TIMEINGUEST: count 399993 min 0 max 3952 sum 19626191 sum_quad 61148587807 avg 49.0663 stddev 387.900 % 71.36
>
> Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
> ---
>
> [diffstat]
> arch/powerpc/include/asm/kvm_host.h | 49 ++++++++
> arch/powerpc/include/asm/kvm_timing_stats.h | 142 ++++++++++++++++++++++++
> arch/powerpc/include/asm/mmu-44x.h | 1
> arch/powerpc/kernel/asm-offsets.c | 11 +
> arch/powerpc/kvm/Kconfig | 9 +
> arch/powerpc/kvm/booke.c | 35 +++---
> arch/powerpc/kvm/booke.h | 7 +
> arch/powerpc/kvm/booke_interrupts.S | 24 ++++
> arch/powerpc/kvm/powerpc.c | 163 +++++++++++++++++++++++++++-
> 9 files changed, 425 insertions(+), 16 deletions(-)
>
> [diff]
>
[...]
--
Grüsse / regards,
Christian Ehrhardt
IBM Linux Technology Center, Open Virtualization
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] [PATCH] kvm: powerpc: add exit timing statistics
2008-11-04 12:33 [PATCH] [PATCH] kvm: powerpc: add exit timing statistics Ehrhardt Christian
2008-11-04 12:42 ` Christian Ehrhardt
@ 2008-11-07 4:58 ` Hollis Blanchard
1 sibling, 0 replies; 3+ messages in thread
From: Hollis Blanchard @ 2008-11-07 4:58 UTC (permalink / raw)
To: kvm-ppc
On Tuesday 04 November 2008 06:33:39 Ehrhardt Christian wrote:
> From: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
>
> Other existing kvm stats are either just counters (kvm_stat) reported for kvm
> generally or trace based aproaches like kvm_trace.
> For kvm on powerpc we had the need to track the timings of the different exit
> types. While this could be achieved parsing data created with a kvm_trace
> extension this adds too muhc overhead (at least on embedded powerpc) slowing
> down the workloads we wanted to measure.
>
> Therefore this patch adds a in kernel exit timing statistic to the powerpc kvm
> code. These statistic is available per vm&vcpu under the kvm debugfs directory.
> As this statistic is low, but still some overhead it can be enabled via a
> .config entry and should be off by default.
It's not just that kvm_trace is "slow" -- I think the important point is that this code is *not* implementing another
trace mechanism. It's simply a set of counters. The reason it doesn't fit naturally into kvm_stat is that
kvm_stat is designed just to record event frequency, so it aggregates the counters from all vcpus. Right?
> Since this patch touched all powerpc kvm_stat code anyway this code is now
> merged and simpliefied together with the exit timing statistic code (still
> working with exit timing disabled in .config).
In general this looks pretty good. I still need to go through it in more detail, but just a couple superficial
comments right now:
> diff --git a/arch/powerpc/include/asm/mmu-44x.h b/arch/powerpc/include/asm/mmu-44x.h
> --- a/arch/powerpc/include/asm/mmu-44x.h
> +++ b/arch/powerpc/include/asm/mmu-44x.h
> @@ -56,6 +56,7 @@
> #ifndef __ASSEMBLY__
>
> extern unsigned int tlb_44x_hwater;
> +extern unsigned int tlb_44x_index;
>
> typedef struct {
> unsigned long id;
This must have accidentally slipped in to your patch.
> @@ -108,6 +258,10 @@
> kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
> if (!kvm)
> return ERR_PTR(-ENOMEM);
> +
> +#ifdef CONFIG_KVM_BOOKE_EXIT_TIMING
> + kvm->arch.vm_id = atomic_inc_return(&vm_count);
> +#endif
>
> return kvm;
> }
This "vm_count" stuff caught my eye as looking strange. Is there no better "ID" value you can use? What
about qemu's PID?
--
Hollis Blanchard
IBM Linux Technology Center
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2008-11-07 4:58 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-11-04 12:33 [PATCH] [PATCH] kvm: powerpc: add exit timing statistics Ehrhardt Christian
2008-11-04 12:42 ` Christian Ehrhardt
2008-11-07 4:58 ` Hollis Blanchard
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.