[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH v2 6/6] xen/PMU: PMU emulation code



Add PMU emulation code that runs when we are processing a PMU interrupt. This 
code
will allow us not to trap to hypervisor on each MSR/LVTPC access (of which there
may be quite a few in the handler).

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
---
 arch/x86/xen/pmu.c             | 209 ++++++++++++++++++++++++++++++++++++-----
 include/xen/interface/xenpmu.h |   1 +
 2 files changed, 187 insertions(+), 23 deletions(-)

diff --git a/arch/x86/xen/pmu.c b/arch/x86/xen/pmu.c
index f92d406..e543629 100644
--- a/arch/x86/xen/pmu.c
+++ b/arch/x86/xen/pmu.c
@@ -13,11 +13,22 @@
 /* x86_pmu.handle_irq definition */
 #include <../kernel/cpu/perf_event.h>
 
+#define XENPMU_IRQ_PROCESSING    1
+struct xenpmu {
+       /* Shared page between hypervisor and domain */
+       struct xen_pmu_data *xenpmu_data;
 
-/* Shared page between hypervisor and domain */
-DEFINE_PER_CPU(struct xen_pmu_data *, xenpmu_shared);
-#define get_xenpmu_data()    per_cpu(xenpmu_shared, smp_processor_id());
+       uint8_t flags;
+};
+DEFINE_PER_CPU(struct xenpmu, xenpmu_shared);
+#define get_xenpmu_data()    ((per_cpu(xenpmu_shared, \
+                                      smp_processor_id())).xenpmu_data)
+#define get_xenpmu_flags()   ((per_cpu(xenpmu_shared, \
+                                      smp_processor_id())).flags)
 
+/* Macro for computing address of a PMU MSR bank */
+#define field_offset(ctxt, field) ((void *)((uintptr_t)ctxt + \
+                                           (uintptr_t)ctxt->field))
 
 /* AMD PMU */
 #define F15H_NUM_COUNTERS   6
@@ -160,18 +171,124 @@ static bool is_intel_pmu_msr(u32 msr_index, int *type, 
int *index)
        return false;
 }
 
-bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
+static bool xen_intel_pmu_emulate(unsigned int msr, u64 *val, int type,
+                                 int index, bool is_read)
+{
+       uint64_t *reg = NULL;
+       struct xen_pmu_intel_ctxt *ctxt;
+       uint64_t *fix_counters;
+       struct xen_pmu_cntr_pair *arch_cntr_pair;
+       struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+       uint8_t xenpmu_flags = get_xenpmu_flags();
+
+
+       if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING))
+               return false;
+
+       ctxt = &xenpmu_data->pmu.intel;
+
+       switch (msr) {
+       case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
+               reg = &ctxt->global_ovf_ctrl;
+               break;
+       case MSR_CORE_PERF_GLOBAL_STATUS:
+               reg = &ctxt->global_status;
+               break;
+       case MSR_CORE_PERF_GLOBAL_CTRL:
+               reg = &ctxt->global_ctrl;
+               break;
+       case MSR_CORE_PERF_FIXED_CTR_CTRL:
+               reg = &ctxt->fixed_ctrl;
+               break;
+       default:
+               switch (type) {
+               case MSR_TYPE_COUNTER:
+                       fix_counters = field_offset(ctxt, fixed_counters);
+                       reg = &fix_counters[index];
+                       break;
+               case MSR_TYPE_ARCH_COUNTER:
+                       arch_cntr_pair = field_offset(ctxt, arch_counters);
+                       reg = &arch_cntr_pair[index].counter;
+                       break;
+               case MSR_TYPE_ARCH_CTRL:
+                       arch_cntr_pair = field_offset(ctxt, arch_counters);
+                       reg = &arch_cntr_pair[index].control;
+                       break;
+               default:
+                       return false;
+               }
+       }
+
+       if (reg) {
+               if (is_read)
+                       *val = *reg;
+               else {
+                       *reg = *val;
+
+                       if (msr == MSR_CORE_PERF_GLOBAL_OVF_CTRL)
+                               ctxt->global_status &= (~(*val));
+               }
+               return true;
+       }
+
+       return false;
+}
+
+static bool xen_amd_pmu_emulate(unsigned int msr, u64 *val, bool is_read)
 {
+       uint64_t *reg = NULL;
+       int i, off = 0;
+       struct xen_pmu_amd_ctxt *ctxt;
+       uint64_t *counter_regs, *ctrl_regs;
+       struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+       uint8_t xenpmu_flags = get_xenpmu_flags();
+
+       if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING))
+               return false;
+
+       if (k7_counters_mirrored &&
+           ((msr >= MSR_K7_EVNTSEL0) && (msr <= MSR_K7_PERFCTR3)))
+               msr = get_fam15h_addr(msr);
 
+       ctxt = &xenpmu_data->pmu.amd;
+       for (i = 0; i < amd_num_counters; i++) {
+               if (msr == amd_ctrls_base + off) {
+                       ctrl_regs = field_offset(ctxt, ctrls);
+                       reg = &ctrl_regs[i];
+                       break;
+               } else if (msr == amd_counters_base + off) {
+                       counter_regs = field_offset(ctxt, counters);
+                       reg = &counter_regs[i];
+                       break;
+               }
+               off += amd_msr_step;
+       }
+
+       if (reg) {
+               if (is_read)
+                       *val = *reg;
+               else
+                       *reg = *val;
+
+               return true;
+       }
+       return false;
+}
+
+bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
+{
        if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
                if (is_amd_pmu_msr(msr)) {
-                       *val = native_read_msr_safe(msr, err);
+                       if (!xen_amd_pmu_emulate(msr, val, 1))
+                               *val = native_read_msr_safe(msr, err);
                        return true;
                }
        } else {
                int type, index;
+
                if (is_intel_pmu_msr(msr, &type, &index)) {
-                       *val = native_read_msr_safe(msr, err);
+                       if (!xen_intel_pmu_emulate(msr, val, type, index, 1))
+                               *val = native_read_msr_safe(msr, err);
                        return true;
                }
        }
@@ -181,16 +298,20 @@ bool pmu_msr_read(unsigned int msr, uint64_t *val, int 
*err)
 
 bool pmu_msr_write(unsigned int msr, uint32_t low, uint32_t high, int *err)
 {
+       uint64_t val = ((uint64_t)high << 32) | low;
+
        if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
                if (is_amd_pmu_msr(msr)) {
-                       *err = native_write_msr_safe(msr, low, high);
+                       if (!xen_amd_pmu_emulate(msr, &val, 0))
+                               *err = native_write_msr_safe(msr, low, high);
                        return true;
                }
        } else {
                int type, index;
 
                if (is_intel_pmu_msr(msr, &type, &index)) {
-                       *err = native_write_msr_safe(msr, low, high);
+                       if (!xen_intel_pmu_emulate(msr, &val, type, index, 0))
+                               *err = native_write_msr_safe(msr, low, high);
                        return true;
                }
        }
@@ -200,24 +321,52 @@ bool pmu_msr_write(unsigned int msr, uint32_t low, 
uint32_t high, int *err)
 
 unsigned long long xen_amd_read_pmc(int counter)
 {
-       uint32_t msr;
-       int err;
+       struct xen_pmu_amd_ctxt *ctxt;
+       uint64_t *counter_regs;
+       struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+       uint8_t xenpmu_flags = get_xenpmu_flags();
+
+       if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING)) {
+               uint32_t msr;
+               int err;
+
+               msr = amd_counters_base + (counter * amd_msr_step);
+               return native_read_msr_safe(msr, &err);
+       }
 
-       msr = amd_counters_base + (counter * amd_msr_step);
-       return native_read_msr_safe(msr, &err);
+       ctxt = &xenpmu_data->pmu.amd;
+       counter_regs = field_offset(ctxt, counters);
+       return counter_regs[counter];
 }
 
 unsigned long long xen_intel_read_pmc(int counter)
 {
-       int err;
-       uint32_t msr;
+       struct xen_pmu_intel_ctxt *ctxt;
+       uint64_t *fixed_counters;
+       struct xen_pmu_cntr_pair *arch_cntr_pair;
+       struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+       uint8_t xenpmu_flags = get_xenpmu_flags();
 
-       if (counter & (1<<30))
-               msr = MSR_CORE_PERF_FIXED_CTR0 + (counter & 0xffff);
-       else
-               msr = MSR_IA32_PERFCTR0 + counter;
+       if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING)) {
+               uint32_t msr;
+               int err;
+
+               if (counter & (1<<30))
+                       msr = MSR_CORE_PERF_FIXED_CTR0 + (counter & 0xffff);
+               else
+                       msr = MSR_IA32_PERFCTR0 + counter;
 
-       return native_read_msr_safe(msr, &err);
+               return native_read_msr_safe(msr, &err);
+       }
+
+       ctxt = &xenpmu_data->pmu.intel;
+       if (counter & (1<<30)) {
+               fixed_counters = field_offset(ctxt, fixed_counters);
+               return fixed_counters[counter & 0xffff];
+       } else {
+               arch_cntr_pair = field_offset(ctxt, arch_counters);
+               return arch_cntr_pair[counter].counter;
+       }
 }
 
 unsigned long long xen_read_pmc(int counter)
@@ -302,24 +451,37 @@ static void xen_convert_regs(struct cpu_user_regs 
*xen_regs,
 irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id)
 {
        int ret = IRQ_NONE;
+       int err;
        struct pt_regs regs;
        struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+       uint8_t xenpmu_flags = get_xenpmu_flags();
 
        if (!xenpmu_data) {
                WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
                return ret;
        }
 
+       per_cpu(xenpmu_shared, smp_processor_id()).flags =
+               xenpmu_flags | XENPMU_IRQ_PROCESSING;
+
        xen_convert_regs(&xenpmu_data->pmu.regs, &regs);
        if (x86_pmu.handle_irq(&regs))
                ret = IRQ_HANDLED;
 
+       /* Write out cached context to HW */
+       err = HYPERVISOR_xenpmu_op(XENPMU_flush, NULL);
+       per_cpu(xenpmu_shared, smp_processor_id()).flags = xenpmu_flags;
+       if (err) {
+               WARN_ONCE(1, "%s failed hypercall, err: %d\n", __func__, err);
+               return IRQ_NONE;
+       }
+
        return ret;
 }
 
 bool is_xen_pmu(int cpu)
 {
-       return (per_cpu(xenpmu_shared, cpu) != NULL);
+       return (get_xenpmu_data() != NULL);
 }
 
 int xen_pmu_init(int cpu)
@@ -345,7 +507,8 @@ int xen_pmu_init(int cpu)
        if (ret)
                goto fail;
 
-       per_cpu(xenpmu_shared, cpu) = xenpmu_data;
+       per_cpu(xenpmu_shared, cpu).xenpmu_data = xenpmu_data;
+       per_cpu(xenpmu_shared, cpu).flags = 0;
 
        if (cpu == 0) {
                perf_register_guest_info_callbacks(&xen_guest_cbs);
@@ -369,6 +532,6 @@ void xen_pmu_finish(int cpu)
 
        (void)HYPERVISOR_xenpmu_op(XENPMU_finish, &xp);
 
-       vfree(per_cpu(xenpmu_shared, cpu));
-       per_cpu(xenpmu_shared, cpu) = NULL;
+       vfree(per_cpu(xenpmu_shared, cpu).xenpmu_data);
+       per_cpu(xenpmu_shared, cpu).xenpmu_data = NULL;
 }
diff --git a/include/xen/interface/xenpmu.h b/include/xen/interface/xenpmu.h
index 79384e4..ec90673 100644
--- a/include/xen/interface/xenpmu.h
+++ b/include/xen/interface/xenpmu.h
@@ -14,6 +14,7 @@
 #define XENPMU_init            4
 #define XENPMU_finish          5
 #define XENPMU_lvtpc_set       6
+#define XENPMU_flush           7 /* Write cached MSR values to HW     */
 
 /* Parameter structure for HYPERVISOR_xenpmu_op call */
 struct xen_pmu_params {
-- 
1.8.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.