Page MenuHomeFreeBSD
Files F165008215

D57630.id179975.diff
No OneTemporary
Actions

D57630.id179975.diff
View Options

diff --git a/sys/sys/pmc.h b/sys/sys/pmc.h
--- a/sys/sys/pmc.h
+++ b/sys/sys/pmc.h
@@ -60,7 +60,7 @@
* The patch version is incremented for every bug fix.
*/
#define PMC_VERSION_MAJOR 0x0A
-#define PMC_VERSION_MINOR 0x01
+#define PMC_VERSION_MINOR 0x02
#define PMC_VERSION_PATCH 0x0000
#define PMC_VERSION (PMC_VERSION_MAJOR << 24 | \
@@ -349,7 +349,10 @@
__PMC_OP(WRITELOG, "Write a cookie to the log file") \
__PMC_OP(CLOSELOG, "Close log file") \
__PMC_OP(GETDYNEVENTINFO, "Get dynamic events list") \
- __PMC_OP(GETCAPS, "Get capabilities")
+ __PMC_OP(GETCAPS, "Get capabilities") \
+ __PMC_OP(PMCGROUPCREATE, "Create a PMC event group") \
+ __PMC_OP(PMCGROUPADD, "Add PMC to a group") \
+ __PMC_OP(PMCGROUPCOMMIT, "Commit a PMC event group")
enum pmc_ops {
#undef __PMC_OP
@@ -361,12 +364,18 @@
* Flags used in operations on PMCs.
*/
-#define PMC_F_UNUSED1 0x00000001 /* unused */
#define PMC_F_DESCENDANTS 0x00000002 /*OP ALLOCATE track descendants */
#define PMC_F_LOG_PROCCSW 0x00000004 /*OP ALLOCATE track ctx switches */
#define PMC_F_LOG_PROCEXIT 0x00000008 /*OP ALLOCATE log proc exits */
#define PMC_F_NEWVALUE 0x00000010 /*OP RW write new value */
#define PMC_F_OLDVALUE 0x00000020 /*OP RW get old value */
+/*
+ * NOTE: bit 0x00000001 is reserved by PMC_PP_ENABLE_MSR_ACCESS in
+ * pp_flags but is incorrectly tested against pm_flags inside
+ * pmc_attach_process(); avoid using it as a pm_flags bit. Bit
+ * 0x00000040 is the first slot that is otherwise free.
+ */
+#define PMC_F_GROUP_DEFER 0x00000040 /* defer HW row assign (group) */
/* V2 API */
#define PMC_F_CALLCHAIN 0x00000080 /*OP ALLOCATE capture callchains */
@@ -380,6 +389,8 @@
* through class-dependent fields
*/
+#define PMC_F_GROUP_MUX 0x00000400 /* multiplex when nevents > slots */
+#define PMC_F_SCALED 0x00000800 /* OP RW return scaled value */
/* internal flags */
#define PMC_F_ATTACHED_TO_OWNER 0x00010000 /*attached to owner*/
#define PMC_F_NEEDS_LOGFILE 0x00020000 /*needs log file */
@@ -436,6 +447,24 @@
#define PMC_CPU_ANY ~0
+#define PMC_ROW_UNASSIGNED 0xFF
+
+struct pmc_op_pmcgroupcreate {
+ uint32_t pm_groupid; /* [return] group id */
+};
+
+struct pmc_op_pmcgroupadd {
+ uint32_t pm_groupid;
+ pmc_id_t pm_pmcid;
+ uint32_t pm_flags; /* PMC_GROUP_F_LEADER */
+};
+
+#define PMC_GROUP_F_LEADER 0x00000001
+
+struct pmc_op_pmcgroupcommit {
+ uint32_t pm_groupid;
+};
+
struct pmc_op_pmcallocate {
uint32_t pm_caps; /* PMC_CAP_* */
uint32_t pm_cpu; /* CPU number or PMC_CPU_ANY */
@@ -748,6 +777,20 @@
uint8_t pps_stalled;
uint8_t pps_cpustate;
} __aligned(CACHE_LINE_SIZE);
+
+#ifdef _KERNEL
+/*
+ * Opaque handles for the hwpmc PMU grouping/multiplex layer
+ * (sys/dev/hwpmc/hwpmc_pmu.h). Declared here as incomplete types so
+ * that struct pmc and struct pmc_process can embed pointers to them;
+ * the full struct definitions (and the matching pmc_sched_constraint_t)
+ * live in the PMU layer header, which is only compiled in on
+ * architectures that provide the grouping backend.
+ */
+typedef struct pmu_event pmu_event_t;
+typedef struct pmu_group pmu_group_t;
+#endif
+
struct pmc {
LIST_HEAD(,pmc_target) pm_targets; /* list of target processes */
LIST_ENTRY(pmc) pm_next; /* owner's list */
@@ -800,8 +843,28 @@
/* md extensions */
union pmc_md_pmc pm_md;
+
+#ifdef _KERNEL
+ pmu_event_t *pm_pmu; /* grouping/multiplex state */
+ pmc_id_t pm_handle; /* stable user-facing id;
+ * pm_id may be rewritten by
+ * the group assigner, but
+ * pm_handle is fixed at
+ * allocate time */
+#endif
};
+/*
+ * Reserved range in the row-index byte for deferred (group) PMC handles
+ * before the assigner places them on real hardware rows. Distinguishes
+ * each deferred PMC for pmc_find_pmc() lookups. Real hardware rows are
+ * always < md->pmd_npmc which is much smaller than 0x80.
+ */
+#define PMC_HANDLE_DEFERRED_BASE 0x80
+#define PMC_HANDLE_DEFERRED_MAX 0xFE
+#define PMC_ROW_IS_DEFERRED_HANDLE(R) \
+ ((R) >= PMC_HANDLE_DEFERRED_BASE && (R) <= PMC_HANDLE_DEFERRED_MAX)
+
/*
* Accessor macros for 'struct pmc'
*/
@@ -857,6 +920,30 @@
int pp_refcnt; /* reference count */
uint32_t pp_flags; /* flags PMC_PP_* */
struct proc *pp_proc; /* target process */
+#ifdef _KERNEL
+ LIST_HEAD(, pmu_group) pp_pmu_groups; /* attached PMU groups (FIFO) */
+ /*
+ * Inter-group multiplex rotation. When the union of events
+ * across pp_pmu_groups exceeds the available HW counters, the
+ * pp-level rotation kthread evicts the front-most scheduled
+ * group every kern.hwpmc.mux_period_ms tick and brings in
+ * any deferred group whose nevents now fits. Within a group
+ * scheduling is always atomic: every sibling is placed on HW
+ * together or none of them is.
+ */
+ struct thread *pp_pmu_rot_td;
+ bool pp_pmu_rot_running;
+ /*
+ * Round-robin cursor: the next group to attempt scheduling-in
+ * at the start of each rotation tick. When a group fails to
+ * fit (ENOSPC) we record it as the next cursor and stop the
+ * tick; the next tick begins from there. When a group is
+ * removed from pp_pmu_groups (release path) it must clear
+ * this if it points at itself, so the kthread does not chase
+ * a freed pmu_group.
+ */
+ pmu_group_t *pp_pmu_rot_cursor;
+#endif
struct pmc_targetstate pp_pmcs[]; /* NHWPMCs */
};
@@ -878,6 +965,10 @@
LIST_ENTRY(pmc_owner) po_next; /* hash chain */
CK_LIST_ENTRY(pmc_owner) po_ssnext; /* (g/p) list of SS PMC owners */
LIST_HEAD(, pmc) po_pmcs; /* owned PMC list */
+#ifdef _KERNEL
+ LIST_HEAD(, pmu_group) po_groups; /* PMU event groups */
+ uint8_t po_next_deferred_serial; /* 0x80..0xFE */
+#endif
TAILQ_HEAD(, pmclog_buffer) po_logbuffers; /* (o) logbuffer list */
struct mtx po_mtx; /* spin lock for (o) */
struct proc *po_owner; /* owner proc */

File Metadata

Mime Type
text/plain
Expires
Thu, Aug 6, 7:45 AM (5 h, 49 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
34068293
Default Alt Text
D57630.id179975.diff (5 KB)

Event Timeline