blob: 4f4e5cf3fed12e0a0d971b018e9ac4be53ec5661 [file] [log] [blame]
/*
* Copyright (c) 2009 The Regents of the University of California
* Barret Rhoden <brho@cs.berkeley.edu>
* See LICENSE for details.
*/
#pragma once
/* SMP related functions */
#include <arch/smp.h>
#include <ros/common.h>
#include <sys/queue.h>
#include <trap.h>
#include <atomic.h>
#include <process.h>
#include <syscall.h>
#include <alarm.h>
#include <trace.h>
#include <core_set.h>
#define CPU_STATE_IRQ 0
#define CPU_STATE_KERNEL 1
#define CPU_STATE_USER 2
#define CPU_STATE_IDLE 3
#define NR_CPU_STATES 4
static char *cpu_state_names[NR_CPU_STATES] =
{
"irq",
"kern",
"user",
"idle",
};
struct per_cpu_info {
#ifdef CONFIG_X86
uintptr_t stacktop; /* must be first */
int coreid; /* must be second */
int nmi_status;
uintptr_t nmi_worker_stacktop;
int vmx_enabled;
int guest_pcoreid;
#endif
spinlock_t lock;
/* Process management */
// cur_proc should be valid on all cores that are not management cores.
struct proc *cur_proc; /* which process context is loaded */
struct proc *owning_proc; /* proc owning the core / cur_ctx */
uint32_t owning_vcoreid; /* of owning proc, if applicable */
struct user_context *cur_ctx; /* user ctx we came in on (can be 0) */
struct user_context actual_ctx; /* storage for cur_ctx */
uint32_t __ctx_depth; /* don't access directly. see trap.h.*/
int __lock_checking_enabled;/* == 1, enables spinlock depth checking */
struct kthread *cur_kthread;/* tracks the running kernel context */
struct kthread *spare; /* useful when restarting */
struct timer_chain tchain; /* for the per-core alarm */
unsigned int lock_depth;
struct trace_ring traces;
int cpu_state;
uint64_t last_tick_cnt;
uint64_t state_ticks[NR_CPU_STATES];
/* TODO: 64b (not sure if we'll need these at all */
#ifdef CONFIG_X86
taskstate_t *tss;
segdesc_t *gdt;
#endif
/* KMSGs */
spinlock_t immed_amsg_lock;
struct kernel_msg_list immed_amsgs;
spinlock_t routine_amsg_lock;
struct kernel_msg_list routine_amsgs;
/* profiling -- opaque to all but the profiling code. */
void *profiling;
}__attribute__((aligned(ARCH_CL_SIZE)));
typedef struct per_cpu_info per_cpu_info_t;
extern per_cpu_info_t per_cpu_info[MAX_NUM_CORES];
#define for_each_core(i) for (int (i) = 0; (i) < num_cores; (i)++)
#define pcpui_ptr(i) &per_cpu_info[(i)]
#define pcpui_var(i, var) per_cpu_info[(i)].var
#define this_pcpui_ptr() pcpui_ptr(core_id())
#define this_pcpui_var(var) pcpui_var(core_id(), var)
/* Allows the kernel to figure out what process is running on this core. Can be
* used just like a pointer to a struct proc. */
#define current this_pcpui_var(cur_proc)
/* Allows the kernel to figure out what *user* ctx is on this core's stack. Can
* be used just like a pointer to a struct user_context. Note the distinction
* between kernel and user contexts. The kernel always returns to its nested,
* interrupted contexts via iret/etc. We never do that for user contexts. */
#define current_ctx this_pcpui_var(cur_ctx)
#define current_kthread this_pcpui_var(cur_kthread)
/* SMP bootup functions */
void smp_boot(void);
void smp_idle(void) __attribute__((noreturn));
void smp_percpu_init(void); // this must be called by each core individually
void __arch_pcpu_init(uint32_t coreid); /* each arch has one of these */
void __set_cpu_state(struct per_cpu_info *pcpui, int state);
void reset_cpu_state_ticks(int coreid);
/* SMP utility functions */
int smp_call_function_self(isr_t handler, void *data,
handler_wrapper_t **wait_wrapper);
int smp_call_function_all(isr_t handler, void *data,
handler_wrapper_t **wait_wrapper);
int smp_call_function_single(uint32_t dest, isr_t handler, void *data,
handler_wrapper_t **wait_wrapper);
int smp_call_wait(handler_wrapper_t *wrapper);
/* PCPUI Trace Rings: */
struct pcpu_trace_event {
int type;
int arg0;
uint64_t arg1;
};
/* If you want to add a type, use the next available number, increment NR_TYPES,
* use your own macro, and provide a handler. Add your handler to
* pcpui_tr_handlers in smp.c. */
#define PCPUI_TR_TYPE_NULL 0
#define PCPUI_TR_TYPE_KMSG 1
#define PCPUI_TR_TYPE_LOCKS 2
#define PCPUI_NR_TYPES 3
#ifdef CONFIG_TRACE_KMSGS
# define pcpui_trace_kmsg(pcpui, pc) \
{ \
struct pcpu_trace_event *e = get_trace_slot_racy(&pcpui->traces); \
if (e) { \
e->type = PCPUI_TR_TYPE_KMSG; \
e->arg1 = pc; \
} \
}
#else
# define pcpui_trace_kmsg(pcpui, pc)
#endif /* CONFIG_TRACE_KMSGS */
#ifdef CONFIG_TRACE_LOCKS
# define pcpui_trace_locks(pcpui, lock) \
{ \
struct pcpu_trace_event *e = get_trace_slot_overwrite(&pcpui->traces); \
if (e) { \
e->type = PCPUI_TR_TYPE_LOCKS; \
e->arg0 = (int)tsc2usec(read_tsc()); \
e->arg1 = (uintptr_t)lock; \
} \
}
#else
# define pcpui_trace_locks(pcpui, lock)
#endif /* CONFIG_TRACE_LOCKS */
void smp_do_in_cores(const struct core_set *cset, void (*func)(void *),
void *opaque);
/* Run the handlers for all events in a pcpui ring. Can run on all cores, or
* just one core. 'type' selects which event type is handled (0 for all). */
void pcpui_tr_foreach(int coreid, int type);
void pcpui_tr_foreach_all(int type);
void pcpui_tr_reset_all(void);
void pcpui_tr_reset_and_clear_all(void);