#include <i386/rtclock_protos.h>
#include <i386/pmCPU.h>
#include <i386/cpu_topology.h>
+#include <i386/seg.h>
#if CONFIG_VMX
#include <i386/vmx/vmx_cpu.h>
#endif
+#if MONOTONIC
+#include <machine/monotonic.h>
+#endif /* MONOTONIC */
+
#include <machine/pal_routines.h>
/*
boolean_t has_expired;
} rtclock_timer_t;
-
typedef struct {
- struct x86_64_tss *cdi_ktss;
- struct __attribute__((packed)) {
- uint16_t size;
- void *ptr;
- } cdi_gdt, cdi_idt;
- struct fake_descriptor *cdi_ldt;
- vm_offset_t cdi_sstk;
+ /* The 'u' suffixed fields store the double-mapped descriptor addresses */
+ struct x86_64_tss *cdi_ktssu;
+ struct x86_64_tss *cdi_ktssb;
+ x86_64_desc_register_t cdi_gdtu;
+ x86_64_desc_register_t cdi_gdtb;
+ x86_64_desc_register_t cdi_idtu;
+ x86_64_desc_register_t cdi_idtb;
+ struct fake_descriptor *cdi_ldtu;
+ struct fake_descriptor *cdi_ldtb;
+ vm_offset_t cdi_sstku;
+ vm_offset_t cdi_sstkb;
} cpu_desc_index_t;
typedef enum {
#define CPU_RTIME_BINS (12)
#define CPU_ITIME_BINS (CPU_RTIME_BINS)
+#define MAXPLFRAMES (16)
+typedef struct {
+ boolean_t pltype;
+ int plevel;
+ uint64_t plbt[MAXPLFRAMES];
+} plrecord_t;
+
/*
* Per-cpu data.
*
* cpu_datap(cpu_number) macro which uses the cpu_data_ptr[] array of per-cpu
* pointers.
*/
+typedef struct {
+ pcid_t cpu_pcid_free_hint;
+#define PMAP_PCID_MAX_PCID (0x800)
+ pcid_ref_t cpu_pcid_refcounts[PMAP_PCID_MAX_PCID];
+ pmap_t cpu_pcid_last_pmap_dispatched[PMAP_PCID_MAX_PCID];
+} pcid_cdata_t;
+
typedef struct cpu_data
{
struct pal_cpu_data cpu_pal_data; /* PAL-specific data */
vm_offset_t cpu_kernel_stack; /* kernel stack top */
vm_offset_t cpu_int_stack_top;
int cpu_interrupt_level;
- int cpu_phys_number; /* Physical CPU */
- cpu_id_t cpu_id; /* Platform Expert */
volatile int cpu_signals; /* IPI events */
volatile int cpu_prior_signals; /* Last set of events,
* debugging
*/
ast_t cpu_pending_ast;
volatile int cpu_running;
+#if !MONOTONIC
boolean_t cpu_fixed_pmcs_enabled;
+#endif /* !MONOTONIC */
rtclock_timer_t rtclock_timer;
+ uint64_t quantum_timer_deadline;
volatile addr64_t cpu_active_cr3 __attribute((aligned(64)));
union {
volatile uint32_t cpu_tlb_invalid;
volatile task_map_t cpu_task_map;
volatile addr64_t cpu_task_cr3;
addr64_t cpu_kernel_cr3;
+ volatile addr64_t cpu_ucr3;
+ boolean_t cpu_pagezero_mapped;
cpu_uber_t cpu_uber;
- void *cpu_chud;
- void *cpu_console_buf;
- struct x86_lcpu lcpu;
+/* Double-mapped per-CPU exception stack address */
+ uintptr_t cd_estack;
+ int cpu_xstate;
+/* Address of shadowed, partially mirrored CPU data structures located
+ * in the double mapped PML4
+ */
+ void *cd_shadow;
struct processor *cpu_processor;
#if NCOPY_WINDOWS > 0
struct cpu_pmap *cpu_pmap;
#endif
+ struct real_descriptor *cpu_ldtp;
struct cpu_desc_table *cpu_desc_tablep;
- struct fake_descriptor *cpu_ldtp;
cpu_desc_index_t cpu_desc_index;
int cpu_ldt;
#if NCOPY_WINDOWS > 0
#define HWINTCNT_SIZE 256
uint32_t cpu_hwIntCnt[HWINTCNT_SIZE]; /* Interrupt counts */
uint64_t cpu_hwIntpexits[HWINTCNT_SIZE];
- uint64_t cpu_hwIntcexits[HWINTCNT_SIZE];
uint64_t cpu_dr7; /* debug control register */
uint64_t cpu_int_event_time; /* intr entry/exit time */
pal_rtc_nanotime_t *cpu_nanotime; /* Nanotime info */
uint64_t *cpu_kpc_shadow;
uint64_t *cpu_kpc_reload;
#endif
+#if MONOTONIC
+ struct mt_cpu cpu_monotonic;
+#endif /* MONOTONIC */
uint32_t cpu_pmap_pcid_enabled;
pcid_t cpu_active_pcid;
pcid_t cpu_last_pcid;
+ pcid_t cpu_kernel_pcid;
volatile pcid_ref_t *cpu_pmap_pcid_coherentp;
volatile pcid_ref_t *cpu_pmap_pcid_coherentp_kernel;
-#define PMAP_PCID_MAX_PCID (0x1000)
- pcid_t cpu_pcid_free_hint;
- pcid_ref_t cpu_pcid_refcounts[PMAP_PCID_MAX_PCID];
- pmap_t cpu_pcid_last_pmap_dispatched[PMAP_PCID_MAX_PCID];
+ pcid_cdata_t *cpu_pcid_data;
#ifdef PCID_STATS
uint64_t cpu_pmap_pcid_flushes;
uint64_t cpu_pmap_pcid_preserves;
uint64_t cpu_rtime_total;
uint64_t cpu_ixtime;
uint64_t cpu_idle_exits;
- uint64_t cpu_rtimes[CPU_RTIME_BINS];
- uint64_t cpu_itimes[CPU_ITIME_BINS];
- uint64_t cpu_cur_insns;
- uint64_t cpu_cur_ucc;
- uint64_t cpu_cur_urc;
+ uint64_t cpu_rtimes[CPU_RTIME_BINS];
+ uint64_t cpu_itimes[CPU_ITIME_BINS];
+#if !MONOTONIC
+ uint64_t cpu_cur_insns;
+ uint64_t cpu_cur_ucc;
+ uint64_t cpu_cur_urc;
+#endif /* !MONOTONIC */
uint64_t cpu_gpmcs[4];
uint64_t cpu_max_observed_int_latency;
int cpu_max_observed_int_latency_vector;
int cpu_threadtype;
boolean_t cpu_iflag;
boolean_t cpu_boot_complete;
- int cpu_hibernate;
+ int cpu_hibernate;
+#define MAX_PREEMPTION_RECORDS (8)
+#if DEVELOPMENT || DEBUG
+ int cpu_plri;
+ plrecord_t plrecords[MAX_PREEMPTION_RECORDS];
+#endif
+ void *cpu_console_buf;
+ struct x86_lcpu lcpu;
+ int cpu_phys_number; /* Physical CPU */
+ cpu_id_t cpu_id; /* Platform Expert */
+#if DEBUG
+ uint64_t cpu_entry_cr3;
+ uint64_t cpu_exit_cr3;
+ uint64_t cpu_pcid_last_cr3;
+#endif
} cpu_data_t;
extern cpu_data_t *cpu_data_ptr[];
CPU_DATA_GET(cpu_phys_number,int)
}
+static inline cpu_data_t *
+current_cpu_datap(void) {
+ CPU_DATA_GET(cpu_this, cpu_data_t *);
+}
+
+/*
+ * Facility to diagnose preemption-level imbalances, which are otherwise
+ * challenging to debug. On each operation that enables or disables preemption,
+ * we record a backtrace into a per-CPU ring buffer, along with the current
+ * preemption level and operation type. Thus, if an imbalance is observed,
+ * one can examine these per-CPU records to determine which codepath failed
+ * to re-enable preemption, enabled premption without a corresponding
+ * disablement etc. The backtracer determines which stack is currently active,
+ * and uses that to perform bounds checks on unterminated stacks.
+ * To enable, sysctl -w machdep.pltrace=1 on DEVELOPMENT or DEBUG kernels (DRK '15)
+ * The bounds check currently doesn't account for non-default thread stack sizes.
+ */
+#if DEVELOPMENT || DEBUG
+static inline void pltrace_bt(uint64_t *rets, int maxframes, uint64_t stacklo, uint64_t stackhi) {
+ uint64_t *cfp = (uint64_t *) __builtin_frame_address(0);
+ int plbtf;
+
+ assert(stacklo !=0 && stackhi !=0);
+
+ for (plbtf = 0; plbtf < maxframes; plbtf++) {
+ if (((uint64_t)cfp == 0) || (((uint64_t)cfp < stacklo) || ((uint64_t)cfp > stackhi))) {
+ rets[plbtf] = 0;
+ continue;
+ }
+ rets[plbtf] = *(cfp + 1);
+ cfp = (uint64_t *) (*cfp);
+ }
+}
+
+
+extern uint32_t low_intstack[]; /* bottom */
+extern uint32_t low_eintstack[]; /* top */
+extern char mp_slave_stack[PAGE_SIZE];
+
+static inline void pltrace_internal(boolean_t enable) {
+ cpu_data_t *cdata = current_cpu_datap();
+ int cpli = cdata->cpu_preemption_level;
+ int cplrecord = cdata->cpu_plri;
+ uint64_t kstackb, kstackt, *plbts;
+
+ assert(cpli >= 0);
+
+ cdata->plrecords[cplrecord].pltype = enable;
+ cdata->plrecords[cplrecord].plevel = cpli;
+
+ plbts = &cdata->plrecords[cplrecord].plbt[0];
+
+ cplrecord++;
+
+ if (cplrecord >= MAX_PREEMPTION_RECORDS) {
+ cplrecord = 0;
+ }
+
+ cdata->cpu_plri = cplrecord;
+ /* Obtain the 'current' program counter, initial backtrace
+ * element. This will also indicate if we were unable to
+ * trace further up the stack for some reason
+ */
+ __asm__ volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
+ : "=m" (plbts[0])
+ :
+ : "rax");
+
+
+ thread_t cplthread = cdata->cpu_active_thread;
+ if (cplthread) {
+ uintptr_t csp;
+ __asm__ __volatile__ ("movq %%rsp, %0": "=r" (csp):);
+ /* Determine which stack we're on to populate stack bounds.
+ * We don't need to trace across stack boundaries for this
+ * routine.
+ */
+ kstackb = cdata->cpu_active_stack;
+ kstackt = kstackb + KERNEL_STACK_SIZE;
+ if (csp < kstackb || csp > kstackt) {
+ kstackt = cdata->cpu_kernel_stack;
+ kstackb = kstackb - KERNEL_STACK_SIZE;
+ if (csp < kstackb || csp > kstackt) {
+ kstackt = cdata->cpu_int_stack_top;
+ kstackb = kstackt - INTSTACK_SIZE;
+ if (csp < kstackb || csp > kstackt) {
+ kstackt = (uintptr_t)low_eintstack;
+ kstackb = (uintptr_t)low_eintstack - INTSTACK_SIZE;
+ if (csp < kstackb || csp > kstackt) {
+ kstackb = (uintptr_t) mp_slave_stack;
+ kstackt = (uintptr_t) mp_slave_stack + PAGE_SIZE;
+ }
+ }
+ }
+ }
+
+ if (kstackb) {
+ pltrace_bt(&plbts[1], MAXPLFRAMES - 1, kstackb, kstackt);
+ }
+ }
+}
+
+extern int plctrace_enabled;
+#endif /* DEVELOPMENT || DEBUG */
+
+static inline void pltrace(boolean_t plenable) {
+#if DEVELOPMENT || DEBUG
+ if (__improbable(plctrace_enabled != 0)) {
+ pltrace_internal(plenable);
+ }
+#else
+ (void)plenable;
+#endif
+}
static inline void
-disable_preemption(void)
-{
+disable_preemption_internal(void) {
+ assert(get_preemption_level() >= 0);
+
#if defined(__clang__)
cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
cpu_data->cpu_preemption_level++;
#else
__asm__ volatile ("incl %%gs:%P0"
- :
- : "i" (offsetof(cpu_data_t, cpu_preemption_level)));
+ :
+ : "i" (offsetof(cpu_data_t, cpu_preemption_level)));
#endif
+ pltrace(FALSE);
}
static inline void
-enable_preemption(void)
-{
+enable_preemption_internal(void) {
assert(get_preemption_level() > 0);
-
+ pltrace(TRUE);
#if defined(__clang__)
cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
if (0 == --cpu_data->cpu_preemption_level)
{
assert(get_preemption_level() > 0);
+ pltrace(TRUE);
#if defined(__clang__)
cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
cpu_data->cpu_preemption_level--;
#endif
}
+static inline void
+_enable_preemption_no_check(void) {
+ enable_preemption_no_check();
+}
+
static inline void
mp_disable_preemption(void)
{
- disable_preemption();
+ disable_preemption_internal();
}
static inline void
-mp_enable_preemption(void)
+_mp_disable_preemption(void)
{
- enable_preemption();
+ disable_preemption_internal();
}
static inline void
-mp_enable_preemption_no_check(void)
+mp_enable_preemption(void)
{
+ enable_preemption_internal();
+}
+
+static inline void
+_mp_enable_preemption(void) {
+ enable_preemption_internal();
+}
+
+static inline void
+mp_enable_preemption_no_check(void) {
enable_preemption_no_check();
}
-static inline cpu_data_t *
-current_cpu_datap(void)
-{
- CPU_DATA_GET(cpu_this, cpu_data_t *);
+static inline void
+_mp_enable_preemption_no_check(void) {
+ enable_preemption_no_check();
}
+#ifdef XNU_KERNEL_PRIVATE
+#define disable_preemption() disable_preemption_internal()
+#define enable_preemption() enable_preemption_internal()
+#define MACHINE_PREEMPTION_MACROS (1)
+#endif
+
static inline cpu_data_t *
-cpu_datap(int cpu)
-{
+cpu_datap(int cpu) {
return cpu_data_ptr[cpu];
}
+static inline int
+cpu_is_running(int cpu) {
+ return ((cpu_datap(cpu) != NULL) && (cpu_datap(cpu)->cpu_running));
+}
+
+#ifdef MACH_KERNEL_PRIVATE
+static inline cpu_data_t *
+cpu_shadowp(int cpu) {
+ return cpu_data_ptr[cpu]->cd_shadow;
+}
+
+#endif
extern cpu_data_t *cpu_data_alloc(boolean_t is_boot_cpu);
extern void cpu_data_realloc(void);
projects / apple / xnu.git / blobdiff