4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License, Version 1.0 only
6 * (the "License"). You may not use this file except in compliance
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or http://www.opensolaris.org/os/licensing.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
26 * Copyright 2005 Sun Microsystems, Inc. All rights reserved.
27 * Use is subject to license terms.
31 * Copyright (c) 2011, Joyent, Inc. All rights reserved.
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/types.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
41 #include <sys/dtrace_impl.h>
42 #include <sys/dtrace_bsd.h>
43 #include <machine/clock.h>
44 #include <machine/cpufunc.h>
45 #include <machine/frame.h>
46 #include <machine/psl.h>
47 #include <machine/trap.h>
50 extern void dtrace_getnanotime(struct timespec *tsp);
52 int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t);
54 typedef struct dtrace_invop_hdlr {
55 int (*dtih_func)(uintptr_t, struct trapframe *, uintptr_t);
56 struct dtrace_invop_hdlr *dtih_next;
57 } dtrace_invop_hdlr_t;
59 dtrace_invop_hdlr_t *dtrace_invop_hdlr;
62 dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t eax)
64 dtrace_invop_hdlr_t *hdlr;
67 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next)
68 if ((rval = hdlr->dtih_func(addr, frame, eax)) != 0)
75 dtrace_invop_add(int (*func)(uintptr_t, struct trapframe *, uintptr_t))
77 dtrace_invop_hdlr_t *hdlr;
79 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP);
80 hdlr->dtih_func = func;
81 hdlr->dtih_next = dtrace_invop_hdlr;
82 dtrace_invop_hdlr = hdlr;
86 dtrace_invop_remove(int (*func)(uintptr_t, struct trapframe *, uintptr_t))
88 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL;
92 panic("attempt to remove non-existent invop handler");
94 if (hdlr->dtih_func == func)
98 hdlr = hdlr->dtih_next;
102 ASSERT(dtrace_invop_hdlr == hdlr);
103 dtrace_invop_hdlr = hdlr->dtih_next;
105 ASSERT(dtrace_invop_hdlr != hdlr);
106 prev->dtih_next = hdlr->dtih_next;
114 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit))
116 (*func)(0, (uintptr_t) addr_PTmap);
120 dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg)
124 if (cpu == DTRACE_CPUALL)
127 CPU_SETOF(cpu, &cpus);
129 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, func,
130 smp_no_rendezvous_barrier, arg);
134 dtrace_sync_func(void)
141 dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL);
146 dtrace_safe_synchronous_signal(void)
148 kthread_t *t = curthread;
149 struct regs *rp = lwptoregs(ttolwp(t));
150 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc;
152 ASSERT(t->t_dtrace_on);
155 * If we're not in the range of scratch addresses, we're not actually
156 * tracing user instructions so turn off the flags. If the instruction
157 * we copied out caused a synchonous trap, reset the pc back to its
158 * original value and turn off the flags.
160 if (rp->r_pc < t->t_dtrace_scrpc ||
161 rp->r_pc > t->t_dtrace_astpc + isz) {
163 } else if (rp->r_pc == t->t_dtrace_scrpc ||
164 rp->r_pc == t->t_dtrace_astpc) {
165 rp->r_pc = t->t_dtrace_pc;
171 dtrace_safe_defer_signal(void)
173 kthread_t *t = curthread;
174 struct regs *rp = lwptoregs(ttolwp(t));
175 size_t isz = t->t_dtrace_npc - t->t_dtrace_pc;
177 ASSERT(t->t_dtrace_on);
180 * If we're not in the range of scratch addresses, we're not actually
181 * tracing user instructions so turn off the flags.
183 if (rp->r_pc < t->t_dtrace_scrpc ||
184 rp->r_pc > t->t_dtrace_astpc + isz) {
190 * If we have executed the original instruction, but we have performed
191 * neither the jmp back to t->t_dtrace_npc nor the clean up of any
192 * registers used to emulate %rip-relative instructions in 64-bit mode,
193 * we'll save ourselves some effort by doing that here and taking the
194 * signal right away. We detect this condition by seeing if the program
195 * counter is the range [scrpc + isz, astpc).
197 if (rp->r_pc >= t->t_dtrace_scrpc + isz &&
198 rp->r_pc < t->t_dtrace_astpc) {
201 * If there is a scratch register and we're on the
202 * instruction immediately after the modified instruction,
203 * restore the value of that scratch register.
205 if (t->t_dtrace_reg != 0 &&
206 rp->r_pc == t->t_dtrace_scrpc + isz) {
207 switch (t->t_dtrace_reg) {
209 rp->r_rax = t->t_dtrace_regv;
212 rp->r_rcx = t->t_dtrace_regv;
215 rp->r_r8 = t->t_dtrace_regv;
218 rp->r_r9 = t->t_dtrace_regv;
223 rp->r_pc = t->t_dtrace_npc;
229 * Otherwise, make sure we'll return to the kernel after executing
230 * the copied out instruction and defer the signal.
232 if (!t->t_dtrace_step) {
233 ASSERT(rp->r_pc < t->t_dtrace_astpc);
234 rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc;
235 t->t_dtrace_step = 1;
244 static int64_t tgt_cpu_tsc;
245 static int64_t hst_cpu_tsc;
246 static int64_t tsc_skew[MAXCPU];
247 static uint64_t nsec_scale;
249 /* See below for the explanation of this macro. */
250 #define SCALE_SHIFT 28
253 dtrace_gethrtime_init_cpu(void *arg)
255 uintptr_t cpu = (uintptr_t) arg;
258 tgt_cpu_tsc = rdtsc();
260 hst_cpu_tsc = rdtsc();
263 #ifdef EARLY_AP_STARTUP
265 dtrace_gethrtime_init(void *arg)
273 * Get the frequency and scale factor as early as possible so that they can be
274 * used for boot-time tracing.
277 dtrace_gethrtime_init_early(void *arg)
283 * Get TSC frequency known at this moment.
284 * This should be constant if TSC is invariant.
285 * Otherwise tick->time conversion will be inaccurate, but
286 * will preserve monotonic property of TSC.
288 tsc_f = atomic_load_acq_64(&tsc_freq);
291 * The following line checks that nsec_scale calculated below
292 * doesn't overflow 32-bit unsigned integer, so that it can multiply
293 * another 32-bit integer without overflowing 64-bit.
294 * Thus minimum supported TSC frequency is 62.5MHz.
296 KASSERT(tsc_f > (NANOSEC >> (32 - SCALE_SHIFT)),
297 ("TSC frequency is too low"));
300 * We scale up NANOSEC/tsc_f ratio to preserve as much precision
302 * 2^28 factor was chosen quite arbitrarily from practical
304 * - it supports TSC frequencies as low as 62.5MHz (see above);
305 * - it provides quite good precision (e < 0.01%) up to THz
306 * (terahertz) values;
308 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tsc_f;
309 #ifndef EARLY_AP_STARTUP
311 SYSINIT(dtrace_gethrtime_init_early, SI_SUB_CPU, SI_ORDER_ANY,
312 dtrace_gethrtime_init_early, NULL);
315 dtrace_gethrtime_init(void *arg)
322 if (vm_guest != VM_GUEST_NO)
325 /* The current CPU is the reference one. */
327 tsc_skew[curcpu] = 0;
333 CPU_SETOF(PCPU_GET(cpuid), &map);
334 CPU_SET(pc->pc_cpuid, &map);
336 smp_rendezvous_cpus(map, NULL,
337 dtrace_gethrtime_init_cpu,
338 smp_no_rendezvous_barrier, (void *)(uintptr_t) i);
340 tsc_skew[i] = tgt_cpu_tsc - hst_cpu_tsc;
344 #ifdef EARLY_AP_STARTUP
345 SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY,
346 dtrace_gethrtime_init, NULL);
348 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init,
353 * DTrace needs a high resolution time function which can
354 * be called from a probe context and guaranteed not to have
355 * instrumented with probes itself.
357 * Returns nanoseconds since boot.
360 dtrace_gethrtime(void)
367 * We split TSC value into lower and higher 32-bit halves and separately
368 * scale them with nsec_scale, then we scale them down by 2^28
369 * (see nsec_scale calculations) taking into account 32-bit shift of
370 * the higher half and finally add.
372 rflags = intr_disable();
373 tsc = rdtsc() - tsc_skew[curcpu];
374 intr_restore(rflags);
378 return (((lo * nsec_scale) >> SCALE_SHIFT) +
379 ((hi * nsec_scale) << (32 - SCALE_SHIFT)));
383 dtrace_gethrestime(void)
385 struct timespec current_time;
387 dtrace_getnanotime(¤t_time);
389 return (current_time.tv_sec * 1000000000ULL + current_time.tv_nsec);
392 /* Function to handle DTrace traps during probes. See amd64/amd64/trap.c. */
394 dtrace_trap(struct trapframe *frame, u_int type)
399 * A trap can occur while DTrace executes a probe. Before
400 * executing the probe, DTrace blocks re-scheduling and sets
401 * a flag in its per-cpu flags to indicate that it doesn't
402 * want to fault. On returning from the probe, the no-fault
403 * flag is cleared and finally re-scheduling is enabled.
405 * Check if DTrace has enabled 'no-fault' mode:
408 nofault = cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT;
411 KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled"));
414 * There are only a couple of trap types that are expected.
415 * All the rest will be handled in the usual way.
418 /* General protection fault. */
420 /* Flag an illegal operation. */
421 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP;
424 * Offset the instruction pointer to the instruction
425 * following the one causing the fault.
427 frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip);
431 /* Flag a bad address. */
432 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR;
433 cpu_core[curcpu].cpuc_dtrace_illval = frame->tf_addr;
436 * Offset the instruction pointer to the instruction
437 * following the one causing the fault.
439 frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip);
442 /* Handle all other traps in the usual way. */
447 /* Handle the trap in the usual way. */