1 /*
2 * CDDL HEADER START
3 *
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
7 * with the License.
8 *
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.
13 *
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]
19 *
20 * CDDL HEADER END
21 *
22 */
23 /*
24 * Copyright 2005 Sun Microsystems, Inc. All rights reserved.
25 * Use is subject to license terms.
26 */
27
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/kernel.h>
31 #include <sys/malloc.h>
32 #include <sys/kmem.h>
33 #include <sys/proc.h>
34 #include <sys/smp.h>
35 #include <sys/dtrace_impl.h>
36 #include <sys/dtrace_bsd.h>
37 #include <cddl/dev/dtrace/dtrace_cddl.h>
38 #include <machine/clock.h>
39 #include <machine/frame.h>
40 #include <machine/trap.h>
41 #include <vm/pmap.h>
42
43 #define DELAYBRANCH(x) ((int)(x) < 0)
44
45 extern dtrace_id_t dtrace_probeid_error;
46 extern int (*dtrace_invop_jump_addr)(struct trapframe *);
47
48 extern void dtrace_getnanotime(struct timespec *tsp);
49
50 int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t);
51 void dtrace_invop_init(void);
52 void dtrace_invop_uninit(void);
53
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;
58
59 dtrace_invop_hdlr_t *dtrace_invop_hdlr;
60
61 int
dtrace_invop(uintptr_t addr,struct trapframe * frame,uintptr_t arg0)62 dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t arg0)
63 {
64 struct thread *td;
65 dtrace_invop_hdlr_t *hdlr;
66 int rval;
67
68 rval = 0;
69 td = curthread;
70 td->t_dtrace_trapframe = frame;
71 for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next)
72 if ((rval = hdlr->dtih_func(addr, frame, arg0)) != 0)
73 break;
74 td->t_dtrace_trapframe = NULL;
75 return (rval);
76 }
77
78 void
dtrace_invop_add(int (* func)(uintptr_t,struct trapframe *,uintptr_t))79 dtrace_invop_add(int (*func)(uintptr_t, struct trapframe *, uintptr_t))
80 {
81 dtrace_invop_hdlr_t *hdlr;
82
83 hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP);
84 hdlr->dtih_func = func;
85 hdlr->dtih_next = dtrace_invop_hdlr;
86 dtrace_invop_hdlr = hdlr;
87 }
88
89 void
dtrace_invop_remove(int (* func)(uintptr_t,struct trapframe *,uintptr_t))90 dtrace_invop_remove(int (*func)(uintptr_t, struct trapframe *, uintptr_t))
91 {
92 dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL;
93
94 for (;;) {
95 if (hdlr == NULL)
96 panic("attempt to remove non-existent invop handler");
97
98 if (hdlr->dtih_func == func)
99 break;
100
101 prev = hdlr;
102 hdlr = hdlr->dtih_next;
103 }
104
105 if (prev == NULL) {
106 ASSERT(dtrace_invop_hdlr == hdlr);
107 dtrace_invop_hdlr = hdlr->dtih_next;
108 } else {
109 ASSERT(dtrace_invop_hdlr != hdlr);
110 prev->dtih_next = hdlr->dtih_next;
111 }
112
113 kmem_free(hdlr, 0);
114 }
115
116
117 /*ARGSUSED*/
118 void
dtrace_toxic_ranges(void (* func)(uintptr_t base,uintptr_t limit))119 dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit))
120 {
121 /*
122 * No toxic regions?
123 */
124 }
125
126 static int64_t tgt_cpu_tsc;
127 static int64_t hst_cpu_tsc;
128 static int64_t timebase_skew[MAXCPU];
129 static uint64_t nsec_scale;
130
131 /* See below for the explanation of this macro. */
132 /* This is taken from the amd64 dtrace_subr, to provide a synchronized timer
133 * between multiple processors in dtrace. Since PowerPC Timebases can be much
134 * lower than x86, the scale shift is 26 instead of 28, allowing for a 15.63MHz
135 * timebase.
136 */
137 #define SCALE_SHIFT 26
138
139 static void
dtrace_gethrtime_init_cpu(void * arg)140 dtrace_gethrtime_init_cpu(void *arg)
141 {
142 uintptr_t cpu = (uintptr_t) arg;
143
144 if (cpu == curcpu)
145 tgt_cpu_tsc = mftb();
146 else
147 hst_cpu_tsc = mftb();
148 }
149
150 static void
dtrace_gethrtime_init(void * arg)151 dtrace_gethrtime_init(void *arg)
152 {
153 struct pcpu *pc;
154 uint64_t tb_f;
155 cpuset_t map;
156 int i;
157
158 tb_f = cpu_tickrate();
159
160 /*
161 * The following line checks that nsec_scale calculated below
162 * doesn't overflow 32-bit unsigned integer, so that it can multiply
163 * another 32-bit integer without overflowing 64-bit.
164 * Thus minimum supported Timebase frequency is 15.63MHz.
165 */
166 KASSERT(tb_f > (NANOSEC >> (32 - SCALE_SHIFT)), ("Timebase frequency is too low"));
167
168 /*
169 * We scale up NANOSEC/tb_f ratio to preserve as much precision
170 * as possible.
171 * 2^26 factor was chosen quite arbitrarily from practical
172 * considerations:
173 * - it supports TSC frequencies as low as 15.63MHz (see above);
174 */
175 nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tb_f;
176
177 /* The current CPU is the reference one. */
178 sched_pin();
179 timebase_skew[curcpu] = 0;
180 CPU_FOREACH(i) {
181 if (i == curcpu)
182 continue;
183
184 pc = pcpu_find(i);
185 CPU_SETOF(PCPU_GET(cpuid), &map);
186 CPU_SET(pc->pc_cpuid, &map);
187
188 smp_rendezvous_cpus(map, NULL,
189 dtrace_gethrtime_init_cpu,
190 smp_no_rendezvous_barrier, (void *)(uintptr_t) i);
191
192 timebase_skew[i] = tgt_cpu_tsc - hst_cpu_tsc;
193 }
194 sched_unpin();
195 }
196 #ifdef EARLY_AP_STARTUP
197 SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY,
198 dtrace_gethrtime_init, NULL);
199 #else
200 SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init,
201 NULL);
202 #endif
203
204 /*
205 * DTrace needs a high resolution time function which can
206 * be called from a probe context and guaranteed not to have
207 * instrumented with probes itself.
208 *
209 * Returns nanoseconds since boot.
210 */
211 uint64_t
dtrace_gethrtime(void)212 dtrace_gethrtime(void)
213 {
214 uint64_t timebase;
215 uint32_t lo;
216 uint32_t hi;
217
218 /*
219 * We split timebase value into lower and higher 32-bit halves and separately
220 * scale them with nsec_scale, then we scale them down by 2^28
221 * (see nsec_scale calculations) taking into account 32-bit shift of
222 * the higher half and finally add.
223 */
224 timebase = mftb() - timebase_skew[curcpu];
225 lo = timebase;
226 hi = timebase >> 32;
227 return (((lo * nsec_scale) >> SCALE_SHIFT) +
228 ((hi * nsec_scale) << (32 - SCALE_SHIFT)));
229 }
230
231 uint64_t
dtrace_gethrestime(void)232 dtrace_gethrestime(void)
233 {
234 struct timespec curtime;
235
236 dtrace_getnanotime(&curtime);
237
238 return (curtime.tv_sec * 1000000000UL + curtime.tv_nsec);
239 }
240
241 /* Function to handle DTrace traps during probes. See powerpc/powerpc/trap.c */
242 int
dtrace_trap(struct trapframe * frame,u_int type)243 dtrace_trap(struct trapframe *frame, u_int type)
244 {
245 uint16_t nofault;
246
247 /*
248 * A trap can occur while DTrace executes a probe. Before
249 * executing the probe, DTrace blocks re-scheduling and sets
250 * a flag in its per-cpu flags to indicate that it doesn't
251 * want to fault. On returning from the probe, the no-fault
252 * flag is cleared and finally re-scheduling is enabled.
253 *
254 * Check if DTrace has enabled 'no-fault' mode:
255 */
256 sched_pin();
257 nofault = cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT;
258 sched_unpin();
259 if (nofault) {
260 KASSERT((frame->srr1 & PSL_EE) == 0, ("interrupts enabled"));
261 /*
262 * There are only a couple of trap types that are expected.
263 * All the rest will be handled in the usual way.
264 */
265 switch (type) {
266 /* Page fault. */
267 case EXC_DSI:
268 case EXC_DSE:
269 /* Flag a bad address. */
270 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR;
271 cpu_core[curcpu].cpuc_dtrace_illval = frame->dar;
272
273 /*
274 * Offset the instruction pointer to the instruction
275 * following the one causing the fault.
276 */
277 frame->srr0 += sizeof(int);
278 return (1);
279 case EXC_ISI:
280 case EXC_ISE:
281 /* Flag a bad address. */
282 cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR;
283 cpu_core[curcpu].cpuc_dtrace_illval = frame->srr0;
284
285 /*
286 * Offset the instruction pointer to the instruction
287 * following the one causing the fault.
288 */
289 frame->srr0 += sizeof(int);
290 return (1);
291 default:
292 /* Handle all other traps in the usual way. */
293 break;
294 }
295 }
296
297 /* Handle the trap in the usual way. */
298 return (0);
299 }
300
301 void
dtrace_probe_error(dtrace_state_t * state,dtrace_epid_t epid,int which,int fault,int fltoffs,uintptr_t illval)302 dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which,
303 int fault, int fltoffs, uintptr_t illval)
304 {
305
306 dtrace_probe(dtrace_probeid_error, (uint64_t)(uintptr_t)state,
307 (uintptr_t)epid,
308 (uintptr_t)which, (uintptr_t)fault, (uintptr_t)fltoffs);
309 }
310
311 static int
dtrace_invop_start(struct trapframe * frame)312 dtrace_invop_start(struct trapframe *frame)
313 {
314
315 switch (dtrace_invop(frame->srr0, frame, frame->fixreg[3])) {
316 case DTRACE_INVOP_JUMP:
317 break;
318 case DTRACE_INVOP_BCTR:
319 frame->srr0 = frame->ctr;
320 break;
321 case DTRACE_INVOP_BLR:
322 frame->srr0 = frame->lr;
323 break;
324 case DTRACE_INVOP_MFLR_R0:
325 frame->fixreg[0] = frame->lr;
326 frame->srr0 = frame->srr0 + 4;
327 break;
328 default:
329 return (-1);
330 }
331 return (0);
332 }
333
dtrace_invop_init(void)334 void dtrace_invop_init(void)
335 {
336 dtrace_invop_jump_addr = dtrace_invop_start;
337 }
338
dtrace_invop_uninit(void)339 void dtrace_invop_uninit(void)
340 {
341 dtrace_invop_jump_addr = 0;
342 }
343