dtrace_subr.c revision 328386
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 * $FreeBSD: stable/11/sys/cddl/dev/dtrace/powerpc/dtrace_subr.c 328386 2018-01-25 02:45:21Z pkelsey $
23 *
24 */
25/*
26 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
27 * Use is subject to license terms.
28 */
29
30#include <sys/cdefs.h>
31__FBSDID("$FreeBSD: stable/11/sys/cddl/dev/dtrace/powerpc/dtrace_subr.c 328386 2018-01-25 02:45:21Z pkelsey $");
32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/types.h>
36#include <sys/kernel.h>
37#include <sys/malloc.h>
38#include <sys/kmem.h>
39#include <sys/smp.h>
40#include <sys/dtrace_impl.h>
41#include <sys/dtrace_bsd.h>
42#include <machine/clock.h>
43#include <machine/frame.h>
44#include <machine/trap.h>
45#include <vm/pmap.h>
46
47#define	DELAYBRANCH(x)	((int)(x) < 0)
48
49extern dtrace_id_t	dtrace_probeid_error;
50extern int (*dtrace_invop_jump_addr)(struct trapframe *);
51
52extern void dtrace_getnanotime(struct timespec *tsp);
53
54int dtrace_invop(uintptr_t, struct trapframe *, uintptr_t);
55void dtrace_invop_init(void);
56void dtrace_invop_uninit(void);
57
58typedef struct dtrace_invop_hdlr {
59	int (*dtih_func)(uintptr_t, struct trapframe *, uintptr_t);
60	struct dtrace_invop_hdlr *dtih_next;
61} dtrace_invop_hdlr_t;
62
63dtrace_invop_hdlr_t *dtrace_invop_hdlr;
64
65int
66dtrace_invop(uintptr_t addr, struct trapframe *frame, uintptr_t arg0)
67{
68	dtrace_invop_hdlr_t *hdlr;
69	int rval;
70
71	for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next)
72		if ((rval = hdlr->dtih_func(addr, frame, arg0)) != 0)
73			return (rval);
74
75	return (0);
76}
77
78void
79dtrace_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
89void
90dtrace_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*/
118void
119dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit))
120{
121	/*
122	 * No toxic regions?
123	 */
124}
125
126void
127dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg)
128{
129	cpuset_t cpus;
130
131	if (cpu == DTRACE_CPUALL)
132		cpus = all_cpus;
133	else
134		CPU_SETOF(cpu, &cpus);
135
136	smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, func,
137			smp_no_rendezvous_barrier, arg);
138}
139
140static void
141dtrace_sync_func(void)
142{
143}
144
145void
146dtrace_sync(void)
147{
148	dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL);
149}
150
151static int64_t	tgt_cpu_tsc;
152static int64_t	hst_cpu_tsc;
153static int64_t	timebase_skew[MAXCPU];
154static uint64_t	nsec_scale;
155
156/* See below for the explanation of this macro. */
157/* This is taken from the amd64 dtrace_subr, to provide a synchronized timer
158 * between multiple processors in dtrace.  Since PowerPC Timebases can be much
159 * lower than x86, the scale shift is 26 instead of 28, allowing for a 15.63MHz
160 * timebase.
161 */
162#define SCALE_SHIFT	26
163
164static void
165dtrace_gethrtime_init_cpu(void *arg)
166{
167	uintptr_t cpu = (uintptr_t) arg;
168
169	if (cpu == curcpu)
170		tgt_cpu_tsc = mftb();
171	else
172		hst_cpu_tsc = mftb();
173}
174
175static void
176dtrace_gethrtime_init(void *arg)
177{
178	struct pcpu *pc;
179	uint64_t tb_f;
180	cpuset_t map;
181	int i;
182
183	tb_f = cpu_tickrate();
184
185	/*
186	 * The following line checks that nsec_scale calculated below
187	 * doesn't overflow 32-bit unsigned integer, so that it can multiply
188	 * another 32-bit integer without overflowing 64-bit.
189	 * Thus minimum supported Timebase frequency is 15.63MHz.
190	 */
191	KASSERT(tb_f > (NANOSEC >> (32 - SCALE_SHIFT)), ("Timebase frequency is too low"));
192
193	/*
194	 * We scale up NANOSEC/tb_f ratio to preserve as much precision
195	 * as possible.
196	 * 2^26 factor was chosen quite arbitrarily from practical
197	 * considerations:
198	 * - it supports TSC frequencies as low as 15.63MHz (see above);
199	 */
200	nsec_scale = ((uint64_t)NANOSEC << SCALE_SHIFT) / tb_f;
201
202	/* The current CPU is the reference one. */
203	sched_pin();
204	timebase_skew[curcpu] = 0;
205	CPU_FOREACH(i) {
206		if (i == curcpu)
207			continue;
208
209		pc = pcpu_find(i);
210		CPU_SETOF(PCPU_GET(cpuid), &map);
211		CPU_SET(pc->pc_cpuid, &map);
212
213		smp_rendezvous_cpus(map, NULL,
214		    dtrace_gethrtime_init_cpu,
215		    smp_no_rendezvous_barrier, (void *)(uintptr_t) i);
216
217		timebase_skew[i] = tgt_cpu_tsc - hst_cpu_tsc;
218	}
219	sched_unpin();
220}
221#ifdef EARLY_AP_STARTUP
222SYSINIT(dtrace_gethrtime_init, SI_SUB_DTRACE, SI_ORDER_ANY,
223    dtrace_gethrtime_init, NULL);
224#else
225SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init,
226    NULL);
227#endif
228
229/*
230 * DTrace needs a high resolution time function which can
231 * be called from a probe context and guaranteed not to have
232 * instrumented with probes itself.
233 *
234 * Returns nanoseconds since boot.
235 */
236uint64_t
237dtrace_gethrtime()
238{
239	uint64_t timebase;
240	uint32_t lo;
241	uint32_t hi;
242
243	/*
244	 * We split timebase value into lower and higher 32-bit halves and separately
245	 * scale them with nsec_scale, then we scale them down by 2^28
246	 * (see nsec_scale calculations) taking into account 32-bit shift of
247	 * the higher half and finally add.
248	 */
249	timebase = mftb() - timebase_skew[curcpu];
250	lo = timebase;
251	hi = timebase >> 32;
252	return (((lo * nsec_scale) >> SCALE_SHIFT) +
253	    ((hi * nsec_scale) << (32 - SCALE_SHIFT)));
254}
255
256uint64_t
257dtrace_gethrestime(void)
258{
259	struct      timespec curtime;
260
261	dtrace_getnanotime(&curtime);
262
263	return (curtime.tv_sec * 1000000000UL + curtime.tv_nsec);
264}
265
266/* Function to handle DTrace traps during probes. See powerpc/powerpc/trap.c */
267int
268dtrace_trap(struct trapframe *frame, u_int type)
269{
270
271	/*
272	 * A trap can occur while DTrace executes a probe. Before
273	 * executing the probe, DTrace blocks re-scheduling and sets
274	 * a flag in its per-cpu flags to indicate that it doesn't
275	 * want to fault. On returning from the probe, the no-fault
276	 * flag is cleared and finally re-scheduling is enabled.
277	 *
278	 * Check if DTrace has enabled 'no-fault' mode:
279	 */
280	if ((cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT) != 0) {
281		/*
282		 * There are only a couple of trap types that are expected.
283		 * All the rest will be handled in the usual way.
284		 */
285		switch (type) {
286		/* Page fault. */
287		case EXC_DSI:
288		case EXC_DSE:
289			/* Flag a bad address. */
290			cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR;
291			cpu_core[curcpu].cpuc_dtrace_illval = frame->dar;
292
293			/*
294			 * Offset the instruction pointer to the instruction
295			 * following the one causing the fault.
296			 */
297			frame->srr0 += sizeof(int);
298			return (1);
299		case EXC_ISI:
300		case EXC_ISE:
301			/* Flag a bad address. */
302			cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR;
303			cpu_core[curcpu].cpuc_dtrace_illval = frame->srr0;
304
305			/*
306			 * Offset the instruction pointer to the instruction
307			 * following the one causing the fault.
308			 */
309			frame->srr0 += sizeof(int);
310			return (1);
311		default:
312			/* Handle all other traps in the usual way. */
313			break;
314		}
315	}
316
317	/* Handle the trap in the usual way. */
318	return (0);
319}
320
321void
322dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which,
323    int fault, int fltoffs, uintptr_t illval)
324{
325
326	dtrace_probe(dtrace_probeid_error, (uint64_t)(uintptr_t)state,
327	    (uintptr_t)epid,
328	    (uintptr_t)which, (uintptr_t)fault, (uintptr_t)fltoffs);
329}
330
331static int
332dtrace_invop_start(struct trapframe *frame)
333{
334
335	switch (dtrace_invop(frame->srr0, frame, frame->fixreg[3])) {
336	case DTRACE_INVOP_JUMP:
337		break;
338	case DTRACE_INVOP_BCTR:
339		frame->srr0 = frame->ctr;
340		break;
341	case DTRACE_INVOP_BLR:
342		frame->srr0 = frame->lr;
343		break;
344	case DTRACE_INVOP_MFLR_R0:
345		frame->fixreg[0] = frame->lr;
346		frame->srr0 = frame->srr0 + 4;
347		break;
348	default:
349		return (-1);
350	}
351	return (0);
352}
353
354void dtrace_invop_init(void)
355{
356	dtrace_invop_jump_addr = dtrace_invop_start;
357}
358
359void dtrace_invop_uninit(void)
360{
361	dtrace_invop_jump_addr = 0;
362}
363