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 (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 /*
28 * Xen event provider for DTrace
29 *
30 * NOTE: This provider is PRIVATE. It is intended as a short-term solution and
31 * may disappear or be re-implemented at anytime.
32 *
33 * This provider isn't suitable as a general-purpose solution for a number of
34 * reasons. First and foremost, we rely on the Xen tracing mechanism and don't
35 * have any way to gather data other than that collected by the Xen trace
36 * buffers. Further, it does not fit into the DTrace model (see "Interacting
37 * with DTrace" below.)
38 *
39 *
40 * Tracing in Xen
41 * --------------
42 *
43 * Xen implements a tracing facility for generating and collecting execution
44 * event traces from the hypervisor. When tracing is enabled, compiled in
45 * probes record events in contiguous per-CPU trace buffers.
46 *
47 * +---------+
48 * +------+ | |
49 * | CPUn |----> | BUFFERn |
50 * +------+ | |
51 * +---------+- tbuf.va + (tbuf.size * n)
52 * : :
53 * +---------+
54 * +------+ | |
55 * | CPU1 |----> | BUFFER1 |
56 * +------+ | |
57 * +---------+- tbuf.va + tbuf.size
58 * +------+ | |
59 * | CPU0 |----> | BUFFER0 |
60 * +------+ | |
61 * +---------+- tbuf.va
62 *
63 * Each CPU buffer consists of a metadata header followed by the trace records.
64 * The metadata consists of a producer/consumer pair of pointers into the buffer
65 * that point to the next record to be written and the next record to be read
66 * respectively.
67 *
68 * A trace record can be in one of two forms, depending on if the TSC is
69 * included. The record header indicates whether or not the TSC field is
70 * present.
71 *
72 * 1. Trace record without TSC:
73 * +------------------------------------------------------------+
74 * | HEADER(uint32_t) | DATA FIELDS |
75 * +------------------------------------------------------------+
76 *
77 * 2. Trace record with TSC:
78 * +--------------------------------------------------------------------------+
79 * | HEADER(uint32_t) | TSC(uint64_t) | DATA FIELDS |
80 * +--------------------------------------------------------------------------+
81 *
82 * Where,
83 *
84 * HEADER bit field:
85 * +--------------------------------------------------------------------------+
86 * | C | NDATA | EVENT |
87 * +--------------------------------------------------------------------------+
88 * 31 30 28 27 0
89 *
90 * EVENT: Event ID.
91 * NDATA: Number of populated data fields.
92 * C: TSC included.
93 *
94 * DATA FIELDS:
95 * +--------------------------------------------------------------------------+
96 * | D1(uint32_t) | D2(uint32_t) | D3(uint32_t) | . . . | D7(uint32_t) |
97 * +--------------------------------------------------------------------------+
98 *
99 *
100 * Interacting with DTrace
101 * -----------------------
102 *
103 * Every xdt_poll_nsec nano-seconds we poll the trace buffers for data and feed
104 * each entry into dtrace_probe() with the corresponding probe ID for the event.
105 * As a result of this periodic collection implementation probe firings are
106 * asynchronous. This is the only sensible way to implement this form of
107 * provider, but because of its asynchronous nature asking things like
108 * "current CPU" and, more importantly, arbitrary questions about the context
109 * surrounding the probe firing are not meaningful. So, consumers should not
110 * attempt to infer anything beyond what is supplied via the probe arguments.
111 */
112
113 #include <sys/xpv_user.h>
114
115 #include <sys/types.h>
116 #include <sys/sysmacros.h>
117 #include <sys/modctl.h>
118 #include <sys/sunddi.h>
119 #include <sys/ddi.h>
120 #include <sys/conf.h>
121 #include <sys/devops.h>
122 #include <sys/stat.h>
123 #include <sys/cmn_err.h>
124 #include <sys/dtrace.h>
125 #include <sys/sdt.h>
126 #include <sys/cyclic.h>
127 #include <vm/seg_kmem.h>
128 #include <vm/hat_i86.h>
129
130 #include <sys/hypervisor.h>
131 #include <xen/public/trace.h>
132 #include <xen/public/sched.h>
133
134 #define XDT_POLL_DEFAULT 100000000 /* default poll interval (ns) */
135 #define XDT_POLL_MIN 10000000 /* min poll interval (ns) */
136 #define XDT_TBUF_RETRY 50 /* tbuf disable retry count */
137
138 /*
139 * The domid must match IDLE_DOMAIN_ID in xen.hg/xen/include/xen/sched.h
140 * in the xVM gate.
141 */
142 #define IS_IDLE_DOM(domid) (domid == 0x7FFFU)
143
144 /* Macros to extract the domid and cpuid from a HVM trace data field */
145 #define HVM_DOMID(d) (d >> 16)
146 #define HVM_VCPUID(d) (d & 0xFFFF)
147
148 /* Flags for shadow page table events */
149 #define SH_GUEST_32 0x000
150 #define SH_GUEST_PAE 0x100
151 #define SH_GUEST_64 0x200
152
153 #define XDT_PROBE5(event, arg0, arg1, arg2, arg3, arg4) { \
154 dtrace_id_t id = xdt_probemap[event]; \
155 if (id) \
156 dtrace_probe(id, arg0, arg1, arg2, arg3, arg4); \
157 } \
158
159 #define XDT_PROBE4(event, arg0, arg1, arg2, arg3) \
160 XDT_PROBE5(event, arg0, arg1, arg2, arg3, 0)
161
162 #define XDT_PROBE3(event, arg0, arg1, arg2) \
163 XDT_PROBE5(event, arg0, arg1, arg2, 0, 0)
164
165 #define XDT_PROBE2(event, arg0, arg1) \
166 XDT_PROBE5(event, arg0, arg1, 0, 0, 0)
167
168 #define XDT_PROBE1(event, arg0) \
169 XDT_PROBE5(event, arg0, 0, 0, 0, 0)
170
171 #define XDT_PROBE0(event) \
172 XDT_PROBE5(event, 0, 0, 0, 0, 0)
173
174 /* Probe classes */
175 #define XDT_SCHED 0
176 #define XDT_MEM 1
177 #define XDT_HVM 2
178 #define XDT_GEN 3
179 #define XDT_PV 4
180 #define XDT_SHADOW 5
181 #define XDT_PM 6
182 #define XDT_NCLASSES 7
183
184 /* Probe events */
185 #define XDT_EVT_INVALID (-(int)1)
186 #define XDT_SCHED_OFF_CPU 0
187 #define XDT_SCHED_ON_CPU 1
188 #define XDT_SCHED_IDLE_OFF_CPU 2
189 #define XDT_SCHED_IDLE_ON_CPU 3
190 #define XDT_SCHED_BLOCK 4
191 #define XDT_SCHED_SLEEP 5
192 #define XDT_SCHED_WAKE 6
193 #define XDT_SCHED_YIELD 7
194 #define XDT_SCHED_SHUTDOWN_POWEROFF 8
195 #define XDT_SCHED_SHUTDOWN_REBOOT 9
196 #define XDT_SCHED_SHUTDOWN_SUSPEND 10
197 #define XDT_SCHED_SHUTDOWN_CRASH 11
198 #define XDT_MEM_PAGE_GRANT_MAP 12
199 #define XDT_MEM_PAGE_GRANT_UNMAP 13
200 #define XDT_MEM_PAGE_GRANT_TRANSFER 14
201 #define XDT_HVM_VMENTRY 15
202 #define XDT_HVM_VMEXIT 16
203 #define XDT_TRC_LOST_RECORDS 17
204 #define XDT_SCHED_ADD_VCPU 18
205 #define XDT_SCHED_REM_VCPU 19 /* unused */
206 #define XDT_SCHED_CTL 20 /* unused */
207 #define XDT_SCHED_ADJDOM 21
208 #define XDT_SCHED_S_TIMER_FN 22 /* unused */
209 #define XDT_SCHED_T_TIMER_FN 23 /* unused */
210 #define XDT_SCHED_DOM_TIMER_FN 24 /* unused */
211 #define XDT_PV_HYPERCALL 25
212 #define XDT_PV_TRAP 26
213 #define XDT_PV_PAGE_FAULT 27
214 #define XDT_PV_FORCED_INVALID_OP 28
215 #define XDT_PV_EMULATE_PRIVOP 29
216 #define XDT_PV_EMULATE_4GB 30 /* unused (32-bit HV only ) */
217 #define XDT_PV_MATH_STATE_RESTORE 31
218 #define XDT_PV_PAGING_FIXUP 32
219 #define XDT_PV_DT_MAPPING_FAULT 33
220 #define XDT_PV_PTWR_EMULATION 34
221 #define XDT_HVM_PF_XEN 35
222 #define XDT_HVM_PF_INJECT 36
223 #define XDT_HVM_EXC_INJECT 37
224 #define XDT_HVM_VIRQ_INJECT 38
225 #define XDT_HVM_VIRQ_REINJECT 39
226 #define XDT_HVM_IO_READ 40 /* unused */
227 #define XDT_HVM_IO_WRITE 41 /* unused */
228 #define XDT_HVM_CR_READ 42
229 #define XDT_HVM_CR_WRITE 43
230 #define XDT_HVM_DR_READ 44 /* unused */
231 #define XDT_HVM_DR_WRITE 45 /* unused */
232 #define XDT_HVM_MSR_READ 46
233 #define XDT_HVM_MSR_WRITE 47
234 #define XDT_HVM_CPUID 48
235 #define XDT_HVM_INTR 49
236 #define XDT_HVM_INTR_WINDOW 50
237 #define XDT_HVM_NMI 51
238 #define XDT_HVM_SMI 52
239 #define XDT_HVM_VMMCALL 53
240 #define XDT_HVM_HLT 54
241 #define XDT_HVM_INVLPG 55
242 #define XDT_HVM_MCE 56
243 #define XDT_HVM_IOPORT_READ 57
244 #define XDT_HVM_IOPORT_WRITE 58
245 #define XDT_HVM_CLTS 59
246 #define XDT_HVM_LMSW 60
247 #define XDT_HVM_IOMEM_READ 61
248 #define XDT_HVM_IOMEM_WRITE 62
249 #define XDT_SHADOW_NOT_SHADOW 63
250 #define XDT_SHADOW_FAST_PROPAGATE 64
251 #define XDT_SHADOW_FAST_MMIO 65
252 #define XDT_SHADOW_FALSE_FAST_PATH 66
253 #define XDT_SHADOW_MMIO 67
254 #define XDT_SHADOW_FIXUP 68
255 #define XDT_SHADOW_DOMF_DYING 69
256 #define XDT_SHADOW_EMULATE 70
257 #define XDT_SHADOW_EMULATE_UNSHADOW_USER 71
258 #define XDT_SHADOW_EMULATE_UNSHADOW_EVTINJ 72
259 #define XDT_SHADOW_EMULATE_UNSHADOW_UNHANDLED 73
260 #define XDT_SHADOW_WRMAP_BF 74
261 #define XDT_SHADOW_PREALLOC_UNPIN 75
262 #define XDT_SHADOW_RESYNC_FULL 76
263 #define XDT_SHADOW_RESYNC_ONLY 77
264 #define XDT_PM_FREQ_CHANGE 78
265 #define XDT_PM_IDLE_ENTRY 79
266 #define XDT_PM_IDLE_EXIT 80
267 #define XDT_SCHED_RUNSTATE_CHANGE 81
268 #define XDT_SCHED_CONTINUE_RUNNING 82
269 #define XDT_NEVENTS 83
270
271 typedef struct {
272 const char *pr_mod; /* probe module */
273 const char *pr_name; /* probe name */
274 int evt_id; /* event id */
275 uint_t class; /* probe class */
276 } xdt_probe_t;
277
278 typedef struct {
279 uint32_t trc_mask; /* trace mask */
280 uint32_t cnt; /* num enabled probes in class */
281 } xdt_classinfo_t;
282
283 typedef struct {
284 ulong_t prev_domid; /* previous dom executed */
285 ulong_t prev_vcpuid; /* previous vcpu executed */
286 ulong_t prev_ctime; /* time spent on cpu */
287 ulong_t next_domid; /* next dom to be scheduled */
288 ulong_t next_vcpuid; /* next vcpu to be scheduled */
289 ulong_t next_wtime; /* time spent waiting to get on cpu */
290 ulong_t next_ts; /* allocated time slice */
291 ulong_t cur_domid; /* current dom */
292 ulong_t cur_vcpuid; /* current vcpuid */
293 int curinfo_valid; /* info is valid */
294 } xdt_schedinfo_t;
295
296 static struct {
297 uint_t cnt; /* total num of trace buffers */
298 size_t size; /* size of each cpu buffer */
299 mfn_t start_mfn; /* starting mfn of buffers */
300 caddr_t va; /* va buffers are mapped into */
301
302 /* per-cpu buffers */
303 struct t_buf **meta; /* buffer metadata */
304 struct t_rec **data; /* buffer data records */
305
306 /* statistics */
307 uint64_t stat_dropped_recs; /* records dropped */
308 uint64_t stat_spurious_cpu; /* recs with garbage cpuids */
309 uint64_t stat_spurious_switch; /* inconsistent vcpu switches */
310 uint64_t stat_unknown_shutdown; /* unknown shutdown code */
311 uint64_t stat_unknown_recs; /* unknown records */
312 } tbuf;
313
314 static size_t tbuf_data_size;
315
316 static char *xdt_stats[] = {
317 "dropped_recs",
318 };
319
320 /*
321 * Tunable variables
322 *
323 * The following may be tuned by adding a line to /etc/system that
324 * includes both the name of the module ("xdt") and the name of the variable.
325 * For example:
326 * set xdt:xdt_tbuf_pages = 40
327 */
328 uint_t xdt_tbuf_pages = 20; /* pages to alloc per-cpu buf */
329
330 /*
331 * The following may be tuned by adding a line to
332 * /platform/i86xpv/kernel/drv/xdt.conf.
333 * For example:
334 * xdt_poll_nsec = 200000000;
335 */
336 static hrtime_t xdt_poll_nsec; /* trace buffer poll interval */
337
338 /*
339 * Another tunable variable: the maximum number of records to process
340 * in one scan. If it is 0 (e.g. not set in /etc/system), it will
341 * be set to ncpu * (bufsize / max_rec_size).
342 *
343 * Having an upper limit avoids a situation where the scan would loop
344 * endlessly in case the hypervisor adds records quicker than we
345 * can process them. It's better to drop records than to loop, obviously.
346 */
347 uint_t xdt_max_recs = 0;
348
349 /*
350 * Internal variables
351 */
352 static dev_info_t *xdt_devi;
353 static dtrace_provider_id_t xdt_id;
354 static uint_t xdt_ncpus; /* total number of phys CPUs */
355 static uint32_t cur_trace_mask; /* current trace mask */
356 static xdt_schedinfo_t *xdt_cpu_schedinfo; /* per-cpu sched info */
357 dtrace_id_t xdt_probemap[XDT_NEVENTS]; /* map of enabled probes */
358 dtrace_id_t xdt_prid[XDT_NEVENTS]; /* IDs of registered events */
359 static cyclic_id_t xdt_cyclic = CYCLIC_NONE;
360 static kstat_t *xdt_kstats;
361 static xdt_classinfo_t xdt_classinfo[XDT_NCLASSES];
362
363 /*
364 * These provide context when probes fire. They can be accessed
365 * from xdt dtrace probe (as `xdt_curdom, etc). It's ok for these
366 * to be global, and not per-cpu, as probes are run strictly in sequence
367 * as the trace buffers are
368 */
369 uint_t xdt_curdom, xdt_curvcpu, xdt_curpcpu;
370 uint64_t xdt_timestamp;
371
372 static xdt_probe_t xdt_probe[] = {
373 /* Sched probes */
374 { "sched", "off-cpu", XDT_SCHED_OFF_CPU, XDT_SCHED },
375 { "sched", "on-cpu", XDT_SCHED_ON_CPU, XDT_SCHED },
376 { "sched", "idle-off-cpu", XDT_SCHED_IDLE_OFF_CPU, XDT_SCHED },
377 { "sched", "idle-on-cpu", XDT_SCHED_IDLE_ON_CPU, XDT_SCHED },
378 { "sched", "block", XDT_SCHED_BLOCK, XDT_SCHED },
379 { "sched", "sleep", XDT_SCHED_SLEEP, XDT_SCHED },
380 { "sched", "wake", XDT_SCHED_WAKE, XDT_SCHED },
381 { "sched", "yield", XDT_SCHED_YIELD, XDT_SCHED },
382 { "sched", "shutdown-poweroff", XDT_SCHED_SHUTDOWN_POWEROFF,
383 XDT_SCHED },
384 { "sched", "shutdown-reboot", XDT_SCHED_SHUTDOWN_REBOOT, XDT_SCHED },
385 { "sched", "shutdown-suspend", XDT_SCHED_SHUTDOWN_SUSPEND, XDT_SCHED },
386 { "sched", "shutdown-crash", XDT_SCHED_SHUTDOWN_CRASH, XDT_SCHED },
387 { "sched", "add", XDT_SCHED_ADD_VCPU, XDT_SCHED },
388 { "sched", "runstate-change", XDT_SCHED_RUNSTATE_CHANGE, XDT_SCHED },
389 { "sched", "continue-running", XDT_SCHED_CONTINUE_RUNNING, XDT_SCHED },
390
391 /* Memory probes */
392 { "mem", "page-grant-map", XDT_MEM_PAGE_GRANT_MAP, XDT_MEM },
393 { "mem", "page-grant-unmap", XDT_MEM_PAGE_GRANT_UNMAP, XDT_MEM },
394 { "mem", "page-grant-transfer", XDT_MEM_PAGE_GRANT_TRANSFER, XDT_MEM },
395
396 {"pv", "hypercall", XDT_PV_HYPERCALL, XDT_PV },
397 {"pv", "trap", XDT_PV_TRAP, XDT_PV },
398 {"pv", "page-fault", XDT_PV_PAGE_FAULT, XDT_PV },
399 {"pv", "forced-invalid-op", XDT_PV_FORCED_INVALID_OP, XDT_PV },
400 {"pv", "emulate-priv-op", XDT_PV_EMULATE_PRIVOP, XDT_PV },
401 {"pv", "math-state-restore", XDT_PV_MATH_STATE_RESTORE, XDT_PV },
402 {"pv", "paging-fixup", XDT_PV_PAGING_FIXUP, XDT_PV },
403 {"pv", "dt-mapping-fault", XDT_PV_DT_MAPPING_FAULT, XDT_PV },
404 {"pv", "pte-write-emul", XDT_PV_PTWR_EMULATION, XDT_PV },
405
406 /* HVM probes */
407 { "hvm", "vmentry", XDT_HVM_VMENTRY, XDT_HVM },
408 { "hvm", "vmexit", XDT_HVM_VMEXIT, XDT_HVM },
409 { "hvm", "pagefault-xen", XDT_HVM_PF_XEN, XDT_HVM },
410 { "hvm", "pagefault-inject", XDT_HVM_PF_INJECT, XDT_HVM },
411 { "hvm", "exception-inject", XDT_HVM_EXC_INJECT, XDT_HVM },
412 { "hvm", "virq-inject", XDT_HVM_VIRQ_INJECT, XDT_HVM },
413 { "hvm", "cr-read", XDT_HVM_CR_READ, XDT_HVM },
414 { "hvm", "cr-write", XDT_HVM_CR_WRITE, XDT_HVM },
415 { "hvm", "msr-read", XDT_HVM_MSR_READ, XDT_HVM },
416 { "hvm", "msr-write", XDT_HVM_MSR_WRITE, XDT_HVM },
417 { "hvm", "cpuid", XDT_HVM_CPUID, XDT_HVM },
418 { "hvm", "intr", XDT_HVM_INTR, XDT_HVM },
419 { "hvm", "intr-window", XDT_HVM_INTR_WINDOW, XDT_HVM },
420 { "hvm", "nmi", XDT_HVM_NMI, XDT_HVM },
421 { "hvm", "smi", XDT_HVM_SMI, XDT_HVM },
422 { "hvm", "vmmcall", XDT_HVM_VMMCALL, XDT_HVM },
423 { "hvm", "hlt", XDT_HVM_HLT, XDT_HVM },
424 { "hvm", "invlpg", XDT_HVM_INVLPG, XDT_HVM },
425 { "hvm", "mce", XDT_HVM_MCE, XDT_HVM },
426 { "hvm", "pio-read", XDT_HVM_IOPORT_READ, XDT_HVM },
427 { "hvm", "pio-write", XDT_HVM_IOPORT_WRITE, XDT_HVM },
428 { "hvm", "mmio-read", XDT_HVM_IOMEM_READ, XDT_HVM },
429 { "hvm", "mmio-write", XDT_HVM_IOMEM_WRITE, XDT_HVM },
430 { "hvm", "clts", XDT_HVM_CLTS, XDT_HVM },
431 { "hvm", "lmsw", XDT_HVM_LMSW, XDT_HVM },
432
433 { "shadow", "fault-not-shadow", XDT_SHADOW_NOT_SHADOW, XDT_SHADOW },
434 { "shadow", "fast-propagate", XDT_SHADOW_FAST_PROPAGATE, XDT_SHADOW },
435 { "shadow", "fast-mmio", XDT_SHADOW_FAST_MMIO, XDT_SHADOW },
436 { "shadow", "false-fast-path", XDT_SHADOW_FALSE_FAST_PATH,
437 XDT_SHADOW },
438 { "shadow", "mmio", XDT_SHADOW_MMIO, XDT_SHADOW },
439 { "shadow", "fixup", XDT_SHADOW_FIXUP, XDT_SHADOW },
440 { "shadow", "domf-dying", XDT_SHADOW_DOMF_DYING, XDT_SHADOW },
441 { "shadow", "emulate", XDT_SHADOW_EMULATE, XDT_SHADOW },
442 { "shadow", "emulate-unshadow-user", XDT_SHADOW_EMULATE_UNSHADOW_USER,
443 XDT_SHADOW },
444 { "shadow", "emulate-unshadow-evtinj",
445 XDT_SHADOW_EMULATE_UNSHADOW_EVTINJ, XDT_SHADOW },
446 { "shadow", "emulate-unshadow-unhandled",
447 XDT_SHADOW_EMULATE_UNSHADOW_UNHANDLED, XDT_SHADOW },
448 { "shadow", "wrmap-bf", XDT_SHADOW_WRMAP_BF, XDT_SHADOW },
449 { "shadow", "prealloc-unpin", XDT_SHADOW_PREALLOC_UNPIN, XDT_SHADOW },
450 { "shadow", "resync-full", XDT_SHADOW_RESYNC_FULL, XDT_SHADOW },
451 { "shadow", "resync-only", XDT_SHADOW_RESYNC_ONLY, XDT_SHADOW },
452
453 { "pm", "freq-change", XDT_PM_FREQ_CHANGE, XDT_PM },
454 { "pm", "idle-entry", XDT_PM_IDLE_ENTRY, XDT_PM },
455 { "pm", "idle-exit", XDT_PM_IDLE_EXIT, XDT_PM },
456
457 /* Trace buffer related probes */
458 { "trace", "records-lost", XDT_TRC_LOST_RECORDS, XDT_GEN },
459
460 { NULL }
461 };
462
463 static inline uint32_t
464 xdt_nr_active_probes()
465 {
466 int i;
467 uint32_t tot = 0;
468
469 for (i = 0; i < XDT_NCLASSES; i++)
470 tot += xdt_classinfo[i].cnt;
471
472 return (tot);
473 }
474
475 static void
476 xdt_init_trace_masks(void)
477 {
478 xdt_classinfo[XDT_SCHED].trc_mask = TRC_SCHED;
479 xdt_classinfo[XDT_MEM].trc_mask = TRC_MEM;
480 xdt_classinfo[XDT_HVM].trc_mask = TRC_HVM;
481 xdt_classinfo[XDT_GEN].trc_mask = TRC_GEN;
482 xdt_classinfo[XDT_PV].trc_mask = TRC_PV;
483 xdt_classinfo[XDT_SHADOW].trc_mask = TRC_SHADOW;
484 xdt_classinfo[XDT_PM].trc_mask = TRC_PM;
485 }
486
487 static int
488 xdt_kstat_update(kstat_t *ksp, int flag)
489 {
490 kstat_named_t *knp;
491
492 if (flag != KSTAT_READ)
493 return (EACCES);
494
495 knp = ksp->ks_data;
496
497 /*
498 * Assignment order should match that of the names in
499 * xdt_stats.
500 */
501 (knp++)->value.ui64 = tbuf.stat_dropped_recs;
502
503 return (0);
504 }
505
506 static void
507 xdt_kstat_init(void)
508 {
509 int nstats = sizeof (xdt_stats) / sizeof (xdt_stats[0]);
510 char **cp = xdt_stats;
511 kstat_named_t *knp;
512
513 if ((xdt_kstats = kstat_create("xdt", 0, "trace_statistics", "misc",
514 KSTAT_TYPE_NAMED, nstats, 0)) == NULL)
515 return;
516
517 xdt_kstats->ks_update = xdt_kstat_update;
518
519 knp = xdt_kstats->ks_data;
520 while (nstats > 0) {
521 kstat_named_init(knp, *cp, KSTAT_DATA_UINT64);
522 knp++;
523 cp++;
524 nstats--;
525 }
526
527 kstat_install(xdt_kstats);
528 }
529
530 static int
531 xdt_sysctl_tbuf(xen_sysctl_tbuf_op_t *tbuf_op)
532 {
533 xen_sysctl_t op;
534 int xerr;
535
536 op.cmd = XEN_SYSCTL_tbuf_op;
537 op.interface_version = XEN_SYSCTL_INTERFACE_VERSION;
538 op.u.tbuf_op = *tbuf_op;
539
540 if ((xerr = HYPERVISOR_sysctl(&op)) != 0)
541 return (xen_xlate_errcode(xerr));
542
543 *tbuf_op = op.u.tbuf_op;
544 return (0);
545 }
546
547 static int
548 xdt_map_trace_buffers(mfn_t mfn, caddr_t va, size_t len)
549 {
550 x86pte_t pte;
551 caddr_t const sva = va;
552 caddr_t const eva = va + len;
553 int xerr;
554
555 ASSERT(mfn != MFN_INVALID);
556 ASSERT(va != NULL);
557 ASSERT(IS_PAGEALIGNED(len));
558
559 for (; va < eva; va += MMU_PAGESIZE) {
560 /*
561 * Ask the HAT to load a throwaway mapping to page zero, then
562 * overwrite it with the hypervisor mapping. It gets removed
563 * later via hat_unload().
564 */
565 hat_devload(kas.a_hat, va, MMU_PAGESIZE, (pfn_t)0,
566 PROT_READ | HAT_UNORDERED_OK,
567 HAT_LOAD_NOCONSIST | HAT_LOAD);
568
569 pte = mmu_ptob((x86pte_t)mfn) | PT_VALID | PT_USER
570 | PT_FOREIGN | PT_WRITABLE;
571
572 xerr = HYPERVISOR_update_va_mapping_otherdomain((ulong_t)va,
573 pte, UVMF_INVLPG | UVMF_LOCAL, DOMID_XEN);
574
575 if (xerr != 0) {
576 /* unmap pages loaded so far */
577 size_t ulen = (uintptr_t)(va + MMU_PAGESIZE) -
578 (uintptr_t)sva;
579 hat_unload(kas.a_hat, sva, ulen, HAT_UNLOAD_UNMAP);
580 return (xen_xlate_errcode(xerr));
581 }
582
583 mfn++;
584 }
585
586 return (0);
587 }
588
589 static int
590 xdt_attach_trace_buffers(void)
591 {
592 xen_sysctl_tbuf_op_t tbuf_op;
593 size_t len;
594 int err;
595 uint_t i;
596
597 /*
598 * Xen does not support trace buffer re-sizing. If the buffers
599 * have already been allocated we just use them as is.
600 */
601 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_get_info;
602 if ((err = xdt_sysctl_tbuf(&tbuf_op)) != 0)
603 return (err);
604
605 if (tbuf_op.size == 0) {
606 /* set trace buffer size */
607 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_set_size;
608 tbuf_op.size = xdt_tbuf_pages;
609 (void) xdt_sysctl_tbuf(&tbuf_op);
610
611 /* get trace buffer info */
612 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_get_info;
613 if ((err = xdt_sysctl_tbuf(&tbuf_op)) != 0)
614 return (err);
615
616 if (tbuf_op.size == 0) {
617 cmn_err(CE_NOTE, "Couldn't allocate trace buffers.");
618 return (ENOBUFS);
619 }
620 }
621
622 tbuf.size = tbuf_op.size;
623 tbuf.start_mfn = (mfn_t)tbuf_op.buffer_mfn;
624 tbuf.cnt = xdt_ncpus;
625
626 ASSERT(tbuf.start_mfn != MFN_INVALID);
627 ASSERT(tbuf.cnt > 0);
628
629 len = tbuf.size * tbuf.cnt;
630 tbuf.va = vmem_alloc(heap_arena, len, VM_SLEEP);
631
632 if ((err = xdt_map_trace_buffers(tbuf.start_mfn, tbuf.va, len)) != 0) {
633 vmem_free(heap_arena, tbuf.va, len);
634 tbuf.va = NULL;
635 return (err);
636 }
637
638 tbuf.meta = (struct t_buf **)kmem_alloc(tbuf.cnt * sizeof (*tbuf.meta),
639 KM_SLEEP);
640 tbuf.data = (struct t_rec **)kmem_alloc(tbuf.cnt * sizeof (*tbuf.data),
641 KM_SLEEP);
642
643 for (i = 0; i < tbuf.cnt; i++) {
644 void *cpu_buf = (void *)(tbuf.va + (tbuf.size * i));
645 tbuf.meta[i] = cpu_buf;
646 tbuf.data[i] = (struct t_rec *)((uintptr_t)cpu_buf +
647 sizeof (struct t_buf));
648
649 /* throw away stale trace records */
650 tbuf.meta[i]->cons = tbuf.meta[i]->prod;
651 }
652
653 tbuf_data_size = tbuf.size - sizeof (struct t_buf);
654 if (xdt_max_recs == 0)
655 xdt_max_recs = (xdt_ncpus * tbuf_data_size)
656 / sizeof (struct t_rec);
657
658 return (0);
659 }
660
661 static void
662 xdt_detach_trace_buffers(void)
663 {
664 size_t len = tbuf.size * tbuf.cnt;
665
666 ASSERT(tbuf.va != NULL);
667
668 hat_unload(kas.a_hat, tbuf.va, len,
669 HAT_UNLOAD_UNMAP | HAT_UNLOAD_UNLOCK);
670 vmem_free(heap_arena, tbuf.va, len);
671 kmem_free(tbuf.meta, tbuf.cnt * sizeof (*tbuf.meta));
672 kmem_free(tbuf.data, tbuf.cnt * sizeof (*tbuf.data));
673 }
674
675 static void
676 xdt_update_sched_context(uint_t cpuid, uint_t dom, uint_t vcpu)
677 {
678 xdt_schedinfo_t *sp = &xdt_cpu_schedinfo[cpuid];
679
680 sp->cur_domid = dom;
681 sp->cur_vcpuid = vcpu;
682 sp->curinfo_valid = 1;
683 }
684
685 static void
686 xdt_update_domain_context(uint_t dom, uint_t vcpu)
687 {
688 xdt_curdom = dom;
689 xdt_curvcpu = vcpu;
690 }
691
692 static size_t
693 xdt_process_rec(uint_t cpuid, struct t_rec *rec)
694 {
695 xdt_schedinfo_t *sp = &xdt_cpu_schedinfo[cpuid];
696 uint_t dom, vcpu;
697 int eid;
698 uint32_t *data;
699 uint64_t tsc, addr64, rip64, val64, pte64;
700 size_t rec_size;
701
702 ASSERT(rec != NULL);
703 ASSERT(xdt_ncpus == xpv_nr_phys_cpus());
704
705 if (cpuid >= xdt_ncpus) {
706 tbuf.stat_spurious_cpu++;
707 goto done;
708 }
709
710 /*
711 * If our current state isn't valid, and if this is not
712 * an event that will update our state, skip it.
713 */
714
715 if (!sp->curinfo_valid &&
716 rec->event != TRC_SCHED_SWITCH &&
717 rec->event != TRC_LOST_RECORDS)
718 goto done;
719
720 if (rec->cycles_included) {
721 data = rec->u.cycles.extra_u32;
722 tsc = (((uint64_t)rec->u.cycles.cycles_hi) << 32)
723 | rec->u.cycles.cycles_lo;
724 } else {
725 data = rec->u.nocycles.extra_u32;
726 tsc = 0;
727 }
728
729 xdt_timestamp = tsc;
730
731 switch (rec->event) {
732 /*
733 * Sched probes
734 */
735 case TRC_SCHED_SWITCH_INFPREV:
736 /*
737 * Info on vCPU being de-scheduled
738 *
739 * data[0] = prev domid
740 * data[1] = time spent on pcpu
741 */
742 sp->prev_domid = data[0];
743 sp->prev_ctime = data[1];
744 break;
745
746 case TRC_SCHED_SWITCH_INFNEXT:
747 /*
748 * Info on next vCPU to be scheduled
749 *
750 * data[0] = next domid
751 * data[1] = time spent waiting to get on cpu
752 * data[2] = time slice
753 */
754 sp->next_domid = data[0];
755 sp->next_wtime = data[1];
756 sp->next_ts = data[2];
757 break;
758
759 case TRC_SCHED_SWITCH:
760 /*
761 * vCPU switch
762 *
763 * data[0] = prev domid
764 * data[1] = prev vcpuid
765 * data[2] = next domid
766 * data[3] = next vcpuid
767 */
768
769 /*
770 * Provide valid context for this probe if there
771 * wasn't one.
772 */
773 if (!sp->curinfo_valid)
774 xdt_update_domain_context(data[0], data[1]);
775
776 xdt_update_sched_context(cpuid, data[0], data[1]);
777
778 if (data[0] != sp->prev_domid &&
779 data[2] != sp->next_domid) {
780 /* prev and next info don't match doms being sched'd */
781 tbuf.stat_spurious_switch++;
782 goto switchdone;
783 }
784
785 sp->prev_vcpuid = data[1];
786 sp->next_vcpuid = data[3];
787
788 XDT_PROBE3(IS_IDLE_DOM(sp->prev_domid)?
789 XDT_SCHED_IDLE_OFF_CPU:XDT_SCHED_OFF_CPU,
790 sp->prev_domid, sp->prev_vcpuid, sp->prev_ctime);
791
792 XDT_PROBE4(IS_IDLE_DOM(sp->next_domid)?
793 XDT_SCHED_IDLE_ON_CPU:XDT_SCHED_ON_CPU,
794 sp->next_domid, sp->next_vcpuid, sp->next_wtime,
795 sp->next_ts);
796 switchdone:
797 xdt_update_sched_context(cpuid, data[2], data[3]);
798 xdt_update_domain_context(data[2], data[3]);
799
800 break;
801
802 case TRC_SCHED_BLOCK:
803 /*
804 * vCPU blocked
805 *
806 * data[0] = domid
807 * data[1] = vcpuid
808 */
809 XDT_PROBE2(XDT_SCHED_BLOCK, data[0], data[1]);
810 break;
811
812 case TRC_SCHED_SLEEP:
813 /*
814 * Put vCPU to sleep
815 *
816 * data[0] = domid
817 * data[1] = vcpuid
818 */
819 XDT_PROBE2(XDT_SCHED_SLEEP, data[0], data[1]);
820 break;
821
822 case TRC_SCHED_WAKE:
823 /*
824 * Wake up vCPU
825 *
826 * data[0] = domid
827 * data[1] = vcpuid
828 */
829 XDT_PROBE2(XDT_SCHED_WAKE, data[0], data[1]);
830 break;
831
832 case TRC_SCHED_YIELD:
833 /*
834 * vCPU yielded
835 *
836 * data[0] = domid
837 * data[1] = vcpuid
838 */
839 XDT_PROBE2(XDT_SCHED_YIELD, data[0], data[1]);
840 break;
841
842 case TRC_SCHED_SHUTDOWN:
843 /*
844 * Guest shutting down
845 *
846 * data[0] = domid
847 * data[1] = initiating vcpu
848 * data[2] = shutdown code
849 */
850 switch (data[2]) {
851 case SHUTDOWN_poweroff:
852 eid = XDT_SCHED_SHUTDOWN_POWEROFF;
853 break;
854 case SHUTDOWN_reboot:
855 eid = XDT_SCHED_SHUTDOWN_REBOOT;
856 break;
857 case SHUTDOWN_suspend:
858 eid = XDT_SCHED_SHUTDOWN_SUSPEND;
859 break;
860 case SHUTDOWN_crash:
861 eid = XDT_SCHED_SHUTDOWN_CRASH;
862 break;
863 default:
864 tbuf.stat_unknown_shutdown++;
865 goto done;
866 }
867
868 XDT_PROBE2(eid, data[0], data[1]);
869 break;
870
871 case TRC_SCHED_DOM_REM:
872 case TRC_SCHED_CTL:
873 case TRC_SCHED_S_TIMER_FN:
874 case TRC_SCHED_T_TIMER_FN:
875 case TRC_SCHED_DOM_TIMER_FN:
876 /* unused */
877 break;
878 case TRC_SCHED_DOM_ADD:
879 /*
880 * Add vcpu to a guest.
881 *
882 * data[0] = domid
883 * data[1] = vcpu
884 */
885 XDT_PROBE2(XDT_SCHED_ADD_VCPU, data[0], data[1]);
886 break;
887 case TRC_SCHED_ADJDOM:
888 /*
889 * Scheduling parameters for a guest
890 * were modified.
891 *
892 * data[0] = domid;
893 */
894 XDT_PROBE1(XDT_SCHED_ADJDOM, data[1]);
895 break;
896 case TRC_SCHED_RUNSTATE_CHANGE:
897 /*
898 * Runstate change for a VCPU.
899 *
900 * data[0] = (domain << 16) | vcpu;
901 * data[1] = oldstate;
902 * data[2] = newstate;
903 */
904 XDT_PROBE4(XDT_SCHED_RUNSTATE_CHANGE, data[0] >> 16,
905 data[0] & 0xffff, data[1], data[2]);
906 break;
907 case TRC_SCHED_CONTINUE_RUNNING:
908 /*
909 * VCPU is back on a physical CPU that it previously
910 * was also running this VCPU.
911 *
912 * data[0] = (domain << 16) | vcpu;
913 */
914 XDT_PROBE2(XDT_SCHED_CONTINUE_RUNNING, data[0] >> 16,
915 data[0] & 0xffff);
916 break;
917 /*
918 * Mem probes
919 */
920 case TRC_MEM_PAGE_GRANT_MAP:
921 /*
922 * Guest mapped page grant
923 *
924 * data[0] = target domid
925 */
926 XDT_PROBE1(XDT_MEM_PAGE_GRANT_MAP, data[0]);
927 break;
928
929 case TRC_MEM_PAGE_GRANT_UNMAP:
930 /*
931 * Guest unmapped page grant
932 *
933 * data[0] = target domid
934 */
935 XDT_PROBE1(XDT_MEM_PAGE_GRANT_UNMAP, data[0]);
936 break;
937
938 case TRC_MEM_PAGE_GRANT_TRANSFER:
939 /*
940 * Page grant is being transferred
941 *
942 * data[0] = target domid
943 */
944 XDT_PROBE1(XDT_MEM_PAGE_GRANT_TRANSFER, data[0]);
945 break;
946
947 /*
948 * Probes for PV domains.
949 */
950 case TRC_PV_HYPERCALL:
951 /*
952 * Hypercall from a 32-bit PV domain.
953 *
954 * data[0] = eip
955 * data[1] = eax
956 */
957 XDT_PROBE2(XDT_PV_HYPERCALL, data[0], data[1]);
958 break;
959 case TRC_PV_HYPERCALL | TRC_64_FLAG:
960 /*
961 * Hypercall from a 64-bit PV domain.
962 *
963 * data[0] = rip(0:31)
964 * data[1] = rip(32:63)
965 * data[2] = eax;
966 */
967 rip64 = (((uint64_t)data[1]) << 32) | data[0];
968 XDT_PROBE2(XDT_PV_HYPERCALL, rip64, data[2]);
969 break;
970 case TRC_PV_TRAP:
971 /*
972 * Trap in a 32-bit PV domain.
973 *
974 * data[0] = eip
975 * data[1] = trapnr | (error_code_valid << 15)
976 * | (error_code << 16);
977 */
978 XDT_PROBE4(XDT_PV_TRAP, data[0], data[1] & 0x7fff,
979 (data[1] >> 15) & 1, data[1] >> 16);
980 break;
981 case TRC_PV_TRAP | TRC_64_FLAG:
982 /*
983 * Trap in a 64-bit PV domain.
984 *
985 * data[0] = rip(0:31)
986 * data[1] = rip(32:63)
987 * data[2] = trapnr | (error_code_valid << 15)
988 * | (error_code << 16);
989 */
990 rip64 = (((uint64_t)data[1]) << 32) | data[2];
991 XDT_PROBE4(XDT_PV_TRAP, rip64, data[2] & 0x7fff,
992 (data[2] >> 15) & 1, data[2] >> 16);
993 break;
994 case TRC_PV_PAGE_FAULT:
995 /*
996 * Page fault in a 32-bit PV domain.
997 *
998 * data[0] = eip
999 * data[1] = vaddr
1000 * data[2] = error code
1001 */
1002 XDT_PROBE3(XDT_PV_PAGE_FAULT, data[0], data[1], data[2]);
1003 break;
1004 case TRC_PV_PAGE_FAULT | TRC_64_FLAG:
1005 /*
1006 * Page fault in a 32-bit PV domain.
1007 *
1008 * data[0] = rip(0:31)
1009 * data[1] = rip(31:63)
1010 * data[2] = vaddr(0:31)
1011 * data[3] = vaddr(31:63)
1012 * data[4] = error code
1013 */
1014 rip64 = (((uint64_t)data[1]) << 32) | data[0];
1015 addr64 = (((uint64_t)data[3]) << 32) | data[2];
1016 XDT_PROBE3(XDT_PV_PAGE_FAULT, rip64, addr64, data[4]);
1017 break;
1018 case TRC_PV_FORCED_INVALID_OP:
1019 /*
1020 * Hypervisor emulated a forced invalid op (ud2)
1021 * in a 32-bit PV domain.
1022 *
1023 * data[1] = eip
1024 */
1025 XDT_PROBE1(XDT_PV_FORCED_INVALID_OP, data[1]);
1026 break;
1027 case TRC_PV_FORCED_INVALID_OP | TRC_64_FLAG:
1028 /*
1029 * Hypervisor emulated a forced invalid op (ud2)
1030 * in a 64-bit PV domain.
1031 *
1032 * data[1] = rip(0:31)
1033 * data[2] = rip(31:63)
1034 *
1035 */
1036 rip64 = (((uint64_t)data[2]) << 32) | data[1];
1037 XDT_PROBE1(XDT_PV_FORCED_INVALID_OP, rip64);
1038 break;
1039 case TRC_PV_EMULATE_PRIVOP:
1040 /*
1041 * Hypervisor emulated a privileged operation
1042 * in a 32-bit PV domain.
1043 *
1044 * data[0] = eip
1045 */
1046 XDT_PROBE1(XDT_PV_EMULATE_PRIVOP, data[0]);
1047 break;
1048 case TRC_PV_EMULATE_PRIVOP | TRC_64_FLAG:
1049 /*
1050 * Hypervisor emulated a privileged operation
1051 * in a 64-bit PV domain.
1052 *
1053 * data[0] = rip(0:31)
1054 * data[1] = rip(31:63)
1055 */
1056 rip64 = (((uint64_t)data[1]) << 32) | data[0];
1057 XDT_PROBE1(XDT_PV_EMULATE_PRIVOP, rip64);
1058 break;
1059 case TRC_PV_EMULATE_4GB:
1060 /* unused, 32-bit hypervisor only */
1061 break;
1062 case TRC_PV_MATH_STATE_RESTORE:
1063 /*
1064 * Hypervisor restores math state after FP DNA trap.
1065 *
1066 * No arguments.
1067 */
1068 XDT_PROBE0(XDT_PV_MATH_STATE_RESTORE);
1069 break;
1070 case TRC_PV_PAGING_FIXUP:
1071 /*
1072 * Hypervisor fixed up a page fault (e.g. it was
1073 * a side-effect of hypervisor guest page table
1074 * bookkeeping, and not propagated to the guest).
1075 *
1076 * data[0] = eip
1077 * data[1] = vaddr
1078 */
1079 XDT_PROBE2(XDT_PV_PAGING_FIXUP, data[0], data[2]);
1080 break;
1081 case TRC_PV_PAGING_FIXUP | TRC_64_FLAG:
1082 /*
1083 * Hypervisor fixed up a page fault (e.g. it was
1084 * a side-effect of hypervisor guest page table
1085 * bookkeeping, and not propagated to the guest).
1086 *
1087 * data[0] = eip(0:31)
1088 * data[1] = eip(31:63)
1089 * data[2] = vaddr(0:31)
1090 * data[3] = vaddr(31:63)
1091 */
1092 rip64 = (((uint64_t)data[1]) << 32) | data[0];
1093 addr64 = (((uint64_t)data[3]) << 32) | data[2];
1094 XDT_PROBE2(XDT_PV_PAGING_FIXUP, rip64, addr64);
1095 break;
1096 case TRC_PV_GDT_LDT_MAPPING_FAULT:
1097 /*
1098 * Descriptor table mapping fault in a 32-bit PV domain.
1099 * data[0] = eip
1100 * data[1] = offset
1101 */
1102 XDT_PROBE2(XDT_PV_DT_MAPPING_FAULT, data[0], data[1]);
1103 break;
1104 case TRC_PV_GDT_LDT_MAPPING_FAULT | TRC_64_FLAG:
1105 /*
1106 * Descriptor table mapping fault in a 64-bit PV domain.
1107 *
1108 * data[0] = eip(0:31)
1109 * data[1] = eip(31:63)
1110 * data[2] = offset(0:31)
1111 * data[3] = offset(31:63)
1112 */
1113 rip64 = (((uint64_t)data[1]) << 32) | data[0];
1114 val64 = (((uint64_t)data[3]) << 32) | data[2];
1115 XDT_PROBE2(XDT_PV_DT_MAPPING_FAULT, rip64, val64);
1116 break;
1117 case TRC_PV_PTWR_EMULATION:
1118 case TRC_PV_PTWR_EMULATION_PAE | TRC_64_FLAG:
1119 /*
1120 * Should only happen on 32-bit hypervisor; unused.
1121 */
1122 break;
1123 case TRC_PV_PTWR_EMULATION_PAE:
1124 /*
1125 * PTE write emulation for a 32-bit PV domain.
1126 *
1127 * data[0] = pte
1128 * data[1] = addr
1129 * data[2] = eip
1130 */
1131 XDT_PROBE3(XDT_PV_PTWR_EMULATION, data[0], data[1], data[2]);
1132 break;
1133 case TRC_PV_PTWR_EMULATION | TRC_64_FLAG:
1134 /*
1135 * PTE write emulation for a 64-bit PV domain.
1136 *
1137 * data[0] = pte(0:31)
1138 * data[1] = pte(32:63)
1139 * data[2] = addr(0:31)
1140 * data[3] = addr(32:63)
1141 * data[4] = rip(0:31)
1142 * data[5] = rip(32:63)
1143 */
1144 pte64 = (((uint64_t)data[1]) << 32) | data[0];
1145 addr64 = (((uint64_t)data[3]) << 32) | data[2];
1146 rip64 = (((uint64_t)data[5]) << 32) | data[4];
1147 XDT_PROBE3(XDT_PV_PTWR_EMULATION, pte64, addr64, rip64);
1148 break;
1149
1150 /*
1151 * HVM probes
1152 */
1153 case TRC_HVM_VMENTRY:
1154 /*
1155 * Return to guest via vmx_launch/vmrun
1156 *
1157 */
1158 XDT_PROBE0(XDT_HVM_VMENTRY);
1159 break;
1160
1161 case TRC_HVM_VMEXIT:
1162 /*
1163 * Entry into VMEXIT handler from 32-bit HVM domain
1164 *
1165 * data[0] = cpu vendor specific exit code
1166 * data[1] = guest eip
1167 */
1168 XDT_PROBE2(XDT_HVM_VMEXIT, data[0], data[1]);
1169 break;
1170 case TRC_HVM_VMEXIT64:
1171 /*
1172 * Entry into VMEXIT handler from 64-bit HVM domain
1173 *
1174 * data[0] = cpu vendor specific exit code
1175 * data[1] = guest rip(0:31)
1176 * data[2] = guest rip(32:64)
1177 */
1178 rip64 = (((uint64_t)data[2]) << 32) | data[1];
1179 XDT_PROBE2(XDT_HVM_VMEXIT, data[0], rip64);
1180 break;
1181
1182 case TRC_HVM_PF_XEN64:
1183 /*
1184 * Pagefault in a guest that is a Xen (e.g. shadow)
1185 * artifact, and is not injected back into the guest.
1186 *
1187 * data[0] = error code
1188 * data[1] = guest VA(0:31)
1189 * data[2] = guest VA(32:64)
1190 */
1191 addr64 = (((uint64_t)data[2]) << 32) | data[1];
1192 XDT_PROBE2(XDT_HVM_PF_XEN, data[0], addr64);
1193 break;
1194
1195 case TRC_HVM_PF_XEN:
1196 /*
1197 * Same as above, but for a 32-bit HVM domain.
1198 *
1199 * data[0] = error code
1200 * data[1] = guest VA
1201 */
1202 XDT_PROBE2(XDT_HVM_PF_XEN, data[0], data[1]);
1203 break;
1204
1205 case TRC_HVM_PF_INJECT:
1206 /*
1207 * 32-bit Xen only.
1208 */
1209 break;
1210 case TRC_HVM_PF_INJECT64:
1211 /*
1212 * Pagefault injected back into a guest (e.g. the shadow
1213 * code found no mapping).
1214 *
1215 * data[0] = error code
1216 * data[1] = guest VA(0:31)
1217 * data[2] = guest VA(32:64)
1218 */
1219 addr64 = (((uint64_t)data[2]) << 32) | data[1];
1220 XDT_PROBE2(XDT_HVM_PF_INJECT, data[0], addr64);
1221 break;
1222
1223 case TRC_HVM_INJ_EXC:
1224 /*
1225 * Exception injected into an HVM guest.
1226 *
1227 * data[0] = trap
1228 * data[1] = error code
1229 */
1230 XDT_PROBE2(XDT_HVM_EXC_INJECT, data[0], data[1]);
1231 break;
1232 case TRC_HVM_INJ_VIRQ:
1233 /*
1234 * Interrupt inject into an HVM guest.
1235 *
1236 * data[0] = vector
1237 */
1238 XDT_PROBE1(XDT_HVM_VIRQ_INJECT, data[0]);
1239 break;
1240 case TRC_HVM_REINJ_VIRQ:
1241 case TRC_HVM_IO_READ:
1242 case TRC_HVM_IO_WRITE:
1243 /* unused */
1244 break;
1245 case TRC_HVM_CR_READ64:
1246 /*
1247 * Control register read. Intel VMX only.
1248 *
1249 * data[0] = control register #
1250 * data[1] = value(0:31)
1251 * data[2] = value(32:63)
1252 */
1253 val64 = (((uint64_t)data[2]) << 32) | data[1];
1254 XDT_PROBE2(XDT_HVM_CR_READ, data[0], val64);
1255 break;
1256 case TRC_HVM_CR_READ:
1257 /*
1258 * unused (32-bit Xen only)
1259 */
1260 break;
1261 case TRC_HVM_CR_WRITE64:
1262 /*
1263 * Control register write. Intel VMX only.
1264 *
1265 * data[0] = control register #
1266 * data[1] = value(0:31)
1267 * data[2] = value(32:63)
1268 */
1269 val64 = (((uint64_t)data[2]) << 32) | data[1];
1270 XDT_PROBE2(XDT_HVM_CR_READ, data[0], val64);
1271 break;
1272 case TRC_HVM_CR_WRITE:
1273 /*
1274 * unused (32-bit Xen only)
1275 */
1276 break;
1277 case TRC_HVM_DR_READ:
1278 /*
1279 * unused.
1280 *
1281 * data[0] = (domid<<16 + vcpuid)
1282 */
1283 break;
1284 case TRC_HVM_DR_WRITE:
1285 /*
1286 * Debug register write. Not too useful; no values,
1287 * so we ignore this.
1288 *
1289 * data[0] = (domid<<16 + vcpuid)
1290 */
1291 break;
1292 case TRC_HVM_MSR_READ:
1293 /*
1294 * MSR read.
1295 *
1296 * data[0] = MSR
1297 * data[1] = value(0:31)
1298 * data[2] = value(32:63)
1299 */
1300 val64 = (((uint64_t)data[3]) << 32) | data[2];
1301 XDT_PROBE2(XDT_HVM_MSR_READ, data[0], val64);
1302 break;
1303 case TRC_HVM_MSR_WRITE:
1304 /*
1305 * MSR write.
1306 *
1307 * data[0] = MSR;
1308 * data[1] = value(0:31)
1309 * data[2] = value(32:63)
1310 */
1311 val64 = (((uint64_t)data[2]) << 32) | data[1];
1312 XDT_PROBE2(XDT_HVM_MSR_WRITE, data[0], val64);
1313 break;
1314 case TRC_HVM_CPUID:
1315 /*
1316 * CPUID insn.
1317 *
1318 * data[0] = %eax (input)
1319 * data[1] = %eax
1320 * data[2] = %ebx
1321 * data[3] = %ecx
1322 * data[4] = %edx
1323 */
1324 XDT_PROBE5(XDT_HVM_CPUID, data[0], data[1], data[2], data[3],
1325 data[4]);
1326 break;
1327 case TRC_HVM_INTR:
1328 /*
1329 * VMEXIT because of an interrupt.
1330 */
1331 XDT_PROBE0(XDT_HVM_INTR);
1332 break;
1333 case TRC_HVM_INTR_WINDOW:
1334 /*
1335 * VMEXIT because of an interrupt window (an interrupt
1336 * can't be delivered immediately to a HVM guest and must
1337 * be delayed).
1338 *
1339 * data[0] = vector
1340 * data[1] = source
1341 * data[2] = info
1342 */
1343 XDT_PROBE3(XDT_HVM_INTR_WINDOW, data[0], data[1], data[2]);
1344 break;
1345 case TRC_HVM_NMI:
1346 /*
1347 * VMEXIT because of an NMI.
1348 */
1349 XDT_PROBE0(XDT_HVM_NMI);
1350 break;
1351 case TRC_HVM_SMI:
1352 /*
1353 * VMEXIT because of an SMI
1354 */
1355 XDT_PROBE0(XDT_HVM_SMI);
1356 break;
1357 case TRC_HVM_VMMCALL:
1358 /*
1359 * VMMCALL insn.
1360 *
1361 * data[0] = %eax
1362 */
1363 XDT_PROBE1(XDT_HVM_VMMCALL, data[0]);
1364 break;
1365 case TRC_HVM_HLT:
1366 /*
1367 * HLT insn.
1368 *
1369 * data[0] = 1 if VCPU runnable, 0 if not
1370 */
1371 XDT_PROBE1(XDT_HVM_HLT, data[0]);
1372 break;
1373 case TRC_HVM_INVLPG64:
1374 /*
1375 *
1376 * data[0] = INVLPGA ? 1 : 0
1377 * data[1] = vaddr(0:31)
1378 * data[2] = vaddr(32:63)
1379 */
1380 addr64 = (((uint64_t)data[2]) << 32) | data[1];
1381 XDT_PROBE2(XDT_HVM_INVLPG, data[0], addr64);
1382 break;
1383 case TRC_HVM_INVLPG:
1384 /*
1385 * unused (32-bit Xen only)
1386 *
1387 * data[0] = (domid<<16 + vcpuid)
1388 */
1389 break;
1390 case TRC_HVM_MCE:
1391 /*
1392 * #MCE VMEXIT
1393 *
1394 */
1395 XDT_PROBE0(XDT_HVM_MCE);
1396 break;
1397 case TRC_HVM_IOPORT_READ:
1398 case TRC_HVM_IOPORT_WRITE:
1399 case TRC_HVM_IOMEM_READ:
1400 case TRC_HVM_IOMEM_WRITE:
1401 /*
1402 * data[0] = addr(0:31)
1403 * data[1] = addr(32:63)
1404 * data[2] = count
1405 * data[3] = size
1406 */
1407 switch (rec->event) {
1408 case TRC_HVM_IOPORT_READ:
1409 eid = XDT_HVM_IOPORT_READ;
1410 break;
1411 case TRC_HVM_IOPORT_WRITE:
1412 eid = XDT_HVM_IOPORT_WRITE;
1413 break;
1414 case TRC_HVM_IOMEM_READ:
1415 eid = XDT_HVM_IOMEM_READ;
1416 break;
1417 case TRC_HVM_IOMEM_WRITE:
1418 eid = XDT_HVM_IOMEM_WRITE;
1419 break;
1420 }
1421 addr64 = (((uint64_t)data[1]) << 32) | data[0];
1422 XDT_PROBE3(eid, addr64, data[2], data[3]);
1423 break;
1424 case TRC_HVM_CLTS:
1425 /*
1426 * CLTS insn (Intel VMX only)
1427 */
1428 XDT_PROBE0(XDT_HVM_CLTS);
1429 break;
1430 case TRC_HVM_LMSW64:
1431 /*
1432 * LMSW insn.
1433 *
1434 * data[0] = value(0:31)
1435 * data[1] = value(32:63)
1436 */
1437 val64 = (((uint64_t)data[1]) << 32) | data[0];
1438 XDT_PROBE1(XDT_HVM_LMSW, val64);
1439 break;
1440 case TRC_HVM_LMSW:
1441 /*
1442 * unused (32-bit Xen only)
1443 */
1444 break;
1445
1446 /*
1447 * Shadow page table probes (mainly used for HVM domains
1448 * without hardware paging support).
1449 */
1450 case TRC_SHADOW_NOT_SHADOW | SH_GUEST_32:
1451 /*
1452 * data[0] = pte(0:31)
1453 * data[1] = pte(32:63)
1454 * data[2] = va
1455 * data[3] = flags
1456 */
1457 pte64 = ((uint64_t)data[1] << 32) | data[0];
1458 XDT_PROBE3(XDT_SHADOW_NOT_SHADOW, pte64, data[2], data[3]);
1459 break;
1460 case TRC_SHADOW_NOT_SHADOW | SH_GUEST_PAE:
1461 case TRC_SHADOW_NOT_SHADOW | SH_GUEST_64:
1462 /*
1463 * data[0] = pte(0:31)
1464 * data[1] = pte(32:63)
1465 * data[2] = va(0:31)
1466 * data[3] = va(32:63)
1467 * data[4] = flags
1468 */
1469 addr64 = ((uint64_t)data[2] << 32) | data[3];
1470 pte64 = ((uint64_t)data[1] << 32) | data[0];
1471 XDT_PROBE3(XDT_SHADOW_NOT_SHADOW, pte64, addr64, data[4]);
1472 break;
1473 case TRC_SHADOW_FAST_PROPAGATE | SH_GUEST_32:
1474 /*
1475 * data[0] = va
1476 */
1477 XDT_PROBE1(XDT_SHADOW_FAST_PROPAGATE, data[0]);
1478 break;
1479 case TRC_SHADOW_FAST_PROPAGATE | SH_GUEST_PAE:
1480 case TRC_SHADOW_FAST_PROPAGATE | SH_GUEST_64:
1481 /*
1482 * data[0] = va(0:31)
1483 * data[1] = va(32:63)
1484 */
1485 addr64 = ((uint64_t)data[1] << 32) | data[0];
1486 XDT_PROBE1(XDT_SHADOW_FAST_PROPAGATE, addr64);
1487 break;
1488 case TRC_SHADOW_FAST_MMIO | SH_GUEST_32:
1489 /*
1490 * data[0] = va
1491 */
1492 XDT_PROBE1(XDT_SHADOW_FAST_MMIO, data[0]);
1493 break;
1494 case TRC_SHADOW_FAST_MMIO | SH_GUEST_PAE:
1495 case TRC_SHADOW_FAST_MMIO | SH_GUEST_64:
1496 /*
1497 * data[0] = va(0:31)
1498 * data[1] = va(32:63)
1499 */
1500 addr64 = ((uint64_t)data[1] << 32) | data[0];
1501 XDT_PROBE1(XDT_SHADOW_FAST_MMIO, addr64);
1502 break;
1503 case TRC_SHADOW_FALSE_FAST_PATH | SH_GUEST_32:
1504 /*
1505 * data[0] = va
1506 */
1507 XDT_PROBE1(XDT_SHADOW_FALSE_FAST_PATH, data[0]);
1508 break;
1509 case TRC_SHADOW_FALSE_FAST_PATH | SH_GUEST_PAE:
1510 case TRC_SHADOW_FALSE_FAST_PATH | SH_GUEST_64:
1511 /*
1512 * data[0] = va(0:31)
1513 * data[1] = va(32:63)
1514 */
1515 addr64 = ((uint64_t)data[1] << 32) | data[0];
1516 XDT_PROBE1(XDT_SHADOW_FALSE_FAST_PATH, addr64);
1517 break;
1518 case TRC_SHADOW_MMIO | SH_GUEST_32:
1519 /*
1520 * data[0] = va
1521 */
1522 XDT_PROBE1(XDT_SHADOW_MMIO, data[0]);
1523 break;
1524 case TRC_SHADOW_MMIO | SH_GUEST_PAE:
1525 case TRC_SHADOW_MMIO | SH_GUEST_64:
1526 /*
1527 * data[0] = va(0:31)
1528 * data[1] = va(32:63)
1529 */
1530 addr64 = ((uint64_t)data[1] << 32) | data[0];
1531 XDT_PROBE1(XDT_SHADOW_MMIO, addr64);
1532 break;
1533 case TRC_SHADOW_FIXUP | SH_GUEST_32:
1534 /*
1535 * data[0] = pte(0:31)
1536 * data[1] = pte(32:63)
1537 * data[2] = va
1538 * data[3] = flags
1539 */
1540 pte64 = ((uint64_t)data[1] << 32) | data[0];
1541 XDT_PROBE3(XDT_SHADOW_FIXUP, pte64, data[2], data[3]);
1542 break;
1543 case TRC_SHADOW_FIXUP | SH_GUEST_64:
1544 case TRC_SHADOW_FIXUP | SH_GUEST_PAE:
1545 /*
1546 * data[0] = pte(0:31)
1547 * data[1] = pte(32:63)
1548 * data[2] = va(0:31)
1549 * data[3] = va(32:63)
1550 * data[4] = flags
1551 */
1552 addr64 = ((uint64_t)data[2] << 32) | data[3];
1553 pte64 = ((uint64_t)data[1] << 32) | data[0];
1554 XDT_PROBE3(XDT_SHADOW_FIXUP, pte64, addr64, data[4]);
1555 break;
1556 case TRC_SHADOW_DOMF_DYING | SH_GUEST_32:
1557 /*
1558 * data[0] = va
1559 */
1560 XDT_PROBE1(XDT_SHADOW_DOMF_DYING, data[0]);
1561 break;
1562 case TRC_SHADOW_DOMF_DYING | SH_GUEST_PAE:
1563 case TRC_SHADOW_DOMF_DYING | SH_GUEST_64:
1564 /*
1565 * data[0] = va(0:31)
1566 * data[1] = va(32:63)
1567 */
1568 addr64 = ((uint64_t)data[1] << 32) | data[0];
1569 XDT_PROBE1(XDT_SHADOW_DOMF_DYING, addr64);
1570 break;
1571 case TRC_SHADOW_EMULATE | SH_GUEST_32:
1572 /*
1573 * data[0] = pte(0:31)
1574 * data[1] = pte(32:63)
1575 * data[2] = val(0:31)
1576 * data[3] = val(32:63)
1577 * data[4] = addr
1578 * data[5] = flags
1579 */
1580 pte64 = ((uint64_t)data[1] << 32) | data[0];
1581 val64 = ((uint64_t)data[3] << 32) | data[2];
1582 XDT_PROBE5(XDT_SHADOW_EMULATE, pte64, val64, data[4],
1583 data[5] & 0x7fffffff, data[5] >> 29);
1584 break;
1585 case TRC_SHADOW_EMULATE | SH_GUEST_PAE:
1586 case TRC_SHADOW_EMULATE | SH_GUEST_64:
1587 /*
1588 * data[0] = pte(0:31)
1589 * data[1] = pte(32:63)
1590 * data[2] = val(0:31)
1591 * data[3] = val(32:63)
1592 * data[4] = addr(0:31)
1593 * data[5] = addr(32:63)
1594 * data[6] = flags
1595 */
1596 pte64 = ((uint64_t)data[1] << 32) | data[0];
1597 val64 = ((uint64_t)data[3] << 32) | data[2];
1598 addr64 = ((uint64_t)data[5] << 32) | data[4];
1599 XDT_PROBE5(XDT_SHADOW_EMULATE, pte64, val64, data[4],
1600 data[6] & 0x7fffffff, data[6] >> 29);
1601 break;
1602 case TRC_SHADOW_EMULATE_UNSHADOW_USER | SH_GUEST_32:
1603 /*
1604 * data[0] = gfn
1605 * data[1] = vaddr
1606 */
1607 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_USER, data[0], data[1]);
1608 break;
1609 case TRC_SHADOW_EMULATE_UNSHADOW_USER | SH_GUEST_PAE:
1610 case TRC_SHADOW_EMULATE_UNSHADOW_USER | SH_GUEST_64:
1611 /*
1612 * data[0] = gfn(0:31)
1613 * data[1] = gfn(32:63)
1614 * data[2] = vaddr(0:31)
1615 * data[3] = vaddr(32:63)
1616 */
1617 val64 = ((uint64_t)data[1] << 32) | data[0];
1618 addr64 = ((uint64_t)data[3] << 32) | data[2];
1619 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_USER, val64, addr64);
1620 break;
1621 case TRC_SHADOW_EMULATE_UNSHADOW_EVTINJ | SH_GUEST_32:
1622 /*
1623 * data[0] = gfn
1624 * data[1] = vaddr
1625 */
1626 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_EVTINJ, data[0],
1627 data[1]);
1628 break;
1629 case TRC_SHADOW_EMULATE_UNSHADOW_EVTINJ | SH_GUEST_PAE:
1630 case TRC_SHADOW_EMULATE_UNSHADOW_EVTINJ | SH_GUEST_64:
1631 /*
1632 * data[0] = gfn(0:31)
1633 * data[1] = gfn(32:63)
1634 * data[2] = vaddr(0:31)
1635 * data[3] = vaddr(32:63)
1636 */
1637 val64 = ((uint64_t)data[1] << 32) | data[0];
1638 addr64 = ((uint64_t)data[3] << 32) | data[2];
1639 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_EVTINJ, val64, addr64);
1640 break;
1641 case TRC_SHADOW_EMULATE_UNSHADOW_UNHANDLED | SH_GUEST_32:
1642 /*
1643 * data[0] = gfn
1644 * data[1] = vaddr
1645 */
1646 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_UNHANDLED, data[0],
1647 data[1]);
1648 break;
1649 case TRC_SHADOW_EMULATE_UNSHADOW_UNHANDLED | SH_GUEST_PAE:
1650 case TRC_SHADOW_EMULATE_UNSHADOW_UNHANDLED | SH_GUEST_64:
1651 /*
1652 * data[0] = gfn(0:31)
1653 * data[1] = gfn(32:63)
1654 * data[2] = vaddr(0:31)
1655 * data[3] = vaddr(32:63)
1656 */
1657 val64 = ((uint64_t)data[1] << 32) | data[0];
1658 addr64 = ((uint64_t)data[3] << 32) | data[2];
1659 XDT_PROBE2(XDT_SHADOW_EMULATE_UNSHADOW_UNHANDLED, val64,
1660 addr64);
1661 break;
1662 case TRC_SHADOW_WRMAP_BF:
1663 /*
1664 * data[0] = gfn(0:31)
1665 * data[1] = gfn(32:63)
1666 */
1667 val64 = ((uint64_t)data[1] << 32) | data[0];
1668 XDT_PROBE1(XDT_SHADOW_WRMAP_BF, val64);
1669 break;
1670 case TRC_SHADOW_PREALLOC_UNPIN:
1671 /*
1672 * data[0] = gfn(0:31)
1673 * data[1] = gfn(32:63)
1674 */
1675 val64 = ((uint64_t)data[1] << 32) | data[0];
1676 XDT_PROBE1(XDT_SHADOW_PREALLOC_UNPIN, val64);
1677 break;
1678 case TRC_SHADOW_RESYNC_FULL:
1679 /*
1680 * data[0] = gmfn(0:31)
1681 * data[1] = gmfn(32:63)
1682 */
1683 val64 = ((uint64_t)data[1] << 32) | data[0];
1684 XDT_PROBE1(XDT_SHADOW_RESYNC_FULL, val64);
1685 break;
1686 case TRC_SHADOW_RESYNC_ONLY:
1687 /*
1688 * data[0] = gmfn(0:31)
1689 * data[1] = gmfn(32:63)
1690 */
1691 val64 = ((uint64_t)data[1] << 32) | data[0];
1692 XDT_PROBE1(XDT_SHADOW_RESYNC_ONLY, val64);
1693 break;
1694
1695 /*
1696 * Power management probes.
1697 */
1698 case TRC_PM_FREQ_CHANGE:
1699 /*
1700 * data[0] = old freq
1701 * data[1] = new freq
1702 */
1703 XDT_PROBE2(XDT_PM_FREQ_CHANGE, data[0], data[1]);
1704 break;
1705 case TRC_PM_IDLE_ENTRY:
1706 /*
1707 * data[0] = C-state
1708 * data[1] = time
1709 */
1710 XDT_PROBE2(XDT_PM_IDLE_ENTRY, data[0], data[1]);
1711 break;
1712 case TRC_PM_IDLE_EXIT:
1713 /*
1714 * data[0] = C-state
1715 * data[1] = time
1716 */
1717 XDT_PROBE2(XDT_PM_IDLE_EXIT, data[0], data[1]);
1718 break;
1719 case TRC_LOST_RECORDS:
1720 vcpu = data[1] >> 16;
1721 dom = data[1] & 0xffff;
1722 xdt_update_sched_context(cpuid, dom, vcpu);
1723 xdt_update_domain_context(dom, vcpu);
1724 XDT_PROBE1(XDT_TRC_LOST_RECORDS, cpuid);
1725 tbuf.stat_dropped_recs++;
1726 break;
1727
1728 default:
1729 tbuf.stat_unknown_recs++;
1730 break;
1731 }
1732
1733 done:
1734 rec_size = 4 + (rec->cycles_included ? 8 : 0) + (rec->extra_u32 * 4);
1735 return (rec_size);
1736 }
1737
1738 /*
1739 * Scan all CPU buffers for the record with the lowest timestamp so
1740 * that the probes will fire in order.
1741 */
1742 static int
1743 xdt_get_first_rec(uint_t *cpuidp, struct t_rec **recp, uint32_t *consp)
1744 {
1745 uint_t cpuid;
1746 uint32_t prod, cons, offset;
1747 struct t_rec *rec;
1748 uint64_t minstamp = ~0ULL, stamp;
1749 uintptr_t data;
1750
1751 for (cpuid = 0; cpuid < tbuf.cnt; cpuid++) {
1752 cons = tbuf.meta[cpuid]->cons;
1753 prod = tbuf.meta[cpuid]->prod;
1754 membar_consumer();
1755 if (prod == cons)
1756 continue;
1757
1758 offset = cons % tbuf_data_size;
1759 data = (uintptr_t)tbuf.data[cpuid] + offset;
1760 rec = (struct t_rec *)data;
1761 ASSERT((caddr_t)rec < tbuf.va + (tbuf.size * (cpuid + 1)));
1762
1763 /*
1764 * All records that we know about have time cycles included.
1765 * If this record doesn't have them, assume it's a type
1766 * that we don't handle. Use a 0 time value, which will make
1767 * it get handled first (it will be thrown away).
1768 */
1769 if (rec->cycles_included)
1770 stamp = (((uint64_t)rec->u.cycles.cycles_hi) << 32)
1771 | rec->u.cycles.cycles_lo;
1772 else
1773 stamp = 0;
1774
1775 if (stamp < minstamp) {
1776 minstamp = stamp;
1777 *cpuidp = cpuid;
1778 *recp = rec;
1779 *consp = cons;
1780 }
1781 }
1782
1783 if (minstamp != ~0ULL)
1784 return (1);
1785
1786 return (0);
1787 }
1788
1789 /*ARGSUSED*/
1790 static void
1791 xdt_tbuf_scan(void *arg)
1792 {
1793 uint32_t bytes_done, cons;
1794 struct t_rec *rec;
1795 xdt_schedinfo_t *sp;
1796 uint_t nrecs, cpuid;
1797
1798 for (nrecs = 0;
1799 nrecs < xdt_max_recs && xdt_get_first_rec(&cpuid, &rec, &cons) > 0;
1800 nrecs++) {
1801 xdt_curpcpu = cpuid;
1802 sp = &xdt_cpu_schedinfo[cpuid];
1803 if (sp->curinfo_valid)
1804 xdt_update_domain_context(sp->cur_domid,
1805 sp->cur_vcpuid);
1806
1807 bytes_done = xdt_process_rec(cpuid, rec);
1808 cons += bytes_done;
1809 /*
1810 * cons and prod are incremented modulo (2 * tbuf_data_size).
1811 * See <xen/public/trace.h>.
1812 */
1813 if (cons >= 2 * tbuf_data_size)
1814 cons -= 2 * tbuf_data_size;
1815 membar_exit();
1816 tbuf.meta[cpuid]->cons = cons;
1817 }
1818 }
1819
1820 static void
1821 xdt_cyclic_enable(void)
1822 {
1823 cyc_handler_t hdlr;
1824 cyc_time_t when;
1825
1826 ASSERT(MUTEX_HELD(&cpu_lock));
1827
1828 hdlr.cyh_func = xdt_tbuf_scan;
1829 hdlr.cyh_arg = NULL;
1830 hdlr.cyh_level = CY_LOW_LEVEL;
1831
1832 when.cyt_interval = xdt_poll_nsec;
1833 when.cyt_when = dtrace_gethrtime() + when.cyt_interval;
1834
1835 xdt_cyclic = cyclic_add(&hdlr, &when);
1836 }
1837
1838 static void
1839 xdt_probe_create(xdt_probe_t *p)
1840 {
1841 ASSERT(p != NULL && p->pr_mod != NULL);
1842
1843 if (dtrace_probe_lookup(xdt_id, p->pr_mod, NULL, p->pr_name) != 0)
1844 return;
1845
1846 xdt_prid[p->evt_id] = dtrace_probe_create(xdt_id, p->pr_mod, NULL,
1847 p->pr_name, dtrace_mach_aframes(), p);
1848 }
1849
1850 /*ARGSUSED*/
1851 static void
1852 xdt_provide(void *arg, const dtrace_probedesc_t *desc)
1853 {
1854 const char *mod, *name;
1855 int i;
1856
1857 if (desc == NULL) {
1858 for (i = 0; xdt_probe[i].pr_mod != NULL; i++) {
1859 xdt_probe_create(&xdt_probe[i]);
1860 }
1861 } else {
1862 mod = desc->dtpd_mod;
1863 name = desc->dtpd_name;
1864 for (i = 0; xdt_probe[i].pr_mod != NULL; i++) {
1865 int l1 = strlen(xdt_probe[i].pr_name);
1866 int l2 = strlen(xdt_probe[i].pr_mod);
1867 if (strncmp(name, xdt_probe[i].pr_name, l1) == 0 &&
1868 strncmp(mod, xdt_probe[i].pr_mod, l2) == 0)
1869 break;
1870 }
1871
1872 if (xdt_probe[i].pr_mod == NULL)
1873 return;
1874 xdt_probe_create(&xdt_probe[i]);
1875 }
1876
1877 }
1878
1879 /*ARGSUSED*/
1880 static void
1881 xdt_destroy(void *arg, dtrace_id_t id, void *parg)
1882 {
1883 xdt_probe_t *p = parg;
1884 xdt_prid[p->evt_id] = 0;
1885 }
1886
1887 static void
1888 xdt_set_trace_mask(uint32_t mask)
1889 {
1890 xen_sysctl_tbuf_op_t tbuf_op;
1891
1892 /* Always need to trace scheduling, for context */
1893 if (mask != 0)
1894 mask |= TRC_SCHED;
1895 tbuf_op.evt_mask = mask;
1896 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_set_evt_mask;
1897 (void) xdt_sysctl_tbuf(&tbuf_op);
1898 }
1899
1900 /*ARGSUSED*/
1901 static int
1902 xdt_enable(void *arg, dtrace_id_t id, void *parg)
1903 {
1904 xdt_probe_t *p = parg;
1905 xen_sysctl_tbuf_op_t tbuf_op;
1906
1907 ASSERT(MUTEX_HELD(&cpu_lock));
1908 ASSERT(xdt_prid[p->evt_id] != 0);
1909
1910 xdt_probemap[p->evt_id] = xdt_prid[p->evt_id];
1911 xdt_classinfo[p->class].cnt++;
1912
1913 if (xdt_classinfo[p->class].cnt == 1) {
1914 /* set the trace mask for this class */
1915 cur_trace_mask |= xdt_classinfo[p->class].trc_mask;
1916 xdt_set_trace_mask(cur_trace_mask);
1917 }
1918
1919 if (xdt_cyclic == CYCLIC_NONE) {
1920 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_enable;
1921 if (xdt_sysctl_tbuf(&tbuf_op) != 0) {
1922 cmn_err(CE_NOTE, "Couldn't enable hypervisor tracing.");
1923 return (-1);
1924 }
1925
1926 xdt_cyclic_enable();
1927 }
1928 return (0);
1929 }
1930
1931 /*ARGSUSED*/
1932 static void
1933 xdt_disable(void *arg, dtrace_id_t id, void *parg)
1934 {
1935 xdt_probe_t *p = parg;
1936 xen_sysctl_tbuf_op_t tbuf_op;
1937 int i, err;
1938
1939 ASSERT(MUTEX_HELD(&cpu_lock));
1940 ASSERT(xdt_probemap[p->evt_id] != 0);
1941 ASSERT(xdt_probemap[p->evt_id] == xdt_prid[p->evt_id]);
1942 ASSERT(xdt_classinfo[p->class].cnt > 0);
1943
1944 /*
1945 * We could be here in the slight window between the cyclic firing and
1946 * a call to dtrace_probe() occurring. We need to be careful if we tear
1947 * down any shared state.
1948 */
1949
1950 xdt_probemap[p->evt_id] = 0;
1951 xdt_classinfo[p->class].cnt--;
1952
1953 if (xdt_nr_active_probes() == 0) {
1954 cur_trace_mask = 0;
1955
1956 if (xdt_cyclic == CYCLIC_NONE)
1957 return;
1958
1959 for (i = 0; i < xdt_ncpus; i++)
1960 xdt_cpu_schedinfo[i].curinfo_valid = 0;
1961
1962 /*
1963 * We will try to disable the trace buffers. If we fail for some
1964 * reason we will try again, up to a count of XDT_TBUF_RETRY.
1965 * If we still aren't successful we try to set the trace mask
1966 * to 0 in order to prevent trace records from being written.
1967 */
1968 tbuf_op.cmd = XEN_SYSCTL_TBUFOP_disable;
1969 i = 0;
1970 do {
1971 err = xdt_sysctl_tbuf(&tbuf_op);
1972 } while ((err != 0) && (++i < XDT_TBUF_RETRY));
1973
1974 if (err != 0) {
1975 cmn_err(CE_NOTE,
1976 "Couldn't disable hypervisor tracing.");
1977 xdt_set_trace_mask(0);
1978 } else {
1979 cyclic_remove(xdt_cyclic);
1980 xdt_cyclic = CYCLIC_NONE;
1981 /*
1982 * We don't bother making the hypercall to set
1983 * the trace mask, since it will be reset when
1984 * tracing is re-enabled.
1985 */
1986 }
1987 } else if (xdt_classinfo[p->class].cnt == 0) {
1988 cur_trace_mask ^= xdt_classinfo[p->class].trc_mask;
1989 /* other probes are enabled, so add the sub-class mask back */
1990 cur_trace_mask |= 0xF000;
1991 xdt_set_trace_mask(cur_trace_mask);
1992 }
1993 }
1994
1995 static dtrace_pattr_t xdt_attr = {
1996 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM },
1997 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM },
1998 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
1999 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM },
2000 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM },
2001 };
2002
2003 static dtrace_pops_t xdt_pops = {
2004 xdt_provide, /* dtps_provide() */
2005 NULL, /* dtps_provide_module() */
2006 xdt_enable, /* dtps_enable() */
2007 xdt_disable, /* dtps_disable() */
2008 NULL, /* dtps_suspend() */
2009 NULL, /* dtps_resume() */
2010 NULL, /* dtps_getargdesc() */
2011 NULL, /* dtps_getargval() */
2012 NULL, /* dtps_usermode() */
2013 xdt_destroy /* dtps_destroy() */
2014 };
2015
2016 static int
2017 xdt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
2018 {
2019 int val;
2020
2021 if (!DOMAIN_IS_INITDOMAIN(xen_info))
2022 return (DDI_FAILURE);
2023
2024 switch (cmd) {
2025 case DDI_ATTACH:
2026 break;
2027
2028 case DDI_RESUME:
2029 /*
2030 * We might support proper suspend/resume in the future, so,
2031 * return DDI_FAILURE for now.
2032 */
2033 return (DDI_FAILURE);
2034
2035 default:
2036 return (DDI_FAILURE);
2037 }
2038
2039 xdt_ncpus = xpv_nr_phys_cpus();
2040 ASSERT(xdt_ncpus > 0);
2041
2042 if (ddi_create_minor_node(devi, "xdt", S_IFCHR, 0, DDI_PSEUDO, 0) ==
2043 DDI_FAILURE || xdt_attach_trace_buffers() != 0 ||
2044 dtrace_register("xdt", &xdt_attr, DTRACE_PRIV_KERNEL, NULL,
2045 &xdt_pops, NULL, &xdt_id) != 0) {
2046 if (tbuf.va != NULL)
2047 xdt_detach_trace_buffers();
2048 ddi_remove_minor_node(devi, NULL);
2049 return (DDI_FAILURE);
2050 }
2051
2052 val = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
2053 "xdt_poll_nsec", XDT_POLL_DEFAULT);
2054 xdt_poll_nsec = MAX(val, XDT_POLL_MIN);
2055
2056 xdt_cpu_schedinfo = (xdt_schedinfo_t *)kmem_zalloc(xdt_ncpus *
2057 sizeof (xdt_schedinfo_t), KM_SLEEP);
2058 xdt_init_trace_masks();
2059 xdt_kstat_init();
2060
2061 xdt_devi = devi;
2062 ddi_report_dev(devi);
2063 return (DDI_SUCCESS);
2064 }
2065
2066 static int
2067 xdt_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
2068 {
2069 switch (cmd) {
2070 case DDI_DETACH:
2071 break;
2072
2073 case DDI_SUSPEND:
2074 /*
2075 * We might support proper suspend/resume in the future. So
2076 * return DDI_FAILURE for now.
2077 */
2078 return (DDI_FAILURE);
2079
2080 default:
2081 return (DDI_FAILURE);
2082 }
2083
2084 if (dtrace_unregister(xdt_id) != 0)
2085 return (DDI_FAILURE);
2086
2087 xdt_detach_trace_buffers();
2088 kmem_free(xdt_cpu_schedinfo, xdt_ncpus * sizeof (xdt_schedinfo_t));
2089 if (xdt_cyclic != CYCLIC_NONE)
2090 cyclic_remove(xdt_cyclic);
2091 if (xdt_kstats != NULL)
2092 kstat_delete(xdt_kstats);
2093 xdt_devi = (void *)0;
2094 ddi_remove_minor_node(devi, NULL);
2095
2096 return (DDI_SUCCESS);
2097 }
2098
2099 /*ARGSUSED*/
2100 static int
2101 xdt_info(dev_info_t *devi, ddi_info_cmd_t infocmd, void *arg, void **result)
2102 {
2103 int error;
2104
2105 switch (infocmd) {
2106 case DDI_INFO_DEVT2DEVINFO:
2107 *result = xdt_devi;
2108 error = DDI_SUCCESS;
2109 break;
2110 case DDI_INFO_DEVT2INSTANCE:
2111 *result = (void *)0;
2112 error = DDI_SUCCESS;
2113 break;
2114 default:
2115 error = DDI_FAILURE;
2116 }
2117 return (error);
2118 }
2119
2120 static struct cb_ops xdt_cb_ops = {
2121 nulldev, /* open(9E) */
2122 nodev, /* close(9E) */
2123 nodev, /* strategy(9E) */
2124 nodev, /* print(9E) */
2125 nodev, /* dump(9E) */
2126 nodev, /* read(9E) */
2127 nodev, /* write(9E) */
2128 nodev, /* ioctl(9E) */
2129 nodev, /* devmap(9E) */
2130 nodev, /* mmap(9E) */
2131 nodev, /* segmap(9E) */
2132 nochpoll, /* chpoll(9E) */
2133 ddi_prop_op, /* prop_op(9E) */
2134 NULL, /* streamtab(9S) */
2135 D_MP | D_64BIT | D_NEW /* cb_flag */
2136 };
2137
2138 static struct dev_ops xdt_ops = {
2139 DEVO_REV, /* devo_rev */
2140 0, /* devo_refcnt */
2141 xdt_info, /* getinfo(9E) */
2142 nulldev, /* identify(9E) */
2143 nulldev, /* probe(9E) */
2144 xdt_attach, /* attach(9E) */
2145 xdt_detach, /* detach(9E) */
2146 nulldev, /* devo_reset */
2147 &xdt_cb_ops, /* devo_cb_ops */
2148 NULL, /* devo_bus_ops */
2149 NULL, /* power(9E) */
2150 ddi_quiesce_not_needed, /* devo_quiesce */
2151 };
2152
2153
2154 static struct modldrv modldrv = {
2155 &mod_driverops,
2156 "Hypervisor event tracing",
2157 &xdt_ops
2158 };
2159
2160 static struct modlinkage modlinkage = {
2161 MODREV_1,
2162 &modldrv,
2163 NULL
2164 };
2165
2166 int
2167 _init(void)
2168 {
2169 return (mod_install(&modlinkage));
2170 }
2171
2172 int
2173 _fini(void)
2174 {
2175 return (mod_remove(&modlinkage));
2176 }
2177
2178 int
2179 _info(struct modinfo *modinfop)
2180 {
2181 return (mod_info(&modlinkage, modinfop));
2182 }