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 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 *
25 * Copyright 2018 Joyent, Inc.
26 */
27
28 /*
29 * Management of KMDB's IDT, which is installed upon KMDB activation.
30 *
31 * Debugger activation has two flavors, which cover the cases where KMDB is
32 * loaded at boot, and when it is loaded after boot. In brief, in both cases,
33 * the KDI needs to interpose upon several handlers in the IDT. When
34 * mod-loaded KMDB is deactivated, we undo the IDT interposition, restoring the
35 * handlers to what they were before we started.
36 *
37 * We also take over the entirety of IDT (except the double-fault handler) on
38 * the active CPU when we're in kmdb so we can handle things like page faults
39 * sensibly.
40 *
41 * Boot-loaded KMDB
42 *
43 * When we're first activated, we're running on boot's IDT. We need to be able
44 * to function in this world, so we'll install our handlers into boot's IDT.
45 * This is a little complicated: we're using the fake cpu_t set up by
46 * boot_kdi_tmpinit(), so we can't access cpu_idt directly. Instead,
47 * kdi_idt_write() notices that cpu_idt is NULL, and works around this problem.
48 *
49 * Later, when we're about to switch to the kernel's IDT, it'll call us via
50 * kdi_idt_sync(), allowing us to add our handlers to the new IDT. While
51 * boot-loaded KMDB can't be unloaded, we still need to save the descriptors we
52 * replace so we can pass traps back to the kernel as necessary.
53 *
54 * The last phase of boot-loaded KMDB activation occurs at non-boot CPU
55 * startup. We will be called on each non-boot CPU, thus allowing us to set up
56 * any watchpoints that may have been configured on the boot CPU and interpose
57 * on the given CPU's IDT. We don't save the interposed descriptors in this
58 * case -- see kdi_cpu_init() for details.
59 *
60 * Mod-loaded KMDB
61 *
62 * This style of activation is much simpler, as the CPUs are already running,
63 * and are using their own copy of the kernel's IDT. We simply interpose upon
64 * each CPU's IDT. We save the handlers we replace, both for deactivation and
65 * for passing traps back to the kernel. Note that for the hypervisors'
66 * benefit, we need to xcall to the other CPUs to do this, since we need to
67 * actively set the trap entries in its virtual IDT from that vcpu's context
68 * rather than just modifying the IDT table from the CPU running kdi_activate().
69 */
70
71 #include <sys/types.h>
72 #include <sys/segments.h>
73 #include <sys/trap.h>
74 #include <sys/cpuvar.h>
75 #include <sys/reboot.h>
76 #include <sys/sunddi.h>
77 #include <sys/archsystm.h>
78 #include <sys/kdi_impl.h>
79 #include <sys/x_call.h>
80 #include <ia32/sys/psw.h>
81 #include <vm/hat_i86.h>
82
83 #define KDI_GATE_NVECS 3
84
85 #define KDI_IDT_NOSAVE 0
86 #define KDI_IDT_SAVE 1
87
88 #define KDI_IDT_DTYPE_KERNEL 0
89 #define KDI_IDT_DTYPE_BOOT 1
90
91 kdi_cpusave_t *kdi_cpusave;
92 int kdi_ncpusave;
93
94 static kdi_main_t kdi_kmdb_main;
95
96 kdi_drreg_t kdi_drreg;
97
98 #ifndef __amd64
99 /* Used to track the current set of valid kernel selectors. */
100 uint32_t kdi_cs;
101 uint32_t kdi_ds;
102 uint32_t kdi_fs;
103 uint32_t kdi_gs;
104 #endif
105
106 uintptr_t kdi_kernel_handler;
107
108 int kdi_trap_switch;
109
110 #define KDI_MEMRANGES_MAX 2
111
112 kdi_memrange_t kdi_memranges[KDI_MEMRANGES_MAX];
113 int kdi_nmemranges;
114
115 typedef void idt_hdlr_f(void);
116
117 extern idt_hdlr_f kdi_trap0, kdi_trap1, kdi_int2, kdi_trap3, kdi_trap4;
118 extern idt_hdlr_f kdi_trap5, kdi_trap6, kdi_trap7, kdi_trap9;
119 extern idt_hdlr_f kdi_traperr10, kdi_traperr11, kdi_traperr12;
120 extern idt_hdlr_f kdi_traperr13, kdi_traperr14, kdi_trap16, kdi_traperr17;
121 extern idt_hdlr_f kdi_trap18, kdi_trap19, kdi_trap20, kdi_ivct32;
122 extern idt_hdlr_f kdi_invaltrap;
123 extern size_t kdi_ivct_size;
124
125 typedef struct kdi_gate_spec {
126 uint_t kgs_vec;
127 uint_t kgs_dpl;
128 } kdi_gate_spec_t;
129
130 /*
131 * Beware: kdi_pass_to_kernel() has unpleasant knowledge of this list.
132 */
133 static const kdi_gate_spec_t kdi_gate_specs[KDI_GATE_NVECS] = {
134 { T_SGLSTP, TRP_KPL },
135 { T_BPTFLT, TRP_UPL },
136 { T_DBGENTR, TRP_KPL }
137 };
138
139 static gate_desc_t kdi_kgates[KDI_GATE_NVECS];
140
141 extern gate_desc_t kdi_idt[NIDT];
142
143 struct idt_description {
144 uint_t id_low;
145 uint_t id_high;
146 idt_hdlr_f *id_basehdlr;
147 size_t *id_incrp;
148 } idt_description[] = {
149 { T_ZERODIV, 0, kdi_trap0, NULL },
150 { T_SGLSTP, 0, kdi_trap1, NULL },
151 { T_NMIFLT, 0, kdi_int2, NULL },
152 { T_BPTFLT, 0, kdi_trap3, NULL },
153 { T_OVFLW, 0, kdi_trap4, NULL },
154 { T_BOUNDFLT, 0, kdi_trap5, NULL },
155 { T_ILLINST, 0, kdi_trap6, NULL },
156 { T_NOEXTFLT, 0, kdi_trap7, NULL },
157 #if !defined(__xpv)
158 { T_DBLFLT, 0, syserrtrap, NULL },
159 #endif
160 { T_EXTOVRFLT, 0, kdi_trap9, NULL },
161 { T_TSSFLT, 0, kdi_traperr10, NULL },
162 { T_SEGFLT, 0, kdi_traperr11, NULL },
163 { T_STKFLT, 0, kdi_traperr12, NULL },
164 { T_GPFLT, 0, kdi_traperr13, NULL },
165 { T_PGFLT, 0, kdi_traperr14, NULL },
166 { 15, 0, kdi_invaltrap, NULL },
167 { T_EXTERRFLT, 0, kdi_trap16, NULL },
168 { T_ALIGNMENT, 0, kdi_traperr17, NULL },
169 { T_MCE, 0, kdi_trap18, NULL },
170 { T_SIMDFPE, 0, kdi_trap19, NULL },
171 { T_DBGENTR, 0, kdi_trap20, NULL },
172 { 21, 31, kdi_invaltrap, NULL },
173 { 32, 255, kdi_ivct32, &kdi_ivct_size },
174 { 0, 0, NULL },
175 };
176
177 void
178 kdi_idt_init(selector_t sel)
179 {
180 struct idt_description *id;
181 int i;
182
183 for (id = idt_description; id->id_basehdlr != NULL; id++) {
184 uint_t high = id->id_high != 0 ? id->id_high : id->id_low;
185 size_t incr = id->id_incrp != NULL ? *id->id_incrp : 0;
186
187 #if !defined(__xpv)
188 if (kpti_enable && sel == KCS_SEL && id->id_low == T_DBLFLT)
189 id->id_basehdlr = tr_syserrtrap;
190 #endif
191
192 for (i = id->id_low; i <= high; i++) {
193 caddr_t hdlr = (caddr_t)id->id_basehdlr +
194 incr * (i - id->id_low);
195 set_gatesegd(&kdi_idt[i], (void (*)())hdlr, sel,
196 SDT_SYSIGT, TRP_KPL, IST_DBG);
197 }
198 }
199 }
200
201 static void
202 kdi_idt_gates_install(selector_t sel, int saveold)
203 {
204 gate_desc_t gates[KDI_GATE_NVECS];
205 int i;
206
207 bzero(gates, sizeof (*gates));
208
209 for (i = 0; i < KDI_GATE_NVECS; i++) {
210 const kdi_gate_spec_t *gs = &kdi_gate_specs[i];
211 uintptr_t func = GATESEG_GETOFFSET(&kdi_idt[gs->kgs_vec]);
212 set_gatesegd(&gates[i], (void (*)())func, sel, SDT_SYSIGT,
213 gs->kgs_dpl, IST_DBG);
214 }
215
216 for (i = 0; i < KDI_GATE_NVECS; i++) {
217 uint_t vec = kdi_gate_specs[i].kgs_vec;
218
219 if (saveold)
220 kdi_kgates[i] = CPU->cpu_m.mcpu_idt[vec];
221
222 kdi_idt_write(&gates[i], vec);
223 }
224 }
225
226 static void
227 kdi_idt_gates_restore(void)
228 {
229 int i;
230
231 for (i = 0; i < KDI_GATE_NVECS; i++)
232 kdi_idt_write(&kdi_kgates[i], kdi_gate_specs[i].kgs_vec);
233 }
234
235 /*
236 * Called when we switch to the kernel's IDT. We need to interpose on the
237 * kernel's IDT entries and stop using KMDBCODE_SEL.
238 */
239 void
240 kdi_idt_sync(void)
241 {
242 kdi_idt_init(KCS_SEL);
243 kdi_idt_gates_install(KCS_SEL, KDI_IDT_SAVE);
244 }
245
246 void
247 kdi_update_drreg(kdi_drreg_t *drreg)
248 {
249 kdi_drreg = *drreg;
250 }
251
252 void
253 kdi_memrange_add(caddr_t base, size_t len)
254 {
255 kdi_memrange_t *mr = &kdi_memranges[kdi_nmemranges];
256
257 ASSERT(kdi_nmemranges != KDI_MEMRANGES_MAX);
258
259 mr->mr_base = base;
260 mr->mr_lim = base + len - 1;
261 kdi_nmemranges++;
262 }
263
264 void
265 kdi_idt_switch(kdi_cpusave_t *cpusave)
266 {
267 if (cpusave == NULL)
268 kdi_idtr_set(kdi_idt, sizeof (kdi_idt) - 1);
269 else
270 kdi_idtr_set(cpusave->krs_idt, (sizeof (*idt0) * NIDT) - 1);
271 }
272
273 /*
274 * Activation for CPUs other than the boot CPU, called from that CPU's
275 * mp_startup(). We saved the kernel's descriptors when we initialized the
276 * boot CPU, so we don't want to do it again. Saving the handlers from this
277 * CPU's IDT would actually be dangerous with the CPU initialization method in
278 * use at the time of this writing. With that method, the startup code creates
279 * the IDTs for slave CPUs by copying the one used by the boot CPU, which has
280 * already been interposed upon by KMDB. Were we to interpose again, we'd
281 * replace the kernel's descriptors with our own in the save area. By not
282 * saving, but still overwriting, we'll work in the current world, and in any
283 * future world where the IDT is generated from scratch.
284 */
285 void
286 kdi_cpu_init(void)
287 {
288 kdi_idt_gates_install(KCS_SEL, KDI_IDT_NOSAVE);
289 /* Load the debug registers. */
290 kdi_cpu_debug_init(&kdi_cpusave[CPU->cpu_id]);
291 }
292
293 /*
294 * Activation for all CPUs for mod-loaded kmdb, i.e. a kmdb that wasn't
295 * loaded at boot.
296 */
297 static int
298 kdi_cpu_activate(void)
299 {
300 kdi_idt_gates_install(KCS_SEL, KDI_IDT_SAVE);
301 return (0);
302 }
303
304 void
305 kdi_activate(kdi_main_t main, kdi_cpusave_t *cpusave, uint_t ncpusave)
306 {
307 int i;
308 cpuset_t cpuset;
309
310 CPUSET_ALL(cpuset);
311
312 kdi_cpusave = cpusave;
313 kdi_ncpusave = ncpusave;
314
315 kdi_kmdb_main = main;
316
317 for (i = 0; i < kdi_ncpusave; i++) {
318 kdi_cpusave[i].krs_cpu_id = i;
319
320 kdi_cpusave[i].krs_curcrumb =
321 &kdi_cpusave[i].krs_crumbs[KDI_NCRUMBS - 1];
322 kdi_cpusave[i].krs_curcrumbidx = KDI_NCRUMBS - 1;
323 }
324
325 if (boothowto & RB_KMDB)
326 kdi_idt_init(KMDBCODE_SEL);
327 else
328 kdi_idt_init(KCS_SEL);
329
330 /* The initial selector set. Updated by the debugger-entry code */
331 #ifndef __amd64
332 kdi_cs = B32CODE_SEL;
333 kdi_ds = kdi_fs = kdi_gs = B32DATA_SEL;
334 #endif
335
336 kdi_memranges[0].mr_base = kdi_segdebugbase;
337 kdi_memranges[0].mr_lim = kdi_segdebugbase + kdi_segdebugsize - 1;
338 kdi_nmemranges = 1;
339
340 kdi_drreg.dr_ctl = KDIREG_DRCTL_RESERVED;
341 kdi_drreg.dr_stat = KDIREG_DRSTAT_RESERVED;
342
343 if (boothowto & RB_KMDB) {
344 kdi_idt_gates_install(KMDBCODE_SEL, KDI_IDT_NOSAVE);
345 } else {
346 xc_call(0, 0, 0, CPUSET2BV(cpuset),
347 (xc_func_t)kdi_cpu_activate);
348 }
349 }
350
351 static int
352 kdi_cpu_deactivate(void)
353 {
354 kdi_idt_gates_restore();
355 return (0);
356 }
357
358 void
359 kdi_deactivate(void)
360 {
361 cpuset_t cpuset;
362 CPUSET_ALL(cpuset);
363
364 xc_call(0, 0, 0, CPUSET2BV(cpuset), (xc_func_t)kdi_cpu_deactivate);
365 kdi_nmemranges = 0;
366 }
367
368 /*
369 * We receive all breakpoints and single step traps. Some of them, including
370 * those from userland and those induced by DTrace providers, are intended for
371 * the kernel, and must be processed there. We adopt this
372 * ours-until-proven-otherwise position due to the painful consequences of
373 * sending the kernel an unexpected breakpoint or single step. Unless someone
374 * can prove to us that the kernel is prepared to handle the trap, we'll assume
375 * there's a problem and will give the user a chance to debug it.
376 *
377 * If we return 2, then the calling code should restore the trap-time %cr3: that
378 * is, it really is a kernel-originated trap.
379 */
380 int
381 kdi_trap_pass(kdi_cpusave_t *cpusave)
382 {
383 greg_t tt = cpusave->krs_gregs[KDIREG_TRAPNO];
384 greg_t pc = cpusave->krs_gregs[KDIREG_PC];
385 greg_t cs = cpusave->krs_gregs[KDIREG_CS];
386
387 if (USERMODE(cs))
388 return (1);
389
390 if (tt != T_BPTFLT && tt != T_SGLSTP)
391 return (0);
392
393 if (tt == T_BPTFLT && kdi_dtrace_get_state() ==
394 KDI_DTSTATE_DTRACE_ACTIVE)
395 return (2);
396
397 /*
398 * See the comments in the kernel's T_SGLSTP handler for why we need to
399 * do this.
400 */
401 #if !defined(__xpv)
402 if (tt == T_SGLSTP &&
403 (pc == (greg_t)sys_sysenter || pc == (greg_t)brand_sys_sysenter ||
404 pc == (greg_t)tr_sys_sysenter ||
405 pc == (greg_t)tr_brand_sys_sysenter)) {
406 #else
407 if (tt == T_SGLSTP &&
408 (pc == (greg_t)sys_sysenter || pc == (greg_t)brand_sys_sysenter)) {
409 #endif
410 return (1);
411 }
412
413 return (0);
414 }
415
416 /*
417 * State has been saved, and all CPUs are on the CPU-specific stacks. All
418 * CPUs enter here, and head off into the debugger proper.
419 */
420 void
421 kdi_debugger_entry(kdi_cpusave_t *cpusave)
422 {
423 /*
424 * BPTFLT gives us control with %eip set to the instruction *after*
425 * the int 3. Back it off, so we're looking at the instruction that
426 * triggered the fault.
427 */
428 if (cpusave->krs_gregs[KDIREG_TRAPNO] == T_BPTFLT)
429 cpusave->krs_gregs[KDIREG_PC]--;
430
431 kdi_kmdb_main(cpusave);
432 }