usr/src/uts/intel/kdi/kdi_idt.c

   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
  82 #define KDI_GATE_NVECS  3
  83
  84 #define KDI_IDT_NOSAVE  0
  85 #define KDI_IDT_SAVE    1
  86
  87 #define KDI_IDT_DTYPE_KERNEL    0
  88 #define KDI_IDT_DTYPE_BOOT      1
  89
  90 kdi_cpusave_t *kdi_cpusave;
  91 int kdi_ncpusave;
  92
  93 static kdi_main_t kdi_kmdb_main;
  94
  95 kdi_drreg_t kdi_drreg;
  96
  97 #ifndef __amd64
  98 /* Used to track the current set of valid kernel selectors. */
  99 uint32_t        kdi_cs;
 100 uint32_t        kdi_ds;
 101 uint32_t        kdi_fs;
 102 uint32_t        kdi_gs;
 103 #endif
 104
 105 uintptr_t       kdi_kernel_handler;
 106
 107 int             kdi_trap_switch;
 108
 109 #define KDI_MEMRANGES_MAX       2
 110
 111 kdi_memrange_t  kdi_memranges[KDI_MEMRANGES_MAX];
 112 int             kdi_nmemranges;
 113
 114 typedef void idt_hdlr_f(void);
 115
 116 extern idt_hdlr_f kdi_trap0, kdi_trap1, kdi_int2, kdi_trap3, kdi_trap4;
 117 extern idt_hdlr_f kdi_trap5, kdi_trap6, kdi_trap7, kdi_trap9;
 118 extern idt_hdlr_f kdi_traperr10, kdi_traperr11, kdi_traperr12;
 119 extern idt_hdlr_f kdi_traperr13, kdi_traperr14, kdi_trap16, kdi_trap17;
 120 extern idt_hdlr_f kdi_trap18, kdi_trap19, kdi_trap20, kdi_ivct32;
 121 extern idt_hdlr_f kdi_invaltrap;
 122 extern size_t kdi_ivct_size;
 123
 124 typedef struct kdi_gate_spec {
 125         uint_t kgs_vec;
 126         uint_t kgs_dpl;
 127 } kdi_gate_spec_t;
 128
 129 /*
 130  * Beware: kdi_pass_to_kernel() has unpleasant knowledge of this list.
 131  */
 132 static const kdi_gate_spec_t kdi_gate_specs[KDI_GATE_NVECS] = {
 133         { T_SGLSTP, TRP_KPL },
 134         { T_BPTFLT, TRP_UPL },
 135         { T_DBGENTR, TRP_KPL }
 136 };
 137
 138 static gate_desc_t kdi_kgates[KDI_GATE_NVECS];
 139
 140 gate_desc_t kdi_idt[NIDT];
 141
 142 struct idt_description {
 143         uint_t id_low;
 144         uint_t id_high;
 145         idt_hdlr_f *id_basehdlr;
 146         size_t *id_incrp;
 147 } idt_description[] = {
 148         { T_ZERODIV, 0,         kdi_trap0, NULL },
 149         { T_SGLSTP, 0,          kdi_trap1, NULL },
 150         { T_NMIFLT, 0,          kdi_int2, NULL },
 151         { T_BPTFLT, 0,          kdi_trap3, NULL },
 152         { T_OVFLW, 0,           kdi_trap4, NULL },
 153         { T_BOUNDFLT, 0,        kdi_trap5, NULL },
 154         { T_ILLINST, 0,         kdi_trap6, NULL },
 155         { T_NOEXTFLT, 0,        kdi_trap7, NULL },
 156 #if !defined(__xpv)
 157         { T_DBLFLT, 0,          syserrtrap, NULL },
 158 #endif
 159         { T_EXTOVRFLT, 0,       kdi_trap9, NULL },
 160         { T_TSSFLT, 0,          kdi_traperr10, NULL },
 161         { T_SEGFLT, 0,          kdi_traperr11, NULL },
 162         { T_STKFLT, 0,          kdi_traperr12, NULL },
 163         { T_GPFLT, 0,           kdi_traperr13, NULL },
 164         { T_PGFLT, 0,           kdi_traperr14, NULL },
 165         { 15, 0,                kdi_invaltrap, NULL },
 166         { T_EXTERRFLT, 0,       kdi_trap16, NULL },
 167         { T_ALIGNMENT, 0,       kdi_trap17, NULL },
 168         { T_MCE, 0,             kdi_trap18, NULL },
 169         { T_SIMDFPE, 0,         kdi_trap19, NULL },
 170         { T_DBGENTR, 0,         kdi_trap20, NULL },
 171         { 21, 31,               kdi_invaltrap, NULL },
 172         { 32, 255,              kdi_ivct32, &kdi_ivct_size },
 173         { 0, 0, NULL },
 174 };
 175
 176 void
 177 kdi_idt_init(selector_t sel)
 178 {
 179         struct idt_description *id;
 180         int i;
 181
 182         for (id = idt_description; id->id_basehdlr != NULL; id++) {
 183                 uint_t high = id->id_high != 0 ? id->id_high : id->id_low;
 184                 size_t incr = id->id_incrp != NULL ? *id->id_incrp : 0;
 185
 186                 for (i = id->id_low; i <= high; i++) {
 187                         caddr_t hdlr = (caddr_t)id->id_basehdlr +
 188                             incr * (i - id->id_low);
 189                         set_gatesegd(&kdi_idt[i], (void (*)())hdlr, sel,
 190                             SDT_SYSIGT, TRP_KPL, i);
 191                 }
 192         }
 193 }
 194
 195 static void
 196 kdi_idt_gates_install(selector_t sel, int saveold)
 197 {
 198         gate_desc_t gates[KDI_GATE_NVECS];
 199         int i;
 200
 201         bzero(gates, sizeof (*gates));
 202
 203         for (i = 0; i < KDI_GATE_NVECS; i++) {
 204                 const kdi_gate_spec_t *gs = &kdi_gate_specs[i];
 205                 uintptr_t func = GATESEG_GETOFFSET(&kdi_idt[gs->kgs_vec]);
 206                 set_gatesegd(&gates[i], (void (*)())func, sel, SDT_SYSIGT,
 207                     gs->kgs_dpl, gs->kgs_vec);
 208         }
 209
 210         for (i = 0; i < KDI_GATE_NVECS; i++) {
 211                 uint_t vec = kdi_gate_specs[i].kgs_vec;
 212
 213                 if (saveold)
 214                         kdi_kgates[i] = CPU->cpu_m.mcpu_idt[vec];
 215
 216                 kdi_idt_write(&gates[i], vec);
 217         }
 218 }
 219
 220 static void
 221 kdi_idt_gates_restore(void)
 222 {
 223         int i;
 224
 225         for (i = 0; i < KDI_GATE_NVECS; i++)
 226                 kdi_idt_write(&kdi_kgates[i], kdi_gate_specs[i].kgs_vec);
 227 }
 228
 229 /*
 230  * Called when we switch to the kernel's IDT.  We need to interpose on the
 231  * kernel's IDT entries and stop using KMDBCODE_SEL.
 232  */
 233 void
 234 kdi_idt_sync(void)
 235 {
 236         kdi_idt_init(KCS_SEL);
 237         kdi_idt_gates_install(KCS_SEL, KDI_IDT_SAVE);
 238 }
 239
 240 void
 241 kdi_update_drreg(kdi_drreg_t *drreg)
 242 {
 243         kdi_drreg = *drreg;
 244 }
 245
 246 void
 247 kdi_memrange_add(caddr_t base, size_t len)
 248 {
 249         kdi_memrange_t *mr = &kdi_memranges[kdi_nmemranges];
 250
 251         ASSERT(kdi_nmemranges != KDI_MEMRANGES_MAX);
 252
 253         mr->mr_base = base;
 254         mr->mr_lim = base + len - 1;
 255         kdi_nmemranges++;
 256 }
 257
 258 void
 259 kdi_idt_switch(kdi_cpusave_t *cpusave)
 260 {
 261         if (cpusave == NULL)
 262                 kdi_idtr_set(kdi_idt, sizeof (kdi_idt) - 1);
 263         else
 264                 kdi_idtr_set(cpusave->krs_idt, (sizeof (*idt0) * NIDT) - 1);
 265 }
 266
 267 /*
 268  * Activation for CPUs other than the boot CPU, called from that CPU's
 269  * mp_startup().  We saved the kernel's descriptors when we initialized the
 270  * boot CPU, so we don't want to do it again.  Saving the handlers from this
 271  * CPU's IDT would actually be dangerous with the CPU initialization method in
 272  * use at the time of this writing.  With that method, the startup code creates
 273  * the IDTs for slave CPUs by copying the one used by the boot CPU, which has
 274  * already been interposed upon by KMDB.  Were we to interpose again, we'd
 275  * replace the kernel's descriptors with our own in the save area.  By not
 276  * saving, but still overwriting, we'll work in the current world, and in any
 277  * future world where the IDT is generated from scratch.
 278  */
 279 void
 280 kdi_cpu_init(void)
 281 {
 282         kdi_idt_gates_install(KCS_SEL, KDI_IDT_NOSAVE);
 283         /* Load the debug registers. */
 284         kdi_cpu_debug_init(&kdi_cpusave[CPU->cpu_id]);
 285 }
 286
 287 /*
 288  * Activation for all CPUs for mod-loaded kmdb, i.e. a kmdb that wasn't
 289  * loaded at boot.
 290  */
 291 static int
 292 kdi_cpu_activate(void)
 293 {
 294         kdi_idt_gates_install(KCS_SEL, KDI_IDT_SAVE);
 295         return (0);
 296 }
 297
 298 void
 299 kdi_activate(kdi_main_t main, kdi_cpusave_t *cpusave, uint_t ncpusave)
 300 {
 301         int i;
 302         cpuset_t cpuset;
 303
 304         CPUSET_ALL(cpuset);
 305
 306         kdi_cpusave = cpusave;
 307         kdi_ncpusave = ncpusave;
 308
 309         kdi_kmdb_main = main;
 310
 311         for (i = 0; i < kdi_ncpusave; i++) {
 312                 kdi_cpusave[i].krs_cpu_id = i;
 313
 314                 kdi_cpusave[i].krs_curcrumb =
 315                     &kdi_cpusave[i].krs_crumbs[KDI_NCRUMBS - 1];
 316                 kdi_cpusave[i].krs_curcrumbidx = KDI_NCRUMBS - 1;
 317         }
 318
 319         if (boothowto & RB_KMDB)
 320                 kdi_idt_init(KMDBCODE_SEL);
 321         else
 322                 kdi_idt_init(KCS_SEL);
 323
 324         /* The initial selector set.  Updated by the debugger-entry code */
 325 #ifndef __amd64
 326         kdi_cs = B32CODE_SEL;
 327         kdi_ds = kdi_fs = kdi_gs = B32DATA_SEL;
 328 #endif
 329
 330         kdi_memranges[0].mr_base = kdi_segdebugbase;
 331         kdi_memranges[0].mr_lim = kdi_segdebugbase + kdi_segdebugsize - 1;
 332         kdi_nmemranges = 1;
 333
 334         kdi_drreg.dr_ctl = KDIREG_DRCTL_RESERVED;
 335         kdi_drreg.dr_stat = KDIREG_DRSTAT_RESERVED;
 336
 337         if (boothowto & RB_KMDB) {
 338                 kdi_idt_gates_install(KMDBCODE_SEL, KDI_IDT_NOSAVE);
 339         } else {
 340                 xc_call(0, 0, 0, CPUSET2BV(cpuset),
 341                     (xc_func_t)kdi_cpu_activate);
 342         }
 343 }
 344
 345 static int
 346 kdi_cpu_deactivate(void)
 347 {
 348         kdi_idt_gates_restore();
 349         return (0);
 350 }
 351
 352 void
 353 kdi_deactivate(void)
 354 {
 355         cpuset_t cpuset;
 356         CPUSET_ALL(cpuset);
 357
 358         xc_call(0, 0, 0, CPUSET2BV(cpuset), (xc_func_t)kdi_cpu_deactivate);
 359         kdi_nmemranges = 0;
 360 }
 361
 362 /*
 363  * We receive all breakpoints and single step traps.  Some of them,
 364  * including those from userland and those induced by DTrace providers,
 365  * are intended for the kernel, and must be processed there.  We adopt
 366  * this ours-until-proven-otherwise position due to the painful
 367  * consequences of sending the kernel an unexpected breakpoint or
 368  * single step.  Unless someone can prove to us that the kernel is
 369  * prepared to handle the trap, we'll assume there's a problem and will
 370  * give the user a chance to debug it.
 371  */
 372 int
 373 kdi_trap_pass(kdi_cpusave_t *cpusave)
 374 {
 375         greg_t tt = cpusave->krs_gregs[KDIREG_TRAPNO];
 376         greg_t pc = cpusave->krs_gregs[KDIREG_PC];
 377         greg_t cs = cpusave->krs_gregs[KDIREG_CS];
 378
 379         if (USERMODE(cs))
 380                 return (1);
 381
 382         if (tt != T_BPTFLT && tt != T_SGLSTP)
 383                 return (0);
 384
 385         if (tt == T_BPTFLT && kdi_dtrace_get_state() ==
 386             KDI_DTSTATE_DTRACE_ACTIVE)
 387                 return (1);
 388
 389         /*
 390          * See the comments in the kernel's T_SGLSTP handler for why we need to
 391          * do this.
 392          */
 393         if (tt == T_SGLSTP &&
 394             (pc == (greg_t)sys_sysenter || pc == (greg_t)brand_sys_sysenter))
 395                 return (1);
 396
 397         return (0);
 398 }
 399
 400 /*
 401  * State has been saved, and all CPUs are on the CPU-specific stacks.  All
 402  * CPUs enter here, and head off into the debugger proper.
 403  */
 404 void
 405 kdi_debugger_entry(kdi_cpusave_t *cpusave)
 406 {
 407         /*
 408          * BPTFLT gives us control with %eip set to the instruction *after*
 409          * the int 3.  Back it off, so we're looking at the instruction that
 410          * triggered the fault.
 411          */
 412         if (cpusave->krs_gregs[KDIREG_TRAPNO] == T_BPTFLT)
 413                 cpusave->krs_gregs[KDIREG_PC]--;
 414
 415         kdi_kmdb_main(cpusave);
 416 }