1.0.23.59: bug 3b has been fixed a while now
[sbcl/tcr.git] / src / runtime / ppc-arch.c
blob5e6b8851ecffa9527553ac38808ef6623cd2e10a
1 /*
2 * This software is part of the SBCL system. See the README file for
3 * more information.
5 * This software is derived from the CMU CL system, which was
6 * written at Carnegie Mellon University and released into the
7 * public domain. The software is in the public domain and is
8 * provided with absolutely no warranty. See the COPYING and CREDITS
9 * files for more information.
12 #include <stdio.h>
14 #include "sbcl.h"
15 #include "arch.h"
16 #include "globals.h"
17 #include "validate.h"
18 #include "os.h"
19 #include "lispregs.h"
20 #include "signal.h"
21 #include "interrupt.h"
22 #include "interr.h"
23 #include "breakpoint.h"
24 #include "alloc.h"
26 #if defined(LISP_FEATURE_GENCGC)
27 #include "gencgc-alloc-region.h"
28 #endif
30 /* The header files may not define PT_DAR/PT_DSISR. This definition
31 is correct for all versions of ppc linux >= 2.0.30
33 As of DR2.1u4, MkLinux doesn't pass these registers to signal
34 handlers correctly; a patch is necessary in order to (partially)
35 correct this.
37 Even with the patch, the DSISR may not have its 'write' bit set
38 correctly (it tends not to be set if the fault was caused by
39 something other than a protection violation.)
41 Caveat callers. */
43 #if defined (LISP_FEATURE_DARWIN) || defined(LISP_FEATURE_LINUX)
44 #ifndef PT_DAR
45 #define PT_DAR 41
46 #endif
48 #ifndef PT_DSISR
49 #define PT_DSISR 42
50 #endif
51 #endif
53 void arch_init() {
56 os_vm_address_t
57 arch_get_bad_addr(int sig, siginfo_t *code, os_context_t *context)
59 os_vm_address_t addr;
61 #if defined(LISP_FEATURE_NETBSD)
62 addr = (os_vm_address_t) (code->si_addr);
63 #else
64 addr = (os_vm_address_t) (*os_context_register_addr(context,PT_DAR));
65 #endif
66 return addr;
70 void
71 arch_skip_instruction(os_context_t *context)
73 char** pcptr;
74 pcptr = (char**) os_context_pc_addr(context);
75 *pcptr += 4;
78 unsigned char *
79 arch_internal_error_arguments(os_context_t *context)
81 return (unsigned char *)(*os_context_pc_addr(context)+4);
85 boolean
86 arch_pseudo_atomic_atomic(os_context_t *context)
88 /* FIXME: this foreign_function_call_active test is dubious at
89 * best. If a foreign call is made in a pseudo atomic section
90 * (?) or more likely a pseudo atomic section is in a foreign
91 * call then an interrupt is executed immediately. Maybe it
92 * has to do with C code not maintaining pseudo atomic
93 * properly. MG - 2005-08-10
95 * The foreign_function_call_active used to live at each call-site
96 * to arch_pseudo_atomic_atomic, but this seems clearer.
97 * --NS 2007-05-15 */
98 return (!foreign_function_call_active)
99 && ((*os_context_register_addr(context,reg_ALLOC)) & 4);
102 void
103 arch_set_pseudo_atomic_interrupted(os_context_t *context)
105 *os_context_register_addr(context,reg_ALLOC) |= 1;
108 void
109 arch_clear_pseudo_atomic_interrupted(os_context_t *context)
111 *os_context_register_addr(context,reg_ALLOC) &= ~1;
114 unsigned int
115 arch_install_breakpoint(void *pc)
117 unsigned int *ptr = (unsigned int *)pc;
118 unsigned int result = *ptr;
119 *ptr = (3<<26) | (5 << 21) | trap_Breakpoint;
120 os_flush_icache((os_vm_address_t) pc, sizeof(unsigned int));
121 return result;
124 void
125 arch_remove_breakpoint(void *pc, unsigned int orig_inst)
127 *(unsigned int *)pc = orig_inst;
128 os_flush_icache((os_vm_address_t) pc, sizeof(unsigned int));
132 * Perform the instruction that we overwrote with a breakpoint. As we
133 * don't have a single-step facility, this means we have to:
134 * - put the instruction back
135 * - put a second breakpoint at the following instruction,
136 * set after_breakpoint and continue execution.
138 * When the second breakpoint is hit (very shortly thereafter, we hope)
139 * sigtrap_handler gets called again, but follows the AfterBreakpoint
140 * arm, which
141 * - puts a bpt back in the first breakpoint place (running across a
142 * breakpoint shouldn't cause it to be uninstalled)
143 * - replaces the second bpt with the instruction it was meant to be
144 * - carries on
146 * Clear?
148 static unsigned int *skipped_break_addr, displaced_after_inst;
149 static sigset_t orig_sigmask;
151 void
152 arch_do_displaced_inst(os_context_t *context, unsigned int orig_inst)
154 /* not sure how we ensure that we get the breakpoint reinstalled
155 * after doing this -dan */
156 unsigned int *pc = (unsigned int *)(*os_context_pc_addr(context));
158 orig_sigmask = *os_context_sigmask_addr(context);
159 sigaddset_blockable(os_context_sigmask_addr(context));
161 *pc = orig_inst;
162 os_flush_icache((os_vm_address_t) pc, sizeof(unsigned int));
163 skipped_break_addr = pc;
165 /* FIXME: we should apparently be installing the after-breakpoint
166 * here, but would need to find the next instruction address for
167 * it first. alpha-arch.c shows how to do it. --NS 2007-04-02 */
170 #ifdef LISP_FEATURE_GENCGC
172 * Return non-zero if the current instruction is an allocation trap
174 static int
175 allocation_trap_p(os_context_t * context)
177 int result;
178 unsigned int *pc;
179 unsigned inst;
180 unsigned opcode;
181 unsigned src;
182 unsigned dst;
184 result = 0;
187 * First, the instruction has to be a TWLGE temp, NL3, which has the
188 * format.
189 * | 6| 5| 5 | 5 | 10|1| width
190 * |31|5 |dst|src| 4|0| field
192 pc = (unsigned int *) (*os_context_pc_addr(context));
193 inst = *pc;
195 #if 0
196 fprintf(stderr, "allocation_trap_p at %p: inst = 0x%08x\n", pc, inst);
197 #endif
199 opcode = inst >> 26;
200 src = (inst >> 11) & 0x1f;
201 dst = (inst >> 16) & 0x1f;
202 if ((opcode == 31) && (src == reg_NL3) && (5 == ((inst >> 21) & 0x1f))
203 && (4 == ((inst >> 1) & 0x3ff))) {
205 * We got the instruction. Now, look back to make sure it was
206 * proceeded by what we expected. 2 instructions back should be
207 * an ADD or ADDI instruction.
209 unsigned int add_inst;
211 add_inst = pc[-3];
212 #if 0
213 fprintf(stderr, " add inst at %p: inst = 0x%08x\n",
214 pc - 3, add_inst);
215 #endif
216 opcode = add_inst >> 26;
217 if ((opcode == 31) && (266 == ((add_inst >> 1) & 0x1ff))) {
218 return 1;
219 } else if ((opcode == 14)) {
220 return 1;
221 } else {
222 fprintf(stderr,
223 "Whoa! Got allocation trap but could not find ADD or ADDI instruction: 0x%08x in the proper place\n",
224 add_inst);
227 return 0;
230 extern struct alloc_region boxed_region;
232 void
233 handle_allocation_trap(os_context_t * context)
235 unsigned int *pc;
236 unsigned int inst;
237 unsigned int target, target_ptr, end_addr;
238 unsigned int opcode;
239 int size;
240 boolean were_in_lisp;
241 char *memory;
243 target = 0;
244 size = 0;
246 #if 0
247 fprintf(stderr, "In handle_allocation_trap\n");
248 #endif
250 /* I don't think it's possible for us NOT to be in lisp when we get
251 * here. Remove this later? */
252 were_in_lisp = !foreign_function_call_active;
254 if (were_in_lisp) {
255 fake_foreign_function_call(context);
256 } else {
257 fprintf(stderr, "**** Whoa! allocation trap and we weren't in lisp!\n");
261 * Look at current instruction: TWNE temp, NL3. We're here because
262 * temp > NL3 and temp is the end of the allocation, and NL3 is
263 * current-region-end-addr.
265 * We need to adjust temp and alloc-tn.
268 pc = (unsigned int *) (*os_context_pc_addr(context));
269 inst = pc[0];
270 end_addr = (inst >> 11) & 0x1f;
271 target = (inst >> 16) & 0x1f;
273 target_ptr = *os_context_register_addr(context, target);
275 #if 0
276 fprintf(stderr, "handle_allocation_trap at %p:\n", pc);
277 fprintf(stderr, "boxed_region.free_pointer: %p\n", boxed_region.free_pointer);
278 fprintf(stderr, "boxed_region.end_addr: %p\n", boxed_region.end_addr);
279 fprintf(stderr, "target reg: %d, end_addr reg: %d\n", target, end_addr);
280 fprintf(stderr, "target: %x\n", *os_context_register_addr(context, target));
281 fprintf(stderr, "end_addr: %x\n", *os_context_register_addr(context, end_addr));
282 #endif
284 #if 0
285 fprintf(stderr, "handle_allocation_trap at %p:\n", pc);
286 fprintf(stderr, " trap inst = 0x%08x\n", inst);
287 fprintf(stderr, " target reg = %s\n", lisp_register_names[target]);
288 #endif
291 * Go back and look at the add/addi instruction. The second src arg
292 * is the size of the allocation. Get it and call alloc to allocate
293 * new space.
295 inst = pc[-3];
296 opcode = inst >> 26;
297 #if 0
298 fprintf(stderr, " add inst = 0x%08x, opcode = %d\n", inst, opcode);
299 #endif
300 if (opcode == 14) {
302 * ADDI temp-tn, alloc-tn, size
304 * Extract the size
306 size = (inst & 0xffff);
307 } else if (opcode == 31) {
309 * ADD temp-tn, alloc-tn, size-tn
311 * Extract the size
313 int reg;
315 reg = (inst >> 11) & 0x1f;
316 #if 0
317 fprintf(stderr, " add, reg = %s\n", lisp_register_names[reg]);
318 #endif
319 size = *os_context_register_addr(context, reg);
323 #if 0
324 fprintf(stderr, "Alloc %d to %s\n", size, lisp_register_names[target]);
325 #endif
327 #if INLINE_ALLOC_DEBUG
328 if ((((unsigned long)boxed_region.end_addr + size) / PAGE_SIZE) ==
329 (((unsigned long)boxed_region.end_addr) / PAGE_SIZE)) {
330 fprintf(stderr,"*** possibly bogus trap allocation of %d bytes at %p\n",
331 size, target_ptr);
332 fprintf(stderr, " dynamic_space_free_pointer: %p, boxed_region.end_addr %p\n",
333 dynamic_space_free_pointer, boxed_region.end_addr);
335 #endif
337 #if 0
338 fprintf(stderr, "Ready to alloc\n");
339 fprintf(stderr, "free_pointer = 0x%08x\n",
340 dynamic_space_free_pointer);
341 #endif
344 * alloc-tn was incremented by size. Need to decrement it by size
345 * to restore its original value. This is not true on GENCGC
346 * anymore. d_s_f_p and reg_alloc get out of sync, but the p_a
347 * bits stay intact and we set it to the proper value when it
348 * needs to be. Keep this comment here for the moment in case
349 * somebody tries to figure out what happened here.
351 /* dynamic_space_free_pointer =
352 (lispobj *) ((long) dynamic_space_free_pointer - size);
354 #if 0
355 fprintf(stderr, "free_pointer = 0x%08x new\n",
356 dynamic_space_free_pointer);
357 #endif
359 memory = (char *) alloc(size);
361 #if 0
362 fprintf(stderr, "alloc returned %p\n", memory);
363 fprintf(stderr, "free_pointer = 0x%08x\n",
364 dynamic_space_free_pointer);
365 #endif
368 * The allocation macro wants the result to point to the end of the
369 * object!
371 memory += size;
373 #if 0
374 fprintf(stderr, "object end at %p\n", memory);
375 #endif
377 *os_context_register_addr(context, target) = (unsigned long) memory;
378 *os_context_register_addr(context, reg_ALLOC) =
379 (unsigned long) dynamic_space_free_pointer
380 | (*os_context_register_addr(context, reg_ALLOC)
381 & LOWTAG_MASK);
383 if (were_in_lisp) {
384 undo_fake_foreign_function_call(context);
389 #endif
391 void
392 arch_handle_breakpoint(os_context_t *context)
394 handle_breakpoint(context);
397 void
398 arch_handle_fun_end_breakpoint(os_context_t *context)
400 *os_context_pc_addr(context)
401 =(int)handle_fun_end_breakpoint(context);
404 void
405 arch_handle_after_breakpoint(os_context_t *context)
407 *skipped_break_addr = trap_Breakpoint;
408 skipped_break_addr = NULL;
409 *(unsigned int *)*os_context_pc_addr(context)
410 = displaced_after_inst;
411 *os_context_sigmask_addr(context)= orig_sigmask;
412 os_flush_icache((os_vm_address_t) *os_context_pc_addr(context),
413 sizeof(unsigned int));
416 void
417 arch_handle_single_step_trap(os_context_t *context, int trap)
419 unsigned int code = *((u32 *)(*os_context_pc_addr(context)));
420 int register_offset = code >> 5 & 0x1f;
421 handle_single_step_trap(context, trap, register_offset);
422 arch_skip_instruction(context);
425 static void
426 sigtrap_handler(int signal, siginfo_t *siginfo, void *void_context)
428 unsigned int code;
429 os_context_t *context = void_context;
431 #ifdef LISP_FEATURE_LINUX
432 os_restore_fp_control(context);
433 #endif
434 code=*((u32 *)(*os_context_pc_addr(context)));
435 if (code == ((3 << 26) | (0x18 << 21) | (reg_NL3 << 16))) {
436 arch_clear_pseudo_atomic_interrupted(context);
437 arch_skip_instruction(context);
438 /* interrupt or GC was requested in PA; now we're done with the
439 PA section we may as well get around to it */
440 interrupt_handle_pending(context);
441 return;
444 #ifdef LISP_FEATURE_GENCGC
445 /* Is this an allocation trap? */
446 if (allocation_trap_p(context)) {
447 handle_allocation_trap(context);
448 arch_skip_instruction(context);
449 return;
451 #endif
453 if ((code >> 16) == ((3 << 10) | (6 << 5))) {
454 /* twllei reg_ZERO,N will always trap if reg_ZERO = 0 */
455 int trap = code & 0x1f;
456 handle_trap(context,trap);
457 return;
459 if (((code >> 26) == 3) && (((code >> 21) & 31) == 24)) {
460 interrupt_internal_error(context, 0);
461 return;
464 interrupt_handle_now(signal, (siginfo_t *)code, context);
468 void arch_install_interrupt_handlers()
470 undoably_install_low_level_interrupt_handler(SIGILL, sigtrap_handler);
471 undoably_install_low_level_interrupt_handler(SIGTRAP, sigtrap_handler);
474 void
475 ppc_flush_icache(os_vm_address_t address, os_vm_size_t length)
477 os_vm_address_t end = (os_vm_address_t) ((int)(address+length+(32-1)) &~(32-1));
478 extern void ppc_flush_cache_line(os_vm_address_t);
480 while (address < end) {
481 ppc_flush_cache_line(address);
482 address += 32;
486 #ifdef LISP_FEATURE_LINKAGE_TABLE
488 /* Linkage tables for PowerPC
490 * Linkage entry size is 16, because we need at least 4 instructions to
491 * implement a jump.
495 * Define the registers to use in the linkage jump table. Can be the
496 * same. Some care must be exercised when choosing these. It has to be
497 * a register that is not otherwise being used. reg_NFP is a good
498 * choice. call_into_c trashes reg_NFP without preserving it, so we can
499 * trash it in the linkage jump table.
501 #define LINKAGE_TEMP_REG reg_NFP
502 #define LINKAGE_ADDR_REG reg_NFP
505 * Insert the necessary jump instructions at the given address.
507 void
508 arch_write_linkage_table_jmp(void* reloc_addr, void *target_addr)
511 * Make JMP to function entry.
513 * The instruction sequence is:
515 * addis 13, 0, (hi part of addr)
516 * ori 13, 13, (low part of addr)
517 * mtctr 13
518 * bctr
521 int* inst_ptr;
522 unsigned long hi; /* Top 16 bits of address */
523 unsigned long lo; /* Low 16 bits of address */
524 unsigned int inst;
526 inst_ptr = (int*) reloc_addr;
529 * Split the target address into hi and lo parts for the sethi
530 * instruction. hi is the top 22 bits. lo is the low 10 bits.
532 hi = (unsigned long) target_addr;
533 lo = hi & 0xffff;
534 hi >>= 16;
537 * addis 13, 0, (hi part)
540 inst = (15 << 26) | (LINKAGE_TEMP_REG << 21) | (0 << 16) | hi;
541 *inst_ptr++ = inst;
544 * ori 13, 13, (lo part)
547 inst = (24 << 26) | (LINKAGE_TEMP_REG << 21) | (LINKAGE_TEMP_REG << 16) | lo;
548 *inst_ptr++ = inst;
551 * mtctr 13
554 inst = (31 << 26) | (LINKAGE_TEMP_REG << 21) | (9 << 16) | (467 << 1);
555 *inst_ptr++ = inst;
558 * bctr
561 inst = (19 << 26) | (20 << 21) | (528 << 1);
562 *inst_ptr++ = inst;
565 *inst_ptr++ = inst;
567 os_flush_icache((os_vm_address_t) reloc_addr, (char*) inst_ptr - (char*) reloc_addr);
570 void
571 arch_write_linkage_table_ref(void * reloc_addr, void *target_addr)
573 *(unsigned long *)reloc_addr = (unsigned long)target_addr;
576 #endif