booke_machdep.c revision 1.21 1 /* $NetBSD: booke_machdep.c,v 1.21 2014/09/22 21:35:15 matt Exp $ */
2 /*-
3 * Copyright (c) 2010, 2011 The NetBSD Foundation, Inc.
4 * All rights reserved.
5 *
6 * This code is derived from software contributed to The NetBSD Foundation
7 * by Raytheon BBN Technologies Corp and Defense Advanced Research Projects
8 * Agency and which was developed by Matt Thomas of 3am Software Foundry.
9 *
10 * This material is based upon work supported by the Defense Advanced Research
11 * Projects Agency and Space and Naval Warfare Systems Center, Pacific, under
12 * Contract No. N66001-09-C-2073.
13 * Approved for Public Release, Distribution Unlimited
14 *
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
17 * are met:
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
25 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
26 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
28 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
31 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
32 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
33 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
34 * POSSIBILITY OF SUCH DAMAGE.
35 */
36
37 #define __INTR_PRIVATE
38 #define _POWERPC_BUS_DMA_PRIVATE
39
40 #include <sys/cdefs.h>
41 __KERNEL_RCSID(0, "$NetBSD: booke_machdep.c,v 1.21 2014/09/22 21:35:15 matt Exp $");
42
43 #include "opt_modular.h"
44
45 #include <sys/param.h>
46 #include <sys/cpu.h>
47 #include <sys/device.h>
48 #include <sys/intr.h>
49 #include <sys/mount.h>
50 #include <sys/msgbuf.h>
51 #include <sys/kernel.h>
52 #include <sys/reboot.h>
53 #include <sys/bus.h>
54 #include <sys/cpu.h>
55
56 #include <uvm/uvm_extern.h>
57
58 #include <powerpc/pcb.h>
59 #include <powerpc/spr.h>
60 #include <powerpc/booke/spr.h>
61 #include <powerpc/booke/cpuvar.h>
62
63 /*
64 * Global variables used here and there
65 */
66 paddr_t msgbuf_paddr;
67 psize_t pmemsize;
68 struct vm_map *phys_map;
69
70 #ifdef MODULAR
71 register_t cpu_psluserset = PSL_USERSET;
72 register_t cpu_pslusermod = PSL_USERMOD;
73 register_t cpu_pslusermask = PSL_USERMASK;
74 #endif
75
76 static bus_addr_t booke_dma_phys_to_bus_mem(bus_dma_tag_t, bus_addr_t);
77 static bus_addr_t booke_dma_bus_mem_to_phys(bus_dma_tag_t, bus_addr_t);
78
79
80 struct powerpc_bus_dma_tag booke_bus_dma_tag = {
81 ._dmamap_create = _bus_dmamap_create,
82 ._dmamap_destroy = _bus_dmamap_destroy,
83 ._dmamap_load = _bus_dmamap_load,
84 ._dmamap_load_mbuf = _bus_dmamap_load_mbuf,
85 ._dmamap_load_uio = _bus_dmamap_load_uio,
86 ._dmamap_load_raw = _bus_dmamap_load_raw,
87 ._dmamap_unload = _bus_dmamap_unload,
88 /*
89 * The caches on BookE are coherent so we don't need to do any special
90 * cache synchronization.
91 */
92 //._dmamap_sync = _bus_dmamap_sync,
93 ._dmamem_alloc = _bus_dmamem_alloc,
94 ._dmamem_free = _bus_dmamem_free,
95 ._dmamem_map = _bus_dmamem_map,
96 ._dmamem_unmap = _bus_dmamem_unmap,
97 ._dmamem_mmap = _bus_dmamem_mmap,
98 ._dma_phys_to_bus_mem = booke_dma_phys_to_bus_mem,
99 ._dma_bus_mem_to_phys = booke_dma_bus_mem_to_phys,
100 };
101
102 static bus_addr_t
103 booke_dma_phys_to_bus_mem(bus_dma_tag_t t, bus_addr_t a)
104 {
105 return a;
106 }
107
108 static bus_addr_t
109 booke_dma_bus_mem_to_phys(bus_dma_tag_t t, bus_addr_t a)
110 {
111 return a;
112 }
113
114 struct cpu_md_ops cpu_md_ops;
115
116 struct cpu_softc cpu_softc[] = {
117 [0] = {
118 .cpu_ci = &cpu_info[0],
119 },
120 #ifdef MULTIPROCESSOR
121 [CPU_MAXNUM-1] = {
122 .cpu_ci = &cpu_info[CPU_MAXNUM-1],
123 },
124 #endif
125 };
126 struct cpu_info cpu_info[] = {
127 [0] = {
128 .ci_curlwp = &lwp0,
129 .ci_tlb_info = &pmap_tlb0_info,
130 .ci_softc = &cpu_softc[0],
131 .ci_cpl = IPL_HIGH,
132 .ci_idepth = -1,
133 },
134 #ifdef MULTIPROCESSOR
135 [CPU_MAXNUM-1] = {
136 .ci_curlwp = NULL,
137 .ci_tlb_info = &pmap_tlb0_info,
138 .ci_softc = &cpu_softc[CPU_MAXNUM-1],
139 .ci_cpl = IPL_HIGH,
140 .ci_idepth = -1,
141 },
142 #endif
143 };
144 __CTASSERT(__arraycount(cpu_info) == __arraycount(cpu_softc));
145
146 /*
147 * This should probably be in autoconf! XXX
148 */
149 char machine[] = MACHINE; /* from <machine/param.h> */
150 char machine_arch[] = MACHINE_ARCH; /* from <machine/param.h> */
151
152 char bootpath[256];
153
154 #if NKSYMS || defined(DDB) || defined(MODULAR)
155 void *startsym, *endsym;
156 #endif
157
158 #if defined(MULTIPROCESSOR)
159 volatile struct cpu_hatch_data cpu_hatch_data __cacheline_aligned;
160 #endif
161
162 int fake_mapiodev = 1;
163
164 void
165 booke_cpu_startup(const char *model)
166 {
167 vaddr_t minaddr, maxaddr;
168 char pbuf[9];
169
170 cpu_setmodel("%s", model);
171
172 printf("%s%s", copyright, version);
173
174 format_bytes(pbuf, sizeof(pbuf), ctob((uint64_t)physmem));
175 printf("total memory = %s\n", pbuf);
176
177 minaddr = 0;
178 /*
179 * Allocate a submap for physio
180 */
181 phys_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
182 VM_PHYS_SIZE, 0, false, NULL);
183
184 /*
185 * No need to allocate an mbuf cluster submap. Mbuf clusters
186 * are allocated via the pool allocator, and we use direct-mapped
187 * pool pages.
188 */
189
190 format_bytes(pbuf, sizeof(pbuf), ptoa(uvmexp.free));
191 printf("avail memory = %s\n", pbuf);
192
193 /*
194 * Register the tlb's evcnts
195 */
196 pmap_tlb_info_evcnt_attach(curcpu()->ci_tlb_info);
197
198 /*
199 * Set up the board properties database.
200 */
201 board_info_init();
202
203 /*
204 * Now that we have VM, malloc()s are OK in bus_space.
205 */
206 bus_space_mallocok();
207 fake_mapiodev = 0;
208
209 #ifdef MULTIPROCESSOR
210 for (size_t i = 1; i < __arraycount(cpu_info); i++) {
211 struct cpu_info * const ci = &cpu_info[i];
212 struct cpu_softc * const cpu = &cpu_softc[i];
213 cpu->cpu_ci = ci;
214 cpu->cpu_bst = cpu_softc[0].cpu_bst;
215 cpu->cpu_le_bst = cpu_softc[0].cpu_le_bst;
216 cpu->cpu_bsh = cpu_softc[0].cpu_bsh;
217 cpu->cpu_highmem = cpu_softc[0].cpu_highmem;
218 ci->ci_softc = cpu;
219 ci->ci_tlb_info = &pmap_tlb0_info;
220 ci->ci_cpl = IPL_HIGH;
221 ci->ci_idepth = -1;
222 ci->ci_pmap_kern_segtab = curcpu()->ci_pmap_kern_segtab;
223 }
224
225 kcpuset_create(&cpuset_info.cpus_running, true);
226 kcpuset_create(&cpuset_info.cpus_hatched, true);
227 kcpuset_create(&cpuset_info.cpus_paused, true);
228 kcpuset_create(&cpuset_info.cpus_resumed, true);
229 kcpuset_create(&cpuset_info.cpus_halted, true);
230 #endif /* MULTIPROCESSOR */
231 }
232
233 static void
234 dumpsys(void)
235 {
236
237 printf("dumpsys: TBD\n");
238 }
239
240 /*
241 * Halt or reboot the machine after syncing/dumping according to howto.
242 */
243 void
244 cpu_reboot(int howto, char *what)
245 {
246 static int syncing;
247 static char str[256];
248 char *ap = str, *ap1 = ap;
249
250 boothowto = howto;
251 if (!cold && !(howto & RB_NOSYNC) && !syncing) {
252 syncing = 1;
253 vfs_shutdown(); /* sync */
254 resettodr(); /* set wall clock */
255 }
256
257 splhigh();
258
259 if (!cold && (howto & RB_DUMP))
260 dumpsys();
261
262 doshutdownhooks();
263
264 pmf_system_shutdown(boothowto);
265
266 if ((howto & RB_POWERDOWN) == RB_POWERDOWN) {
267 /* Power off here if we know how...*/
268 }
269
270 if (howto & RB_HALT) {
271 printf("halted\n\n");
272
273 goto reboot; /* XXX for now... */
274
275 #ifdef DDB
276 printf("dropping to debugger\n");
277 while(1)
278 Debugger();
279 #endif
280 }
281
282 printf("rebooting\n\n");
283 if (what && *what) {
284 if (strlen(what) > sizeof str - 5)
285 printf("boot string too large, ignored\n");
286 else {
287 strcpy(str, what);
288 ap1 = ap = str + strlen(str);
289 *ap++ = ' ';
290 }
291 }
292 *ap++ = '-';
293 if (howto & RB_SINGLE)
294 *ap++ = 's';
295 if (howto & RB_KDB)
296 *ap++ = 'd';
297 *ap++ = 0;
298 if (ap[-2] == '-')
299 *ap1 = 0;
300
301 /* flush cache for msgbuf */
302 dcache_wb(msgbuf_paddr, round_page(MSGBUFSIZE));
303
304 reboot:
305 __asm volatile("msync; isync");
306 (*cpu_md_ops.md_cpu_reset)();
307
308 printf("%s: md_cpu_reset() failed!\n", __func__);
309 #ifdef DDB
310 for (;;)
311 Debugger();
312 #else
313 for (;;)
314 /* nothing */;
315 #endif
316 }
317
318 /*
319 * mapiodev:
320 *
321 * Allocate vm space and mapin the I/O address. Use reserved TLB
322 * mapping if one is found.
323 */
324 void *
325 mapiodev(paddr_t pa, psize_t len, bool prefetchable)
326 {
327 const vsize_t off = pa & PAGE_MASK;
328
329 /*
330 * See if we have reserved TLB entry for the pa. This needs to be
331 * true for console as we can't use uvm during early bootstrap.
332 */
333 void * const p = tlb_mapiodev(pa, len, prefetchable);
334 if (p != NULL)
335 return p;
336
337 if (fake_mapiodev)
338 panic("mapiodev: no TLB entry reserved for %llx+%llx",
339 (long long)pa, (long long)len);
340
341 const paddr_t orig_pa = pa;
342 const psize_t orig_len = len;
343 vsize_t align = 0;
344 pa = trunc_page(pa);
345 len = round_page(off + len);
346 /*
347 * If we are allocating a large amount (>= 1MB) try to get an
348 * aligned VA region for it so try to do a large mapping for it.
349 */
350 if ((len & (len - 1)) == 0 && len >= 0x100000)
351 align = len;
352
353 vaddr_t va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
354
355 if (va == 0 && align > 0) {
356 /*
357 * Large aligned request failed. Let's just get anything.
358 */
359 align = 0;
360 va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
361 }
362 if (va == 0)
363 return NULL;
364
365 if (align) {
366 /*
367 * Now try to map that via one big TLB entry.
368 */
369 pt_entry_t pte = pte_make_kenter_pa(pa, NULL,
370 VM_PROT_READ|VM_PROT_WRITE,
371 prefetchable ? 0 : PMAP_NOCACHE);
372 if (!tlb_ioreserve(va, len, pte)) {
373 void * const p0 = tlb_mapiodev(orig_pa, orig_len,
374 prefetchable);
375 KASSERT(p0 != NULL);
376 return p0;
377 }
378 }
379
380 for (va += len, pa += len; len > 0; len -= PAGE_SIZE) {
381 va -= PAGE_SIZE;
382 pa -= PAGE_SIZE;
383 pmap_kenter_pa(va, pa, VM_PROT_READ|VM_PROT_WRITE,
384 prefetchable ? 0 : PMAP_NOCACHE);
385 }
386 pmap_update(pmap_kernel());
387 return (void *)(va + off);
388 }
389
390 void
391 unmapiodev(vaddr_t va, vsize_t len)
392 {
393 /* Nothing to do for reserved (ie. not uvm_km_alloc'd) mappings. */
394 if (va < VM_MIN_KERNEL_ADDRESS || va > VM_MAX_KERNEL_ADDRESS) {
395 tlb_unmapiodev(va, len);
396 return;
397 }
398
399 len = round_page((va & PAGE_MASK) + len);
400 va = trunc_page(va);
401
402 pmap_kremove(va, len);
403 uvm_km_free(kernel_map, va, len, UVM_KMF_VAONLY);
404 }
405
406 void
407 cpu_evcnt_attach(struct cpu_info *ci)
408 {
409 struct cpu_softc * const cpu = ci->ci_softc;
410 const char * const xname = ci->ci_data.cpu_name;
411
412 evcnt_attach_dynamic_nozero(&ci->ci_ev_clock, EVCNT_TYPE_INTR,
413 NULL, xname, "clock");
414 evcnt_attach_dynamic_nozero(&cpu->cpu_ev_late_clock, EVCNT_TYPE_INTR,
415 NULL, xname, "late clock");
416 evcnt_attach_dynamic_nozero(&cpu->cpu_ev_exec_trap_sync, EVCNT_TYPE_TRAP,
417 NULL, xname, "exec pages synced (trap)");
418 evcnt_attach_dynamic_nozero(&ci->ci_ev_traps, EVCNT_TYPE_TRAP,
419 NULL, xname, "traps");
420 evcnt_attach_dynamic_nozero(&ci->ci_ev_kdsi, EVCNT_TYPE_TRAP,
421 &ci->ci_ev_traps, xname, "kernel DSI traps");
422 evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi, EVCNT_TYPE_TRAP,
423 &ci->ci_ev_traps, xname, "user DSI traps");
424 evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi_fatal, EVCNT_TYPE_TRAP,
425 &ci->ci_ev_udsi, xname, "user DSI failures");
426 evcnt_attach_dynamic_nozero(&ci->ci_ev_kisi, EVCNT_TYPE_TRAP,
427 &ci->ci_ev_traps, xname, "kernel ISI traps");
428 evcnt_attach_dynamic_nozero(&ci->ci_ev_isi, EVCNT_TYPE_TRAP,
429 &ci->ci_ev_traps, xname, "user ISI traps");
430 evcnt_attach_dynamic_nozero(&ci->ci_ev_isi_fatal, EVCNT_TYPE_TRAP,
431 &ci->ci_ev_isi, xname, "user ISI failures");
432 evcnt_attach_dynamic_nozero(&ci->ci_ev_scalls, EVCNT_TYPE_TRAP,
433 &ci->ci_ev_traps, xname, "system call traps");
434 evcnt_attach_dynamic_nozero(&ci->ci_ev_pgm, EVCNT_TYPE_TRAP,
435 &ci->ci_ev_traps, xname, "PGM traps");
436 evcnt_attach_dynamic_nozero(&ci->ci_ev_debug, EVCNT_TYPE_TRAP,
437 &ci->ci_ev_traps, xname, "debug traps");
438 evcnt_attach_dynamic_nozero(&ci->ci_ev_fpu, EVCNT_TYPE_TRAP,
439 &ci->ci_ev_traps, xname, "FPU unavailable traps");
440 evcnt_attach_dynamic_nozero(&ci->ci_ev_fpusw, EVCNT_TYPE_MISC,
441 &ci->ci_ev_fpu, xname, "FPU context switches");
442 evcnt_attach_dynamic_nozero(&ci->ci_ev_ali, EVCNT_TYPE_TRAP,
443 &ci->ci_ev_traps, xname, "user alignment traps");
444 evcnt_attach_dynamic_nozero(&ci->ci_ev_ali_fatal, EVCNT_TYPE_TRAP,
445 &ci->ci_ev_ali, xname, "user alignment traps");
446 evcnt_attach_dynamic_nozero(&ci->ci_ev_umchk, EVCNT_TYPE_TRAP,
447 &ci->ci_ev_umchk, xname, "user MCHK failures");
448 evcnt_attach_dynamic_nozero(&ci->ci_ev_vec, EVCNT_TYPE_TRAP,
449 &ci->ci_ev_traps, xname, "SPE unavailable");
450 evcnt_attach_dynamic_nozero(&ci->ci_ev_vecsw, EVCNT_TYPE_MISC,
451 &ci->ci_ev_vec, xname, "SPE context switches");
452 evcnt_attach_dynamic_nozero(&ci->ci_ev_ipi, EVCNT_TYPE_INTR,
453 NULL, xname, "IPIs");
454 evcnt_attach_dynamic_nozero(&ci->ci_ev_tlbmiss_soft, EVCNT_TYPE_TRAP,
455 &ci->ci_ev_traps, xname, "soft tlb misses");
456 evcnt_attach_dynamic_nozero(&ci->ci_ev_dtlbmiss_hard, EVCNT_TYPE_TRAP,
457 &ci->ci_ev_traps, xname, "data tlb misses");
458 evcnt_attach_dynamic_nozero(&ci->ci_ev_itlbmiss_hard, EVCNT_TYPE_TRAP,
459 &ci->ci_ev_traps, xname, "inst tlb misses");
460 }
461
462 #ifdef MULTIPROCESSOR
463 register_t
464 cpu_hatch(void)
465 {
466 struct cpuset_info * const csi = &cpuset_info;
467 const size_t id = cpu_number();
468
469 /*
470 * We've hatched so tell the spinup code.
471 */
472 kcpuset_set(csi->cpus_hatched, id);
473
474 /*
475 * Loop until running bit for this cpu is set.
476 */
477 while (!kcpuset_isset(csi->cpus_running, id)) {
478 continue;
479 }
480
481 /*
482 * Now that we are active, start the clocks.
483 */
484 cpu_initclocks();
485
486 /*
487 * Return sp of the idlelwp. Which we should be already using but ...
488 */
489 return curcpu()->ci_curpcb->pcb_sp;
490 }
491
492 void
493 cpu_boot_secondary_processors(void)
494 {
495 volatile struct cpuset_info * const csi = &cpuset_info;
496 CPU_INFO_ITERATOR cii;
497 struct cpu_info *ci;
498 kcpuset_t *running;
499
500 kcpuset_create(&running, true);
501
502 for (CPU_INFO_FOREACH(cii, ci)) {
503 /*
504 * Skip this CPU if it didn't sucessfully hatch.
505 */
506 if (!kcpuset_isset(csi->cpus_hatched, cpu_index(ci)))
507 continue;
508
509 KASSERT(!CPU_IS_PRIMARY(ci));
510 KASSERT(ci->ci_data.cpu_idlelwp);
511
512 kcpuset_set(running, cpu_index(ci));
513 }
514 KASSERT(kcpuset_match(csi->cpus_hatched, running));
515 if (!kcpuset_iszero(running)) {
516 kcpuset_merge(csi->cpus_running, running);
517 }
518 kcpuset_destroy(running);
519 }
520 #endif
521
522 uint32_t
523 cpu_read_4(bus_addr_t a)
524 {
525 struct cpu_softc * const cpu = curcpu()->ci_softc;
526 // printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
527 return bus_space_read_4(cpu->cpu_bst, cpu->cpu_bsh, a);
528 }
529
530 uint8_t
531 cpu_read_1(bus_addr_t a)
532 {
533 struct cpu_softc * const cpu = curcpu()->ci_softc;
534 // printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
535 return bus_space_read_1(cpu->cpu_bst, cpu->cpu_bsh, a);
536 }
537
538 void
539 cpu_write_4(bus_addr_t a, uint32_t v)
540 {
541 struct cpu_softc * const cpu = curcpu()->ci_softc;
542 bus_space_write_4(cpu->cpu_bst, cpu->cpu_bsh, a, v);
543 }
544
545 void
546 cpu_write_1(bus_addr_t a, uint8_t v)
547 {
548 struct cpu_softc * const cpu = curcpu()->ci_softc;
549 bus_space_write_1(cpu->cpu_bst, cpu->cpu_bsh, a, v);
550 }
551
552 void
553 booke_sstep(struct trapframe *tf)
554 {
555 KASSERT(tf->tf_srr1 & PSL_DE);
556 const uint32_t insn = ufetch_32((const void *)tf->tf_srr0);
557 register_t dbcr0 = DBCR0_IAC1 | DBCR0_IDM;
558 register_t dbcr1 = DBCR1_IAC1US_USER | DBCR1_IAC1ER_DS1;
559 if ((insn >> 28) == 4) {
560 uint32_t iac2 = 0;
561 if ((insn >> 26) == 0x12) {
562 const int32_t off = (((int32_t)insn << 6) >> 6) & ~3;
563 iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
564 dbcr0 |= DBCR0_IAC2;
565 } else if ((insn >> 26) == 0x10) {
566 const int16_t off = insn & ~3;
567 iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
568 dbcr0 |= DBCR0_IAC2;
569 } else if ((insn & 0xfc00fffe) == 0x4c000420) {
570 iac2 = tf->tf_ctr;
571 dbcr0 |= DBCR0_IAC2;
572 } else if ((insn & 0xfc00fffe) == 0x4c000020) {
573 iac2 = tf->tf_lr;
574 dbcr0 |= DBCR0_IAC2;
575 }
576 if (dbcr0 & DBCR0_IAC2) {
577 dbcr1 |= DBCR1_IAC2US_USER | DBCR1_IAC2ER_DS1;
578 mtspr(SPR_IAC2, iac2);
579 }
580 }
581 mtspr(SPR_IAC1, tf->tf_srr0 + 4);
582 mtspr(SPR_DBCR1, dbcr1);
583 mtspr(SPR_DBCR0, dbcr0);
584 }
585
586 #ifdef DIAGNOSTIC
587 static inline void
588 swap_data(uint64_t *data, size_t a, size_t b)
589 {
590 uint64_t swap = data[a];
591 data[a] = data[b];
592 data[b] = swap;
593 }
594
595 static void
596 sort_data(uint64_t *data, size_t count)
597 {
598 #if 0
599 /*
600 * Mostly classic bubble sort
601 */
602 do {
603 size_t new_count = 0;
604 for (size_t i = 1; i < count; i++) {
605 if (tbs[i - 1] > tbs[i]) {
606 swap_tbs(tbs, i - 1, i);
607 new_count = i;
608 }
609 }
610 count = new_count;
611 } while (count > 0);
612 #else
613 /*
614 * Comb sort
615 */
616 size_t gap = count;
617 bool swapped = false;
618 while (gap > 1 || swapped) {
619 if (gap > 1) {
620 /*
621 * phi = (1 + sqrt(5)) / 2 [golden ratio]
622 * N = 1 / (1 - e^-phi)) = 1.247330950103979
623 *
624 * We want to but can't use floating point to calculate
625 * gap = (size_t)((double)gap / N)
626 *
627 * So we will use the multicative inverse of N
628 * (module 65536) to achieve the division.
629 *
630 * iN = 2^16 / 1.24733... = 52540
631 * x / N == (x * iN) / 65536
632 */
633 gap = (gap * 52540) / 65536;
634 }
635
636 swapped = false;
637
638 for (size_t i = 0; gap + i < count; i++) {
639 if (data[i] > data[i + gap]) {
640 swap_data(data, i, i + gap);
641 swapped = true;
642 }
643 }
644 }
645 #endif
646 }
647 #endif
648
649 void
650 dump_splhist(struct cpu_info *ci, void (*pr)(const char *, ...))
651 {
652 #ifdef DIAGNOSTIC
653 struct cpu_softc * const cpu = ci->ci_softc;
654 uint64_t tbs[NIPL*NIPL];
655 size_t ntbs = 0;
656 for (size_t to = 0; to < NIPL; to++) {
657 for (size_t from = 0; from < NIPL; from++) {
658 uint64_t tb = cpu->cpu_spl_tb[to][from];
659 if (tb == 0)
660 continue;
661 tbs[ntbs++] = (tb << 8) | (to << 4) | from;
662 }
663 }
664 sort_data(tbs, ntbs);
665
666 if (pr == NULL)
667 pr = printf;
668 uint64_t last_tb = 0;
669 for (size_t i = 0; i < ntbs; i++) {
670 uint64_t tb = tbs[i];
671 size_t from = tb & 15;
672 size_t to = (tb >> 4) & 15;
673 tb >>= 8;
674 (*pr)("%s(%zu) from %zu at %"PRId64"",
675 from < to ? "splraise" : "splx",
676 to, from, tb);
677 if (last_tb && from != IPL_NONE)
678 (*pr)(" (+%"PRId64")", tb - last_tb);
679 (*pr)("\n");
680 last_tb = tb;
681 }
682 #endif
683 }
684