Home | History | Annotate | Line # | Download | only in booke
booke_machdep.c revision 1.26
      1 /*	$NetBSD: booke_machdep.c,v 1.26 2019/04/07 05:25:55 thorpej 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 1.26 2019/04/07 05:25:55 thorpej 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 <dev/cons.h>
     59 
     60 #include <powerpc/pcb.h>
     61 #include <powerpc/spr.h>
     62 #include <powerpc/booke/spr.h>
     63 #include <powerpc/booke/cpuvar.h>
     64 
     65 /*
     66  * Global variables used here and there
     67  */
     68 paddr_t msgbuf_paddr;
     69 psize_t pmemsize;
     70 struct vm_map *phys_map;
     71 
     72 #ifdef MODULAR
     73 register_t cpu_psluserset = PSL_USERSET;
     74 register_t cpu_pslusermod = PSL_USERMOD;
     75 register_t cpu_pslusermask = PSL_USERMASK;
     76 #endif
     77 
     78 static bus_addr_t booke_dma_phys_to_bus_mem(bus_dma_tag_t, bus_addr_t);
     79 static bus_addr_t booke_dma_bus_mem_to_phys(bus_dma_tag_t, bus_addr_t);
     80 
     81 
     82 struct powerpc_bus_dma_tag booke_bus_dma_tag = {
     83 	._dmamap_create = _bus_dmamap_create,
     84 	._dmamap_destroy = _bus_dmamap_destroy,
     85 	._dmamap_load = _bus_dmamap_load,
     86 	._dmamap_load_mbuf = _bus_dmamap_load_mbuf,
     87 	._dmamap_load_uio = _bus_dmamap_load_uio,
     88 	._dmamap_load_raw = _bus_dmamap_load_raw,
     89 	._dmamap_unload = _bus_dmamap_unload,
     90 	/*
     91 	 * The caches on BookE are coherent so we don't need to do any special
     92 	 * cache synchronization.
     93 	 */
     94 	//._dmamap_sync = _bus_dmamap_sync,
     95 	._dmamem_alloc = _bus_dmamem_alloc,
     96 	._dmamem_free = _bus_dmamem_free,
     97 	._dmamem_map = _bus_dmamem_map,
     98 	._dmamem_unmap = _bus_dmamem_unmap,
     99 	._dmamem_mmap = _bus_dmamem_mmap,
    100 	._dma_phys_to_bus_mem = booke_dma_phys_to_bus_mem,
    101 	._dma_bus_mem_to_phys = booke_dma_bus_mem_to_phys,
    102 };
    103 
    104 static bus_addr_t
    105 booke_dma_phys_to_bus_mem(bus_dma_tag_t t, bus_addr_t a)
    106 {
    107 	return a;
    108 }
    109 
    110 static bus_addr_t
    111 booke_dma_bus_mem_to_phys(bus_dma_tag_t t, bus_addr_t a)
    112 {
    113 	return a;
    114 }
    115 
    116 struct cpu_md_ops cpu_md_ops;
    117 
    118 struct cpu_softc cpu_softc[] = {
    119 	[0] = {
    120 		.cpu_ci = &cpu_info[0],
    121 	},
    122 #ifdef MULTIPROCESSOR
    123 	[CPU_MAXNUM-1] = {
    124 		.cpu_ci = &cpu_info[CPU_MAXNUM-1],
    125 	},
    126 #endif
    127 };
    128 struct cpu_info cpu_info[] = {
    129 	[0] = {
    130 		.ci_curlwp = &lwp0,
    131 		.ci_tlb_info = &pmap_tlb0_info,
    132 		.ci_softc = &cpu_softc[0],
    133 		.ci_cpl = IPL_HIGH,
    134 		.ci_idepth = -1,
    135 		.ci_pmap_kern_segtab = &pmap_kern_segtab,
    136 	},
    137 #ifdef MULTIPROCESSOR
    138 	[CPU_MAXNUM-1] = {
    139 		.ci_curlwp = NULL,
    140 		.ci_tlb_info = &pmap_tlb0_info,
    141 		.ci_softc = &cpu_softc[CPU_MAXNUM-1],
    142 		.ci_cpl = IPL_HIGH,
    143 		.ci_idepth = -1,
    144 		.ci_pmap_kern_segtab = &pmap_kern_segtab,
    145 	},
    146 #endif
    147 };
    148 __CTASSERT(__arraycount(cpu_info) == __arraycount(cpu_softc));
    149 
    150 /*
    151  * This should probably be in autoconf!				XXX
    152  */
    153 char machine[] = MACHINE;		/* from <machine/param.h> */
    154 char machine_arch[] = MACHINE_ARCH;	/* from <machine/param.h> */
    155 
    156 char bootpath[256];
    157 
    158 #if NKSYMS || defined(DDB) || defined(MODULAR)
    159 void *startsym, *endsym;
    160 #endif
    161 
    162 #if defined(MULTIPROCESSOR)
    163 volatile struct cpu_hatch_data cpu_hatch_data __cacheline_aligned;
    164 #endif
    165 
    166 int fake_mapiodev = 1;
    167 
    168 void
    169 booke_cpu_startup(const char *model)
    170 {
    171 	vaddr_t 	minaddr, maxaddr;
    172 	char 		pbuf[9];
    173 
    174 	cpu_setmodel("%s", model);
    175 
    176 	printf("%s%s", copyright, version);
    177 
    178 	format_bytes(pbuf, sizeof(pbuf), ctob((uint64_t)physmem));
    179 	printf("total memory = %s\n", pbuf);
    180 
    181 	minaddr = 0;
    182 	/*
    183 	 * Allocate a submap for physio
    184 	 */
    185 	phys_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
    186 				 VM_PHYS_SIZE, 0, false, NULL);
    187 
    188 	/*
    189 	 * No need to allocate an mbuf cluster submap.  Mbuf clusters
    190 	 * are allocated via the pool allocator, and we use direct-mapped
    191 	 * pool pages.
    192 	 */
    193 
    194 	format_bytes(pbuf, sizeof(pbuf), ptoa(uvmexp.free));
    195 	printf("avail memory = %s\n", pbuf);
    196 
    197 	/*
    198 	 * Register the tlb's evcnts
    199 	 */
    200 	pmap_tlb_info_evcnt_attach(curcpu()->ci_tlb_info);
    201 
    202 	/*
    203 	 * Set up the board properties database.
    204 	 */
    205 	board_info_init();
    206 
    207 	/*
    208 	 * Now that we have VM, malloc()s are OK in bus_space.
    209 	 */
    210 	bus_space_mallocok();
    211 	fake_mapiodev = 0;
    212 
    213 #ifdef MULTIPROCESSOR
    214 	pmap_kernel()->pm_active = kcpuset_running;
    215 	pmap_kernel()->pm_onproc = kcpuset_running;
    216 
    217 	for (size_t i = 1; i < __arraycount(cpu_info); i++) {
    218 		struct cpu_info * const ci = &cpu_info[i];
    219 		struct cpu_softc * const cpu = &cpu_softc[i];
    220 		cpu->cpu_ci = ci;
    221 		cpu->cpu_bst = cpu_softc[0].cpu_bst;
    222 		cpu->cpu_le_bst = cpu_softc[0].cpu_le_bst;
    223 		cpu->cpu_bsh = cpu_softc[0].cpu_bsh;
    224 		cpu->cpu_highmem = cpu_softc[0].cpu_highmem;
    225 		ci->ci_softc = cpu;
    226 		ci->ci_tlb_info = &pmap_tlb0_info;
    227 		ci->ci_cpl = IPL_HIGH;
    228 		ci->ci_idepth = -1;
    229 		ci->ci_pmap_kern_segtab = curcpu()->ci_pmap_kern_segtab;
    230 	}
    231 
    232 	kcpuset_create(&cpuset_info.cpus_running, true);
    233 	kcpuset_create(&cpuset_info.cpus_hatched, true);
    234 	kcpuset_create(&cpuset_info.cpus_paused, true);
    235 	kcpuset_create(&cpuset_info.cpus_resumed, true);
    236 	kcpuset_create(&cpuset_info.cpus_halted, true);
    237 
    238 	kcpuset_set(cpuset_info.cpus_running, cpu_number());
    239 #endif /* MULTIPROCESSOR */
    240 }
    241 
    242 static void
    243 dumpsys(void)
    244 {
    245 
    246 	printf("dumpsys: TBD\n");
    247 }
    248 
    249 /*
    250  * Halt or reboot the machine after syncing/dumping according to howto.
    251  */
    252 void
    253 cpu_reboot(int howto, char *what)
    254 {
    255 	static int syncing;
    256 	static char str[256];
    257 	char *ap = str, *ap1 = ap;
    258 
    259 	boothowto = howto;
    260 	if (!cold && !(howto & RB_NOSYNC) && !syncing) {
    261 		syncing = 1;
    262 		vfs_shutdown();		/* sync */
    263 		resettodr();		/* set wall clock */
    264 	}
    265 
    266 	splhigh();
    267 
    268 	if (!cold && (howto & RB_DUMP))
    269 		dumpsys();
    270 
    271 	doshutdownhooks();
    272 
    273 	pmf_system_shutdown(boothowto);
    274 
    275 	if ((howto & RB_POWERDOWN) == RB_POWERDOWN) {
    276 	  /* Power off here if we know how...*/
    277 	}
    278 
    279 	if (howto & RB_HALT) {
    280 		printf("The operating system has halted.\n"
    281 		    "Press any key to reboot.\n\n");
    282 		cnpollc(1);	/* For proper keyboard command handling */
    283 		cngetc();
    284 		cnpollc(0);
    285 
    286 		printf("rebooting...\n\n");
    287 		goto reboot;	/* XXX for now... */
    288 
    289 #ifdef DDB
    290 		printf("dropping to debugger\n");
    291 		while(1)
    292 			Debugger();
    293 #endif
    294 	}
    295 
    296 	printf("rebooting\n\n");
    297 	if (what && *what) {
    298 		if (strlen(what) > sizeof str - 5)
    299 			printf("boot string too large, ignored\n");
    300 		else {
    301 			strcpy(str, what);
    302 			ap1 = ap = str + strlen(str);
    303 			*ap++ = ' ';
    304 		}
    305 	}
    306 	*ap++ = '-';
    307 	if (howto & RB_SINGLE)
    308 		*ap++ = 's';
    309 	if (howto & RB_KDB)
    310 		*ap++ = 'd';
    311 	*ap++ = 0;
    312 	if (ap[-2] == '-')
    313 		*ap1 = 0;
    314 
    315 	/* flush cache for msgbuf */
    316 	dcache_wb(msgbuf_paddr, round_page(MSGBUFSIZE));
    317 
    318  reboot:
    319 	__asm volatile("msync; isync");
    320 	(*cpu_md_ops.md_cpu_reset)();
    321 
    322 	printf("%s: md_cpu_reset() failed!\n", __func__);
    323 #ifdef DDB
    324 	for (;;)
    325 		Debugger();
    326 #else
    327 	for (;;)
    328 		/* nothing */;
    329 #endif
    330 }
    331 
    332 /*
    333  * mapiodev:
    334  *
    335  * 	Allocate vm space and mapin the I/O address. Use reserved TLB
    336  * 	mapping if one is found.
    337  */
    338 void *
    339 mapiodev(paddr_t pa, psize_t len, bool prefetchable)
    340 {
    341 	const vsize_t off = pa & PAGE_MASK;
    342 
    343 	/*
    344 	 * See if we have reserved TLB entry for the pa. This needs to be
    345 	 * true for console as we can't use uvm during early bootstrap.
    346 	 */
    347 	void * const p = tlb_mapiodev(pa, len, prefetchable);
    348 	if (p != NULL)
    349 		return p;
    350 
    351 	if (fake_mapiodev)
    352 		panic("mapiodev: no TLB entry reserved for %llx+%llx",
    353 		    (long long)pa, (long long)len);
    354 
    355 	const paddr_t orig_pa = pa;
    356 	const psize_t orig_len = len;
    357 	vsize_t align = 0;
    358 	pa = trunc_page(pa);
    359 	len = round_page(off + len);
    360 	/*
    361 	 * If we are allocating a large amount (>= 1MB) try to get an
    362 	 * aligned VA region for it so try to do a large mapping for it.
    363 	 */
    364 	if ((len & (len - 1)) == 0 && len >= 0x100000)
    365 		align = len;
    366 
    367 	vaddr_t va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
    368 
    369 	if (va == 0 && align > 0) {
    370 		/*
    371 		 * Large aligned request failed.  Let's just get anything.
    372 		 */
    373 		align = 0;
    374 		va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
    375 	}
    376 	if (va == 0)
    377 		return NULL;
    378 
    379 	if (align) {
    380 		/*
    381 		 * Now try to map that via one big TLB entry.
    382 		 */
    383 		pt_entry_t pte = pte_make_kenter_pa(pa, NULL,
    384 		    VM_PROT_READ|VM_PROT_WRITE,
    385 		    prefetchable ? 0 : PMAP_NOCACHE);
    386 		if (!tlb_ioreserve(va, len, pte)) {
    387 			void * const p0 = tlb_mapiodev(orig_pa, orig_len,
    388 			    prefetchable);
    389 			KASSERT(p0 != NULL);
    390 			return p0;
    391 		}
    392 	}
    393 
    394 	for (va += len, pa += len; len > 0; len -= PAGE_SIZE) {
    395 		va -= PAGE_SIZE;
    396 		pa -= PAGE_SIZE;
    397 		pmap_kenter_pa(va, pa, VM_PROT_READ|VM_PROT_WRITE,
    398 		    prefetchable ? 0 : PMAP_NOCACHE);
    399 	}
    400 	pmap_update(pmap_kernel());
    401 	return (void *)(va + off);
    402 }
    403 
    404 void
    405 unmapiodev(vaddr_t va, vsize_t len)
    406 {
    407 	/* Nothing to do for reserved (ie. not uvm_km_alloc'd) mappings. */
    408 	if (va < VM_MIN_KERNEL_ADDRESS || va > VM_MAX_KERNEL_ADDRESS) {
    409 		tlb_unmapiodev(va, len);
    410 		return;
    411 	}
    412 
    413 	len = round_page((va & PAGE_MASK) + len);
    414 	va = trunc_page(va);
    415 
    416 	pmap_kremove(va, len);
    417 	uvm_km_free(kernel_map, va, len, UVM_KMF_VAONLY);
    418 }
    419 
    420 void
    421 cpu_evcnt_attach(struct cpu_info *ci)
    422 {
    423 	struct cpu_softc * const cpu = ci->ci_softc;
    424 	const char * const xname = ci->ci_data.cpu_name;
    425 
    426 	evcnt_attach_dynamic_nozero(&ci->ci_ev_clock, EVCNT_TYPE_INTR,
    427 		NULL, xname, "clock");
    428 	evcnt_attach_dynamic_nozero(&cpu->cpu_ev_late_clock, EVCNT_TYPE_INTR,
    429 		NULL, xname, "late clock");
    430 	evcnt_attach_dynamic_nozero(&cpu->cpu_ev_exec_trap_sync, EVCNT_TYPE_TRAP,
    431 		NULL, xname, "exec pages synced (trap)");
    432 	evcnt_attach_dynamic_nozero(&ci->ci_ev_traps, EVCNT_TYPE_TRAP,
    433 		NULL, xname, "traps");
    434 	evcnt_attach_dynamic_nozero(&ci->ci_ev_kdsi, EVCNT_TYPE_TRAP,
    435 		&ci->ci_ev_traps, xname, "kernel DSI traps");
    436 	evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi, EVCNT_TYPE_TRAP,
    437 		&ci->ci_ev_traps, xname, "user DSI traps");
    438 	evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi_fatal, EVCNT_TYPE_TRAP,
    439 		&ci->ci_ev_udsi, xname, "user DSI failures");
    440 	evcnt_attach_dynamic_nozero(&ci->ci_ev_kisi, EVCNT_TYPE_TRAP,
    441 		&ci->ci_ev_traps, xname, "kernel ISI traps");
    442 	evcnt_attach_dynamic_nozero(&ci->ci_ev_isi, EVCNT_TYPE_TRAP,
    443 		&ci->ci_ev_traps, xname, "user ISI traps");
    444 	evcnt_attach_dynamic_nozero(&ci->ci_ev_isi_fatal, EVCNT_TYPE_TRAP,
    445 		&ci->ci_ev_isi, xname, "user ISI failures");
    446 	evcnt_attach_dynamic_nozero(&ci->ci_ev_scalls, EVCNT_TYPE_TRAP,
    447 		&ci->ci_ev_traps, xname, "system call traps");
    448 	evcnt_attach_dynamic_nozero(&ci->ci_ev_pgm, EVCNT_TYPE_TRAP,
    449 		&ci->ci_ev_traps, xname, "PGM traps");
    450 	evcnt_attach_dynamic_nozero(&ci->ci_ev_debug, EVCNT_TYPE_TRAP,
    451 		&ci->ci_ev_traps, xname, "debug traps");
    452 	evcnt_attach_dynamic_nozero(&ci->ci_ev_fpu, EVCNT_TYPE_TRAP,
    453 		&ci->ci_ev_traps, xname, "FPU unavailable traps");
    454 	evcnt_attach_dynamic_nozero(&ci->ci_ev_fpusw, EVCNT_TYPE_MISC,
    455 		&ci->ci_ev_fpu, xname, "FPU context switches");
    456 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ali, EVCNT_TYPE_TRAP,
    457 		&ci->ci_ev_traps, xname, "user alignment traps");
    458 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ali_fatal, EVCNT_TYPE_TRAP,
    459 		&ci->ci_ev_ali, xname, "user alignment traps");
    460 	evcnt_attach_dynamic_nozero(&ci->ci_ev_umchk, EVCNT_TYPE_TRAP,
    461 		&ci->ci_ev_umchk, xname, "user MCHK failures");
    462 	evcnt_attach_dynamic_nozero(&ci->ci_ev_vec, EVCNT_TYPE_TRAP,
    463 		&ci->ci_ev_traps, xname, "SPE unavailable");
    464 	evcnt_attach_dynamic_nozero(&ci->ci_ev_vecsw, EVCNT_TYPE_MISC,
    465 	    &ci->ci_ev_vec, xname, "SPE context switches");
    466 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ipi, EVCNT_TYPE_INTR,
    467 		NULL, xname, "IPIs");
    468 	evcnt_attach_dynamic_nozero(&ci->ci_ev_tlbmiss_soft, EVCNT_TYPE_TRAP,
    469 		&ci->ci_ev_traps, xname, "soft tlb misses");
    470 	evcnt_attach_dynamic_nozero(&ci->ci_ev_dtlbmiss_hard, EVCNT_TYPE_TRAP,
    471 		&ci->ci_ev_traps, xname, "data tlb misses");
    472 	evcnt_attach_dynamic_nozero(&ci->ci_ev_itlbmiss_hard, EVCNT_TYPE_TRAP,
    473 		&ci->ci_ev_traps, xname, "inst tlb misses");
    474 }
    475 
    476 #ifdef MULTIPROCESSOR
    477 register_t
    478 cpu_hatch(void)
    479 {
    480 	struct cpuset_info * const csi = &cpuset_info;
    481 	const size_t id = cpu_number();
    482 
    483 	/*
    484 	 * We've hatched so tell the spinup code.
    485 	 */
    486 	kcpuset_set(csi->cpus_hatched, id);
    487 
    488 	/*
    489 	 * Loop until running bit for this cpu is set.
    490 	 */
    491 	while (!kcpuset_isset(csi->cpus_running, id)) {
    492 		continue;
    493 	}
    494 
    495 	/*
    496 	 * Now that we are active, start the clocks.
    497 	 */
    498 	cpu_initclocks();
    499 
    500 	/*
    501 	 * Return sp of the idlelwp.  Which we should be already using but ...
    502 	 */
    503 	return curcpu()->ci_curpcb->pcb_sp;
    504 }
    505 
    506 void
    507 cpu_boot_secondary_processors(void)
    508 {
    509 	volatile struct cpuset_info * const csi = &cpuset_info;
    510 	CPU_INFO_ITERATOR cii;
    511 	struct cpu_info *ci;
    512 	kcpuset_t *running;
    513 
    514 	kcpuset_create(&running, true);
    515 
    516 	for (CPU_INFO_FOREACH(cii, ci)) {
    517 		/*
    518 		 * Skip this CPU if it didn't sucessfully hatch.
    519 		 */
    520 		if (!kcpuset_isset(csi->cpus_hatched, cpu_index(ci)))
    521 			continue;
    522 
    523 		KASSERT(!CPU_IS_PRIMARY(ci));
    524 		KASSERT(ci->ci_data.cpu_idlelwp);
    525 
    526 		kcpuset_set(running, cpu_index(ci));
    527 	}
    528 	KASSERT(kcpuset_match(csi->cpus_hatched, running));
    529 	if (!kcpuset_iszero(running)) {
    530 		kcpuset_merge(csi->cpus_running, running);
    531 	}
    532 	kcpuset_destroy(running);
    533 }
    534 #endif
    535 
    536 uint32_t
    537 cpu_read_4(bus_addr_t a)
    538 {
    539 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    540 //	printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
    541 	return bus_space_read_4(cpu->cpu_bst, cpu->cpu_bsh, a);
    542 }
    543 
    544 uint8_t
    545 cpu_read_1(bus_addr_t a)
    546 {
    547 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    548 //	printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
    549 	return bus_space_read_1(cpu->cpu_bst, cpu->cpu_bsh, a);
    550 }
    551 
    552 void
    553 cpu_write_4(bus_addr_t a, uint32_t v)
    554 {
    555 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    556 	bus_space_write_4(cpu->cpu_bst, cpu->cpu_bsh, a, v);
    557 }
    558 
    559 void
    560 cpu_write_1(bus_addr_t a, uint8_t v)
    561 {
    562 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    563 	bus_space_write_1(cpu->cpu_bst, cpu->cpu_bsh, a, v);
    564 }
    565 
    566 void
    567 booke_sstep(struct trapframe *tf)
    568 {
    569 	uint32_t insn;
    570 
    571 	KASSERT(tf->tf_srr1 & PSL_DE);
    572 	if (ufetch_32((const void *)tf->tf_srr0, &insn) != 0)
    573 		return;
    574 
    575 	register_t dbcr0 = DBCR0_IAC1 | DBCR0_IDM;
    576 	register_t dbcr1 = DBCR1_IAC1US_USER | DBCR1_IAC1ER_DS1;
    577 	if ((insn >> 28) == 4) {
    578 		uint32_t iac2 = 0;
    579 		if ((insn >> 26) == 0x12) {
    580 			const int32_t off = (((int32_t)insn << 6) >> 6) & ~3;
    581 			iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
    582 			dbcr0 |= DBCR0_IAC2;
    583 		} else if ((insn >> 26) == 0x10) {
    584 			const int16_t off = insn & ~3;
    585 			iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
    586 			dbcr0 |= DBCR0_IAC2;
    587 		} else if ((insn & 0xfc00fffe) == 0x4c000420) {
    588 			iac2 = tf->tf_ctr;
    589 			dbcr0 |= DBCR0_IAC2;
    590 		} else if ((insn & 0xfc00fffe) == 0x4c000020) {
    591 			iac2 = tf->tf_lr;
    592 			dbcr0 |= DBCR0_IAC2;
    593 		}
    594 		if (dbcr0 & DBCR0_IAC2) {
    595 			dbcr1 |= DBCR1_IAC2US_USER | DBCR1_IAC2ER_DS1;
    596 			mtspr(SPR_IAC2, iac2);
    597 		}
    598 	}
    599 	mtspr(SPR_IAC1, tf->tf_srr0 + 4);
    600 	mtspr(SPR_DBCR1, dbcr1);
    601 	mtspr(SPR_DBCR0, dbcr0);
    602 }
    603 
    604 #ifdef DIAGNOSTIC
    605 static inline void
    606 swap_data(uint64_t *data, size_t a, size_t b)
    607 {
    608 	uint64_t swap = data[a];
    609 	data[a] = data[b];
    610 	data[b] = swap;
    611 }
    612 
    613 static void
    614 sort_data(uint64_t *data, size_t count)
    615 {
    616 #if 0
    617 	/*
    618 	 * Mostly classic bubble sort
    619 	 */
    620 	do {
    621 		size_t new_count = 0;
    622 		for (size_t i = 1; i < count; i++) {
    623 			if (tbs[i - 1] > tbs[i]) {
    624 				swap_tbs(tbs, i - 1, i);
    625 				new_count = i;
    626 			}
    627 		}
    628 		count = new_count;
    629 	} while (count > 0);
    630 #else
    631 	/*
    632 	 * Comb sort
    633 	 */
    634 	size_t gap = count;
    635 	bool swapped = false;
    636 	while (gap > 1 || swapped) {
    637 		if (gap > 1) {
    638 			/*
    639 			 * phi = (1 + sqrt(5)) / 2 [golden ratio]
    640 			 * N = 1 / (1 - e^-phi)) = 1.247330950103979
    641 			 *
    642 			 * We want to but can't use floating point to calculate
    643 			 *	gap = (size_t)((double)gap / N)
    644 			 *
    645 			 * So we will use the multicative inverse of N
    646 			 * (module 65536) to achieve the division.
    647 			 *
    648 			 * iN = 2^16 / 1.24733... = 52540
    649 			 * x / N == (x * iN) / 65536
    650 			 */
    651 			gap = (gap * 52540) / 65536;
    652 		}
    653 
    654 		swapped = false;
    655 
    656 		for (size_t i = 0; gap + i < count; i++) {
    657 			if (data[i] > data[i + gap]) {
    658 				swap_data(data, i, i + gap);
    659 				swapped = true;
    660 			}
    661 		}
    662 	}
    663 #endif
    664 }
    665 #endif
    666 
    667 void
    668 dump_splhist(struct cpu_info *ci, void (*pr)(const char *, ...))
    669 {
    670 #ifdef DIAGNOSTIC
    671 	struct cpu_softc * const cpu = ci->ci_softc;
    672 	uint64_t tbs[NIPL*NIPL];
    673 	size_t ntbs = 0;
    674 	for (size_t to = 0; to < NIPL; to++) {
    675 		for (size_t from = 0; from < NIPL; from++) {
    676 			uint64_t tb = cpu->cpu_spl_tb[to][from];
    677 			if (tb == 0)
    678 				continue;
    679 			tbs[ntbs++] = (tb << 8) | (to << 4) | from;
    680 		}
    681 	}
    682 	sort_data(tbs, ntbs);
    683 
    684 	if (pr == NULL)
    685 		pr = printf;
    686 	uint64_t last_tb = 0;
    687 	for (size_t i = 0; i < ntbs; i++) {
    688 		uint64_t tb = tbs[i];
    689 		size_t from = tb & 15;
    690 		size_t to = (tb >> 4) & 15;
    691 		tb >>= 8;
    692 		(*pr)("%s(%zu) from %zu at %"PRId64"",
    693 		     from < to ? "splraise" : "splx",
    694 		     to, from, tb);
    695 		if (last_tb && from != IPL_NONE)
    696 			(*pr)(" (+%"PRId64")", tb - last_tb);
    697 		(*pr)("\n");
    698 		last_tb = tb;
    699 	}
    700 #endif
    701 }
    702