Home | History | Annotate | Line # | Download | only in booke
booke_machdep.c revision 1.24
      1 /*	$NetBSD: booke_machdep.c,v 1.24 2016/07/11 16:06:52 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 1.24 2016/07/11 16:06:52 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 		.ci_pmap_kern_segtab = &pmap_kern_segtab,
    134 	},
    135 #ifdef MULTIPROCESSOR
    136 	[CPU_MAXNUM-1] = {
    137 		.ci_curlwp = NULL,
    138 		.ci_tlb_info = &pmap_tlb0_info,
    139 		.ci_softc = &cpu_softc[CPU_MAXNUM-1],
    140 		.ci_cpl = IPL_HIGH,
    141 		.ci_idepth = -1,
    142 		.ci_pmap_kern_segtab = &pmap_kern_segtab,
    143 	},
    144 #endif
    145 };
    146 __CTASSERT(__arraycount(cpu_info) == __arraycount(cpu_softc));
    147 
    148 /*
    149  * This should probably be in autoconf!				XXX
    150  */
    151 char machine[] = MACHINE;		/* from <machine/param.h> */
    152 char machine_arch[] = MACHINE_ARCH;	/* from <machine/param.h> */
    153 
    154 char bootpath[256];
    155 
    156 #if NKSYMS || defined(DDB) || defined(MODULAR)
    157 void *startsym, *endsym;
    158 #endif
    159 
    160 #if defined(MULTIPROCESSOR)
    161 volatile struct cpu_hatch_data cpu_hatch_data __cacheline_aligned;
    162 #endif
    163 
    164 int fake_mapiodev = 1;
    165 
    166 void
    167 booke_cpu_startup(const char *model)
    168 {
    169 	vaddr_t 	minaddr, maxaddr;
    170 	char 		pbuf[9];
    171 
    172 	cpu_setmodel("%s", model);
    173 
    174 	printf("%s%s", copyright, version);
    175 
    176 	format_bytes(pbuf, sizeof(pbuf), ctob((uint64_t)physmem));
    177 	printf("total memory = %s\n", pbuf);
    178 
    179 	minaddr = 0;
    180 	/*
    181 	 * Allocate a submap for physio
    182 	 */
    183 	phys_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
    184 				 VM_PHYS_SIZE, 0, false, NULL);
    185 
    186 	/*
    187 	 * No need to allocate an mbuf cluster submap.  Mbuf clusters
    188 	 * are allocated via the pool allocator, and we use direct-mapped
    189 	 * pool pages.
    190 	 */
    191 
    192 	format_bytes(pbuf, sizeof(pbuf), ptoa(uvmexp.free));
    193 	printf("avail memory = %s\n", pbuf);
    194 
    195 	/*
    196 	 * Register the tlb's evcnts
    197 	 */
    198 	pmap_tlb_info_evcnt_attach(curcpu()->ci_tlb_info);
    199 
    200 	/*
    201 	 * Set up the board properties database.
    202 	 */
    203 	board_info_init();
    204 
    205 	/*
    206 	 * Now that we have VM, malloc()s are OK in bus_space.
    207 	 */
    208 	bus_space_mallocok();
    209 	fake_mapiodev = 0;
    210 
    211 #ifdef MULTIPROCESSOR
    212 	pmap_kernel()->pm_active = kcpuset_running;
    213 	pmap_kernel()->pm_onproc = kcpuset_running;
    214 
    215 	for (size_t i = 1; i < __arraycount(cpu_info); i++) {
    216 		struct cpu_info * const ci = &cpu_info[i];
    217 		struct cpu_softc * const cpu = &cpu_softc[i];
    218 		cpu->cpu_ci = ci;
    219 		cpu->cpu_bst = cpu_softc[0].cpu_bst;
    220 		cpu->cpu_le_bst = cpu_softc[0].cpu_le_bst;
    221 		cpu->cpu_bsh = cpu_softc[0].cpu_bsh;
    222 		cpu->cpu_highmem = cpu_softc[0].cpu_highmem;
    223 		ci->ci_softc = cpu;
    224 		ci->ci_tlb_info = &pmap_tlb0_info;
    225 		ci->ci_cpl = IPL_HIGH;
    226 		ci->ci_idepth = -1;
    227 		ci->ci_pmap_kern_segtab = curcpu()->ci_pmap_kern_segtab;
    228 	}
    229 
    230 	kcpuset_create(&cpuset_info.cpus_running, true);
    231 	kcpuset_create(&cpuset_info.cpus_hatched, true);
    232 	kcpuset_create(&cpuset_info.cpus_paused, true);
    233 	kcpuset_create(&cpuset_info.cpus_resumed, true);
    234 	kcpuset_create(&cpuset_info.cpus_halted, true);
    235 
    236 	kcpuset_set(cpuset_info.cpus_running, cpu_number());
    237 #endif /* MULTIPROCESSOR */
    238 }
    239 
    240 static void
    241 dumpsys(void)
    242 {
    243 
    244 	printf("dumpsys: TBD\n");
    245 }
    246 
    247 /*
    248  * Halt or reboot the machine after syncing/dumping according to howto.
    249  */
    250 void
    251 cpu_reboot(int howto, char *what)
    252 {
    253 	static int syncing;
    254 	static char str[256];
    255 	char *ap = str, *ap1 = ap;
    256 
    257 	boothowto = howto;
    258 	if (!cold && !(howto & RB_NOSYNC) && !syncing) {
    259 		syncing = 1;
    260 		vfs_shutdown();		/* sync */
    261 		resettodr();		/* set wall clock */
    262 	}
    263 
    264 	splhigh();
    265 
    266 	if (!cold && (howto & RB_DUMP))
    267 		dumpsys();
    268 
    269 	doshutdownhooks();
    270 
    271 	pmf_system_shutdown(boothowto);
    272 
    273 	if ((howto & RB_POWERDOWN) == RB_POWERDOWN) {
    274 	  /* Power off here if we know how...*/
    275 	}
    276 
    277 	if (howto & RB_HALT) {
    278 		printf("halted\n\n");
    279 
    280 		goto reboot;	/* XXX for now... */
    281 
    282 #ifdef DDB
    283 		printf("dropping to debugger\n");
    284 		while(1)
    285 			Debugger();
    286 #endif
    287 	}
    288 
    289 	printf("rebooting\n\n");
    290 	if (what && *what) {
    291 		if (strlen(what) > sizeof str - 5)
    292 			printf("boot string too large, ignored\n");
    293 		else {
    294 			strcpy(str, what);
    295 			ap1 = ap = str + strlen(str);
    296 			*ap++ = ' ';
    297 		}
    298 	}
    299 	*ap++ = '-';
    300 	if (howto & RB_SINGLE)
    301 		*ap++ = 's';
    302 	if (howto & RB_KDB)
    303 		*ap++ = 'd';
    304 	*ap++ = 0;
    305 	if (ap[-2] == '-')
    306 		*ap1 = 0;
    307 
    308 	/* flush cache for msgbuf */
    309 	dcache_wb(msgbuf_paddr, round_page(MSGBUFSIZE));
    310 
    311  reboot:
    312 	__asm volatile("msync; isync");
    313 	(*cpu_md_ops.md_cpu_reset)();
    314 
    315 	printf("%s: md_cpu_reset() failed!\n", __func__);
    316 #ifdef DDB
    317 	for (;;)
    318 		Debugger();
    319 #else
    320 	for (;;)
    321 		/* nothing */;
    322 #endif
    323 }
    324 
    325 /*
    326  * mapiodev:
    327  *
    328  * 	Allocate vm space and mapin the I/O address. Use reserved TLB
    329  * 	mapping if one is found.
    330  */
    331 void *
    332 mapiodev(paddr_t pa, psize_t len, bool prefetchable)
    333 {
    334 	const vsize_t off = pa & PAGE_MASK;
    335 
    336 	/*
    337 	 * See if we have reserved TLB entry for the pa. This needs to be
    338 	 * true for console as we can't use uvm during early bootstrap.
    339 	 */
    340 	void * const p = tlb_mapiodev(pa, len, prefetchable);
    341 	if (p != NULL)
    342 		return p;
    343 
    344 	if (fake_mapiodev)
    345 		panic("mapiodev: no TLB entry reserved for %llx+%llx",
    346 		    (long long)pa, (long long)len);
    347 
    348 	const paddr_t orig_pa = pa;
    349 	const psize_t orig_len = len;
    350 	vsize_t align = 0;
    351 	pa = trunc_page(pa);
    352 	len = round_page(off + len);
    353 	/*
    354 	 * If we are allocating a large amount (>= 1MB) try to get an
    355 	 * aligned VA region for it so try to do a large mapping for it.
    356 	 */
    357 	if ((len & (len - 1)) == 0 && len >= 0x100000)
    358 		align = len;
    359 
    360 	vaddr_t va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
    361 
    362 	if (va == 0 && align > 0) {
    363 		/*
    364 		 * Large aligned request failed.  Let's just get anything.
    365 		 */
    366 		align = 0;
    367 		va = uvm_km_alloc(kernel_map, len, align, UVM_KMF_VAONLY);
    368 	}
    369 	if (va == 0)
    370 		return NULL;
    371 
    372 	if (align) {
    373 		/*
    374 		 * Now try to map that via one big TLB entry.
    375 		 */
    376 		pt_entry_t pte = pte_make_kenter_pa(pa, NULL,
    377 		    VM_PROT_READ|VM_PROT_WRITE,
    378 		    prefetchable ? 0 : PMAP_NOCACHE);
    379 		if (!tlb_ioreserve(va, len, pte)) {
    380 			void * const p0 = tlb_mapiodev(orig_pa, orig_len,
    381 			    prefetchable);
    382 			KASSERT(p0 != NULL);
    383 			return p0;
    384 		}
    385 	}
    386 
    387 	for (va += len, pa += len; len > 0; len -= PAGE_SIZE) {
    388 		va -= PAGE_SIZE;
    389 		pa -= PAGE_SIZE;
    390 		pmap_kenter_pa(va, pa, VM_PROT_READ|VM_PROT_WRITE,
    391 		    prefetchable ? 0 : PMAP_NOCACHE);
    392 	}
    393 	pmap_update(pmap_kernel());
    394 	return (void *)(va + off);
    395 }
    396 
    397 void
    398 unmapiodev(vaddr_t va, vsize_t len)
    399 {
    400 	/* Nothing to do for reserved (ie. not uvm_km_alloc'd) mappings. */
    401 	if (va < VM_MIN_KERNEL_ADDRESS || va > VM_MAX_KERNEL_ADDRESS) {
    402 		tlb_unmapiodev(va, len);
    403 		return;
    404 	}
    405 
    406 	len = round_page((va & PAGE_MASK) + len);
    407 	va = trunc_page(va);
    408 
    409 	pmap_kremove(va, len);
    410 	uvm_km_free(kernel_map, va, len, UVM_KMF_VAONLY);
    411 }
    412 
    413 void
    414 cpu_evcnt_attach(struct cpu_info *ci)
    415 {
    416 	struct cpu_softc * const cpu = ci->ci_softc;
    417 	const char * const xname = ci->ci_data.cpu_name;
    418 
    419 	evcnt_attach_dynamic_nozero(&ci->ci_ev_clock, EVCNT_TYPE_INTR,
    420 		NULL, xname, "clock");
    421 	evcnt_attach_dynamic_nozero(&cpu->cpu_ev_late_clock, EVCNT_TYPE_INTR,
    422 		NULL, xname, "late clock");
    423 	evcnt_attach_dynamic_nozero(&cpu->cpu_ev_exec_trap_sync, EVCNT_TYPE_TRAP,
    424 		NULL, xname, "exec pages synced (trap)");
    425 	evcnt_attach_dynamic_nozero(&ci->ci_ev_traps, EVCNT_TYPE_TRAP,
    426 		NULL, xname, "traps");
    427 	evcnt_attach_dynamic_nozero(&ci->ci_ev_kdsi, EVCNT_TYPE_TRAP,
    428 		&ci->ci_ev_traps, xname, "kernel DSI traps");
    429 	evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi, EVCNT_TYPE_TRAP,
    430 		&ci->ci_ev_traps, xname, "user DSI traps");
    431 	evcnt_attach_dynamic_nozero(&ci->ci_ev_udsi_fatal, EVCNT_TYPE_TRAP,
    432 		&ci->ci_ev_udsi, xname, "user DSI failures");
    433 	evcnt_attach_dynamic_nozero(&ci->ci_ev_kisi, EVCNT_TYPE_TRAP,
    434 		&ci->ci_ev_traps, xname, "kernel ISI traps");
    435 	evcnt_attach_dynamic_nozero(&ci->ci_ev_isi, EVCNT_TYPE_TRAP,
    436 		&ci->ci_ev_traps, xname, "user ISI traps");
    437 	evcnt_attach_dynamic_nozero(&ci->ci_ev_isi_fatal, EVCNT_TYPE_TRAP,
    438 		&ci->ci_ev_isi, xname, "user ISI failures");
    439 	evcnt_attach_dynamic_nozero(&ci->ci_ev_scalls, EVCNT_TYPE_TRAP,
    440 		&ci->ci_ev_traps, xname, "system call traps");
    441 	evcnt_attach_dynamic_nozero(&ci->ci_ev_pgm, EVCNT_TYPE_TRAP,
    442 		&ci->ci_ev_traps, xname, "PGM traps");
    443 	evcnt_attach_dynamic_nozero(&ci->ci_ev_debug, EVCNT_TYPE_TRAP,
    444 		&ci->ci_ev_traps, xname, "debug traps");
    445 	evcnt_attach_dynamic_nozero(&ci->ci_ev_fpu, EVCNT_TYPE_TRAP,
    446 		&ci->ci_ev_traps, xname, "FPU unavailable traps");
    447 	evcnt_attach_dynamic_nozero(&ci->ci_ev_fpusw, EVCNT_TYPE_MISC,
    448 		&ci->ci_ev_fpu, xname, "FPU context switches");
    449 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ali, EVCNT_TYPE_TRAP,
    450 		&ci->ci_ev_traps, xname, "user alignment traps");
    451 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ali_fatal, EVCNT_TYPE_TRAP,
    452 		&ci->ci_ev_ali, xname, "user alignment traps");
    453 	evcnt_attach_dynamic_nozero(&ci->ci_ev_umchk, EVCNT_TYPE_TRAP,
    454 		&ci->ci_ev_umchk, xname, "user MCHK failures");
    455 	evcnt_attach_dynamic_nozero(&ci->ci_ev_vec, EVCNT_TYPE_TRAP,
    456 		&ci->ci_ev_traps, xname, "SPE unavailable");
    457 	evcnt_attach_dynamic_nozero(&ci->ci_ev_vecsw, EVCNT_TYPE_MISC,
    458 	    &ci->ci_ev_vec, xname, "SPE context switches");
    459 	evcnt_attach_dynamic_nozero(&ci->ci_ev_ipi, EVCNT_TYPE_INTR,
    460 		NULL, xname, "IPIs");
    461 	evcnt_attach_dynamic_nozero(&ci->ci_ev_tlbmiss_soft, EVCNT_TYPE_TRAP,
    462 		&ci->ci_ev_traps, xname, "soft tlb misses");
    463 	evcnt_attach_dynamic_nozero(&ci->ci_ev_dtlbmiss_hard, EVCNT_TYPE_TRAP,
    464 		&ci->ci_ev_traps, xname, "data tlb misses");
    465 	evcnt_attach_dynamic_nozero(&ci->ci_ev_itlbmiss_hard, EVCNT_TYPE_TRAP,
    466 		&ci->ci_ev_traps, xname, "inst tlb misses");
    467 }
    468 
    469 #ifdef MULTIPROCESSOR
    470 register_t
    471 cpu_hatch(void)
    472 {
    473 	struct cpuset_info * const csi = &cpuset_info;
    474 	const size_t id = cpu_number();
    475 
    476 	/*
    477 	 * We've hatched so tell the spinup code.
    478 	 */
    479 	kcpuset_set(csi->cpus_hatched, id);
    480 
    481 	/*
    482 	 * Loop until running bit for this cpu is set.
    483 	 */
    484 	while (!kcpuset_isset(csi->cpus_running, id)) {
    485 		continue;
    486 	}
    487 
    488 	/*
    489 	 * Now that we are active, start the clocks.
    490 	 */
    491 	cpu_initclocks();
    492 
    493 	/*
    494 	 * Return sp of the idlelwp.  Which we should be already using but ...
    495 	 */
    496 	return curcpu()->ci_curpcb->pcb_sp;
    497 }
    498 
    499 void
    500 cpu_boot_secondary_processors(void)
    501 {
    502 	volatile struct cpuset_info * const csi = &cpuset_info;
    503 	CPU_INFO_ITERATOR cii;
    504 	struct cpu_info *ci;
    505 	kcpuset_t *running;
    506 
    507 	kcpuset_create(&running, true);
    508 
    509 	for (CPU_INFO_FOREACH(cii, ci)) {
    510 		/*
    511 		 * Skip this CPU if it didn't sucessfully hatch.
    512 		 */
    513 		if (!kcpuset_isset(csi->cpus_hatched, cpu_index(ci)))
    514 			continue;
    515 
    516 		KASSERT(!CPU_IS_PRIMARY(ci));
    517 		KASSERT(ci->ci_data.cpu_idlelwp);
    518 
    519 		kcpuset_set(running, cpu_index(ci));
    520 	}
    521 	KASSERT(kcpuset_match(csi->cpus_hatched, running));
    522 	if (!kcpuset_iszero(running)) {
    523 		kcpuset_merge(csi->cpus_running, running);
    524 	}
    525 	kcpuset_destroy(running);
    526 }
    527 #endif
    528 
    529 uint32_t
    530 cpu_read_4(bus_addr_t a)
    531 {
    532 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    533 //	printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
    534 	return bus_space_read_4(cpu->cpu_bst, cpu->cpu_bsh, a);
    535 }
    536 
    537 uint8_t
    538 cpu_read_1(bus_addr_t a)
    539 {
    540 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    541 //	printf(" %s(%p, %x, %x)", __func__, cpu->cpu_bst, cpu->cpu_bsh, a);
    542 	return bus_space_read_1(cpu->cpu_bst, cpu->cpu_bsh, a);
    543 }
    544 
    545 void
    546 cpu_write_4(bus_addr_t a, uint32_t v)
    547 {
    548 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    549 	bus_space_write_4(cpu->cpu_bst, cpu->cpu_bsh, a, v);
    550 }
    551 
    552 void
    553 cpu_write_1(bus_addr_t a, uint8_t v)
    554 {
    555 	struct cpu_softc * const cpu = curcpu()->ci_softc;
    556 	bus_space_write_1(cpu->cpu_bst, cpu->cpu_bsh, a, v);
    557 }
    558 
    559 void
    560 booke_sstep(struct trapframe *tf)
    561 {
    562 	KASSERT(tf->tf_srr1 & PSL_DE);
    563 	const uint32_t insn = ufetch_32((const void *)tf->tf_srr0);
    564 	register_t dbcr0 = DBCR0_IAC1 | DBCR0_IDM;
    565 	register_t dbcr1 = DBCR1_IAC1US_USER | DBCR1_IAC1ER_DS1;
    566 	if ((insn >> 28) == 4) {
    567 		uint32_t iac2 = 0;
    568 		if ((insn >> 26) == 0x12) {
    569 			const int32_t off = (((int32_t)insn << 6) >> 6) & ~3;
    570 			iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
    571 			dbcr0 |= DBCR0_IAC2;
    572 		} else if ((insn >> 26) == 0x10) {
    573 			const int16_t off = insn & ~3;
    574 			iac2 = ((insn & 2) ? 0 : tf->tf_srr0) + off;
    575 			dbcr0 |= DBCR0_IAC2;
    576 		} else if ((insn & 0xfc00fffe) == 0x4c000420) {
    577 			iac2 = tf->tf_ctr;
    578 			dbcr0 |= DBCR0_IAC2;
    579 		} else if ((insn & 0xfc00fffe) == 0x4c000020) {
    580 			iac2 = tf->tf_lr;
    581 			dbcr0 |= DBCR0_IAC2;
    582 		}
    583 		if (dbcr0 & DBCR0_IAC2) {
    584 			dbcr1 |= DBCR1_IAC2US_USER | DBCR1_IAC2ER_DS1;
    585 			mtspr(SPR_IAC2, iac2);
    586 		}
    587 	}
    588 	mtspr(SPR_IAC1, tf->tf_srr0 + 4);
    589 	mtspr(SPR_DBCR1, dbcr1);
    590 	mtspr(SPR_DBCR0, dbcr0);
    591 }
    592 
    593 #ifdef DIAGNOSTIC
    594 static inline void
    595 swap_data(uint64_t *data, size_t a, size_t b)
    596 {
    597 	uint64_t swap = data[a];
    598 	data[a] = data[b];
    599 	data[b] = swap;
    600 }
    601 
    602 static void
    603 sort_data(uint64_t *data, size_t count)
    604 {
    605 #if 0
    606 	/*
    607 	 * Mostly classic bubble sort
    608 	 */
    609 	do {
    610 		size_t new_count = 0;
    611 		for (size_t i = 1; i < count; i++) {
    612 			if (tbs[i - 1] > tbs[i]) {
    613 				swap_tbs(tbs, i - 1, i);
    614 				new_count = i;
    615 			}
    616 		}
    617 		count = new_count;
    618 	} while (count > 0);
    619 #else
    620 	/*
    621 	 * Comb sort
    622 	 */
    623 	size_t gap = count;
    624 	bool swapped = false;
    625 	while (gap > 1 || swapped) {
    626 		if (gap > 1) {
    627 			/*
    628 			 * phi = (1 + sqrt(5)) / 2 [golden ratio]
    629 			 * N = 1 / (1 - e^-phi)) = 1.247330950103979
    630 			 *
    631 			 * We want to but can't use floating point to calculate
    632 			 *	gap = (size_t)((double)gap / N)
    633 			 *
    634 			 * So we will use the multicative inverse of N
    635 			 * (module 65536) to achieve the division.
    636 			 *
    637 			 * iN = 2^16 / 1.24733... = 52540
    638 			 * x / N == (x * iN) / 65536
    639 			 */
    640 			gap = (gap * 52540) / 65536;
    641 		}
    642 
    643 		swapped = false;
    644 
    645 		for (size_t i = 0; gap + i < count; i++) {
    646 			if (data[i] > data[i + gap]) {
    647 				swap_data(data, i, i + gap);
    648 				swapped = true;
    649 			}
    650 		}
    651 	}
    652 #endif
    653 }
    654 #endif
    655 
    656 void
    657 dump_splhist(struct cpu_info *ci, void (*pr)(const char *, ...))
    658 {
    659 #ifdef DIAGNOSTIC
    660 	struct cpu_softc * const cpu = ci->ci_softc;
    661 	uint64_t tbs[NIPL*NIPL];
    662 	size_t ntbs = 0;
    663 	for (size_t to = 0; to < NIPL; to++) {
    664 		for (size_t from = 0; from < NIPL; from++) {
    665 			uint64_t tb = cpu->cpu_spl_tb[to][from];
    666 			if (tb == 0)
    667 				continue;
    668 			tbs[ntbs++] = (tb << 8) | (to << 4) | from;
    669 		}
    670 	}
    671 	sort_data(tbs, ntbs);
    672 
    673 	if (pr == NULL)
    674 		pr = printf;
    675 	uint64_t last_tb = 0;
    676 	for (size_t i = 0; i < ntbs; i++) {
    677 		uint64_t tb = tbs[i];
    678 		size_t from = tb & 15;
    679 		size_t to = (tb >> 4) & 15;
    680 		tb >>= 8;
    681 		(*pr)("%s(%zu) from %zu at %"PRId64"",
    682 		     from < to ? "splraise" : "splx",
    683 		     to, from, tb);
    684 		if (last_tb && from != IPL_NONE)
    685 			(*pr)(" (+%"PRId64")", tb - last_tb);
    686 		(*pr)("\n");
    687 		last_tb = tb;
    688 	}
    689 #endif
    690 }
    691