2 * SPDX-License-Identifier: BSD-4-Clause
4 * Copyright (C) 1996 Wolfgang Solfrank.
5 * Copyright (C) 1996 TooLs GmbH.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by TooLs GmbH.
19 * 4. The name of TooLs GmbH may not be used to endorse or promote products
20 * derived from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
27 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
28 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
29 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
30 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
31 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 * $NetBSD: ofw_machdep.c,v 1.5 2000/05/23 13:25:43 tsubai Exp $
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include "opt_platform.h"
40 #include <sys/param.h>
42 #include <sys/systm.h>
45 #include <sys/fcntl.h>
46 #include <sys/malloc.h>
49 #include <sys/endian.h>
51 #include <net/ethernet.h>
53 #include <dev/fdt/fdt_common.h>
54 #include <dev/ofw/openfirm.h>
55 #include <dev/ofw/ofw_pci.h>
56 #include <dev/ofw/ofw_bus.h>
57 #include <dev/ofw/ofw_subr.h>
60 #include <vm/vm_param.h>
61 #include <vm/vm_page.h>
63 #include <machine/bus.h>
64 #include <machine/cpu.h>
65 #include <machine/md_var.h>
66 #include <machine/platform.h>
67 #include <machine/ofw_machdep.h>
68 #include <machine/trap.h>
70 #include <contrib/libfdt/libfdt.h>
73 #include <powerpc/powernv/opal.h>
80 extern register_t ofmsr[5];
81 extern void *openfirmware_entry;
82 char save_trap_init[0x2f00]; /* EXC_LAST */
83 char save_trap_of[0x2f00]; /* EXC_LAST */
86 static int openfirmware(void *args);
89 ofw_save_trap_vec(char *save_trap_vec)
91 if (!ofw_real_mode || !hw_direct_map)
94 bcopy((void *)PHYS_TO_DMAP(EXC_RST), save_trap_vec, EXC_LAST - EXC_RST);
98 ofw_restore_trap_vec(char *restore_trap_vec)
100 if (!ofw_real_mode || !hw_direct_map)
103 bcopy(restore_trap_vec, (void *)PHYS_TO_DMAP(EXC_RST),
105 __syncicache((void *)PHYS_TO_DMAP(EXC_RSVD), EXC_LAST - EXC_RSVD);
109 * Saved SPRG0-3 from OpenFirmware. Will be restored prior to the callback.
111 register_t ofw_sprg0_save;
114 ofw_sprg_prepare(void)
120 * Assume that interrupt are disabled at this point, or
121 * SPRG1-3 could be trashed
124 __asm __volatile("mtsprg1 %0\n\t"
132 __asm __volatile("mfsprg0 %0\n\t"
137 : "=&r"(ofw_sprg0_save)
146 ofw_sprg_restore(void)
152 * Note that SPRG1-3 contents are irrelevant. They are scratch
153 * registers used in the early portion of trap handling when
154 * interrupts are disabled.
156 * PCPU data cannot be used until this routine is called !
158 #ifndef __powerpc64__
159 __asm __volatile("mtsprg0 %0" :: "r"(ofw_sprg0_save));
165 parse_ofw_memory(phandle_t node, const char *prop, struct mem_region *output)
167 cell_t address_cells, size_cells;
168 cell_t OFmem[4 * PHYS_AVAIL_SZ];
175 * Get #address-cells from root node, defaulting to 1 if it cannot
178 phandle = OF_finddevice("/");
179 if (OF_getencprop(phandle, "#address-cells", &address_cells,
180 sizeof(address_cells)) < (ssize_t)sizeof(address_cells))
182 if (OF_getencprop(phandle, "#size-cells", &size_cells,
183 sizeof(size_cells)) < (ssize_t)sizeof(size_cells))
189 if (node == -1 || (sz = OF_getencprop(node, prop,
190 OFmem, sizeof(OFmem))) <= 0)
191 panic("Physical memory map not found");
195 while (i < sz/sizeof(cell_t)) {
196 output[j].mr_start = OFmem[i++];
197 if (address_cells == 2) {
198 output[j].mr_start <<= 32;
199 output[j].mr_start += OFmem[i++];
202 output[j].mr_size = OFmem[i++];
203 if (size_cells == 2) {
204 output[j].mr_size <<= 32;
205 output[j].mr_size += OFmem[i++];
208 if (output[j].mr_start > BUS_SPACE_MAXADDR)
212 * Constrain memory to that which we can access.
213 * 32-bit AIM can only reference 32 bits of address currently,
214 * but Book-E can access 36 bits.
216 if (((uint64_t)output[j].mr_start +
217 (uint64_t)output[j].mr_size - 1) >
219 output[j].mr_size = BUS_SPACE_MAXADDR -
220 output[j].mr_start + 1;
225 sz = j*sizeof(output[0]);
232 excise_reserved_regions(struct mem_region *avail, int asz,
233 struct mem_region *exclude, int esz)
237 for (i = 0; i < asz; i++) {
238 for (j = 0; j < esz; j++) {
240 * Case 1: Exclusion region encloses complete
241 * available entry. Drop it and move on.
243 if (exclude[j].mr_start <= avail[i].mr_start &&
244 exclude[j].mr_start + exclude[j].mr_size >=
245 avail[i].mr_start + avail[i].mr_size) {
246 for (k = i+1; k < asz; k++)
247 avail[k-1] = avail[k];
249 i--; /* Repeat some entries */
254 * Case 2: Exclusion region starts in available entry.
255 * Trim it to where the entry begins and append
256 * a new available entry with the region after
257 * the excluded region, if any.
259 if (exclude[j].mr_start >= avail[i].mr_start &&
260 exclude[j].mr_start < avail[i].mr_start +
262 if (exclude[j].mr_start + exclude[j].mr_size <
263 avail[i].mr_start + avail[i].mr_size) {
264 avail[asz].mr_start =
265 exclude[j].mr_start + exclude[j].mr_size;
266 avail[asz].mr_size = avail[i].mr_start +
272 avail[i].mr_size = exclude[j].mr_start -
277 * Case 3: Exclusion region ends in available entry.
278 * Move start point to where the exclusion zone ends.
279 * The case of a contained exclusion zone has already
280 * been caught in case 2.
282 if (exclude[j].mr_start + exclude[j].mr_size >=
283 avail[i].mr_start && exclude[j].mr_start +
284 exclude[j].mr_size < avail[i].mr_start +
286 avail[i].mr_size += avail[i].mr_start;
288 exclude[j].mr_start + exclude[j].mr_size;
289 avail[i].mr_size -= avail[i].mr_start;
298 excise_initrd_region(struct mem_region *avail, int asz)
303 struct mem_region initrdmap[1];
306 chosen = OF_finddevice("/chosen");
308 size = OF_getencprop(chosen, "linux,initrd-start", cell, sizeof(cell));
314 start = (uint64_t)cell[0] << 32 | cell[1];
316 /* Invalid value length */
317 printf("WARNING: linux,initrd-start must be either 4 or 8 bytes long\n");
321 size = OF_getencprop(chosen, "linux,initrd-end", cell, sizeof(cell));
327 end = (uint64_t)cell[0] << 32 | cell[1];
329 /* Invalid value length */
330 printf("WARNING: linux,initrd-end must be either 4 or 8 bytes long\n");
337 initrdmap[0].mr_start = start;
338 initrdmap[0].mr_size = end - start;
340 asz = excise_reserved_regions(avail, asz, initrdmap, 1);
347 excise_msi_region(struct mem_region *avail, int asz)
350 struct mem_region initrdmap[1];
353 * This range of physical addresses is used to implement optimized
354 * 32 bit MSI interrupts on POWER9. Exclude it to avoid accidentally
355 * using it for DMA, as this will cause an immediate PHB fence.
356 * While we could theoretically turn off this behavior in the ETU,
357 * doing so would break 32-bit MSI, so just reserve the range in
358 * the physical map instead.
359 * See section 4.4.2.8 of the PHB4 specification.
361 start = 0x00000000ffff0000ul;
362 end = 0x00000000fffffffful;
364 initrdmap[0].mr_start = start;
365 initrdmap[0].mr_size = end - start;
367 asz = excise_reserved_regions(avail, asz, initrdmap, 1);
374 excise_fdt_reserved(struct mem_region *avail, int asz)
376 struct mem_region fdtmap[32];
381 chosen = OF_finddevice("/chosen");
382 fdtmapsize = OF_getprop(chosen, "fdtmemreserv", fdtmap, sizeof(fdtmap));
384 for (j = 0; j < fdtmapsize/sizeof(fdtmap[0]); j++) {
385 fdtmap[j].mr_start = be64toh(fdtmap[j].mr_start) & ~PAGE_MASK;
386 fdtmap[j].mr_size = round_page(be64toh(fdtmap[j].mr_size));
389 KASSERT(j*sizeof(fdtmap[0]) < sizeof(fdtmap),
390 ("Exceeded number of FDT reservations"));
391 /* Add a virtual entry for the FDT itself */
393 fdtmap[j].mr_start = (vm_offset_t)fdt & ~PAGE_MASK;
394 fdtmap[j].mr_size = round_page(fdt_totalsize(fdt));
395 fdtmapsize += sizeof(fdtmap[0]);
398 fdtentries = fdtmapsize/sizeof(fdtmap[0]);
399 asz = excise_reserved_regions(avail, asz, fdtmap, fdtentries);
406 * This is called during powerpc_init, before the system is really initialized.
407 * It shall provide the total and the available regions of RAM.
408 * The available regions need not take the kernel into account.
411 ofw_mem_regions(struct mem_region *memp, int *memsz,
412 struct mem_region *availp, int *availsz)
422 * Get memory from all the /memory nodes.
424 for (phandle = OF_child(OF_peer(0)); phandle != 0;
425 phandle = OF_peer(phandle)) {
426 if (OF_getprop(phandle, "name", name, sizeof(name)) <= 0)
428 if (strncmp(name, "memory", sizeof(name)) != 0 &&
429 strncmp(name, "memory@", strlen("memory@")) != 0)
432 res = parse_ofw_memory(phandle, "reg", &memp[msz]);
433 msz += res/sizeof(struct mem_region);
436 * On POWER9 Systems we might have both linux,usable-memory and
437 * reg properties. 'reg' denotes all available memory, but we
438 * must use 'linux,usable-memory', a subset, as some memory
439 * regions are reserved for NVLink.
441 if (OF_getproplen(phandle, "linux,usable-memory") >= 0)
442 res = parse_ofw_memory(phandle, "linux,usable-memory",
444 else if (OF_getproplen(phandle, "available") >= 0)
445 res = parse_ofw_memory(phandle, "available",
448 res = parse_ofw_memory(phandle, "reg", &availp[asz]);
449 asz += res/sizeof(struct mem_region);
453 phandle = OF_finddevice("/chosen");
454 if (OF_hasprop(phandle, "fdtmemreserv"))
455 asz = excise_fdt_reserved(availp, asz);
457 /* If the kernel is being loaded through kexec, initrd region is listed
458 * in /chosen but the region is not marked as reserved, so, we might exclude
461 if (OF_hasprop(phandle, "linux,initrd-start"))
462 asz = excise_initrd_region(availp, asz);
466 if (opal_check() == 0)
467 asz = excise_msi_region(availp, asz);
475 OF_initial_setup(void *fdt_ptr, void *junk, int (*openfirm)(void *))
482 __asm __volatile("mfsprg0 %0" : "=&r"(ofmsr[1]));
484 __asm __volatile("mfsprg1 %0" : "=&r"(ofmsr[2]));
485 __asm __volatile("mfsprg2 %0" : "=&r"(ofmsr[3]));
486 __asm __volatile("mfsprg3 %0" : "=&r"(ofmsr[4]));
487 openfirmware_entry = openfirm;
489 if (ofmsr[0] & PSL_DR)
494 ofw_save_trap_vec(save_trap_init);
505 boolean_t status = FALSE;
509 if (openfirmware_entry != NULL) {
511 status = OF_install(OFW_STD_REAL, 0);
514 status = OF_install(OFW_STD_32BIT, 0);
516 status = OF_install(OFW_STD_DIRECT, 0);
523 err = OF_init(openfirmware);
529 bus_space_tag_t fdt_bt;
530 vm_offset_t tmp_fdt_ptr;
535 status = OF_install(OFW_FDT, 0);
539 #ifdef AIM /* AIM-only for now -- Book-E does this remapping in early init */
540 /* Get the FDT size for mapping if we can */
541 tmp_fdt_ptr = pmap_early_io_map((vm_paddr_t)fdt, PAGE_SIZE);
542 if (fdt_check_header((void *)tmp_fdt_ptr) != 0) {
543 pmap_early_io_unmap(tmp_fdt_ptr, PAGE_SIZE);
546 fdt_size = fdt_totalsize((void *)tmp_fdt_ptr);
547 pmap_early_io_unmap(tmp_fdt_ptr, PAGE_SIZE);
550 * Map this for real. Use bus_space_map() to take advantage
551 * of its auto-remapping function once the kernel is loaded.
552 * This is a dirty hack, but what we have.
554 #ifdef _LITTLE_ENDIAN
559 bus_space_map(fdt_bt, (vm_paddr_t)fdt, fdt_size, 0, &fdt_va);
561 err = OF_init((void *)fdt_va);
568 #ifdef FDT_DTB_STATIC
570 * Check for a statically included blob already in the kernel and
571 * needing no mapping.
574 status = OF_install(OFW_FDT, 0);
577 err = OF_init(&fdt_static_dtb);
599 KASSERT(!pmap_bootstrapped, ("Cannot call ofw_quiesce after VM is up"));
601 args.name = (cell_t)(uintptr_t)"quiesce";
608 openfirmware_core(void *args)
613 if (openfirmware_entry == NULL)
617 * Turn off exceptions - we really don't want to end up
618 * anywhere unexpected with PCPU set to something strange
619 * or the stack pointer wrong.
621 oldmsr = intr_disable();
625 /* Save trap vectors */
626 ofw_save_trap_vec(save_trap_of);
628 /* Restore initially saved trap vectors */
629 ofw_restore_trap_vec(save_trap_init);
631 #ifndef __powerpc64__
633 * Clear battable[] translations
635 if (!(cpu_features & PPC_FEATURE_64))
636 __asm __volatile("mtdbatu 2, %0\n"
637 "mtdbatu 3, %0" : : "r" (0));
641 result = ofwcall(args);
643 /* Restore trap vecotrs */
644 ofw_restore_trap_vec(save_trap_of);
648 intr_restore(oldmsr);
657 volatile int in_progress;
661 ofw_rendezvous_dispatch(void *xargs)
663 struct ofw_rv_args *rv_args = xargs;
665 /* NOTE: Interrupts are disabled here */
667 if (PCPU_GET(cpuid) == 0) {
669 * Execute all OF calls on CPU 0
671 rv_args->retval = openfirmware_core(rv_args->args);
672 rv_args->in_progress = 0;
675 * Spin with interrupts off on other CPUs while OF has
676 * control of the machine.
678 while (rv_args->in_progress)
685 openfirmware(void *args)
689 struct ofw_rv_args rv_args;
692 if (openfirmware_entry == NULL)
697 result = openfirmware_core(args);
700 rv_args.in_progress = 1;
701 smp_rendezvous(smp_no_rendezvous_barrier,
702 ofw_rendezvous_dispatch, smp_no_rendezvous_barrier,
704 result = rv_args.retval;
707 result = openfirmware_core(args);
723 args.name = (cell_t)(uintptr_t)"interpret";
726 args.arg = (cell_t)(uintptr_t)"reset-all";
727 openfirmware_core(&args); /* Don't do rendezvous! */
729 for (;;); /* just in case */
735 OF_getetheraddr(device_t dev, u_char *addr)
739 node = ofw_bus_get_node(dev);
740 OF_getprop(node, "local-mac-address", addr, ETHER_ADDR_LEN);
744 * Return a bus handle and bus tag that corresponds to the register
745 * numbered regno for the device referenced by the package handle
746 * dev. This function is intended to be used by console drivers in
747 * early boot only. It works by mapping the address of the device's
748 * register in the address space of its parent and recursively walk
749 * the device tree upward this way.
752 OF_decode_addr(phandle_t dev, int regno, bus_space_tag_t *tag,
753 bus_space_handle_t *handle, bus_size_t *sz)
760 res = ofw_reg_to_paddr(dev, regno, &addr, &size, &pci_hi);
764 if (pci_hi == OFW_PADDR_NOT_PCI) {
769 flags = (pci_hi & OFW_PCI_PHYS_HI_PREFETCHABLE) ?
770 BUS_SPACE_MAP_PREFETCHABLE: 0;
776 return (bus_space_map(*tag, addr, size, flags, handle));