3 * Copyright (C) 2001 Todd Inglett, IBM Corporation
5 * pSeries LPAR support.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 /* Enables debugging of low-level hash table routines - careful! */
25 #include <linux/kernel.h>
26 #include <linux/dma-mapping.h>
27 #include <linux/console.h>
28 #include <asm/processor.h>
31 #include <asm/pgtable.h>
32 #include <asm/machdep.h>
33 #include <asm/abs_addr.h>
34 #include <asm/mmu_context.h>
35 #include <asm/iommu.h>
36 #include <asm/tlbflush.h>
39 #include <asm/cputable.h>
43 #include "plpar_wrappers.h"
48 EXPORT_SYMBOL(plpar_hcall);
49 EXPORT_SYMBOL(plpar_hcall9);
50 EXPORT_SYMBOL(plpar_hcall_norets);
52 extern void pSeries_find_serial_port(void);
55 static int vtermno; /* virtual terminal# for udbg */
57 #define __ALIGNED__ __attribute__((__aligned__(sizeof(long))))
58 static void udbg_hvsi_putc(char c)
60 /* packet's seqno isn't used anyways */
61 uint8_t packet[] __ALIGNED__ = { 0xff, 5, 0, 0, c };
68 rc = plpar_put_term_char(vtermno, sizeof(packet), packet);
69 } while (rc == H_BUSY);
72 static long hvsi_udbg_buf_len;
73 static uint8_t hvsi_udbg_buf[256];
75 static int udbg_hvsi_getc_poll(void)
80 if (hvsi_udbg_buf_len == 0) {
81 rc = plpar_get_term_char(vtermno, &hvsi_udbg_buf_len, hvsi_udbg_buf);
82 if (rc != H_SUCCESS || hvsi_udbg_buf[0] != 0xff) {
83 /* bad read or non-data packet */
84 hvsi_udbg_buf_len = 0;
86 /* remove the packet header */
87 for (i = 4; i < hvsi_udbg_buf_len; i++)
88 hvsi_udbg_buf[i-4] = hvsi_udbg_buf[i];
89 hvsi_udbg_buf_len -= 4;
93 if (hvsi_udbg_buf_len <= 0 || hvsi_udbg_buf_len > 256) {
95 hvsi_udbg_buf_len = 0;
99 ch = hvsi_udbg_buf[0];
100 /* shift remaining data down */
101 for (i = 1; i < hvsi_udbg_buf_len; i++) {
102 hvsi_udbg_buf[i-1] = hvsi_udbg_buf[i];
109 static int udbg_hvsi_getc(void)
113 ch = udbg_hvsi_getc_poll();
115 /* This shouldn't be needed...but... */
116 volatile unsigned long delay;
117 for (delay=0; delay < 2000000; delay++)
125 static void udbg_putcLP(char c)
135 rc = plpar_put_term_char(vtermno, 1, buf);
136 } while(rc == H_BUSY);
139 /* Buffered chars getc */
140 static long inbuflen;
141 static long inbuf[2]; /* must be 2 longs */
143 static int udbg_getc_pollLP(void)
145 /* The interface is tricky because it may return up to 16 chars.
146 * We save them statically for future calls to udbg_getc().
148 char ch, *buf = (char *)inbuf;
152 /* get some more chars. */
154 rc = plpar_get_term_char(vtermno, &inbuflen, buf);
156 inbuflen = 0; /* otherwise inbuflen is garbage */
158 if (inbuflen <= 0 || inbuflen > 16) {
159 /* Catch error case as well as other oddities (corruption) */
164 for (i = 1; i < inbuflen; i++) /* shuffle them down. */
170 static int udbg_getcLP(void)
174 ch = udbg_getc_pollLP();
176 /* This shouldn't be needed...but... */
177 volatile unsigned long delay;
178 for (delay=0; delay < 2000000; delay++)
186 /* call this from early_init() for a working debug console on
187 * vterm capable LPAR machines
189 void __init udbg_init_debug_lpar(void)
192 udbg_putc = udbg_putcLP;
193 udbg_getc = udbg_getcLP;
194 udbg_getc_poll = udbg_getc_pollLP;
196 register_early_udbg_console();
199 /* returns 0 if couldn't find or use /chosen/stdout as console */
200 void __init find_udbg_vterm(void)
202 struct device_node *stdout_node;
206 /* find the boot console from /chosen/stdout */
209 name = of_get_property(of_chosen, "linux,stdout-path", NULL);
212 stdout_node = of_find_node_by_path(name);
215 name = of_get_property(stdout_node, "name", NULL);
217 printk(KERN_WARNING "stdout node missing 'name' property!\n");
221 /* Check if it's a virtual terminal */
222 if (strncmp(name, "vty", 3) != 0)
224 termno = of_get_property(stdout_node, "reg", NULL);
229 if (of_device_is_compatible(stdout_node, "hvterm1")) {
230 udbg_putc = udbg_putcLP;
231 udbg_getc = udbg_getcLP;
232 udbg_getc_poll = udbg_getc_pollLP;
233 add_preferred_console("hvc", termno[0] & 0xff, NULL);
234 } else if (of_device_is_compatible(stdout_node, "hvterm-protocol")) {
236 udbg_putc = udbg_hvsi_putc;
237 udbg_getc = udbg_hvsi_getc;
238 udbg_getc_poll = udbg_hvsi_getc_poll;
239 add_preferred_console("hvsi", termno[0] & 0xff, NULL);
242 of_node_put(stdout_node);
245 void vpa_init(int cpu)
247 int hwcpu = get_hard_smp_processor_id(cpu);
251 if (cpu_has_feature(CPU_FTR_ALTIVEC))
252 lppaca[cpu].vmxregs_in_use = 1;
254 addr = __pa(&lppaca[cpu]);
255 ret = register_vpa(hwcpu, addr);
258 printk(KERN_ERR "WARNING: vpa_init: VPA registration for "
259 "cpu %d (hw %d) of area %lx returns %ld\n",
260 cpu, hwcpu, addr, ret);
264 * PAPR says this feature is SLB-Buffer but firmware never
265 * reports that. All SPLPAR support SLB shadow buffer.
267 addr = __pa(&slb_shadow[cpu]);
268 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
269 ret = register_slb_shadow(hwcpu, addr);
272 "WARNING: vpa_init: SLB shadow buffer "
273 "registration for cpu %d (hw %d) of area %lx "
274 "returns %ld\n", cpu, hwcpu, addr, ret);
278 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
279 unsigned long va, unsigned long pa,
280 unsigned long rflags, unsigned long vflags,
281 int psize, int ssize)
283 unsigned long lpar_rc;
286 unsigned long hpte_v, hpte_r;
288 if (!(vflags & HPTE_V_BOLTED))
289 pr_debug("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
290 "rflags=%lx, vflags=%lx, psize=%d)\n",
291 hpte_group, va, pa, rflags, vflags, psize);
293 hpte_v = hpte_encode_v(va, psize, ssize) | vflags | HPTE_V_VALID;
294 hpte_r = hpte_encode_r(pa, psize) | rflags;
296 if (!(vflags & HPTE_V_BOLTED))
297 pr_debug(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
299 /* Now fill in the actual HPTE */
300 /* Set CEC cookie to 0 */
302 /* I-cache Invalidate = 0 */
303 /* I-cache synchronize = 0 */
307 /* Make pHyp happy */
308 if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
309 hpte_r &= ~_PAGE_COHERENT;
311 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
312 if (unlikely(lpar_rc == H_PTEG_FULL)) {
313 if (!(vflags & HPTE_V_BOLTED))
319 * Since we try and ioremap PHBs we don't own, the pte insert
320 * will fail. However we must catch the failure in hash_page
321 * or we will loop forever, so return -2 in this case.
323 if (unlikely(lpar_rc != H_SUCCESS)) {
324 if (!(vflags & HPTE_V_BOLTED))
325 pr_debug(" lpar err %lu\n", lpar_rc);
328 if (!(vflags & HPTE_V_BOLTED))
329 pr_debug(" -> slot: %lu\n", slot & 7);
331 /* Because of iSeries, we have to pass down the secondary
332 * bucket bit here as well
334 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
337 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
339 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
341 unsigned long slot_offset;
342 unsigned long lpar_rc;
344 unsigned long dummy1, dummy2;
346 /* pick a random slot to start at */
347 slot_offset = mftb() & 0x7;
349 for (i = 0; i < HPTES_PER_GROUP; i++) {
351 /* don't remove a bolted entry */
352 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
353 (0x1UL << 4), &dummy1, &dummy2);
354 if (lpar_rc == H_SUCCESS)
356 BUG_ON(lpar_rc != H_NOT_FOUND);
365 static void pSeries_lpar_hptab_clear(void)
367 unsigned long size_bytes = 1UL << ppc64_pft_size;
368 unsigned long hpte_count = size_bytes >> 4;
369 unsigned long dummy1, dummy2, dword0;
373 /* TODO: Use bulk call */
374 for (i = 0; i < hpte_count; i++) {
375 /* dont remove HPTEs with VRMA mappings */
376 lpar_rc = plpar_pte_remove_raw(H_ANDCOND, i, HPTE_V_1TB_SEG,
378 if (lpar_rc == H_NOT_FOUND) {
379 lpar_rc = plpar_pte_read_raw(0, i, &dword0, &dummy1);
380 if (!lpar_rc && ((dword0 & HPTE_V_VRMA_MASK)
381 != HPTE_V_VRMA_MASK))
382 /* Can be hpte for 1TB Seg. So remove it */
383 plpar_pte_remove_raw(0, i, 0, &dummy1, &dummy2);
389 * This computes the AVPN and B fields of the first dword of a HPTE,
390 * for use when we want to match an existing PTE. The bottom 7 bits
391 * of the returned value are zero.
393 static inline unsigned long hpte_encode_avpn(unsigned long va, int psize,
398 v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
399 v <<= HPTE_V_AVPN_SHIFT;
400 v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
405 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
406 * the low 3 bits of flags happen to line up. So no transform is needed.
407 * We can probably optimize here and assume the high bits of newpp are
408 * already zero. For now I am paranoid.
410 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
413 int psize, int ssize, int local)
415 unsigned long lpar_rc;
416 unsigned long flags = (newpp & 7) | H_AVPN;
417 unsigned long want_v;
419 want_v = hpte_encode_avpn(va, psize, ssize);
421 pr_debug(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
422 want_v, slot, flags, psize);
424 lpar_rc = plpar_pte_protect(flags, slot, want_v);
426 if (lpar_rc == H_NOT_FOUND) {
427 pr_debug("not found !\n");
433 BUG_ON(lpar_rc != H_SUCCESS);
438 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
440 unsigned long dword0;
441 unsigned long lpar_rc;
442 unsigned long dummy_word1;
445 /* Read 1 pte at a time */
446 /* Do not need RPN to logical page translation */
447 /* No cross CEC PFT access */
450 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
452 BUG_ON(lpar_rc != H_SUCCESS);
457 static long pSeries_lpar_hpte_find(unsigned long va, int psize, int ssize)
462 unsigned long want_v, hpte_v;
464 hash = hpt_hash(va, mmu_psize_defs[psize].shift, ssize);
465 want_v = hpte_encode_avpn(va, psize, ssize);
467 /* Bolted entries are always in the primary group */
468 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
469 for (i = 0; i < HPTES_PER_GROUP; i++) {
470 hpte_v = pSeries_lpar_hpte_getword0(slot);
472 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
481 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
483 int psize, int ssize)
485 unsigned long lpar_rc, slot, vsid, va, flags;
487 vsid = get_kernel_vsid(ea, ssize);
488 va = hpt_va(ea, vsid, ssize);
490 slot = pSeries_lpar_hpte_find(va, psize, ssize);
494 lpar_rc = plpar_pte_protect(flags, slot, 0);
496 BUG_ON(lpar_rc != H_SUCCESS);
499 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
500 int psize, int ssize, int local)
502 unsigned long want_v;
503 unsigned long lpar_rc;
504 unsigned long dummy1, dummy2;
506 pr_debug(" inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
507 slot, va, psize, local);
509 want_v = hpte_encode_avpn(va, psize, ssize);
510 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
511 if (lpar_rc == H_NOT_FOUND)
514 BUG_ON(lpar_rc != H_SUCCESS);
517 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
518 int psize, int ssize)
520 unsigned long slot, vsid, va;
522 vsid = get_kernel_vsid(ea, ssize);
523 va = hpt_va(ea, vsid, ssize);
525 slot = pSeries_lpar_hpte_find(va, psize, ssize);
528 pSeries_lpar_hpte_invalidate(slot, va, psize, ssize, 0);
531 /* Flag bits for H_BULK_REMOVE */
532 #define HBR_REQUEST 0x4000000000000000UL
533 #define HBR_RESPONSE 0x8000000000000000UL
534 #define HBR_END 0xc000000000000000UL
535 #define HBR_AVPN 0x0200000000000000UL
536 #define HBR_ANDCOND 0x0100000000000000UL
539 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
542 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
544 unsigned long i, pix, rc;
545 unsigned long flags = 0;
546 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
547 int lock_tlbie = !cpu_has_feature(CPU_FTR_LOCKLESS_TLBIE);
548 unsigned long param[9];
550 unsigned long hash, index, shift, hidx, slot;
555 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
557 psize = batch->psize;
558 ssize = batch->ssize;
560 for (i = 0; i < number; i++) {
561 va = batch->vaddr[i];
563 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
564 hash = hpt_hash(va, shift, ssize);
565 hidx = __rpte_to_hidx(pte, index);
566 if (hidx & _PTEIDX_SECONDARY)
568 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
569 slot += hidx & _PTEIDX_GROUP_IX;
570 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
571 pSeries_lpar_hpte_invalidate(slot, va, psize,
574 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
575 param[pix+1] = hpte_encode_avpn(va, psize,
579 rc = plpar_hcall9(H_BULK_REMOVE, param,
580 param[0], param[1], param[2],
581 param[3], param[4], param[5],
583 BUG_ON(rc != H_SUCCESS);
587 } pte_iterate_hashed_end();
590 param[pix] = HBR_END;
591 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
592 param[2], param[3], param[4], param[5],
594 BUG_ON(rc != H_SUCCESS);
598 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
601 void __init hpte_init_lpar(void)
603 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
604 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
605 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
606 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
607 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
608 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
609 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
610 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
613 #ifdef CONFIG_PPC_SMLPAR
614 #define CMO_FREE_HINT_DEFAULT 1
615 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
617 static int __init cmo_free_hint(char *str)
620 parm = strstrip(str);
622 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
623 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
624 cmo_free_hint_flag = 0;
628 cmo_free_hint_flag = 1;
629 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
631 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
637 __setup("cmo_free_hint=", cmo_free_hint);
639 static void pSeries_set_page_state(struct page *page, int order,
643 unsigned long cmo_page_sz, addr;
645 cmo_page_sz = cmo_get_page_size();
646 addr = __pa((unsigned long)page_address(page));
648 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
649 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
650 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
654 void arch_free_page(struct page *page, int order)
656 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
659 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
661 EXPORT_SYMBOL(arch_free_page);