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 <linux/export.h>
29 #include <linux/static_key.h>
30 #include <asm/processor.h>
33 #include <asm/pgtable.h>
34 #include <asm/machdep.h>
35 #include <asm/mmu_context.h>
36 #include <asm/iommu.h>
37 #include <asm/tlbflush.h>
40 #include <asm/cputable.h>
43 #include <asm/trace.h>
44 #include <asm/firmware.h>
45 #include <asm/plpar_wrappers.h>
49 /* Flag bits for H_BULK_REMOVE */
50 #define HBR_REQUEST 0x4000000000000000UL
51 #define HBR_RESPONSE 0x8000000000000000UL
52 #define HBR_END 0xc000000000000000UL
53 #define HBR_AVPN 0x0200000000000000UL
54 #define HBR_ANDCOND 0x0100000000000000UL
58 EXPORT_SYMBOL(plpar_hcall);
59 EXPORT_SYMBOL(plpar_hcall9);
60 EXPORT_SYMBOL(plpar_hcall_norets);
62 void vpa_init(int cpu)
64 int hwcpu = get_hard_smp_processor_id(cpu);
67 struct paca_struct *pp;
68 struct dtl_entry *dtl;
71 * The spec says it "may be problematic" if CPU x registers the VPA of
72 * CPU y. We should never do that, but wail if we ever do.
74 WARN_ON(cpu != smp_processor_id());
76 if (cpu_has_feature(CPU_FTR_ALTIVEC))
77 lppaca_of(cpu).vmxregs_in_use = 1;
79 if (cpu_has_feature(CPU_FTR_ARCH_207S))
80 lppaca_of(cpu).ebb_regs_in_use = 1;
82 addr = __pa(&lppaca_of(cpu));
83 ret = register_vpa(hwcpu, addr);
86 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
87 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
91 * PAPR says this feature is SLB-Buffer but firmware never
92 * reports that. All SPLPAR support SLB shadow buffer.
94 addr = __pa(paca[cpu].slb_shadow_ptr);
95 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
96 ret = register_slb_shadow(hwcpu, addr);
98 pr_err("WARNING: SLB shadow buffer registration for "
99 "cpu %d (hw %d) of area %lx failed with %ld\n",
100 cpu, hwcpu, addr, ret);
104 * Register dispatch trace log, if one has been allocated.
107 dtl = pp->dispatch_log;
111 lppaca_of(cpu).dtl_idx = 0;
113 /* hypervisor reads buffer length from this field */
114 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
115 ret = register_dtl(hwcpu, __pa(dtl));
117 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
118 "failed with %ld\n", smp_processor_id(),
120 lppaca_of(cpu).dtl_enable_mask = 2;
124 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
125 unsigned long vpn, unsigned long pa,
126 unsigned long rflags, unsigned long vflags,
127 int psize, int apsize, int ssize)
129 unsigned long lpar_rc;
132 unsigned long hpte_v, hpte_r;
134 if (!(vflags & HPTE_V_BOLTED))
135 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
136 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
137 hpte_group, vpn, pa, rflags, vflags, psize);
139 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
140 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
142 if (!(vflags & HPTE_V_BOLTED))
143 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
145 /* Now fill in the actual HPTE */
146 /* Set CEC cookie to 0 */
148 /* I-cache Invalidate = 0 */
149 /* I-cache synchronize = 0 */
153 /* Make pHyp happy */
154 if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
157 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
158 flags |= H_COALESCE_CAND;
160 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
161 if (unlikely(lpar_rc == H_PTEG_FULL)) {
162 if (!(vflags & HPTE_V_BOLTED))
168 * Since we try and ioremap PHBs we don't own, the pte insert
169 * will fail. However we must catch the failure in hash_page
170 * or we will loop forever, so return -2 in this case.
172 if (unlikely(lpar_rc != H_SUCCESS)) {
173 if (!(vflags & HPTE_V_BOLTED))
174 pr_devel(" lpar err %ld\n", lpar_rc);
177 if (!(vflags & HPTE_V_BOLTED))
178 pr_devel(" -> slot: %lu\n", slot & 7);
180 /* Because of iSeries, we have to pass down the secondary
181 * bucket bit here as well
183 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
186 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
188 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
190 unsigned long slot_offset;
191 unsigned long lpar_rc;
193 unsigned long dummy1, dummy2;
195 /* pick a random slot to start at */
196 slot_offset = mftb() & 0x7;
198 for (i = 0; i < HPTES_PER_GROUP; i++) {
200 /* don't remove a bolted entry */
201 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
202 (0x1UL << 4), &dummy1, &dummy2);
203 if (lpar_rc == H_SUCCESS)
207 * The test for adjunct partition is performed before the
208 * ANDCOND test. H_RESOURCE may be returned, so we need to
209 * check for that as well.
211 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
220 static void pSeries_lpar_hptab_clear(void)
222 unsigned long size_bytes = 1UL << ppc64_pft_size;
223 unsigned long hpte_count = size_bytes >> 4;
231 /* Read in batches of 4,
232 * invalidate only valid entries not in the VRMA
233 * hpte_count will be a multiple of 4
235 for (i = 0; i < hpte_count; i += 4) {
236 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
237 if (lpar_rc != H_SUCCESS)
239 for (j = 0; j < 4; j++){
240 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
243 if (ptes[j].pteh & HPTE_V_VALID)
244 plpar_pte_remove_raw(0, i + j, 0,
245 &(ptes[j].pteh), &(ptes[j].ptel));
249 #ifdef __LITTLE_ENDIAN__
250 /* Reset exceptions to big endian */
251 if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
254 rc = pseries_big_endian_exceptions();
256 * At this point it is unlikely panic() will get anything
257 * out to the user, but at least this will stop us from
258 * continuing on further and creating an even more
259 * difficult to debug situation.
262 panic("Could not enable big endian exceptions");
268 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
269 * the low 3 bits of flags happen to line up. So no transform is needed.
270 * We can probably optimize here and assume the high bits of newpp are
271 * already zero. For now I am paranoid.
273 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
276 int psize, int apsize,
277 int ssize, int local)
279 unsigned long lpar_rc;
280 unsigned long flags = (newpp & 7) | H_AVPN;
281 unsigned long want_v;
283 want_v = hpte_encode_avpn(vpn, psize, ssize);
285 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
286 want_v, slot, flags, psize);
288 lpar_rc = plpar_pte_protect(flags, slot, want_v);
290 if (lpar_rc == H_NOT_FOUND) {
291 pr_devel("not found !\n");
297 BUG_ON(lpar_rc != H_SUCCESS);
302 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
304 unsigned long dword0;
305 unsigned long lpar_rc;
306 unsigned long dummy_word1;
309 /* Read 1 pte at a time */
310 /* Do not need RPN to logical page translation */
311 /* No cross CEC PFT access */
314 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
316 BUG_ON(lpar_rc != H_SUCCESS);
321 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
326 unsigned long want_v, hpte_v;
328 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
329 want_v = hpte_encode_avpn(vpn, psize, ssize);
331 /* Bolted entries are always in the primary group */
332 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
333 for (i = 0; i < HPTES_PER_GROUP; i++) {
334 hpte_v = pSeries_lpar_hpte_getword0(slot);
336 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
345 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
347 int psize, int ssize)
350 unsigned long lpar_rc, slot, vsid, flags;
352 vsid = get_kernel_vsid(ea, ssize);
353 vpn = hpt_vpn(ea, vsid, ssize);
355 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
359 lpar_rc = plpar_pte_protect(flags, slot, 0);
361 BUG_ON(lpar_rc != H_SUCCESS);
364 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
365 int psize, int apsize,
366 int ssize, int local)
368 unsigned long want_v;
369 unsigned long lpar_rc;
370 unsigned long dummy1, dummy2;
372 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
373 slot, vpn, psize, local);
375 want_v = hpte_encode_avpn(vpn, psize, ssize);
376 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
377 if (lpar_rc == H_NOT_FOUND)
380 BUG_ON(lpar_rc != H_SUCCESS);
384 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
385 * to make sure that we avoid bouncing the hypervisor tlbie lock.
387 #define PPC64_HUGE_HPTE_BATCH 12
389 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
390 unsigned long *vpn, int count,
391 int psize, int ssize)
393 unsigned long param[8];
394 int i = 0, pix = 0, rc;
395 unsigned long flags = 0;
396 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
399 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
401 for (i = 0; i < count; i++) {
403 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
404 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
407 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
408 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
411 rc = plpar_hcall9(H_BULK_REMOVE, param,
412 param[0], param[1], param[2],
413 param[3], param[4], param[5],
415 BUG_ON(rc != H_SUCCESS);
421 param[pix] = HBR_END;
422 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
423 param[2], param[3], param[4], param[5],
425 BUG_ON(rc != H_SUCCESS);
429 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
432 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
434 unsigned char *hpte_slot_array,
435 int psize, int ssize)
438 unsigned long s_addr = addr;
439 unsigned int max_hpte_count, valid;
440 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
441 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
442 unsigned long shift, hidx, vpn = 0, hash, slot;
444 shift = mmu_psize_defs[psize].shift;
445 max_hpte_count = 1U << (PMD_SHIFT - shift);
447 for (i = 0; i < max_hpte_count; i++) {
448 valid = hpte_valid(hpte_slot_array, i);
451 hidx = hpte_hash_index(hpte_slot_array, i);
454 addr = s_addr + (i * (1ul << shift));
455 vpn = hpt_vpn(addr, vsid, ssize);
456 hash = hpt_hash(vpn, shift, ssize);
457 if (hidx & _PTEIDX_SECONDARY)
460 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
461 slot += hidx & _PTEIDX_GROUP_IX;
463 slot_array[index] = slot;
464 vpn_array[index] = vpn;
465 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
467 * Now do a bluk invalidate
469 __pSeries_lpar_hugepage_invalidate(slot_array,
471 PPC64_HUGE_HPTE_BATCH,
478 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
479 index, psize, ssize);
482 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
483 int psize, int ssize)
486 unsigned long slot, vsid;
488 vsid = get_kernel_vsid(ea, ssize);
489 vpn = hpt_vpn(ea, vsid, ssize);
491 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
494 * lpar doesn't use the passed actual page size
496 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
500 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
503 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
506 unsigned long i, pix, rc;
507 unsigned long flags = 0;
508 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
509 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
510 unsigned long param[9];
511 unsigned long hash, index, shift, hidx, slot;
516 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
518 psize = batch->psize;
519 ssize = batch->ssize;
521 for (i = 0; i < number; i++) {
524 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
525 hash = hpt_hash(vpn, shift, ssize);
526 hidx = __rpte_to_hidx(pte, index);
527 if (hidx & _PTEIDX_SECONDARY)
529 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
530 slot += hidx & _PTEIDX_GROUP_IX;
531 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
533 * lpar doesn't use the passed actual page size
535 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
538 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
539 param[pix+1] = hpte_encode_avpn(vpn, psize,
543 rc = plpar_hcall9(H_BULK_REMOVE, param,
544 param[0], param[1], param[2],
545 param[3], param[4], param[5],
547 BUG_ON(rc != H_SUCCESS);
551 } pte_iterate_hashed_end();
554 param[pix] = HBR_END;
555 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
556 param[2], param[3], param[4], param[5],
558 BUG_ON(rc != H_SUCCESS);
562 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
565 static int __init disable_bulk_remove(char *str)
567 if (strcmp(str, "off") == 0 &&
568 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
569 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
570 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
575 __setup("bulk_remove=", disable_bulk_remove);
577 void __init hpte_init_lpar(void)
579 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
580 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
581 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
582 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
583 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
584 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
585 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
586 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
587 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
590 #ifdef CONFIG_PPC_SMLPAR
591 #define CMO_FREE_HINT_DEFAULT 1
592 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
594 static int __init cmo_free_hint(char *str)
597 parm = strstrip(str);
599 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
600 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
601 cmo_free_hint_flag = 0;
605 cmo_free_hint_flag = 1;
606 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
608 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
614 __setup("cmo_free_hint=", cmo_free_hint);
616 static void pSeries_set_page_state(struct page *page, int order,
620 unsigned long cmo_page_sz, addr;
622 cmo_page_sz = cmo_get_page_size();
623 addr = __pa((unsigned long)page_address(page));
625 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
626 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
627 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
631 void arch_free_page(struct page *page, int order)
633 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
636 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
638 EXPORT_SYMBOL(arch_free_page);
642 #ifdef CONFIG_TRACEPOINTS
643 #ifdef HAVE_JUMP_LABEL
644 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
646 void hcall_tracepoint_regfunc(void)
648 static_key_slow_inc(&hcall_tracepoint_key);
651 void hcall_tracepoint_unregfunc(void)
653 static_key_slow_dec(&hcall_tracepoint_key);
657 * We optimise our hcall path by placing hcall_tracepoint_refcount
658 * directly in the TOC so we can check if the hcall tracepoints are
659 * enabled via a single load.
662 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
663 extern long hcall_tracepoint_refcount;
665 void hcall_tracepoint_regfunc(void)
667 hcall_tracepoint_refcount++;
670 void hcall_tracepoint_unregfunc(void)
672 hcall_tracepoint_refcount--;
677 * Since the tracing code might execute hcalls we need to guard against
678 * recursion. One example of this are spinlocks calling H_YIELD on
679 * shared processor partitions.
681 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
684 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
690 * We cannot call tracepoints inside RCU idle regions which
691 * means we must not trace H_CEDE.
693 if (opcode == H_CEDE)
696 local_irq_save(flags);
698 depth = &__get_cpu_var(hcall_trace_depth);
705 trace_hcall_entry(opcode, args);
709 local_irq_restore(flags);
712 void __trace_hcall_exit(long opcode, unsigned long retval,
713 unsigned long *retbuf)
718 if (opcode == H_CEDE)
721 local_irq_save(flags);
723 depth = &__get_cpu_var(hcall_trace_depth);
729 trace_hcall_exit(opcode, retval, retbuf);
734 local_irq_restore(flags);
740 * H_GET_MPP hcall returns info in 7 parms
742 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
745 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
747 rc = plpar_hcall9(H_GET_MPP, retbuf);
749 mpp_data->entitled_mem = retbuf[0];
750 mpp_data->mapped_mem = retbuf[1];
752 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
753 mpp_data->pool_num = retbuf[2] & 0xffff;
755 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
756 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
757 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
759 mpp_data->pool_size = retbuf[4];
760 mpp_data->loan_request = retbuf[5];
761 mpp_data->backing_mem = retbuf[6];
765 EXPORT_SYMBOL(h_get_mpp);
767 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
770 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
772 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
774 mpp_x_data->coalesced_bytes = retbuf[0];
775 mpp_x_data->pool_coalesced_bytes = retbuf[1];
776 mpp_x_data->pool_purr_cycles = retbuf[2];
777 mpp_x_data->pool_spurr_cycles = retbuf[3];