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/jump_label.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>
46 #include <asm/kexec.h>
47 #include <asm/fadump.h>
51 /* Flag bits for H_BULK_REMOVE */
52 #define HBR_REQUEST 0x4000000000000000UL
53 #define HBR_RESPONSE 0x8000000000000000UL
54 #define HBR_END 0xc000000000000000UL
55 #define HBR_AVPN 0x0200000000000000UL
56 #define HBR_ANDCOND 0x0100000000000000UL
60 EXPORT_SYMBOL(plpar_hcall);
61 EXPORT_SYMBOL(plpar_hcall9);
62 EXPORT_SYMBOL(plpar_hcall_norets);
64 void vpa_init(int cpu)
66 int hwcpu = get_hard_smp_processor_id(cpu);
69 struct paca_struct *pp;
70 struct dtl_entry *dtl;
73 * The spec says it "may be problematic" if CPU x registers the VPA of
74 * CPU y. We should never do that, but wail if we ever do.
76 WARN_ON(cpu != smp_processor_id());
78 if (cpu_has_feature(CPU_FTR_ALTIVEC))
79 lppaca_of(cpu).vmxregs_in_use = 1;
81 if (cpu_has_feature(CPU_FTR_ARCH_207S))
82 lppaca_of(cpu).ebb_regs_in_use = 1;
84 addr = __pa(&lppaca_of(cpu));
85 ret = register_vpa(hwcpu, addr);
88 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
89 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
93 * PAPR says this feature is SLB-Buffer but firmware never
94 * reports that. All SPLPAR support SLB shadow buffer.
96 addr = __pa(paca[cpu].slb_shadow_ptr);
97 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
98 ret = register_slb_shadow(hwcpu, addr);
100 pr_err("WARNING: SLB shadow buffer registration for "
101 "cpu %d (hw %d) of area %lx failed with %ld\n",
102 cpu, hwcpu, addr, ret);
106 * Register dispatch trace log, if one has been allocated.
109 dtl = pp->dispatch_log;
113 lppaca_of(cpu).dtl_idx = 0;
115 /* hypervisor reads buffer length from this field */
116 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
117 ret = register_dtl(hwcpu, __pa(dtl));
119 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
120 "failed with %ld\n", smp_processor_id(),
122 lppaca_of(cpu).dtl_enable_mask = 2;
126 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
127 unsigned long vpn, unsigned long pa,
128 unsigned long rflags, unsigned long vflags,
129 int psize, int apsize, int ssize)
131 unsigned long lpar_rc;
134 unsigned long hpte_v, hpte_r;
136 if (!(vflags & HPTE_V_BOLTED))
137 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
138 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
139 hpte_group, vpn, pa, rflags, vflags, psize);
141 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
142 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
144 if (!(vflags & HPTE_V_BOLTED))
145 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
147 /* Now fill in the actual HPTE */
148 /* Set CEC cookie to 0 */
150 /* I-cache Invalidate = 0 */
151 /* I-cache synchronize = 0 */
155 /* Make pHyp happy */
156 if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
159 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
160 flags |= H_COALESCE_CAND;
162 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
163 if (unlikely(lpar_rc == H_PTEG_FULL)) {
164 if (!(vflags & HPTE_V_BOLTED))
170 * Since we try and ioremap PHBs we don't own, the pte insert
171 * will fail. However we must catch the failure in hash_page
172 * or we will loop forever, so return -2 in this case.
174 if (unlikely(lpar_rc != H_SUCCESS)) {
175 if (!(vflags & HPTE_V_BOLTED))
176 pr_devel(" lpar err %ld\n", lpar_rc);
179 if (!(vflags & HPTE_V_BOLTED))
180 pr_devel(" -> slot: %lu\n", slot & 7);
182 /* Because of iSeries, we have to pass down the secondary
183 * bucket bit here as well
185 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
188 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
190 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
192 unsigned long slot_offset;
193 unsigned long lpar_rc;
195 unsigned long dummy1, dummy2;
197 /* pick a random slot to start at */
198 slot_offset = mftb() & 0x7;
200 for (i = 0; i < HPTES_PER_GROUP; i++) {
202 /* don't remove a bolted entry */
203 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
204 (0x1UL << 4), &dummy1, &dummy2);
205 if (lpar_rc == H_SUCCESS)
209 * The test for adjunct partition is performed before the
210 * ANDCOND test. H_RESOURCE may be returned, so we need to
211 * check for that as well.
213 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
222 static void pSeries_lpar_hptab_clear(void)
224 unsigned long size_bytes = 1UL << ppc64_pft_size;
225 unsigned long hpte_count = size_bytes >> 4;
233 /* Read in batches of 4,
234 * invalidate only valid entries not in the VRMA
235 * hpte_count will be a multiple of 4
237 for (i = 0; i < hpte_count; i += 4) {
238 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
239 if (lpar_rc != H_SUCCESS)
241 for (j = 0; j < 4; j++){
242 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
245 if (ptes[j].pteh & HPTE_V_VALID)
246 plpar_pte_remove_raw(0, i + j, 0,
247 &(ptes[j].pteh), &(ptes[j].ptel));
251 #ifdef __LITTLE_ENDIAN__
253 * Reset exceptions to big endian.
255 * FIXME this is a hack for kexec, we need to reset the exception
256 * endian before starting the new kernel and this is a convenient place
259 * This is also called on boot when a fadump happens. In that case we
260 * must not change the exception endian mode.
262 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) {
265 rc = pseries_big_endian_exceptions();
267 * At this point it is unlikely panic() will get anything
268 * out to the user, but at least this will stop us from
269 * continuing on further and creating an even more
270 * difficult to debug situation.
272 * There is a known problem when kdump'ing, if cpus are offline
273 * the above call will fail. Rather than panicking again, keep
274 * going and hope the kdump kernel is also little endian, which
277 if (rc && !kdump_in_progress())
278 panic("Could not enable big endian exceptions");
284 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
285 * the low 3 bits of flags happen to line up. So no transform is needed.
286 * We can probably optimize here and assume the high bits of newpp are
287 * already zero. For now I am paranoid.
289 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
292 int psize, int apsize,
293 int ssize, unsigned long inv_flags)
295 unsigned long lpar_rc;
296 unsigned long flags = (newpp & 7) | H_AVPN;
297 unsigned long want_v;
299 want_v = hpte_encode_avpn(vpn, psize, ssize);
301 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
302 want_v, slot, flags, psize);
304 lpar_rc = plpar_pte_protect(flags, slot, want_v);
306 if (lpar_rc == H_NOT_FOUND) {
307 pr_devel("not found !\n");
313 BUG_ON(lpar_rc != H_SUCCESS);
318 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
320 unsigned long dword0;
321 unsigned long lpar_rc;
322 unsigned long dummy_word1;
325 /* Read 1 pte at a time */
326 /* Do not need RPN to logical page translation */
327 /* No cross CEC PFT access */
330 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
332 BUG_ON(lpar_rc != H_SUCCESS);
337 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
342 unsigned long want_v, hpte_v;
344 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
345 want_v = hpte_encode_avpn(vpn, psize, ssize);
347 /* Bolted entries are always in the primary group */
348 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
349 for (i = 0; i < HPTES_PER_GROUP; i++) {
350 hpte_v = pSeries_lpar_hpte_getword0(slot);
352 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
361 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
363 int psize, int ssize)
366 unsigned long lpar_rc, slot, vsid, flags;
368 vsid = get_kernel_vsid(ea, ssize);
369 vpn = hpt_vpn(ea, vsid, ssize);
371 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
375 lpar_rc = plpar_pte_protect(flags, slot, 0);
377 BUG_ON(lpar_rc != H_SUCCESS);
380 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
381 int psize, int apsize,
382 int ssize, int local)
384 unsigned long want_v;
385 unsigned long lpar_rc;
386 unsigned long dummy1, dummy2;
388 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
389 slot, vpn, psize, local);
391 want_v = hpte_encode_avpn(vpn, psize, ssize);
392 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
393 if (lpar_rc == H_NOT_FOUND)
396 BUG_ON(lpar_rc != H_SUCCESS);
400 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
401 * to make sure that we avoid bouncing the hypervisor tlbie lock.
403 #define PPC64_HUGE_HPTE_BATCH 12
405 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
406 unsigned long *vpn, int count,
407 int psize, int ssize)
409 unsigned long param[8];
410 int i = 0, pix = 0, rc;
411 unsigned long flags = 0;
412 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
415 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
417 for (i = 0; i < count; i++) {
419 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
420 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
423 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
424 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
427 rc = plpar_hcall9(H_BULK_REMOVE, param,
428 param[0], param[1], param[2],
429 param[3], param[4], param[5],
431 BUG_ON(rc != H_SUCCESS);
437 param[pix] = HBR_END;
438 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
439 param[2], param[3], param[4], param[5],
441 BUG_ON(rc != H_SUCCESS);
445 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
448 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
450 unsigned char *hpte_slot_array,
451 int psize, int ssize, int local)
454 unsigned long s_addr = addr;
455 unsigned int max_hpte_count, valid;
456 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
457 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
458 unsigned long shift, hidx, vpn = 0, hash, slot;
460 shift = mmu_psize_defs[psize].shift;
461 max_hpte_count = 1U << (PMD_SHIFT - shift);
463 for (i = 0; i < max_hpte_count; i++) {
464 valid = hpte_valid(hpte_slot_array, i);
467 hidx = hpte_hash_index(hpte_slot_array, i);
470 addr = s_addr + (i * (1ul << shift));
471 vpn = hpt_vpn(addr, vsid, ssize);
472 hash = hpt_hash(vpn, shift, ssize);
473 if (hidx & _PTEIDX_SECONDARY)
476 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
477 slot += hidx & _PTEIDX_GROUP_IX;
479 slot_array[index] = slot;
480 vpn_array[index] = vpn;
481 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
483 * Now do a bluk invalidate
485 __pSeries_lpar_hugepage_invalidate(slot_array,
487 PPC64_HUGE_HPTE_BATCH,
494 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
495 index, psize, ssize);
498 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
499 int psize, int ssize)
502 unsigned long slot, vsid;
504 vsid = get_kernel_vsid(ea, ssize);
505 vpn = hpt_vpn(ea, vsid, ssize);
507 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
510 * lpar doesn't use the passed actual page size
512 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
516 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
519 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
522 unsigned long i, pix, rc;
523 unsigned long flags = 0;
524 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
525 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
526 unsigned long param[9];
527 unsigned long hash, index, shift, hidx, slot;
532 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
534 psize = batch->psize;
535 ssize = batch->ssize;
537 for (i = 0; i < number; i++) {
540 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
541 hash = hpt_hash(vpn, shift, ssize);
542 hidx = __rpte_to_hidx(pte, index);
543 if (hidx & _PTEIDX_SECONDARY)
545 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
546 slot += hidx & _PTEIDX_GROUP_IX;
547 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
549 * lpar doesn't use the passed actual page size
551 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
554 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
555 param[pix+1] = hpte_encode_avpn(vpn, psize,
559 rc = plpar_hcall9(H_BULK_REMOVE, param,
560 param[0], param[1], param[2],
561 param[3], param[4], param[5],
563 BUG_ON(rc != H_SUCCESS);
567 } pte_iterate_hashed_end();
570 param[pix] = HBR_END;
571 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
572 param[2], param[3], param[4], param[5],
574 BUG_ON(rc != H_SUCCESS);
578 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
581 static int __init disable_bulk_remove(char *str)
583 if (strcmp(str, "off") == 0 &&
584 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
585 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
586 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
591 __setup("bulk_remove=", disable_bulk_remove);
593 void __init hpte_init_lpar(void)
595 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
596 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
597 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
598 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
599 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
600 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
601 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
602 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
603 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
606 #ifdef CONFIG_PPC_SMLPAR
607 #define CMO_FREE_HINT_DEFAULT 1
608 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
610 static int __init cmo_free_hint(char *str)
613 parm = strstrip(str);
615 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
616 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
617 cmo_free_hint_flag = 0;
621 cmo_free_hint_flag = 1;
622 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
624 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
630 __setup("cmo_free_hint=", cmo_free_hint);
632 static void pSeries_set_page_state(struct page *page, int order,
636 unsigned long cmo_page_sz, addr;
638 cmo_page_sz = cmo_get_page_size();
639 addr = __pa((unsigned long)page_address(page));
641 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
642 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
643 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
647 void arch_free_page(struct page *page, int order)
649 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
652 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
654 EXPORT_SYMBOL(arch_free_page);
658 #ifdef CONFIG_TRACEPOINTS
659 #ifdef HAVE_JUMP_LABEL
660 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
662 void hcall_tracepoint_regfunc(void)
664 static_key_slow_inc(&hcall_tracepoint_key);
667 void hcall_tracepoint_unregfunc(void)
669 static_key_slow_dec(&hcall_tracepoint_key);
673 * We optimise our hcall path by placing hcall_tracepoint_refcount
674 * directly in the TOC so we can check if the hcall tracepoints are
675 * enabled via a single load.
678 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
679 extern long hcall_tracepoint_refcount;
681 void hcall_tracepoint_regfunc(void)
683 hcall_tracepoint_refcount++;
686 void hcall_tracepoint_unregfunc(void)
688 hcall_tracepoint_refcount--;
693 * Since the tracing code might execute hcalls we need to guard against
694 * recursion. One example of this are spinlocks calling H_YIELD on
695 * shared processor partitions.
697 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
700 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
706 * We cannot call tracepoints inside RCU idle regions which
707 * means we must not trace H_CEDE.
709 if (opcode == H_CEDE)
712 local_irq_save(flags);
714 depth = this_cpu_ptr(&hcall_trace_depth);
721 trace_hcall_entry(opcode, args);
725 local_irq_restore(flags);
728 void __trace_hcall_exit(long opcode, unsigned long retval,
729 unsigned long *retbuf)
734 if (opcode == H_CEDE)
737 local_irq_save(flags);
739 depth = this_cpu_ptr(&hcall_trace_depth);
745 trace_hcall_exit(opcode, retval, retbuf);
750 local_irq_restore(flags);
756 * H_GET_MPP hcall returns info in 7 parms
758 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
761 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
763 rc = plpar_hcall9(H_GET_MPP, retbuf);
765 mpp_data->entitled_mem = retbuf[0];
766 mpp_data->mapped_mem = retbuf[1];
768 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
769 mpp_data->pool_num = retbuf[2] & 0xffff;
771 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
772 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
773 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
775 mpp_data->pool_size = retbuf[4];
776 mpp_data->loan_request = retbuf[5];
777 mpp_data->backing_mem = retbuf[6];
781 EXPORT_SYMBOL(h_get_mpp);
783 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
786 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
788 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
790 mpp_x_data->coalesced_bytes = retbuf[0];
791 mpp_x_data->pool_coalesced_bytes = retbuf[1];
792 mpp_x_data->pool_purr_cycles = retbuf[2];
793 mpp_x_data->pool_spurr_cycles = retbuf[3];