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>
48 #include <asm/asm-prototypes.h>
52 /* Flag bits for H_BULK_REMOVE */
53 #define HBR_REQUEST 0x4000000000000000UL
54 #define HBR_RESPONSE 0x8000000000000000UL
55 #define HBR_END 0xc000000000000000UL
56 #define HBR_AVPN 0x0200000000000000UL
57 #define HBR_ANDCOND 0x0100000000000000UL
61 EXPORT_SYMBOL(plpar_hcall);
62 EXPORT_SYMBOL(plpar_hcall9);
63 EXPORT_SYMBOL(plpar_hcall_norets);
65 void vpa_init(int cpu)
67 int hwcpu = get_hard_smp_processor_id(cpu);
70 struct paca_struct *pp;
71 struct dtl_entry *dtl;
74 * The spec says it "may be problematic" if CPU x registers the VPA of
75 * CPU y. We should never do that, but wail if we ever do.
77 WARN_ON(cpu != smp_processor_id());
79 if (cpu_has_feature(CPU_FTR_ALTIVEC))
80 lppaca_of(cpu).vmxregs_in_use = 1;
82 if (cpu_has_feature(CPU_FTR_ARCH_207S))
83 lppaca_of(cpu).ebb_regs_in_use = 1;
85 addr = __pa(&lppaca_of(cpu));
86 ret = register_vpa(hwcpu, addr);
89 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
90 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
94 #ifdef CONFIG_PPC_STD_MMU_64
96 * PAPR says this feature is SLB-Buffer but firmware never
97 * reports that. All SPLPAR support SLB shadow buffer.
99 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
100 addr = __pa(paca[cpu].slb_shadow_ptr);
101 ret = register_slb_shadow(hwcpu, addr);
103 pr_err("WARNING: SLB shadow buffer registration for "
104 "cpu %d (hw %d) of area %lx failed with %ld\n",
105 cpu, hwcpu, addr, ret);
107 #endif /* CONFIG_PPC_STD_MMU_64 */
110 * Register dispatch trace log, if one has been allocated.
113 dtl = pp->dispatch_log;
117 lppaca_of(cpu).dtl_idx = 0;
119 /* hypervisor reads buffer length from this field */
120 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
121 ret = register_dtl(hwcpu, __pa(dtl));
123 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
124 "failed with %ld\n", smp_processor_id(),
126 lppaca_of(cpu).dtl_enable_mask = 2;
130 #ifdef CONFIG_PPC_STD_MMU_64
132 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
133 unsigned long vpn, unsigned long pa,
134 unsigned long rflags, unsigned long vflags,
135 int psize, int apsize, int ssize)
137 unsigned long lpar_rc;
140 unsigned long hpte_v, hpte_r;
142 if (!(vflags & HPTE_V_BOLTED))
143 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
144 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
145 hpte_group, vpn, pa, rflags, vflags, psize);
147 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
148 hpte_r = hpte_encode_r(pa, psize, apsize, ssize) | rflags;
150 if (!(vflags & HPTE_V_BOLTED))
151 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
153 /* Now fill in the actual HPTE */
154 /* Set CEC cookie to 0 */
156 /* I-cache Invalidate = 0 */
157 /* I-cache synchronize = 0 */
161 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
162 flags |= H_COALESCE_CAND;
164 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
165 if (unlikely(lpar_rc == H_PTEG_FULL)) {
166 if (!(vflags & HPTE_V_BOLTED))
172 * Since we try and ioremap PHBs we don't own, the pte insert
173 * will fail. However we must catch the failure in hash_page
174 * or we will loop forever, so return -2 in this case.
176 if (unlikely(lpar_rc != H_SUCCESS)) {
177 if (!(vflags & HPTE_V_BOLTED))
178 pr_devel(" lpar err %ld\n", lpar_rc);
181 if (!(vflags & HPTE_V_BOLTED))
182 pr_devel(" -> slot: %lu\n", slot & 7);
184 /* Because of iSeries, we have to pass down the secondary
185 * bucket bit here as well
187 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
190 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
192 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
194 unsigned long slot_offset;
195 unsigned long lpar_rc;
197 unsigned long dummy1, dummy2;
199 /* pick a random slot to start at */
200 slot_offset = mftb() & 0x7;
202 for (i = 0; i < HPTES_PER_GROUP; i++) {
204 /* don't remove a bolted entry */
205 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
206 (0x1UL << 4), &dummy1, &dummy2);
207 if (lpar_rc == H_SUCCESS)
211 * The test for adjunct partition is performed before the
212 * ANDCOND test. H_RESOURCE may be returned, so we need to
213 * check for that as well.
215 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
224 static void pSeries_lpar_hptab_clear(void)
226 unsigned long size_bytes = 1UL << ppc64_pft_size;
227 unsigned long hpte_count = size_bytes >> 4;
235 /* Read in batches of 4,
236 * invalidate only valid entries not in the VRMA
237 * hpte_count will be a multiple of 4
239 for (i = 0; i < hpte_count; i += 4) {
240 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
241 if (lpar_rc != H_SUCCESS)
243 for (j = 0; j < 4; j++){
244 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
247 if (ptes[j].pteh & HPTE_V_VALID)
248 plpar_pte_remove_raw(0, i + j, 0,
249 &(ptes[j].pteh), &(ptes[j].ptel));
253 #ifdef __LITTLE_ENDIAN__
255 * Reset exceptions to big endian.
257 * FIXME this is a hack for kexec, we need to reset the exception
258 * endian before starting the new kernel and this is a convenient place
261 * This is also called on boot when a fadump happens. In that case we
262 * must not change the exception endian mode.
264 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) {
267 rc = pseries_big_endian_exceptions();
269 * At this point it is unlikely panic() will get anything
270 * out to the user, but at least this will stop us from
271 * continuing on further and creating an even more
272 * difficult to debug situation.
274 * There is a known problem when kdump'ing, if cpus are offline
275 * the above call will fail. Rather than panicking again, keep
276 * going and hope the kdump kernel is also little endian, which
279 if (rc && !kdump_in_progress())
280 panic("Could not enable big endian exceptions");
286 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
287 * the low 3 bits of flags happen to line up. So no transform is needed.
288 * We can probably optimize here and assume the high bits of newpp are
289 * already zero. For now I am paranoid.
291 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
294 int psize, int apsize,
295 int ssize, unsigned long inv_flags)
297 unsigned long lpar_rc;
298 unsigned long flags = (newpp & 7) | H_AVPN;
299 unsigned long want_v;
301 want_v = hpte_encode_avpn(vpn, psize, ssize);
303 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
304 want_v, slot, flags, psize);
306 lpar_rc = plpar_pte_protect(flags, slot, want_v);
308 if (lpar_rc == H_NOT_FOUND) {
309 pr_devel("not found !\n");
315 BUG_ON(lpar_rc != H_SUCCESS);
320 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
329 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
331 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
332 if (lpar_rc != H_SUCCESS)
335 for (j = 0; j < 4; j++) {
336 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
337 (ptes[j].pteh & HPTE_V_VALID))
345 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
349 unsigned long want_v;
350 unsigned long hpte_group;
352 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
353 want_v = hpte_encode_avpn(vpn, psize, ssize);
355 /* Bolted entries are always in the primary group */
356 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
357 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
360 return hpte_group + slot;
363 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
365 int psize, int ssize)
368 unsigned long lpar_rc, slot, vsid, flags;
370 vsid = get_kernel_vsid(ea, ssize);
371 vpn = hpt_vpn(ea, vsid, ssize);
373 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
377 lpar_rc = plpar_pte_protect(flags, slot, 0);
379 BUG_ON(lpar_rc != H_SUCCESS);
382 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
383 int psize, int apsize,
384 int ssize, int local)
386 unsigned long want_v;
387 unsigned long lpar_rc;
388 unsigned long dummy1, dummy2;
390 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
391 slot, vpn, psize, local);
393 want_v = hpte_encode_avpn(vpn, psize, ssize);
394 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
395 if (lpar_rc == H_NOT_FOUND)
398 BUG_ON(lpar_rc != H_SUCCESS);
401 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
403 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
404 * to make sure that we avoid bouncing the hypervisor tlbie lock.
406 #define PPC64_HUGE_HPTE_BATCH 12
408 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
409 unsigned long *vpn, int count,
410 int psize, int ssize)
412 unsigned long param[8];
413 int i = 0, pix = 0, rc;
414 unsigned long flags = 0;
415 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
418 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
420 for (i = 0; i < count; i++) {
422 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
423 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
426 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
427 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
430 rc = plpar_hcall9(H_BULK_REMOVE, param,
431 param[0], param[1], param[2],
432 param[3], param[4], param[5],
434 BUG_ON(rc != H_SUCCESS);
440 param[pix] = HBR_END;
441 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
442 param[2], param[3], param[4], param[5],
444 BUG_ON(rc != H_SUCCESS);
448 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
451 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
453 unsigned char *hpte_slot_array,
454 int psize, int ssize, int local)
457 unsigned long s_addr = addr;
458 unsigned int max_hpte_count, valid;
459 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
460 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
461 unsigned long shift, hidx, vpn = 0, hash, slot;
463 shift = mmu_psize_defs[psize].shift;
464 max_hpte_count = 1U << (PMD_SHIFT - shift);
466 for (i = 0; i < max_hpte_count; i++) {
467 valid = hpte_valid(hpte_slot_array, i);
470 hidx = hpte_hash_index(hpte_slot_array, i);
473 addr = s_addr + (i * (1ul << shift));
474 vpn = hpt_vpn(addr, vsid, ssize);
475 hash = hpt_hash(vpn, shift, ssize);
476 if (hidx & _PTEIDX_SECONDARY)
479 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
480 slot += hidx & _PTEIDX_GROUP_IX;
482 slot_array[index] = slot;
483 vpn_array[index] = vpn;
484 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
486 * Now do a bluk invalidate
488 __pSeries_lpar_hugepage_invalidate(slot_array,
490 PPC64_HUGE_HPTE_BATCH,
497 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
498 index, psize, ssize);
501 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
503 unsigned char *hpte_slot_array,
504 int psize, int ssize, int local)
506 WARN(1, "%s called without THP support\n", __func__);
510 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
511 int psize, int ssize)
514 unsigned long slot, vsid;
516 vsid = get_kernel_vsid(ea, ssize);
517 vpn = hpt_vpn(ea, vsid, ssize);
519 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
524 * lpar doesn't use the passed actual page size
526 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
531 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
534 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
537 unsigned long i, pix, rc;
538 unsigned long flags = 0;
539 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
540 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
541 unsigned long param[9];
542 unsigned long hash, index, shift, hidx, slot;
547 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
549 psize = batch->psize;
550 ssize = batch->ssize;
552 for (i = 0; i < number; i++) {
555 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
556 hash = hpt_hash(vpn, shift, ssize);
557 hidx = __rpte_to_hidx(pte, index);
558 if (hidx & _PTEIDX_SECONDARY)
560 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
561 slot += hidx & _PTEIDX_GROUP_IX;
562 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
564 * lpar doesn't use the passed actual page size
566 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
569 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
570 param[pix+1] = hpte_encode_avpn(vpn, psize,
574 rc = plpar_hcall9(H_BULK_REMOVE, param,
575 param[0], param[1], param[2],
576 param[3], param[4], param[5],
578 BUG_ON(rc != H_SUCCESS);
582 } pte_iterate_hashed_end();
585 param[pix] = HBR_END;
586 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
587 param[2], param[3], param[4], param[5],
589 BUG_ON(rc != H_SUCCESS);
593 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
596 static int __init disable_bulk_remove(char *str)
598 if (strcmp(str, "off") == 0 &&
599 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
600 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
601 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
606 __setup("bulk_remove=", disable_bulk_remove);
608 void __init hpte_init_lpar(void)
610 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
611 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
612 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
613 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
614 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
615 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
616 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
617 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
618 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
621 #ifdef CONFIG_PPC_SMLPAR
622 #define CMO_FREE_HINT_DEFAULT 1
623 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
625 static int __init cmo_free_hint(char *str)
628 parm = strstrip(str);
630 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
631 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
632 cmo_free_hint_flag = 0;
636 cmo_free_hint_flag = 1;
637 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
639 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
645 __setup("cmo_free_hint=", cmo_free_hint);
647 static void pSeries_set_page_state(struct page *page, int order,
651 unsigned long cmo_page_sz, addr;
653 cmo_page_sz = cmo_get_page_size();
654 addr = __pa((unsigned long)page_address(page));
656 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
657 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
658 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
662 void arch_free_page(struct page *page, int order)
666 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
669 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
671 EXPORT_SYMBOL(arch_free_page);
673 #endif /* CONFIG_PPC_SMLPAR */
674 #endif /* CONFIG_PPC_STD_MMU_64 */
676 #ifdef CONFIG_TRACEPOINTS
677 #ifdef HAVE_JUMP_LABEL
678 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
680 void hcall_tracepoint_regfunc(void)
682 static_key_slow_inc(&hcall_tracepoint_key);
685 void hcall_tracepoint_unregfunc(void)
687 static_key_slow_dec(&hcall_tracepoint_key);
691 * We optimise our hcall path by placing hcall_tracepoint_refcount
692 * directly in the TOC so we can check if the hcall tracepoints are
693 * enabled via a single load.
696 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
697 extern long hcall_tracepoint_refcount;
699 void hcall_tracepoint_regfunc(void)
701 hcall_tracepoint_refcount++;
704 void hcall_tracepoint_unregfunc(void)
706 hcall_tracepoint_refcount--;
711 * Since the tracing code might execute hcalls we need to guard against
712 * recursion. One example of this are spinlocks calling H_YIELD on
713 * shared processor partitions.
715 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
718 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
724 * We cannot call tracepoints inside RCU idle regions which
725 * means we must not trace H_CEDE.
727 if (opcode == H_CEDE)
730 local_irq_save(flags);
732 depth = this_cpu_ptr(&hcall_trace_depth);
739 trace_hcall_entry(opcode, args);
743 local_irq_restore(flags);
746 void __trace_hcall_exit(long opcode, unsigned long retval,
747 unsigned long *retbuf)
752 if (opcode == H_CEDE)
755 local_irq_save(flags);
757 depth = this_cpu_ptr(&hcall_trace_depth);
763 trace_hcall_exit(opcode, retval, retbuf);
768 local_irq_restore(flags);
774 * H_GET_MPP hcall returns info in 7 parms
776 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
779 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
781 rc = plpar_hcall9(H_GET_MPP, retbuf);
783 mpp_data->entitled_mem = retbuf[0];
784 mpp_data->mapped_mem = retbuf[1];
786 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
787 mpp_data->pool_num = retbuf[2] & 0xffff;
789 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
790 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
791 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
793 mpp_data->pool_size = retbuf[4];
794 mpp_data->loan_request = retbuf[5];
795 mpp_data->backing_mem = retbuf[6];
799 EXPORT_SYMBOL(h_get_mpp);
801 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
804 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
806 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
808 mpp_x_data->coalesced_bytes = retbuf[0];
809 mpp_x_data->pool_coalesced_bytes = retbuf[1];
810 mpp_x_data->pool_purr_cycles = retbuf[2];
811 mpp_x_data->pool_spurr_cycles = retbuf[3];