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>
46 #include <asm/fadump.h>
50 /* Flag bits for H_BULK_REMOVE */
51 #define HBR_REQUEST 0x4000000000000000UL
52 #define HBR_RESPONSE 0x8000000000000000UL
53 #define HBR_END 0xc000000000000000UL
54 #define HBR_AVPN 0x0200000000000000UL
55 #define HBR_ANDCOND 0x0100000000000000UL
59 EXPORT_SYMBOL(plpar_hcall);
60 EXPORT_SYMBOL(plpar_hcall9);
61 EXPORT_SYMBOL(plpar_hcall_norets);
63 void vpa_init(int cpu)
65 int hwcpu = get_hard_smp_processor_id(cpu);
68 struct paca_struct *pp;
69 struct dtl_entry *dtl;
72 * The spec says it "may be problematic" if CPU x registers the VPA of
73 * CPU y. We should never do that, but wail if we ever do.
75 WARN_ON(cpu != smp_processor_id());
77 if (cpu_has_feature(CPU_FTR_ALTIVEC))
78 lppaca_of(cpu).vmxregs_in_use = 1;
80 if (cpu_has_feature(CPU_FTR_ARCH_207S))
81 lppaca_of(cpu).ebb_regs_in_use = 1;
83 addr = __pa(&lppaca_of(cpu));
84 ret = register_vpa(hwcpu, addr);
87 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
88 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
92 * PAPR says this feature is SLB-Buffer but firmware never
93 * reports that. All SPLPAR support SLB shadow buffer.
95 addr = __pa(paca[cpu].slb_shadow_ptr);
96 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
97 ret = register_slb_shadow(hwcpu, addr);
99 pr_err("WARNING: SLB shadow buffer registration for "
100 "cpu %d (hw %d) of area %lx failed with %ld\n",
101 cpu, hwcpu, addr, ret);
105 * Register dispatch trace log, if one has been allocated.
108 dtl = pp->dispatch_log;
112 lppaca_of(cpu).dtl_idx = 0;
114 /* hypervisor reads buffer length from this field */
115 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
116 ret = register_dtl(hwcpu, __pa(dtl));
118 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
119 "failed with %ld\n", smp_processor_id(),
121 lppaca_of(cpu).dtl_enable_mask = 2;
125 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
126 unsigned long vpn, unsigned long pa,
127 unsigned long rflags, unsigned long vflags,
128 int psize, int apsize, int ssize)
130 unsigned long lpar_rc;
133 unsigned long hpte_v, hpte_r;
135 if (!(vflags & HPTE_V_BOLTED))
136 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
137 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
138 hpte_group, vpn, pa, rflags, vflags, psize);
140 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
141 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
143 if (!(vflags & HPTE_V_BOLTED))
144 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
146 /* Now fill in the actual HPTE */
147 /* Set CEC cookie to 0 */
149 /* I-cache Invalidate = 0 */
150 /* I-cache synchronize = 0 */
154 /* Make pHyp happy */
155 if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
158 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
159 flags |= H_COALESCE_CAND;
161 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
162 if (unlikely(lpar_rc == H_PTEG_FULL)) {
163 if (!(vflags & HPTE_V_BOLTED))
169 * Since we try and ioremap PHBs we don't own, the pte insert
170 * will fail. However we must catch the failure in hash_page
171 * or we will loop forever, so return -2 in this case.
173 if (unlikely(lpar_rc != H_SUCCESS)) {
174 if (!(vflags & HPTE_V_BOLTED))
175 pr_devel(" lpar err %ld\n", lpar_rc);
178 if (!(vflags & HPTE_V_BOLTED))
179 pr_devel(" -> slot: %lu\n", slot & 7);
181 /* Because of iSeries, we have to pass down the secondary
182 * bucket bit here as well
184 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
187 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
189 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
191 unsigned long slot_offset;
192 unsigned long lpar_rc;
194 unsigned long dummy1, dummy2;
196 /* pick a random slot to start at */
197 slot_offset = mftb() & 0x7;
199 for (i = 0; i < HPTES_PER_GROUP; i++) {
201 /* don't remove a bolted entry */
202 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
203 (0x1UL << 4), &dummy1, &dummy2);
204 if (lpar_rc == H_SUCCESS)
208 * The test for adjunct partition is performed before the
209 * ANDCOND test. H_RESOURCE may be returned, so we need to
210 * check for that as well.
212 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
221 static void pSeries_lpar_hptab_clear(void)
223 unsigned long size_bytes = 1UL << ppc64_pft_size;
224 unsigned long hpte_count = size_bytes >> 4;
232 /* Read in batches of 4,
233 * invalidate only valid entries not in the VRMA
234 * hpte_count will be a multiple of 4
236 for (i = 0; i < hpte_count; i += 4) {
237 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
238 if (lpar_rc != H_SUCCESS)
240 for (j = 0; j < 4; j++){
241 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
244 if (ptes[j].pteh & HPTE_V_VALID)
245 plpar_pte_remove_raw(0, i + j, 0,
246 &(ptes[j].pteh), &(ptes[j].ptel));
250 #ifdef __LITTLE_ENDIAN__
252 * Reset exceptions to big endian.
254 * FIXME this is a hack for kexec, we need to reset the exception
255 * endian before starting the new kernel and this is a convenient place
258 * This is also called on boot when a fadump happens. In that case we
259 * must not change the exception endian mode.
261 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) {
264 rc = pseries_big_endian_exceptions();
266 * At this point it is unlikely panic() will get anything
267 * out to the user, but at least this will stop us from
268 * continuing on further and creating an even more
269 * difficult to debug situation.
272 panic("Could not enable big endian exceptions");
278 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
279 * the low 3 bits of flags happen to line up. So no transform is needed.
280 * We can probably optimize here and assume the high bits of newpp are
281 * already zero. For now I am paranoid.
283 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
286 int psize, int apsize,
287 int ssize, unsigned long inv_flags)
289 unsigned long lpar_rc;
290 unsigned long flags = (newpp & 7) | H_AVPN;
291 unsigned long want_v;
293 want_v = hpte_encode_avpn(vpn, psize, ssize);
295 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
296 want_v, slot, flags, psize);
298 lpar_rc = plpar_pte_protect(flags, slot, want_v);
300 if (lpar_rc == H_NOT_FOUND) {
301 pr_devel("not found !\n");
307 BUG_ON(lpar_rc != H_SUCCESS);
312 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
314 unsigned long dword0;
315 unsigned long lpar_rc;
316 unsigned long dummy_word1;
319 /* Read 1 pte at a time */
320 /* Do not need RPN to logical page translation */
321 /* No cross CEC PFT access */
324 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
326 BUG_ON(lpar_rc != H_SUCCESS);
331 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
336 unsigned long want_v, hpte_v;
338 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
339 want_v = hpte_encode_avpn(vpn, psize, ssize);
341 /* Bolted entries are always in the primary group */
342 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
343 for (i = 0; i < HPTES_PER_GROUP; i++) {
344 hpte_v = pSeries_lpar_hpte_getword0(slot);
346 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
355 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
357 int psize, int ssize)
360 unsigned long lpar_rc, slot, vsid, flags;
362 vsid = get_kernel_vsid(ea, ssize);
363 vpn = hpt_vpn(ea, vsid, ssize);
365 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
369 lpar_rc = plpar_pte_protect(flags, slot, 0);
371 BUG_ON(lpar_rc != H_SUCCESS);
374 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
375 int psize, int apsize,
376 int ssize, int local)
378 unsigned long want_v;
379 unsigned long lpar_rc;
380 unsigned long dummy1, dummy2;
382 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
383 slot, vpn, psize, local);
385 want_v = hpte_encode_avpn(vpn, psize, ssize);
386 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
387 if (lpar_rc == H_NOT_FOUND)
390 BUG_ON(lpar_rc != H_SUCCESS);
394 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
395 * to make sure that we avoid bouncing the hypervisor tlbie lock.
397 #define PPC64_HUGE_HPTE_BATCH 12
399 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
400 unsigned long *vpn, int count,
401 int psize, int ssize)
403 unsigned long param[8];
404 int i = 0, pix = 0, rc;
405 unsigned long flags = 0;
406 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
409 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
411 for (i = 0; i < count; i++) {
413 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
414 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
417 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
418 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
421 rc = plpar_hcall9(H_BULK_REMOVE, param,
422 param[0], param[1], param[2],
423 param[3], param[4], param[5],
425 BUG_ON(rc != H_SUCCESS);
431 param[pix] = HBR_END;
432 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
433 param[2], param[3], param[4], param[5],
435 BUG_ON(rc != H_SUCCESS);
439 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
442 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
444 unsigned char *hpte_slot_array,
445 int psize, int ssize, int local)
448 unsigned long s_addr = addr;
449 unsigned int max_hpte_count, valid;
450 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
451 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
452 unsigned long shift, hidx, vpn = 0, hash, slot;
454 shift = mmu_psize_defs[psize].shift;
455 max_hpte_count = 1U << (PMD_SHIFT - shift);
457 for (i = 0; i < max_hpte_count; i++) {
458 valid = hpte_valid(hpte_slot_array, i);
461 hidx = hpte_hash_index(hpte_slot_array, i);
464 addr = s_addr + (i * (1ul << shift));
465 vpn = hpt_vpn(addr, vsid, ssize);
466 hash = hpt_hash(vpn, shift, ssize);
467 if (hidx & _PTEIDX_SECONDARY)
470 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
471 slot += hidx & _PTEIDX_GROUP_IX;
473 slot_array[index] = slot;
474 vpn_array[index] = vpn;
475 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
477 * Now do a bluk invalidate
479 __pSeries_lpar_hugepage_invalidate(slot_array,
481 PPC64_HUGE_HPTE_BATCH,
488 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
489 index, psize, ssize);
492 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
493 int psize, int ssize)
496 unsigned long slot, vsid;
498 vsid = get_kernel_vsid(ea, ssize);
499 vpn = hpt_vpn(ea, vsid, ssize);
501 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
504 * lpar doesn't use the passed actual page size
506 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
510 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
513 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
516 unsigned long i, pix, rc;
517 unsigned long flags = 0;
518 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
519 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
520 unsigned long param[9];
521 unsigned long hash, index, shift, hidx, slot;
526 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
528 psize = batch->psize;
529 ssize = batch->ssize;
531 for (i = 0; i < number; i++) {
534 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
535 hash = hpt_hash(vpn, shift, ssize);
536 hidx = __rpte_to_hidx(pte, index);
537 if (hidx & _PTEIDX_SECONDARY)
539 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
540 slot += hidx & _PTEIDX_GROUP_IX;
541 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
543 * lpar doesn't use the passed actual page size
545 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
548 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
549 param[pix+1] = hpte_encode_avpn(vpn, psize,
553 rc = plpar_hcall9(H_BULK_REMOVE, param,
554 param[0], param[1], param[2],
555 param[3], param[4], param[5],
557 BUG_ON(rc != H_SUCCESS);
561 } pte_iterate_hashed_end();
564 param[pix] = HBR_END;
565 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
566 param[2], param[3], param[4], param[5],
568 BUG_ON(rc != H_SUCCESS);
572 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
575 static int __init disable_bulk_remove(char *str)
577 if (strcmp(str, "off") == 0 &&
578 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
579 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
580 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
585 __setup("bulk_remove=", disable_bulk_remove);
587 void __init hpte_init_lpar(void)
589 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
590 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
591 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
592 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
593 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
594 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
595 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
596 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
597 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
600 #ifdef CONFIG_PPC_SMLPAR
601 #define CMO_FREE_HINT_DEFAULT 1
602 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
604 static int __init cmo_free_hint(char *str)
607 parm = strstrip(str);
609 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
610 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
611 cmo_free_hint_flag = 0;
615 cmo_free_hint_flag = 1;
616 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
618 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
624 __setup("cmo_free_hint=", cmo_free_hint);
626 static void pSeries_set_page_state(struct page *page, int order,
630 unsigned long cmo_page_sz, addr;
632 cmo_page_sz = cmo_get_page_size();
633 addr = __pa((unsigned long)page_address(page));
635 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
636 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
637 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
641 void arch_free_page(struct page *page, int order)
643 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
646 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
648 EXPORT_SYMBOL(arch_free_page);
652 #ifdef CONFIG_TRACEPOINTS
653 #ifdef HAVE_JUMP_LABEL
654 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
656 void hcall_tracepoint_regfunc(void)
658 static_key_slow_inc(&hcall_tracepoint_key);
661 void hcall_tracepoint_unregfunc(void)
663 static_key_slow_dec(&hcall_tracepoint_key);
667 * We optimise our hcall path by placing hcall_tracepoint_refcount
668 * directly in the TOC so we can check if the hcall tracepoints are
669 * enabled via a single load.
672 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
673 extern long hcall_tracepoint_refcount;
675 void hcall_tracepoint_regfunc(void)
677 hcall_tracepoint_refcount++;
680 void hcall_tracepoint_unregfunc(void)
682 hcall_tracepoint_refcount--;
687 * Since the tracing code might execute hcalls we need to guard against
688 * recursion. One example of this are spinlocks calling H_YIELD on
689 * shared processor partitions.
691 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
694 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
700 * We cannot call tracepoints inside RCU idle regions which
701 * means we must not trace H_CEDE.
703 if (opcode == H_CEDE)
706 local_irq_save(flags);
708 depth = this_cpu_ptr(&hcall_trace_depth);
715 trace_hcall_entry(opcode, args);
719 local_irq_restore(flags);
722 void __trace_hcall_exit(long opcode, unsigned long retval,
723 unsigned long *retbuf)
728 if (opcode == H_CEDE)
731 local_irq_save(flags);
733 depth = this_cpu_ptr(&hcall_trace_depth);
739 trace_hcall_exit(opcode, retval, retbuf);
744 local_irq_restore(flags);
750 * H_GET_MPP hcall returns info in 7 parms
752 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
755 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
757 rc = plpar_hcall9(H_GET_MPP, retbuf);
759 mpp_data->entitled_mem = retbuf[0];
760 mpp_data->mapped_mem = retbuf[1];
762 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
763 mpp_data->pool_num = retbuf[2] & 0xffff;
765 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
766 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
767 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
769 mpp_data->pool_size = retbuf[4];
770 mpp_data->loan_request = retbuf[5];
771 mpp_data->backing_mem = retbuf[6];
775 EXPORT_SYMBOL(h_get_mpp);
777 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
780 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
782 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
784 mpp_x_data->coalesced_bytes = retbuf[0];
785 mpp_x_data->pool_coalesced_bytes = retbuf[1];
786 mpp_x_data->pool_purr_cycles = retbuf[2];
787 mpp_x_data->pool_spurr_cycles = retbuf[3];