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! */
24 #define pr_fmt(fmt) "lpar: " fmt
26 #include <linux/kernel.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/console.h>
29 #include <linux/export.h>
30 #include <linux/jump_label.h>
31 #include <linux/delay.h>
32 #include <linux/stop_machine.h>
33 #include <asm/processor.h>
36 #include <asm/pgtable.h>
37 #include <asm/machdep.h>
38 #include <asm/mmu_context.h>
39 #include <asm/iommu.h>
42 #include <asm/cputable.h>
45 #include <asm/trace.h>
46 #include <asm/firmware.h>
47 #include <asm/plpar_wrappers.h>
48 #include <asm/kexec.h>
49 #include <asm/fadump.h>
50 #include <asm/asm-prototypes.h>
51 #include <asm/debugfs.h>
55 /* Flag bits for H_BULK_REMOVE */
56 #define HBR_REQUEST 0x4000000000000000UL
57 #define HBR_RESPONSE 0x8000000000000000UL
58 #define HBR_END 0xc000000000000000UL
59 #define HBR_AVPN 0x0200000000000000UL
60 #define HBR_ANDCOND 0x0100000000000000UL
64 EXPORT_SYMBOL(plpar_hcall);
65 EXPORT_SYMBOL(plpar_hcall9);
66 EXPORT_SYMBOL(plpar_hcall_norets);
68 void vpa_init(int cpu)
70 int hwcpu = get_hard_smp_processor_id(cpu);
73 struct paca_struct *pp;
74 struct dtl_entry *dtl;
77 * The spec says it "may be problematic" if CPU x registers the VPA of
78 * CPU y. We should never do that, but wail if we ever do.
80 WARN_ON(cpu != smp_processor_id());
82 if (cpu_has_feature(CPU_FTR_ALTIVEC))
83 lppaca_of(cpu).vmxregs_in_use = 1;
85 if (cpu_has_feature(CPU_FTR_ARCH_207S))
86 lppaca_of(cpu).ebb_regs_in_use = 1;
88 addr = __pa(&lppaca_of(cpu));
89 ret = register_vpa(hwcpu, addr);
92 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
93 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
97 #ifdef CONFIG_PPC_BOOK3S_64
99 * PAPR says this feature is SLB-Buffer but firmware never
100 * reports that. All SPLPAR support SLB shadow buffer.
102 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
103 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
104 ret = register_slb_shadow(hwcpu, addr);
106 pr_err("WARNING: SLB shadow buffer registration for "
107 "cpu %d (hw %d) of area %lx failed with %ld\n",
108 cpu, hwcpu, addr, ret);
110 #endif /* CONFIG_PPC_BOOK3S_64 */
113 * Register dispatch trace log, if one has been allocated.
116 dtl = pp->dispatch_log;
120 lppaca_of(cpu).dtl_idx = 0;
122 /* hypervisor reads buffer length from this field */
123 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
124 ret = register_dtl(hwcpu, __pa(dtl));
126 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
127 "failed with %ld\n", smp_processor_id(),
129 lppaca_of(cpu).dtl_enable_mask = 2;
133 #ifdef CONFIG_PPC_BOOK3S_64
135 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
136 unsigned long vpn, unsigned long pa,
137 unsigned long rflags, unsigned long vflags,
138 int psize, int apsize, int ssize)
140 unsigned long lpar_rc;
143 unsigned long hpte_v, hpte_r;
145 if (!(vflags & HPTE_V_BOLTED))
146 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
147 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
148 hpte_group, vpn, pa, rflags, vflags, psize);
150 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
151 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
153 if (!(vflags & HPTE_V_BOLTED))
154 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
156 /* Now fill in the actual HPTE */
157 /* Set CEC cookie to 0 */
159 /* I-cache Invalidate = 0 */
160 /* I-cache synchronize = 0 */
164 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
165 flags |= H_COALESCE_CAND;
167 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
168 if (unlikely(lpar_rc == H_PTEG_FULL)) {
169 pr_devel("Hash table group is full\n");
174 * Since we try and ioremap PHBs we don't own, the pte insert
175 * will fail. However we must catch the failure in hash_page
176 * or we will loop forever, so return -2 in this case.
178 if (unlikely(lpar_rc != H_SUCCESS)) {
179 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
182 if (!(vflags & HPTE_V_BOLTED))
183 pr_devel(" -> slot: %lu\n", slot & 7);
185 /* Because of iSeries, we have to pass down the secondary
186 * bucket bit here as well
188 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
191 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
193 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
195 unsigned long slot_offset;
196 unsigned long lpar_rc;
198 unsigned long dummy1, dummy2;
200 /* pick a random slot to start at */
201 slot_offset = mftb() & 0x7;
203 for (i = 0; i < HPTES_PER_GROUP; i++) {
205 /* don't remove a bolted entry */
206 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
207 (0x1UL << 4), &dummy1, &dummy2);
208 if (lpar_rc == H_SUCCESS)
212 * The test for adjunct partition is performed before the
213 * ANDCOND test. H_RESOURCE may be returned, so we need to
214 * check for that as well.
216 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
225 static void manual_hpte_clear_all(void)
227 unsigned long size_bytes = 1UL << ppc64_pft_size;
228 unsigned long hpte_count = size_bytes >> 4;
236 /* Read in batches of 4,
237 * invalidate only valid entries not in the VRMA
238 * hpte_count will be a multiple of 4
240 for (i = 0; i < hpte_count; i += 4) {
241 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
242 if (lpar_rc != H_SUCCESS) {
243 pr_info("Failed to read hash page table at %ld err %ld\n",
247 for (j = 0; j < 4; j++){
248 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
251 if (ptes[j].pteh & HPTE_V_VALID)
252 plpar_pte_remove_raw(0, i + j, 0,
253 &(ptes[j].pteh), &(ptes[j].ptel));
258 static int hcall_hpte_clear_all(void)
263 rc = plpar_hcall_norets(H_CLEAR_HPT);
264 } while (rc == H_CONTINUE);
269 static void pseries_hpte_clear_all(void)
273 rc = hcall_hpte_clear_all();
275 manual_hpte_clear_all();
277 #ifdef __LITTLE_ENDIAN__
279 * Reset exceptions to big endian.
281 * FIXME this is a hack for kexec, we need to reset the exception
282 * endian before starting the new kernel and this is a convenient place
285 * This is also called on boot when a fadump happens. In that case we
286 * must not change the exception endian mode.
288 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
289 pseries_big_endian_exceptions();
294 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
295 * the low 3 bits of flags happen to line up. So no transform is needed.
296 * We can probably optimize here and assume the high bits of newpp are
297 * already zero. For now I am paranoid.
299 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
302 int psize, int apsize,
303 int ssize, unsigned long inv_flags)
305 unsigned long lpar_rc;
307 unsigned long want_v;
309 want_v = hpte_encode_avpn(vpn, psize, ssize);
311 flags = (newpp & 7) | H_AVPN;
312 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
313 /* Move pp0 into bit 8 (IBM 55) */
314 flags |= (newpp & HPTE_R_PP0) >> 55;
316 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
317 want_v, slot, flags, psize);
319 lpar_rc = plpar_pte_protect(flags, slot, want_v);
321 if (lpar_rc == H_NOT_FOUND) {
322 pr_devel("not found !\n");
328 BUG_ON(lpar_rc != H_SUCCESS);
333 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
342 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
344 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
345 if (lpar_rc != H_SUCCESS) {
346 pr_info("Failed to read hash page table at %ld err %ld\n",
347 hpte_group, lpar_rc);
351 for (j = 0; j < 4; j++) {
352 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
353 (ptes[j].pteh & HPTE_V_VALID))
361 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
365 unsigned long want_v;
366 unsigned long hpte_group;
368 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
369 want_v = hpte_encode_avpn(vpn, psize, ssize);
371 /* Bolted entries are always in the primary group */
372 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
373 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
376 return hpte_group + slot;
379 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
381 int psize, int ssize)
384 unsigned long lpar_rc, slot, vsid, flags;
386 vsid = get_kernel_vsid(ea, ssize);
387 vpn = hpt_vpn(ea, vsid, ssize);
389 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
393 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
394 /* Move pp0 into bit 8 (IBM 55) */
395 flags |= (newpp & HPTE_R_PP0) >> 55;
397 lpar_rc = plpar_pte_protect(flags, slot, 0);
399 BUG_ON(lpar_rc != H_SUCCESS);
402 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
403 int psize, int apsize,
404 int ssize, int local)
406 unsigned long want_v;
407 unsigned long lpar_rc;
408 unsigned long dummy1, dummy2;
410 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
411 slot, vpn, psize, local);
413 want_v = hpte_encode_avpn(vpn, psize, ssize);
414 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
415 if (lpar_rc == H_NOT_FOUND)
418 BUG_ON(lpar_rc != H_SUCCESS);
423 * As defined in the PAPR's section 14.5.4.1.8
424 * The control mask doesn't include the returned reference and change bit from
427 #define HBLKR_AVPN 0x0100000000000000UL
428 #define HBLKR_CTRL_MASK 0xf800000000000000UL
429 #define HBLKR_CTRL_SUCCESS 0x8000000000000000UL
430 #define HBLKR_CTRL_ERRNOTFOUND 0x8800000000000000UL
431 #define HBLKR_CTRL_ERRBUSY 0xa000000000000000UL
434 * H_BLOCK_REMOVE caller.
435 * @idx should point to the latest @param entry set with a PTEX.
436 * If PTE cannot be processed because another CPUs has already locked that
437 * group, those entries are put back in @param starting at index 1.
438 * If entries has to be retried and @retry_busy is set to true, these entries
439 * are retried until success. If @retry_busy is set to false, the returned
440 * is the number of entries yet to process.
442 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
445 unsigned long i, rc, new_idx;
446 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
449 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
454 if (idx > PLPAR_HCALL9_BUFSIZE) {
455 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
456 idx = PLPAR_HCALL9_BUFSIZE;
457 } else if (idx < PLPAR_HCALL9_BUFSIZE)
458 param[idx] = HBR_END;
460 rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
462 param[1], param[2], param[3], param[4], /* TS0-7 */
463 param[5], param[6], param[7], param[8]);
467 BUG_ON(rc != H_PARTIAL);
469 /* Check that the unprocessed entries were 'not found' or 'busy' */
470 for (i = 0; i < idx-1; i++) {
471 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
473 if (ctrl == HBLKR_CTRL_ERRBUSY) {
474 param[++new_idx] = param[i+1];
478 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
479 && ctrl != HBLKR_CTRL_ERRNOTFOUND);
483 * If there were entries found busy, retry these entries if requested,
484 * of if all the entries have to be retried.
486 if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
494 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
496 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
497 * to make sure that we avoid bouncing the hypervisor tlbie lock.
499 #define PPC64_HUGE_HPTE_BATCH 12
501 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
502 int count, int psize, int ssize)
504 unsigned long param[PLPAR_HCALL9_BUFSIZE];
505 unsigned long shift, current_vpgb, vpgb;
508 shift = mmu_psize_defs[psize].shift;
510 for (i = 0; i < count; i++) {
512 * Shifting 3 bits more on the right to get a
513 * 8 pages aligned virtual addresse.
515 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
516 if (!pix || vpgb != current_vpgb) {
518 * Need to start a new 8 pages block, flush
519 * the current one if needed.
522 (void)call_block_remove(pix, param, true);
524 param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
528 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
529 if (pix == PLPAR_HCALL9_BUFSIZE) {
530 pix = call_block_remove(pix, param, false);
532 * pix = 0 means that all the entries were
533 * removed, we can start a new block.
534 * Otherwise, this means that there are entries
535 * to retry, and pix points to latest one, so
536 * we should increment it and try to continue
544 (void)call_block_remove(pix, param, true);
547 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
548 int count, int psize, int ssize)
550 unsigned long param[PLPAR_HCALL9_BUFSIZE];
551 int i = 0, pix = 0, rc;
553 for (i = 0; i < count; i++) {
555 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
556 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
559 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
560 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
563 rc = plpar_hcall9(H_BULK_REMOVE, param,
564 param[0], param[1], param[2],
565 param[3], param[4], param[5],
567 BUG_ON(rc != H_SUCCESS);
573 param[pix] = HBR_END;
574 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
575 param[2], param[3], param[4], param[5],
577 BUG_ON(rc != H_SUCCESS);
581 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
583 int count, int psize,
586 unsigned long flags = 0;
587 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
590 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
592 if (firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
593 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
595 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
598 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
601 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
603 unsigned char *hpte_slot_array,
604 int psize, int ssize, int local)
607 unsigned long s_addr = addr;
608 unsigned int max_hpte_count, valid;
609 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
610 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
611 unsigned long shift, hidx, vpn = 0, hash, slot;
613 shift = mmu_psize_defs[psize].shift;
614 max_hpte_count = 1U << (PMD_SHIFT - shift);
616 for (i = 0; i < max_hpte_count; i++) {
617 valid = hpte_valid(hpte_slot_array, i);
620 hidx = hpte_hash_index(hpte_slot_array, i);
623 addr = s_addr + (i * (1ul << shift));
624 vpn = hpt_vpn(addr, vsid, ssize);
625 hash = hpt_hash(vpn, shift, ssize);
626 if (hidx & _PTEIDX_SECONDARY)
629 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
630 slot += hidx & _PTEIDX_GROUP_IX;
632 slot_array[index] = slot;
633 vpn_array[index] = vpn;
634 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
636 * Now do a bluk invalidate
638 __pSeries_lpar_hugepage_invalidate(slot_array,
640 PPC64_HUGE_HPTE_BATCH,
647 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
648 index, psize, ssize);
651 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
653 unsigned char *hpte_slot_array,
654 int psize, int ssize, int local)
656 WARN(1, "%s called without THP support\n", __func__);
660 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
661 int psize, int ssize)
664 unsigned long slot, vsid;
666 vsid = get_kernel_vsid(ea, ssize);
667 vpn = hpt_vpn(ea, vsid, ssize);
669 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
674 * lpar doesn't use the passed actual page size
676 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
681 static inline unsigned long compute_slot(real_pte_t pte,
687 unsigned long slot, hash, hidx;
689 hash = hpt_hash(vpn, shift, ssize);
690 hidx = __rpte_to_hidx(pte, index);
691 if (hidx & _PTEIDX_SECONDARY)
693 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
694 slot += hidx & _PTEIDX_GROUP_IX;
699 * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
700 * "all within the same naturally aligned 8 page virtual address block".
702 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
703 unsigned long *param)
706 unsigned long i, pix = 0;
707 unsigned long index, shift, slot, current_vpgb, vpgb;
711 psize = batch->psize;
712 ssize = batch->ssize;
714 for (i = 0; i < number; i++) {
717 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
719 * Shifting 3 bits more on the right to get a
720 * 8 pages aligned virtual addresse.
722 vpgb = (vpn >> (shift - VPN_SHIFT + 3));
723 if (!pix || vpgb != current_vpgb) {
725 * Need to start a new 8 pages block, flush
726 * the current one if needed.
729 (void)call_block_remove(pix, param,
732 param[0] = hpte_encode_avpn(vpn, psize,
737 slot = compute_slot(pte, vpn, index, shift, ssize);
738 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
740 if (pix == PLPAR_HCALL9_BUFSIZE) {
741 pix = call_block_remove(pix, param, false);
743 * pix = 0 means that all the entries were
744 * removed, we can start a new block.
745 * Otherwise, this means that there are entries
746 * to retry, and pix points to latest one, so
747 * we should increment it and try to continue
753 } pte_iterate_hashed_end();
757 (void)call_block_remove(pix, param, true);
761 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
764 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
767 unsigned long i, pix, rc;
768 unsigned long flags = 0;
769 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
770 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
771 unsigned long param[PLPAR_HCALL9_BUFSIZE];
772 unsigned long index, shift, slot;
777 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
779 if (firmware_has_feature(FW_FEATURE_BLOCK_REMOVE)) {
780 do_block_remove(number, batch, param);
784 psize = batch->psize;
785 ssize = batch->ssize;
787 for (i = 0; i < number; i++) {
790 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
791 slot = compute_slot(pte, vpn, index, shift, ssize);
792 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
794 * lpar doesn't use the passed actual page size
796 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
799 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
800 param[pix+1] = hpte_encode_avpn(vpn, psize,
804 rc = plpar_hcall9(H_BULK_REMOVE, param,
805 param[0], param[1], param[2],
806 param[3], param[4], param[5],
808 BUG_ON(rc != H_SUCCESS);
812 } pte_iterate_hashed_end();
815 param[pix] = HBR_END;
816 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
817 param[2], param[3], param[4], param[5],
819 BUG_ON(rc != H_SUCCESS);
824 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
827 static int __init disable_bulk_remove(char *str)
829 if (strcmp(str, "off") == 0 &&
830 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
831 pr_info("Disabling BULK_REMOVE firmware feature");
832 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
837 __setup("bulk_remove=", disable_bulk_remove);
839 #define HPT_RESIZE_TIMEOUT 10000 /* ms */
841 struct hpt_resize_state {
846 static int pseries_lpar_resize_hpt_commit(void *data)
848 struct hpt_resize_state *state = data;
850 state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
851 if (state->commit_rc != H_SUCCESS)
854 /* Hypervisor has transitioned the HTAB, update our globals */
855 ppc64_pft_size = state->shift;
856 htab_size_bytes = 1UL << ppc64_pft_size;
857 htab_hash_mask = (htab_size_bytes >> 7) - 1;
862 /* Must be called in user context */
863 static int pseries_lpar_resize_hpt(unsigned long shift)
865 struct hpt_resize_state state = {
867 .commit_rc = H_FUNCTION,
869 unsigned int delay, total_delay = 0;
875 if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
878 pr_info("Attempting to resize HPT to shift %lu\n", shift);
882 rc = plpar_resize_hpt_prepare(0, shift);
883 while (H_IS_LONG_BUSY(rc)) {
884 delay = get_longbusy_msecs(rc);
885 total_delay += delay;
886 if (total_delay > HPT_RESIZE_TIMEOUT) {
887 /* prepare with shift==0 cancels an in-progress resize */
888 rc = plpar_resize_hpt_prepare(0, 0);
890 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
895 rc = plpar_resize_hpt_prepare(0, shift);
904 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
907 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
910 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
916 rc = stop_machine(pseries_lpar_resize_hpt_commit, &state, NULL);
921 switch (state.commit_rc) {
926 pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
932 pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
933 shift, (long long) ktime_ms_delta(t1, t0),
934 (long long) ktime_ms_delta(t2, t1));
939 static int pseries_lpar_register_process_table(unsigned long base,
940 unsigned long page_size, unsigned long table_size)
943 unsigned long flags = 0;
946 flags |= PROC_TABLE_NEW;
948 flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
950 flags |= PROC_TABLE_HPT_SLB;
952 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
953 page_size, table_size);
954 if (!H_IS_LONG_BUSY(rc))
956 mdelay(get_longbusy_msecs(rc));
958 if (rc != H_SUCCESS) {
959 pr_err("Failed to register process table (rc=%ld)\n", rc);
965 void __init hpte_init_pseries(void)
967 mmu_hash_ops.hpte_invalidate = pSeries_lpar_hpte_invalidate;
968 mmu_hash_ops.hpte_updatepp = pSeries_lpar_hpte_updatepp;
969 mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
970 mmu_hash_ops.hpte_insert = pSeries_lpar_hpte_insert;
971 mmu_hash_ops.hpte_remove = pSeries_lpar_hpte_remove;
972 mmu_hash_ops.hpte_removebolted = pSeries_lpar_hpte_removebolted;
973 mmu_hash_ops.flush_hash_range = pSeries_lpar_flush_hash_range;
974 mmu_hash_ops.hpte_clear_all = pseries_hpte_clear_all;
975 mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
976 register_process_table = pseries_lpar_register_process_table;
978 if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
979 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
982 void radix_init_pseries(void)
984 pr_info("Using radix MMU under hypervisor\n");
985 register_process_table = pseries_lpar_register_process_table;
988 #ifdef CONFIG_PPC_SMLPAR
989 #define CMO_FREE_HINT_DEFAULT 1
990 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
992 static int __init cmo_free_hint(char *str)
995 parm = strstrip(str);
997 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
998 pr_info("%s: CMO free page hinting is not active.\n", __func__);
999 cmo_free_hint_flag = 0;
1003 cmo_free_hint_flag = 1;
1004 pr_info("%s: CMO free page hinting is active.\n", __func__);
1006 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1012 __setup("cmo_free_hint=", cmo_free_hint);
1014 static void pSeries_set_page_state(struct page *page, int order,
1015 unsigned long state)
1018 unsigned long cmo_page_sz, addr;
1020 cmo_page_sz = cmo_get_page_size();
1021 addr = __pa((unsigned long)page_address(page));
1023 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1024 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1025 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1029 void arch_free_page(struct page *page, int order)
1031 if (radix_enabled())
1033 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1036 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1038 EXPORT_SYMBOL(arch_free_page);
1040 #endif /* CONFIG_PPC_SMLPAR */
1041 #endif /* CONFIG_PPC_BOOK3S_64 */
1043 #ifdef CONFIG_TRACEPOINTS
1044 #ifdef CONFIG_JUMP_LABEL
1045 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1047 int hcall_tracepoint_regfunc(void)
1049 static_key_slow_inc(&hcall_tracepoint_key);
1053 void hcall_tracepoint_unregfunc(void)
1055 static_key_slow_dec(&hcall_tracepoint_key);
1059 * We optimise our hcall path by placing hcall_tracepoint_refcount
1060 * directly in the TOC so we can check if the hcall tracepoints are
1061 * enabled via a single load.
1064 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1065 extern long hcall_tracepoint_refcount;
1067 int hcall_tracepoint_regfunc(void)
1069 hcall_tracepoint_refcount++;
1073 void hcall_tracepoint_unregfunc(void)
1075 hcall_tracepoint_refcount--;
1080 * Since the tracing code might execute hcalls we need to guard against
1081 * recursion. One example of this are spinlocks calling H_YIELD on
1082 * shared processor partitions.
1084 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1087 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1089 unsigned long flags;
1090 unsigned int *depth;
1093 * We cannot call tracepoints inside RCU idle regions which
1094 * means we must not trace H_CEDE.
1096 if (opcode == H_CEDE)
1099 local_irq_save(flags);
1101 depth = this_cpu_ptr(&hcall_trace_depth);
1108 trace_hcall_entry(opcode, args);
1112 local_irq_restore(flags);
1115 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1117 unsigned long flags;
1118 unsigned int *depth;
1120 if (opcode == H_CEDE)
1123 local_irq_save(flags);
1125 depth = this_cpu_ptr(&hcall_trace_depth);
1131 trace_hcall_exit(opcode, retval, retbuf);
1136 local_irq_restore(flags);
1142 * H_GET_MPP hcall returns info in 7 parms
1144 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1147 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1149 rc = plpar_hcall9(H_GET_MPP, retbuf);
1151 mpp_data->entitled_mem = retbuf[0];
1152 mpp_data->mapped_mem = retbuf[1];
1154 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1155 mpp_data->pool_num = retbuf[2] & 0xffff;
1157 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1158 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1159 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1161 mpp_data->pool_size = retbuf[4];
1162 mpp_data->loan_request = retbuf[5];
1163 mpp_data->backing_mem = retbuf[6];
1167 EXPORT_SYMBOL(h_get_mpp);
1169 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1172 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1174 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1176 mpp_x_data->coalesced_bytes = retbuf[0];
1177 mpp_x_data->pool_coalesced_bytes = retbuf[1];
1178 mpp_x_data->pool_purr_cycles = retbuf[2];
1179 mpp_x_data->pool_spurr_cycles = retbuf[3];
1184 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
1186 unsigned long protovsid;
1187 unsigned long va_bits = VA_BITS;
1188 unsigned long modinv, vsid_modulus;
1189 unsigned long max_mod_inv, tmp_modinv;
1191 if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1194 if (ssize == MMU_SEGSIZE_256M) {
1195 modinv = VSID_MULINV_256M;
1196 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1198 modinv = VSID_MULINV_1T;
1199 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1203 * vsid outside our range.
1205 if (vsid >= vsid_modulus)
1209 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1210 * and vsid = (protovsid * x) % vsid_modulus, then we say:
1211 * protovsid = (vsid * modinv) % vsid_modulus
1214 /* Check if (vsid * modinv) overflow (63 bits) */
1215 max_mod_inv = 0x7fffffffffffffffull / vsid;
1216 if (modinv < max_mod_inv)
1217 return (vsid * modinv) % vsid_modulus;
1219 tmp_modinv = modinv/max_mod_inv;
1220 modinv %= max_mod_inv;
1222 protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1223 protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1228 static int __init reserve_vrma_context_id(void)
1230 unsigned long protovsid;
1233 * Reserve context ids which map to reserved virtual addresses. For now
1234 * we only reserve the context id which maps to the VRMA VSID. We ignore
1235 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1236 * enable adjunct support via the "ibm,client-architecture-support"
1239 protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1240 hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1243 machine_device_initcall(pseries, reserve_vrma_context_id);
1245 #ifdef CONFIG_DEBUG_FS
1246 /* debugfs file interface for vpa data */
1247 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1250 int cpu = (long)filp->private_data;
1251 struct lppaca *lppaca = &lppaca_of(cpu);
1253 return simple_read_from_buffer(buf, len, pos, lppaca,
1254 sizeof(struct lppaca));
1257 static const struct file_operations vpa_fops = {
1258 .open = simple_open,
1259 .read = vpa_file_read,
1260 .llseek = default_llseek,
1263 static int __init vpa_debugfs_init(void)
1267 static struct dentry *vpa_dir;
1269 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
1272 vpa_dir = debugfs_create_dir("vpa", powerpc_debugfs_root);
1274 pr_warn("%s: can't create vpa root dir\n", __func__);
1278 /* set up the per-cpu vpa file*/
1279 for_each_possible_cpu(i) {
1282 sprintf(name, "cpu-%ld", i);
1284 d = debugfs_create_file(name, 0400, vpa_dir, (void *)i,
1287 pr_warn("%s: can't create per-cpu vpa file\n",
1295 machine_arch_initcall(pseries, vpa_debugfs_init);
1296 #endif /* CONFIG_DEBUG_FS */