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 <asm/processor.h>
32 #include <asm/pgtable.h>
33 #include <asm/machdep.h>
34 #include <asm/mmu_context.h>
35 #include <asm/iommu.h>
36 #include <asm/tlbflush.h>
39 #include <asm/cputable.h>
42 #include <asm/trace.h>
43 #include <asm/firmware.h>
44 #include <asm/plpar_wrappers.h>
48 /* Flag bits for H_BULK_REMOVE */
49 #define HBR_REQUEST 0x4000000000000000UL
50 #define HBR_RESPONSE 0x8000000000000000UL
51 #define HBR_END 0xc000000000000000UL
52 #define HBR_AVPN 0x0200000000000000UL
53 #define HBR_ANDCOND 0x0100000000000000UL
57 EXPORT_SYMBOL(plpar_hcall);
58 EXPORT_SYMBOL(plpar_hcall9);
59 EXPORT_SYMBOL(plpar_hcall_norets);
61 extern void pSeries_find_serial_port(void);
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(&slb_shadow[cpu]);
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))
156 hpte_r &= ~_PAGE_COHERENT;
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));
251 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
252 * the low 3 bits of flags happen to line up. So no transform is needed.
253 * We can probably optimize here and assume the high bits of newpp are
254 * already zero. For now I am paranoid.
256 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
259 int psize, int apsize,
260 int ssize, int local)
262 unsigned long lpar_rc;
263 unsigned long flags = (newpp & 7) | H_AVPN;
264 unsigned long want_v;
266 want_v = hpte_encode_avpn(vpn, psize, ssize);
268 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
269 want_v, slot, flags, psize);
271 lpar_rc = plpar_pte_protect(flags, slot, want_v);
273 if (lpar_rc == H_NOT_FOUND) {
274 pr_devel("not found !\n");
280 BUG_ON(lpar_rc != H_SUCCESS);
285 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
287 unsigned long dword0;
288 unsigned long lpar_rc;
289 unsigned long dummy_word1;
292 /* Read 1 pte at a time */
293 /* Do not need RPN to logical page translation */
294 /* No cross CEC PFT access */
297 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
299 BUG_ON(lpar_rc != H_SUCCESS);
304 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
309 unsigned long want_v, hpte_v;
311 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
312 want_v = hpte_encode_avpn(vpn, psize, ssize);
314 /* Bolted entries are always in the primary group */
315 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
316 for (i = 0; i < HPTES_PER_GROUP; i++) {
317 hpte_v = pSeries_lpar_hpte_getword0(slot);
319 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
328 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
330 int psize, int ssize)
333 unsigned long lpar_rc, slot, vsid, flags;
335 vsid = get_kernel_vsid(ea, ssize);
336 vpn = hpt_vpn(ea, vsid, ssize);
338 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
342 lpar_rc = plpar_pte_protect(flags, slot, 0);
344 BUG_ON(lpar_rc != H_SUCCESS);
347 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
348 int psize, int apsize,
349 int ssize, int local)
351 unsigned long want_v;
352 unsigned long lpar_rc;
353 unsigned long dummy1, dummy2;
355 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
356 slot, vpn, psize, local);
358 want_v = hpte_encode_avpn(vpn, psize, ssize);
359 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
360 if (lpar_rc == H_NOT_FOUND)
363 BUG_ON(lpar_rc != H_SUCCESS);
367 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
368 * to make sure that we avoid bouncing the hypervisor tlbie lock.
370 #define PPC64_HUGE_HPTE_BATCH 12
372 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
373 unsigned long *vpn, int count,
374 int psize, int ssize)
376 unsigned long param[8];
377 int i = 0, pix = 0, rc;
378 unsigned long flags = 0;
379 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
382 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
384 for (i = 0; i < count; i++) {
386 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
387 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
390 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
391 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
394 rc = plpar_hcall9(H_BULK_REMOVE, param,
395 param[0], param[1], param[2],
396 param[3], param[4], param[5],
398 BUG_ON(rc != H_SUCCESS);
404 param[pix] = HBR_END;
405 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
406 param[2], param[3], param[4], param[5],
408 BUG_ON(rc != H_SUCCESS);
412 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
415 static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
416 unsigned char *hpte_slot_array,
417 unsigned long addr, int psize)
419 int ssize = 0, i, index = 0;
420 unsigned long s_addr = addr;
421 unsigned int max_hpte_count, valid;
422 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
423 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
424 unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
426 shift = mmu_psize_defs[psize].shift;
427 max_hpte_count = 1U << (PMD_SHIFT - shift);
429 for (i = 0; i < max_hpte_count; i++) {
430 valid = hpte_valid(hpte_slot_array, i);
433 hidx = hpte_hash_index(hpte_slot_array, i);
436 addr = s_addr + (i * (1ul << shift));
437 if (!is_kernel_addr(addr)) {
438 ssize = user_segment_size(addr);
439 vsid = get_vsid(mm->context.id, addr, ssize);
442 vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
443 ssize = mmu_kernel_ssize;
446 vpn = hpt_vpn(addr, vsid, ssize);
447 hash = hpt_hash(vpn, shift, ssize);
448 if (hidx & _PTEIDX_SECONDARY)
451 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
452 slot += hidx & _PTEIDX_GROUP_IX;
454 slot_array[index] = slot;
455 vpn_array[index] = vpn;
456 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
458 * Now do a bluk invalidate
460 __pSeries_lpar_hugepage_invalidate(slot_array,
462 PPC64_HUGE_HPTE_BATCH,
469 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
470 index, psize, ssize);
473 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
474 int psize, int ssize)
477 unsigned long slot, vsid;
479 vsid = get_kernel_vsid(ea, ssize);
480 vpn = hpt_vpn(ea, vsid, ssize);
482 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
485 * lpar doesn't use the passed actual page size
487 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
491 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
494 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
497 unsigned long i, pix, rc;
498 unsigned long flags = 0;
499 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
500 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
501 unsigned long param[9];
502 unsigned long hash, index, shift, hidx, slot;
507 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
509 psize = batch->psize;
510 ssize = batch->ssize;
512 for (i = 0; i < number; i++) {
515 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
516 hash = hpt_hash(vpn, shift, ssize);
517 hidx = __rpte_to_hidx(pte, index);
518 if (hidx & _PTEIDX_SECONDARY)
520 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
521 slot += hidx & _PTEIDX_GROUP_IX;
522 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
524 * lpar doesn't use the passed actual page size
526 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
529 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
530 param[pix+1] = hpte_encode_avpn(vpn, psize,
534 rc = plpar_hcall9(H_BULK_REMOVE, param,
535 param[0], param[1], param[2],
536 param[3], param[4], param[5],
538 BUG_ON(rc != H_SUCCESS);
542 } pte_iterate_hashed_end();
545 param[pix] = HBR_END;
546 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
547 param[2], param[3], param[4], param[5],
549 BUG_ON(rc != H_SUCCESS);
553 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
556 static int __init disable_bulk_remove(char *str)
558 if (strcmp(str, "off") == 0 &&
559 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
560 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
561 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
566 __setup("bulk_remove=", disable_bulk_remove);
568 void __init hpte_init_lpar(void)
570 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
571 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
572 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
573 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
574 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
575 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
576 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
577 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
578 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
581 #ifdef CONFIG_PPC_SMLPAR
582 #define CMO_FREE_HINT_DEFAULT 1
583 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
585 static int __init cmo_free_hint(char *str)
588 parm = strstrip(str);
590 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
591 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
592 cmo_free_hint_flag = 0;
596 cmo_free_hint_flag = 1;
597 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
599 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
605 __setup("cmo_free_hint=", cmo_free_hint);
607 static void pSeries_set_page_state(struct page *page, int order,
611 unsigned long cmo_page_sz, addr;
613 cmo_page_sz = cmo_get_page_size();
614 addr = __pa((unsigned long)page_address(page));
616 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
617 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
618 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
622 void arch_free_page(struct page *page, int order)
624 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
627 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
629 EXPORT_SYMBOL(arch_free_page);
633 #ifdef CONFIG_TRACEPOINTS
635 * We optimise our hcall path by placing hcall_tracepoint_refcount
636 * directly in the TOC so we can check if the hcall tracepoints are
637 * enabled via a single load.
640 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
641 extern long hcall_tracepoint_refcount;
644 * Since the tracing code might execute hcalls we need to guard against
645 * recursion. One example of this are spinlocks calling H_YIELD on
646 * shared processor partitions.
648 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
650 void hcall_tracepoint_regfunc(void)
652 hcall_tracepoint_refcount++;
655 void hcall_tracepoint_unregfunc(void)
657 hcall_tracepoint_refcount--;
660 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
666 * We cannot call tracepoints inside RCU idle regions which
667 * means we must not trace H_CEDE.
669 if (opcode == H_CEDE)
672 local_irq_save(flags);
674 depth = &__get_cpu_var(hcall_trace_depth);
681 trace_hcall_entry(opcode, args);
685 local_irq_restore(flags);
688 void __trace_hcall_exit(long opcode, unsigned long retval,
689 unsigned long *retbuf)
694 if (opcode == H_CEDE)
697 local_irq_save(flags);
699 depth = &__get_cpu_var(hcall_trace_depth);
705 trace_hcall_exit(opcode, retval, retbuf);
710 local_irq_restore(flags);
716 * H_GET_MPP hcall returns info in 7 parms
718 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
721 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
723 rc = plpar_hcall9(H_GET_MPP, retbuf);
725 mpp_data->entitled_mem = retbuf[0];
726 mpp_data->mapped_mem = retbuf[1];
728 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
729 mpp_data->pool_num = retbuf[2] & 0xffff;
731 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
732 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
733 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
735 mpp_data->pool_size = retbuf[4];
736 mpp_data->loan_request = retbuf[5];
737 mpp_data->backing_mem = retbuf[6];
741 EXPORT_SYMBOL(h_get_mpp);
743 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
746 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
748 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
750 mpp_x_data->coalesced_bytes = retbuf[0];
751 mpp_x_data->pool_coalesced_bytes = retbuf[1];
752 mpp_x_data->pool_purr_cycles = retbuf[2];
753 mpp_x_data->pool_spurr_cycles = retbuf[3];