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>
45 #include "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;
71 if (cpu_has_feature(CPU_FTR_ALTIVEC))
72 lppaca_of(cpu).vmxregs_in_use = 1;
74 addr = __pa(&lppaca_of(cpu));
75 ret = register_vpa(hwcpu, addr);
78 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
79 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
83 * PAPR says this feature is SLB-Buffer but firmware never
84 * reports that. All SPLPAR support SLB shadow buffer.
86 addr = __pa(&slb_shadow[cpu]);
87 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
88 ret = register_slb_shadow(hwcpu, addr);
90 pr_err("WARNING: SLB shadow buffer registration for "
91 "cpu %d (hw %d) of area %lx failed with %ld\n",
92 cpu, hwcpu, addr, ret);
96 * Register dispatch trace log, if one has been allocated.
99 dtl = pp->dispatch_log;
103 lppaca_of(cpu).dtl_idx = 0;
105 /* hypervisor reads buffer length from this field */
106 dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
107 ret = register_dtl(hwcpu, __pa(dtl));
109 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
110 "failed with %ld\n", smp_processor_id(),
112 lppaca_of(cpu).dtl_enable_mask = 2;
116 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
117 unsigned long vpn, unsigned long pa,
118 unsigned long rflags, unsigned long vflags,
119 int psize, int apsize, int ssize)
121 unsigned long lpar_rc;
124 unsigned long hpte_v, hpte_r;
126 if (!(vflags & HPTE_V_BOLTED))
127 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
128 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
129 hpte_group, vpn, pa, rflags, vflags, psize);
131 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
132 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
134 if (!(vflags & HPTE_V_BOLTED))
135 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
137 /* Now fill in the actual HPTE */
138 /* Set CEC cookie to 0 */
140 /* I-cache Invalidate = 0 */
141 /* I-cache synchronize = 0 */
145 /* Make pHyp happy */
146 if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
147 hpte_r &= ~_PAGE_COHERENT;
148 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
149 flags |= H_COALESCE_CAND;
151 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
152 if (unlikely(lpar_rc == H_PTEG_FULL)) {
153 if (!(vflags & HPTE_V_BOLTED))
159 * Since we try and ioremap PHBs we don't own, the pte insert
160 * will fail. However we must catch the failure in hash_page
161 * or we will loop forever, so return -2 in this case.
163 if (unlikely(lpar_rc != H_SUCCESS)) {
164 if (!(vflags & HPTE_V_BOLTED))
165 pr_devel(" lpar err %ld\n", lpar_rc);
168 if (!(vflags & HPTE_V_BOLTED))
169 pr_devel(" -> slot: %lu\n", slot & 7);
171 /* Because of iSeries, we have to pass down the secondary
172 * bucket bit here as well
174 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
177 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
179 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
181 unsigned long slot_offset;
182 unsigned long lpar_rc;
184 unsigned long dummy1, dummy2;
186 /* pick a random slot to start at */
187 slot_offset = mftb() & 0x7;
189 for (i = 0; i < HPTES_PER_GROUP; i++) {
191 /* don't remove a bolted entry */
192 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
193 (0x1UL << 4), &dummy1, &dummy2);
194 if (lpar_rc == H_SUCCESS)
198 * The test for adjunct partition is performed before the
199 * ANDCOND test. H_RESOURCE may be returned, so we need to
200 * check for that as well.
202 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
211 static void pSeries_lpar_hptab_clear(void)
213 unsigned long size_bytes = 1UL << ppc64_pft_size;
214 unsigned long hpte_count = size_bytes >> 4;
222 /* Read in batches of 4,
223 * invalidate only valid entries not in the VRMA
224 * hpte_count will be a multiple of 4
226 for (i = 0; i < hpte_count; i += 4) {
227 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
228 if (lpar_rc != H_SUCCESS)
230 for (j = 0; j < 4; j++){
231 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
234 if (ptes[j].pteh & HPTE_V_VALID)
235 plpar_pte_remove_raw(0, i + j, 0,
236 &(ptes[j].pteh), &(ptes[j].ptel));
242 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
243 * the low 3 bits of flags happen to line up. So no transform is needed.
244 * We can probably optimize here and assume the high bits of newpp are
245 * already zero. For now I am paranoid.
247 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
250 int psize, int apsize,
251 int ssize, int local)
253 unsigned long lpar_rc;
254 unsigned long flags = (newpp & 7) | H_AVPN;
255 unsigned long want_v;
257 want_v = hpte_encode_avpn(vpn, psize, ssize);
259 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
260 want_v, slot, flags, psize);
262 lpar_rc = plpar_pte_protect(flags, slot, want_v);
264 if (lpar_rc == H_NOT_FOUND) {
265 pr_devel("not found !\n");
271 BUG_ON(lpar_rc != H_SUCCESS);
276 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
278 unsigned long dword0;
279 unsigned long lpar_rc;
280 unsigned long dummy_word1;
283 /* Read 1 pte at a time */
284 /* Do not need RPN to logical page translation */
285 /* No cross CEC PFT access */
288 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
290 BUG_ON(lpar_rc != H_SUCCESS);
295 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
300 unsigned long want_v, hpte_v;
302 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
303 want_v = hpte_encode_avpn(vpn, psize, ssize);
305 /* Bolted entries are always in the primary group */
306 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
307 for (i = 0; i < HPTES_PER_GROUP; i++) {
308 hpte_v = pSeries_lpar_hpte_getword0(slot);
310 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
319 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
321 int psize, int ssize)
324 unsigned long lpar_rc, slot, vsid, flags;
326 vsid = get_kernel_vsid(ea, ssize);
327 vpn = hpt_vpn(ea, vsid, ssize);
329 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
333 lpar_rc = plpar_pte_protect(flags, slot, 0);
335 BUG_ON(lpar_rc != H_SUCCESS);
338 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
339 int psize, int apsize,
340 int ssize, int local)
342 unsigned long want_v;
343 unsigned long lpar_rc;
344 unsigned long dummy1, dummy2;
346 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
347 slot, vpn, psize, local);
349 want_v = hpte_encode_avpn(vpn, psize, ssize);
350 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
351 if (lpar_rc == H_NOT_FOUND)
354 BUG_ON(lpar_rc != H_SUCCESS);
358 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
359 * to make sure that we avoid bouncing the hypervisor tlbie lock.
361 #define PPC64_HUGE_HPTE_BATCH 12
363 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
364 unsigned long *vpn, int count,
365 int psize, int ssize)
367 unsigned long param[8];
368 int i = 0, pix = 0, rc;
369 unsigned long flags = 0;
370 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
373 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
375 for (i = 0; i < count; i++) {
377 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
378 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
381 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
382 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
385 rc = plpar_hcall9(H_BULK_REMOVE, param,
386 param[0], param[1], param[2],
387 param[3], param[4], param[5],
389 BUG_ON(rc != H_SUCCESS);
395 param[pix] = HBR_END;
396 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
397 param[2], param[3], param[4], param[5],
399 BUG_ON(rc != H_SUCCESS);
403 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
406 static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
407 unsigned char *hpte_slot_array,
408 unsigned long addr, int psize)
410 int ssize = 0, i, index = 0;
411 unsigned long s_addr = addr;
412 unsigned int max_hpte_count, valid;
413 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
414 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
415 unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
417 shift = mmu_psize_defs[psize].shift;
418 max_hpte_count = 1U << (PMD_SHIFT - shift);
420 for (i = 0; i < max_hpte_count; i++) {
421 valid = hpte_valid(hpte_slot_array, i);
424 hidx = hpte_hash_index(hpte_slot_array, i);
427 addr = s_addr + (i * (1ul << shift));
428 if (!is_kernel_addr(addr)) {
429 ssize = user_segment_size(addr);
430 vsid = get_vsid(mm->context.id, addr, ssize);
433 vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
434 ssize = mmu_kernel_ssize;
437 vpn = hpt_vpn(addr, vsid, ssize);
438 hash = hpt_hash(vpn, shift, ssize);
439 if (hidx & _PTEIDX_SECONDARY)
442 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
443 slot += hidx & _PTEIDX_GROUP_IX;
445 slot_array[index] = slot;
446 vpn_array[index] = vpn;
447 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
449 * Now do a bluk invalidate
451 __pSeries_lpar_hugepage_invalidate(slot_array,
453 PPC64_HUGE_HPTE_BATCH,
460 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
461 index, psize, ssize);
464 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
465 int psize, int ssize)
468 unsigned long slot, vsid;
470 vsid = get_kernel_vsid(ea, ssize);
471 vpn = hpt_vpn(ea, vsid, ssize);
473 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
476 * lpar doesn't use the passed actual page size
478 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
482 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
485 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
488 unsigned long i, pix, rc;
489 unsigned long flags = 0;
490 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
491 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
492 unsigned long param[9];
493 unsigned long hash, index, shift, hidx, slot;
498 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
500 psize = batch->psize;
501 ssize = batch->ssize;
503 for (i = 0; i < number; i++) {
506 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
507 hash = hpt_hash(vpn, shift, ssize);
508 hidx = __rpte_to_hidx(pte, index);
509 if (hidx & _PTEIDX_SECONDARY)
511 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
512 slot += hidx & _PTEIDX_GROUP_IX;
513 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
515 * lpar doesn't use the passed actual page size
517 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
520 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
521 param[pix+1] = hpte_encode_avpn(vpn, psize,
525 rc = plpar_hcall9(H_BULK_REMOVE, param,
526 param[0], param[1], param[2],
527 param[3], param[4], param[5],
529 BUG_ON(rc != H_SUCCESS);
533 } pte_iterate_hashed_end();
536 param[pix] = HBR_END;
537 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
538 param[2], param[3], param[4], param[5],
540 BUG_ON(rc != H_SUCCESS);
544 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
547 static int __init disable_bulk_remove(char *str)
549 if (strcmp(str, "off") == 0 &&
550 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
551 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
552 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
557 __setup("bulk_remove=", disable_bulk_remove);
559 void __init hpte_init_lpar(void)
561 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
562 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
563 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
564 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
565 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
566 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
567 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
568 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
569 ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
572 #ifdef CONFIG_PPC_SMLPAR
573 #define CMO_FREE_HINT_DEFAULT 1
574 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
576 static int __init cmo_free_hint(char *str)
579 parm = strstrip(str);
581 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
582 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
583 cmo_free_hint_flag = 0;
587 cmo_free_hint_flag = 1;
588 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
590 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
596 __setup("cmo_free_hint=", cmo_free_hint);
598 static void pSeries_set_page_state(struct page *page, int order,
602 unsigned long cmo_page_sz, addr;
604 cmo_page_sz = cmo_get_page_size();
605 addr = __pa((unsigned long)page_address(page));
607 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
608 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
609 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
613 void arch_free_page(struct page *page, int order)
615 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
618 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
620 EXPORT_SYMBOL(arch_free_page);
624 #ifdef CONFIG_TRACEPOINTS
626 * We optimise our hcall path by placing hcall_tracepoint_refcount
627 * directly in the TOC so we can check if the hcall tracepoints are
628 * enabled via a single load.
631 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
632 extern long hcall_tracepoint_refcount;
635 * Since the tracing code might execute hcalls we need to guard against
636 * recursion. One example of this are spinlocks calling H_YIELD on
637 * shared processor partitions.
639 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
641 void hcall_tracepoint_regfunc(void)
643 hcall_tracepoint_refcount++;
646 void hcall_tracepoint_unregfunc(void)
648 hcall_tracepoint_refcount--;
651 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
657 * We cannot call tracepoints inside RCU idle regions which
658 * means we must not trace H_CEDE.
660 if (opcode == H_CEDE)
663 local_irq_save(flags);
665 depth = &__get_cpu_var(hcall_trace_depth);
672 trace_hcall_entry(opcode, args);
676 local_irq_restore(flags);
679 void __trace_hcall_exit(long opcode, unsigned long retval,
680 unsigned long *retbuf)
685 if (opcode == H_CEDE)
688 local_irq_save(flags);
690 depth = &__get_cpu_var(hcall_trace_depth);
696 trace_hcall_exit(opcode, retval, retbuf);
701 local_irq_restore(flags);
707 * H_GET_MPP hcall returns info in 7 parms
709 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
712 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
714 rc = plpar_hcall9(H_GET_MPP, retbuf);
716 mpp_data->entitled_mem = retbuf[0];
717 mpp_data->mapped_mem = retbuf[1];
719 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
720 mpp_data->pool_num = retbuf[2] & 0xffff;
722 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
723 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
724 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
726 mpp_data->pool_size = retbuf[4];
727 mpp_data->loan_request = retbuf[5];
728 mpp_data->backing_mem = retbuf[6];
732 EXPORT_SYMBOL(h_get_mpp);
734 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
737 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
739 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
741 mpp_x_data->coalesced_bytes = retbuf[0];
742 mpp_x_data->pool_coalesced_bytes = retbuf[1];
743 mpp_x_data->pool_purr_cycles = retbuf[2];
744 mpp_x_data->pool_spurr_cycles = retbuf[3];