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"
50 EXPORT_SYMBOL(plpar_hcall);
51 EXPORT_SYMBOL(plpar_hcall9);
52 EXPORT_SYMBOL(plpar_hcall_norets);
54 extern void pSeries_find_serial_port(void);
56 void vpa_init(int cpu)
58 int hwcpu = get_hard_smp_processor_id(cpu);
61 struct paca_struct *pp;
62 struct dtl_entry *dtl;
64 if (cpu_has_feature(CPU_FTR_ALTIVEC))
65 lppaca_of(cpu).vmxregs_in_use = 1;
67 addr = __pa(&lppaca_of(cpu));
68 ret = register_vpa(hwcpu, addr);
71 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
72 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
76 * PAPR says this feature is SLB-Buffer but firmware never
77 * reports that. All SPLPAR support SLB shadow buffer.
79 addr = __pa(&slb_shadow[cpu]);
80 if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
81 ret = register_slb_shadow(hwcpu, addr);
83 pr_err("WARNING: SLB shadow buffer registration for "
84 "cpu %d (hw %d) of area %lx failed with %ld\n",
85 cpu, hwcpu, addr, ret);
89 * Register dispatch trace log, if one has been allocated.
92 dtl = pp->dispatch_log;
96 lppaca_of(cpu).dtl_idx = 0;
98 /* hypervisor reads buffer length from this field */
99 dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
100 ret = register_dtl(hwcpu, __pa(dtl));
102 pr_err("WARNING: DTL registration of cpu %d (hw %d) "
103 "failed with %ld\n", smp_processor_id(),
105 lppaca_of(cpu).dtl_enable_mask = 2;
109 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
110 unsigned long vpn, unsigned long pa,
111 unsigned long rflags, unsigned long vflags,
112 int psize, int apsize, int ssize)
114 unsigned long lpar_rc;
117 unsigned long hpte_v, hpte_r;
119 if (!(vflags & HPTE_V_BOLTED))
120 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
121 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
122 hpte_group, vpn, pa, rflags, vflags, psize);
124 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
125 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
127 if (!(vflags & HPTE_V_BOLTED))
128 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
130 /* Now fill in the actual HPTE */
131 /* Set CEC cookie to 0 */
133 /* I-cache Invalidate = 0 */
134 /* I-cache synchronize = 0 */
138 /* Make pHyp happy */
139 if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
140 hpte_r &= ~_PAGE_COHERENT;
141 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
142 flags |= H_COALESCE_CAND;
144 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
145 if (unlikely(lpar_rc == H_PTEG_FULL)) {
146 if (!(vflags & HPTE_V_BOLTED))
152 * Since we try and ioremap PHBs we don't own, the pte insert
153 * will fail. However we must catch the failure in hash_page
154 * or we will loop forever, so return -2 in this case.
156 if (unlikely(lpar_rc != H_SUCCESS)) {
157 if (!(vflags & HPTE_V_BOLTED))
158 pr_devel(" lpar err %ld\n", lpar_rc);
161 if (!(vflags & HPTE_V_BOLTED))
162 pr_devel(" -> slot: %lu\n", slot & 7);
164 /* Because of iSeries, we have to pass down the secondary
165 * bucket bit here as well
167 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
170 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
172 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
174 unsigned long slot_offset;
175 unsigned long lpar_rc;
177 unsigned long dummy1, dummy2;
179 /* pick a random slot to start at */
180 slot_offset = mftb() & 0x7;
182 for (i = 0; i < HPTES_PER_GROUP; i++) {
184 /* don't remove a bolted entry */
185 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
186 (0x1UL << 4), &dummy1, &dummy2);
187 if (lpar_rc == H_SUCCESS)
191 * The test for adjunct partition is performed before the
192 * ANDCOND test. H_RESOURCE may be returned, so we need to
193 * check for that as well.
195 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
204 static void pSeries_lpar_hptab_clear(void)
206 unsigned long size_bytes = 1UL << ppc64_pft_size;
207 unsigned long hpte_count = size_bytes >> 4;
215 /* Read in batches of 4,
216 * invalidate only valid entries not in the VRMA
217 * hpte_count will be a multiple of 4
219 for (i = 0; i < hpte_count; i += 4) {
220 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
221 if (lpar_rc != H_SUCCESS)
223 for (j = 0; j < 4; j++){
224 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
227 if (ptes[j].pteh & HPTE_V_VALID)
228 plpar_pte_remove_raw(0, i + j, 0,
229 &(ptes[j].pteh), &(ptes[j].ptel));
235 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
236 * the low 3 bits of flags happen to line up. So no transform is needed.
237 * We can probably optimize here and assume the high bits of newpp are
238 * already zero. For now I am paranoid.
240 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
243 int psize, int apsize,
244 int ssize, int local)
246 unsigned long lpar_rc;
247 unsigned long flags = (newpp & 7) | H_AVPN;
248 unsigned long want_v;
250 want_v = hpte_encode_avpn(vpn, psize, ssize);
252 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
253 want_v, slot, flags, psize);
255 lpar_rc = plpar_pte_protect(flags, slot, want_v);
257 if (lpar_rc == H_NOT_FOUND) {
258 pr_devel("not found !\n");
264 BUG_ON(lpar_rc != H_SUCCESS);
269 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
271 unsigned long dword0;
272 unsigned long lpar_rc;
273 unsigned long dummy_word1;
276 /* Read 1 pte at a time */
277 /* Do not need RPN to logical page translation */
278 /* No cross CEC PFT access */
281 lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
283 BUG_ON(lpar_rc != H_SUCCESS);
288 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
293 unsigned long want_v, hpte_v;
295 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
296 want_v = hpte_encode_avpn(vpn, psize, ssize);
298 /* Bolted entries are always in the primary group */
299 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
300 for (i = 0; i < HPTES_PER_GROUP; i++) {
301 hpte_v = pSeries_lpar_hpte_getword0(slot);
303 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
312 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
314 int psize, int ssize)
317 unsigned long lpar_rc, slot, vsid, flags;
319 vsid = get_kernel_vsid(ea, ssize);
320 vpn = hpt_vpn(ea, vsid, ssize);
322 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
326 lpar_rc = plpar_pte_protect(flags, slot, 0);
328 BUG_ON(lpar_rc != H_SUCCESS);
331 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
332 int psize, int apsize,
333 int ssize, int local)
335 unsigned long want_v;
336 unsigned long lpar_rc;
337 unsigned long dummy1, dummy2;
339 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
340 slot, vpn, psize, local);
342 want_v = hpte_encode_avpn(vpn, psize, ssize);
343 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
344 if (lpar_rc == H_NOT_FOUND)
347 BUG_ON(lpar_rc != H_SUCCESS);
350 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
351 int psize, int ssize)
354 unsigned long slot, vsid;
356 vsid = get_kernel_vsid(ea, ssize);
357 vpn = hpt_vpn(ea, vsid, ssize);
359 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
362 * lpar doesn't use the passed actual page size
364 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
367 /* Flag bits for H_BULK_REMOVE */
368 #define HBR_REQUEST 0x4000000000000000UL
369 #define HBR_RESPONSE 0x8000000000000000UL
370 #define HBR_END 0xc000000000000000UL
371 #define HBR_AVPN 0x0200000000000000UL
372 #define HBR_ANDCOND 0x0100000000000000UL
375 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
378 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
381 unsigned long i, pix, rc;
382 unsigned long flags = 0;
383 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
384 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
385 unsigned long param[9];
386 unsigned long hash, index, shift, hidx, slot;
391 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
393 psize = batch->psize;
394 ssize = batch->ssize;
396 for (i = 0; i < number; i++) {
399 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
400 hash = hpt_hash(vpn, shift, ssize);
401 hidx = __rpte_to_hidx(pte, index);
402 if (hidx & _PTEIDX_SECONDARY)
404 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
405 slot += hidx & _PTEIDX_GROUP_IX;
406 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
408 * lpar doesn't use the passed actual page size
410 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
413 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
414 param[pix+1] = hpte_encode_avpn(vpn, psize,
418 rc = plpar_hcall9(H_BULK_REMOVE, param,
419 param[0], param[1], param[2],
420 param[3], param[4], param[5],
422 BUG_ON(rc != H_SUCCESS);
426 } pte_iterate_hashed_end();
429 param[pix] = HBR_END;
430 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
431 param[2], param[3], param[4], param[5],
433 BUG_ON(rc != H_SUCCESS);
437 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
440 static int __init disable_bulk_remove(char *str)
442 if (strcmp(str, "off") == 0 &&
443 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
444 printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
445 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
450 __setup("bulk_remove=", disable_bulk_remove);
452 void __init hpte_init_lpar(void)
454 ppc_md.hpte_invalidate = pSeries_lpar_hpte_invalidate;
455 ppc_md.hpte_updatepp = pSeries_lpar_hpte_updatepp;
456 ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
457 ppc_md.hpte_insert = pSeries_lpar_hpte_insert;
458 ppc_md.hpte_remove = pSeries_lpar_hpte_remove;
459 ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
460 ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
461 ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
464 #ifdef CONFIG_PPC_SMLPAR
465 #define CMO_FREE_HINT_DEFAULT 1
466 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
468 static int __init cmo_free_hint(char *str)
471 parm = strstrip(str);
473 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
474 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
475 cmo_free_hint_flag = 0;
479 cmo_free_hint_flag = 1;
480 printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
482 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
488 __setup("cmo_free_hint=", cmo_free_hint);
490 static void pSeries_set_page_state(struct page *page, int order,
494 unsigned long cmo_page_sz, addr;
496 cmo_page_sz = cmo_get_page_size();
497 addr = __pa((unsigned long)page_address(page));
499 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
500 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
501 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
505 void arch_free_page(struct page *page, int order)
507 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
510 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
512 EXPORT_SYMBOL(arch_free_page);
516 #ifdef CONFIG_TRACEPOINTS
518 * We optimise our hcall path by placing hcall_tracepoint_refcount
519 * directly in the TOC so we can check if the hcall tracepoints are
520 * enabled via a single load.
523 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
524 extern long hcall_tracepoint_refcount;
527 * Since the tracing code might execute hcalls we need to guard against
528 * recursion. One example of this are spinlocks calling H_YIELD on
529 * shared processor partitions.
531 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
533 void hcall_tracepoint_regfunc(void)
535 hcall_tracepoint_refcount++;
538 void hcall_tracepoint_unregfunc(void)
540 hcall_tracepoint_refcount--;
543 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
549 * We cannot call tracepoints inside RCU idle regions which
550 * means we must not trace H_CEDE.
552 if (opcode == H_CEDE)
555 local_irq_save(flags);
557 depth = &__get_cpu_var(hcall_trace_depth);
564 trace_hcall_entry(opcode, args);
568 local_irq_restore(flags);
571 void __trace_hcall_exit(long opcode, unsigned long retval,
572 unsigned long *retbuf)
577 if (opcode == H_CEDE)
580 local_irq_save(flags);
582 depth = &__get_cpu_var(hcall_trace_depth);
588 trace_hcall_exit(opcode, retval, retbuf);
593 local_irq_restore(flags);
599 * H_GET_MPP hcall returns info in 7 parms
601 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
604 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
606 rc = plpar_hcall9(H_GET_MPP, retbuf);
608 mpp_data->entitled_mem = retbuf[0];
609 mpp_data->mapped_mem = retbuf[1];
611 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
612 mpp_data->pool_num = retbuf[2] & 0xffff;
614 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
615 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
616 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
618 mpp_data->pool_size = retbuf[4];
619 mpp_data->loan_request = retbuf[5];
620 mpp_data->backing_mem = retbuf[6];
624 EXPORT_SYMBOL(h_get_mpp);
626 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
629 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
631 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
633 mpp_x_data->coalesced_bytes = retbuf[0];
634 mpp_x_data->pool_coalesced_bytes = retbuf[1];
635 mpp_x_data->pool_purr_cycles = retbuf[2];
636 mpp_x_data->pool_spurr_cycles = retbuf[3];