lpar.c revision 92c08a0d522c7e62c01a63e42597f0c2b02c4245
1/*
2 * pSeries_lpar.c
3 * Copyright (C) 2001 Todd Inglett, IBM Corporation
4 *
5 * pSeries LPAR support.
6 *
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.
11 *
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.
16 *
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
20 */
21
22/* Enables debugging of low-level hash table routines - careful! */
23#undef DEBUG
24
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>
30#include <asm/mmu.h>
31#include <asm/page.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>
37#include <asm/tlb.h>
38#include <asm/prom.h>
39#include <asm/cputable.h>
40#include <asm/udbg.h>
41#include <asm/smp.h>
42#include <asm/trace.h>
43#include <asm/firmware.h>
44#include <asm/plpar_wrappers.h>
45
46#include "pseries.h"
47
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
54
55
56/* in hvCall.S */
57EXPORT_SYMBOL(plpar_hcall);
58EXPORT_SYMBOL(plpar_hcall9);
59EXPORT_SYMBOL(plpar_hcall_norets);
60
61extern void pSeries_find_serial_port(void);
62
63void vpa_init(int cpu)
64{
65	int hwcpu = get_hard_smp_processor_id(cpu);
66	unsigned long addr;
67	long ret;
68	struct paca_struct *pp;
69	struct dtl_entry *dtl;
70
71	/*
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.
74	 */
75	WARN_ON(cpu != smp_processor_id());
76
77	if (cpu_has_feature(CPU_FTR_ALTIVEC))
78		lppaca_of(cpu).vmxregs_in_use = 1;
79
80	if (cpu_has_feature(CPU_FTR_ARCH_207S))
81		lppaca_of(cpu).ebb_regs_in_use = 1;
82
83	addr = __pa(&lppaca_of(cpu));
84	ret = register_vpa(hwcpu, addr);
85
86	if (ret) {
87		pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
88		       "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
89		return;
90	}
91	/*
92	 * PAPR says this feature is SLB-Buffer but firmware never
93	 * reports that.  All SPLPAR support SLB shadow buffer.
94	 */
95	addr = __pa(paca[cpu].slb_shadow_ptr);
96	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
97		ret = register_slb_shadow(hwcpu, addr);
98		if (ret)
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);
102	}
103
104	/*
105	 * Register dispatch trace log, if one has been allocated.
106	 */
107	pp = &paca[cpu];
108	dtl = pp->dispatch_log;
109	if (dtl) {
110		pp->dtl_ridx = 0;
111		pp->dtl_curr = dtl;
112		lppaca_of(cpu).dtl_idx = 0;
113
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));
117		if (ret)
118			pr_err("WARNING: DTL registration of cpu %d (hw %d) "
119			       "failed with %ld\n", smp_processor_id(),
120			       hwcpu, ret);
121		lppaca_of(cpu).dtl_enable_mask = 2;
122	}
123}
124
125static 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)
129{
130	unsigned long lpar_rc;
131	unsigned long flags;
132	unsigned long slot;
133	unsigned long hpte_v, hpte_r;
134
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);
139
140	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
141	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
142
143	if (!(vflags & HPTE_V_BOLTED))
144		pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
145
146	/* Now fill in the actual HPTE */
147	/* Set CEC cookie to 0         */
148	/* Zero page = 0               */
149	/* I-cache Invalidate = 0      */
150	/* I-cache synchronize = 0     */
151	/* Exact = 0                   */
152	flags = 0;
153
154	/* Make pHyp happy */
155	if ((rflags & _PAGE_NO_CACHE) && !(rflags & _PAGE_WRITETHRU))
156		hpte_r &= ~HPTE_R_M;
157
158	if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
159		flags |= H_COALESCE_CAND;
160
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))
164			pr_devel(" full\n");
165		return -1;
166	}
167
168	/*
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.
172	 */
173	if (unlikely(lpar_rc != H_SUCCESS)) {
174		if (!(vflags & HPTE_V_BOLTED))
175			pr_devel(" lpar err %ld\n", lpar_rc);
176		return -2;
177	}
178	if (!(vflags & HPTE_V_BOLTED))
179		pr_devel(" -> slot: %lu\n", slot & 7);
180
181	/* Because of iSeries, we have to pass down the secondary
182	 * bucket bit here as well
183	 */
184	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
185}
186
187static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
188
189static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
190{
191	unsigned long slot_offset;
192	unsigned long lpar_rc;
193	int i;
194	unsigned long dummy1, dummy2;
195
196	/* pick a random slot to start at */
197	slot_offset = mftb() & 0x7;
198
199	for (i = 0; i < HPTES_PER_GROUP; i++) {
200
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)
205			return i;
206
207		/*
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.
211		 */
212		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
213
214		slot_offset++;
215		slot_offset &= 0x7;
216	}
217
218	return -1;
219}
220
221static void pSeries_lpar_hptab_clear(void)
222{
223	unsigned long size_bytes = 1UL << ppc64_pft_size;
224	unsigned long hpte_count = size_bytes >> 4;
225	struct {
226		unsigned long pteh;
227		unsigned long ptel;
228	} ptes[4];
229	long lpar_rc;
230	unsigned long i, j;
231
232	/* Read in batches of 4,
233	 * invalidate only valid entries not in the VRMA
234	 * hpte_count will be a multiple of 4
235         */
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)
239			continue;
240		for (j = 0; j < 4; j++){
241			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
242				HPTE_V_VRMA_MASK)
243				continue;
244			if (ptes[j].pteh & HPTE_V_VALID)
245				plpar_pte_remove_raw(0, i + j, 0,
246					&(ptes[j].pteh), &(ptes[j].ptel));
247		}
248	}
249
250#ifdef __LITTLE_ENDIAN__
251	/* Reset exceptions to big endian */
252	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
253		long rc;
254
255		rc = pseries_big_endian_exceptions();
256		/*
257		 * At this point it is unlikely panic() will get anything
258		 * out to the user, but at least this will stop us from
259		 * continuing on further and creating an even more
260		 * difficult to debug situation.
261		 */
262		if (rc)
263			panic("Could not enable big endian exceptions");
264	}
265#endif
266}
267
268/*
269 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
270 * the low 3 bits of flags happen to line up.  So no transform is needed.
271 * We can probably optimize here and assume the high bits of newpp are
272 * already zero.  For now I am paranoid.
273 */
274static long pSeries_lpar_hpte_updatepp(unsigned long slot,
275				       unsigned long newpp,
276				       unsigned long vpn,
277				       int psize, int apsize,
278				       int ssize, int local)
279{
280	unsigned long lpar_rc;
281	unsigned long flags = (newpp & 7) | H_AVPN;
282	unsigned long want_v;
283
284	want_v = hpte_encode_avpn(vpn, psize, ssize);
285
286	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
287		 want_v, slot, flags, psize);
288
289	lpar_rc = plpar_pte_protect(flags, slot, want_v);
290
291	if (lpar_rc == H_NOT_FOUND) {
292		pr_devel("not found !\n");
293		return -1;
294	}
295
296	pr_devel("ok\n");
297
298	BUG_ON(lpar_rc != H_SUCCESS);
299
300	return 0;
301}
302
303static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
304{
305	unsigned long dword0;
306	unsigned long lpar_rc;
307	unsigned long dummy_word1;
308	unsigned long flags;
309
310	/* Read 1 pte at a time                        */
311	/* Do not need RPN to logical page translation */
312	/* No cross CEC PFT access                     */
313	flags = 0;
314
315	lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
316
317	BUG_ON(lpar_rc != H_SUCCESS);
318
319	return dword0;
320}
321
322static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
323{
324	unsigned long hash;
325	unsigned long i;
326	long slot;
327	unsigned long want_v, hpte_v;
328
329	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
330	want_v = hpte_encode_avpn(vpn, psize, ssize);
331
332	/* Bolted entries are always in the primary group */
333	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
334	for (i = 0; i < HPTES_PER_GROUP; i++) {
335		hpte_v = pSeries_lpar_hpte_getword0(slot);
336
337		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
338			/* HPTE matches */
339			return slot;
340		++slot;
341	}
342
343	return -1;
344}
345
346static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
347					     unsigned long ea,
348					     int psize, int ssize)
349{
350	unsigned long vpn;
351	unsigned long lpar_rc, slot, vsid, flags;
352
353	vsid = get_kernel_vsid(ea, ssize);
354	vpn = hpt_vpn(ea, vsid, ssize);
355
356	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
357	BUG_ON(slot == -1);
358
359	flags = newpp & 7;
360	lpar_rc = plpar_pte_protect(flags, slot, 0);
361
362	BUG_ON(lpar_rc != H_SUCCESS);
363}
364
365static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
366					 int psize, int apsize,
367					 int ssize, int local)
368{
369	unsigned long want_v;
370	unsigned long lpar_rc;
371	unsigned long dummy1, dummy2;
372
373	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
374		 slot, vpn, psize, local);
375
376	want_v = hpte_encode_avpn(vpn, psize, ssize);
377	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
378	if (lpar_rc == H_NOT_FOUND)
379		return;
380
381	BUG_ON(lpar_rc != H_SUCCESS);
382}
383
384/*
385 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
386 * to make sure that we avoid bouncing the hypervisor tlbie lock.
387 */
388#define PPC64_HUGE_HPTE_BATCH 12
389
390static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
391					     unsigned long *vpn, int count,
392					     int psize, int ssize)
393{
394	unsigned long param[8];
395	int i = 0, pix = 0, rc;
396	unsigned long flags = 0;
397	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
398
399	if (lock_tlbie)
400		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
401
402	for (i = 0; i < count; i++) {
403
404		if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
405			pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
406						     ssize, 0);
407		} else {
408			param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
409			param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
410			pix += 2;
411			if (pix == 8) {
412				rc = plpar_hcall9(H_BULK_REMOVE, param,
413						  param[0], param[1], param[2],
414						  param[3], param[4], param[5],
415						  param[6], param[7]);
416				BUG_ON(rc != H_SUCCESS);
417				pix = 0;
418			}
419		}
420	}
421	if (pix) {
422		param[pix] = HBR_END;
423		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
424				  param[2], param[3], param[4], param[5],
425				  param[6], param[7]);
426		BUG_ON(rc != H_SUCCESS);
427	}
428
429	if (lock_tlbie)
430		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
431}
432
433static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
434				       unsigned char *hpte_slot_array,
435				       unsigned long addr, int psize)
436{
437	int ssize = 0, i, index = 0;
438	unsigned long s_addr = addr;
439	unsigned int max_hpte_count, valid;
440	unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
441	unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
442	unsigned long shift, hidx, vpn = 0, vsid, hash, slot;
443
444	shift = mmu_psize_defs[psize].shift;
445	max_hpte_count = 1U << (PMD_SHIFT - shift);
446
447	for (i = 0; i < max_hpte_count; i++) {
448		valid = hpte_valid(hpte_slot_array, i);
449		if (!valid)
450			continue;
451		hidx =  hpte_hash_index(hpte_slot_array, i);
452
453		/* get the vpn */
454		addr = s_addr + (i * (1ul << shift));
455		if (!is_kernel_addr(addr)) {
456			ssize = user_segment_size(addr);
457			vsid = get_vsid(mm->context.id, addr, ssize);
458			WARN_ON(vsid == 0);
459		} else {
460			vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
461			ssize = mmu_kernel_ssize;
462		}
463
464		vpn = hpt_vpn(addr, vsid, ssize);
465		hash = hpt_hash(vpn, shift, ssize);
466		if (hidx & _PTEIDX_SECONDARY)
467			hash = ~hash;
468
469		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
470		slot += hidx & _PTEIDX_GROUP_IX;
471
472		slot_array[index] = slot;
473		vpn_array[index] = vpn;
474		if (index == PPC64_HUGE_HPTE_BATCH - 1) {
475			/*
476			 * Now do a bluk invalidate
477			 */
478			__pSeries_lpar_hugepage_invalidate(slot_array,
479							   vpn_array,
480							   PPC64_HUGE_HPTE_BATCH,
481							   psize, ssize);
482			index = 0;
483		} else
484			index++;
485	}
486	if (index)
487		__pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
488						   index, psize, ssize);
489}
490
491static void pSeries_lpar_hpte_removebolted(unsigned long ea,
492					   int psize, int ssize)
493{
494	unsigned long vpn;
495	unsigned long slot, vsid;
496
497	vsid = get_kernel_vsid(ea, ssize);
498	vpn = hpt_vpn(ea, vsid, ssize);
499
500	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
501	BUG_ON(slot == -1);
502	/*
503	 * lpar doesn't use the passed actual page size
504	 */
505	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
506}
507
508/*
509 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
510 * lock.
511 */
512static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
513{
514	unsigned long vpn;
515	unsigned long i, pix, rc;
516	unsigned long flags = 0;
517	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
518	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
519	unsigned long param[9];
520	unsigned long hash, index, shift, hidx, slot;
521	real_pte_t pte;
522	int psize, ssize;
523
524	if (lock_tlbie)
525		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
526
527	psize = batch->psize;
528	ssize = batch->ssize;
529	pix = 0;
530	for (i = 0; i < number; i++) {
531		vpn = batch->vpn[i];
532		pte = batch->pte[i];
533		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
534			hash = hpt_hash(vpn, shift, ssize);
535			hidx = __rpte_to_hidx(pte, index);
536			if (hidx & _PTEIDX_SECONDARY)
537				hash = ~hash;
538			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
539			slot += hidx & _PTEIDX_GROUP_IX;
540			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
541				/*
542				 * lpar doesn't use the passed actual page size
543				 */
544				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
545							     0, ssize, local);
546			} else {
547				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
548				param[pix+1] = hpte_encode_avpn(vpn, psize,
549								ssize);
550				pix += 2;
551				if (pix == 8) {
552					rc = plpar_hcall9(H_BULK_REMOVE, param,
553						param[0], param[1], param[2],
554						param[3], param[4], param[5],
555						param[6], param[7]);
556					BUG_ON(rc != H_SUCCESS);
557					pix = 0;
558				}
559			}
560		} pte_iterate_hashed_end();
561	}
562	if (pix) {
563		param[pix] = HBR_END;
564		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
565				  param[2], param[3], param[4], param[5],
566				  param[6], param[7]);
567		BUG_ON(rc != H_SUCCESS);
568	}
569
570	if (lock_tlbie)
571		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
572}
573
574static int __init disable_bulk_remove(char *str)
575{
576	if (strcmp(str, "off") == 0 &&
577	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
578			printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
579			powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
580	}
581	return 1;
582}
583
584__setup("bulk_remove=", disable_bulk_remove);
585
586void __init hpte_init_lpar(void)
587{
588	ppc_md.hpte_invalidate	= pSeries_lpar_hpte_invalidate;
589	ppc_md.hpte_updatepp	= pSeries_lpar_hpte_updatepp;
590	ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
591	ppc_md.hpte_insert	= pSeries_lpar_hpte_insert;
592	ppc_md.hpte_remove	= pSeries_lpar_hpte_remove;
593	ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
594	ppc_md.flush_hash_range	= pSeries_lpar_flush_hash_range;
595	ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
596	ppc_md.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
597}
598
599#ifdef CONFIG_PPC_SMLPAR
600#define CMO_FREE_HINT_DEFAULT 1
601static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
602
603static int __init cmo_free_hint(char *str)
604{
605	char *parm;
606	parm = strstrip(str);
607
608	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
609		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
610		cmo_free_hint_flag = 0;
611		return 1;
612	}
613
614	cmo_free_hint_flag = 1;
615	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
616
617	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
618		return 1;
619
620	return 0;
621}
622
623__setup("cmo_free_hint=", cmo_free_hint);
624
625static void pSeries_set_page_state(struct page *page, int order,
626				   unsigned long state)
627{
628	int i, j;
629	unsigned long cmo_page_sz, addr;
630
631	cmo_page_sz = cmo_get_page_size();
632	addr = __pa((unsigned long)page_address(page));
633
634	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
635		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
636			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
637	}
638}
639
640void arch_free_page(struct page *page, int order)
641{
642	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
643		return;
644
645	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
646}
647EXPORT_SYMBOL(arch_free_page);
648
649#endif
650
651#ifdef CONFIG_TRACEPOINTS
652/*
653 * We optimise our hcall path by placing hcall_tracepoint_refcount
654 * directly in the TOC so we can check if the hcall tracepoints are
655 * enabled via a single load.
656 */
657
658/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
659extern long hcall_tracepoint_refcount;
660
661/*
662 * Since the tracing code might execute hcalls we need to guard against
663 * recursion. One example of this are spinlocks calling H_YIELD on
664 * shared processor partitions.
665 */
666static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
667
668void hcall_tracepoint_regfunc(void)
669{
670	hcall_tracepoint_refcount++;
671}
672
673void hcall_tracepoint_unregfunc(void)
674{
675	hcall_tracepoint_refcount--;
676}
677
678void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
679{
680	unsigned long flags;
681	unsigned int *depth;
682
683	/*
684	 * We cannot call tracepoints inside RCU idle regions which
685	 * means we must not trace H_CEDE.
686	 */
687	if (opcode == H_CEDE)
688		return;
689
690	local_irq_save(flags);
691
692	depth = &__get_cpu_var(hcall_trace_depth);
693
694	if (*depth)
695		goto out;
696
697	(*depth)++;
698	preempt_disable();
699	trace_hcall_entry(opcode, args);
700	(*depth)--;
701
702out:
703	local_irq_restore(flags);
704}
705
706void __trace_hcall_exit(long opcode, unsigned long retval,
707			unsigned long *retbuf)
708{
709	unsigned long flags;
710	unsigned int *depth;
711
712	if (opcode == H_CEDE)
713		return;
714
715	local_irq_save(flags);
716
717	depth = &__get_cpu_var(hcall_trace_depth);
718
719	if (*depth)
720		goto out;
721
722	(*depth)++;
723	trace_hcall_exit(opcode, retval, retbuf);
724	preempt_enable();
725	(*depth)--;
726
727out:
728	local_irq_restore(flags);
729}
730#endif
731
732/**
733 * h_get_mpp
734 * H_GET_MPP hcall returns info in 7 parms
735 */
736int h_get_mpp(struct hvcall_mpp_data *mpp_data)
737{
738	int rc;
739	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
740
741	rc = plpar_hcall9(H_GET_MPP, retbuf);
742
743	mpp_data->entitled_mem = retbuf[0];
744	mpp_data->mapped_mem = retbuf[1];
745
746	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
747	mpp_data->pool_num = retbuf[2] & 0xffff;
748
749	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
750	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
751	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
752
753	mpp_data->pool_size = retbuf[4];
754	mpp_data->loan_request = retbuf[5];
755	mpp_data->backing_mem = retbuf[6];
756
757	return rc;
758}
759EXPORT_SYMBOL(h_get_mpp);
760
761int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
762{
763	int rc;
764	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
765
766	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
767
768	mpp_x_data->coalesced_bytes = retbuf[0];
769	mpp_x_data->pool_coalesced_bytes = retbuf[1];
770	mpp_x_data->pool_purr_cycles = retbuf[2];
771	mpp_x_data->pool_spurr_cycles = retbuf[3];
772
773	return rc;
774}
775