lpar.c revision 57cdfdf829a850a317425ed93c6a576c9ee6329c
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 <asm/processor.h>
29#include <asm/mmu.h>
30#include <asm/page.h>
31#include <asm/pgtable.h>
32#include <asm/machdep.h>
33#include <asm/abs_addr.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
44#include "plpar_wrappers.h"
45#include "pseries.h"
46
47
48/* in hvCall.S */
49EXPORT_SYMBOL(plpar_hcall);
50EXPORT_SYMBOL(plpar_hcall9);
51EXPORT_SYMBOL(plpar_hcall_norets);
52
53extern void pSeries_find_serial_port(void);
54
55
56static int vtermno;	/* virtual terminal# for udbg  */
57
58#define __ALIGNED__ __attribute__((__aligned__(sizeof(long))))
59static void udbg_hvsi_putc(char c)
60{
61	/* packet's seqno isn't used anyways */
62	uint8_t packet[] __ALIGNED__ = { 0xff, 5, 0, 0, c };
63	int rc;
64
65	if (c == '\n')
66		udbg_hvsi_putc('\r');
67
68	do {
69		rc = plpar_put_term_char(vtermno, sizeof(packet), packet);
70	} while (rc == H_BUSY);
71}
72
73static long hvsi_udbg_buf_len;
74static uint8_t hvsi_udbg_buf[256];
75
76static int udbg_hvsi_getc_poll(void)
77{
78	unsigned char ch;
79	int rc, i;
80
81	if (hvsi_udbg_buf_len == 0) {
82		rc = plpar_get_term_char(vtermno, &hvsi_udbg_buf_len, hvsi_udbg_buf);
83		if (rc != H_SUCCESS || hvsi_udbg_buf[0] != 0xff) {
84			/* bad read or non-data packet */
85			hvsi_udbg_buf_len = 0;
86		} else {
87			/* remove the packet header */
88			for (i = 4; i < hvsi_udbg_buf_len; i++)
89				hvsi_udbg_buf[i-4] = hvsi_udbg_buf[i];
90			hvsi_udbg_buf_len -= 4;
91		}
92	}
93
94	if (hvsi_udbg_buf_len <= 0 || hvsi_udbg_buf_len > 256) {
95		/* no data ready */
96		hvsi_udbg_buf_len = 0;
97		return -1;
98	}
99
100	ch = hvsi_udbg_buf[0];
101	/* shift remaining data down */
102	for (i = 1; i < hvsi_udbg_buf_len; i++) {
103		hvsi_udbg_buf[i-1] = hvsi_udbg_buf[i];
104	}
105	hvsi_udbg_buf_len--;
106
107	return ch;
108}
109
110static int udbg_hvsi_getc(void)
111{
112	int ch;
113	for (;;) {
114		ch = udbg_hvsi_getc_poll();
115		if (ch == -1) {
116			/* This shouldn't be needed...but... */
117			volatile unsigned long delay;
118			for (delay=0; delay < 2000000; delay++)
119				;
120		} else {
121			return ch;
122		}
123	}
124}
125
126static void udbg_putcLP(char c)
127{
128	char buf[16];
129	unsigned long rc;
130
131	if (c == '\n')
132		udbg_putcLP('\r');
133
134	buf[0] = c;
135	do {
136		rc = plpar_put_term_char(vtermno, 1, buf);
137	} while(rc == H_BUSY);
138}
139
140/* Buffered chars getc */
141static long inbuflen;
142static long inbuf[2];	/* must be 2 longs */
143
144static int udbg_getc_pollLP(void)
145{
146	/* The interface is tricky because it may return up to 16 chars.
147	 * We save them statically for future calls to udbg_getc().
148	 */
149	char ch, *buf = (char *)inbuf;
150	int i;
151	long rc;
152	if (inbuflen == 0) {
153		/* get some more chars. */
154		inbuflen = 0;
155		rc = plpar_get_term_char(vtermno, &inbuflen, buf);
156		if (rc != H_SUCCESS)
157			inbuflen = 0;	/* otherwise inbuflen is garbage */
158	}
159	if (inbuflen <= 0 || inbuflen > 16) {
160		/* Catch error case as well as other oddities (corruption) */
161		inbuflen = 0;
162		return -1;
163	}
164	ch = buf[0];
165	for (i = 1; i < inbuflen; i++)	/* shuffle them down. */
166		buf[i-1] = buf[i];
167	inbuflen--;
168	return ch;
169}
170
171static int udbg_getcLP(void)
172{
173	int ch;
174	for (;;) {
175		ch = udbg_getc_pollLP();
176		if (ch == -1) {
177			/* This shouldn't be needed...but... */
178			volatile unsigned long delay;
179			for (delay=0; delay < 2000000; delay++)
180				;
181		} else {
182			return ch;
183		}
184	}
185}
186
187/* call this from early_init() for a working debug console on
188 * vterm capable LPAR machines
189 */
190void __init udbg_init_debug_lpar(void)
191{
192	vtermno = 0;
193	udbg_putc = udbg_putcLP;
194	udbg_getc = udbg_getcLP;
195	udbg_getc_poll = udbg_getc_pollLP;
196
197	register_early_udbg_console();
198}
199
200/* returns 0 if couldn't find or use /chosen/stdout as console */
201void __init find_udbg_vterm(void)
202{
203	struct device_node *stdout_node;
204	const u32 *termno;
205	const char *name;
206
207	/* find the boot console from /chosen/stdout */
208	if (!of_chosen)
209		return;
210	name = of_get_property(of_chosen, "linux,stdout-path", NULL);
211	if (name == NULL)
212		return;
213	stdout_node = of_find_node_by_path(name);
214	if (!stdout_node)
215		return;
216	name = of_get_property(stdout_node, "name", NULL);
217	if (!name) {
218		printk(KERN_WARNING "stdout node missing 'name' property!\n");
219		goto out;
220	}
221
222	/* Check if it's a virtual terminal */
223	if (strncmp(name, "vty", 3) != 0)
224		goto out;
225	termno = of_get_property(stdout_node, "reg", NULL);
226	if (termno == NULL)
227		goto out;
228	vtermno = termno[0];
229
230	if (of_device_is_compatible(stdout_node, "hvterm1")) {
231		udbg_putc = udbg_putcLP;
232		udbg_getc = udbg_getcLP;
233		udbg_getc_poll = udbg_getc_pollLP;
234		add_preferred_console("hvc", termno[0] & 0xff, NULL);
235	} else if (of_device_is_compatible(stdout_node, "hvterm-protocol")) {
236		vtermno = termno[0];
237		udbg_putc = udbg_hvsi_putc;
238		udbg_getc = udbg_hvsi_getc;
239		udbg_getc_poll = udbg_hvsi_getc_poll;
240		add_preferred_console("hvsi", termno[0] & 0xff, NULL);
241	}
242out:
243	of_node_put(stdout_node);
244}
245
246void vpa_init(int cpu)
247{
248	int hwcpu = get_hard_smp_processor_id(cpu);
249	unsigned long addr;
250	long ret;
251	struct paca_struct *pp;
252	struct dtl_entry *dtl;
253
254	if (cpu_has_feature(CPU_FTR_ALTIVEC))
255		lppaca_of(cpu).vmxregs_in_use = 1;
256
257	addr = __pa(&lppaca_of(cpu));
258	ret = register_vpa(hwcpu, addr);
259
260	if (ret) {
261		printk(KERN_ERR "WARNING: vpa_init: VPA registration for "
262				"cpu %d (hw %d) of area %lx returns %ld\n",
263				cpu, hwcpu, addr, ret);
264		return;
265	}
266	/*
267	 * PAPR says this feature is SLB-Buffer but firmware never
268	 * reports that.  All SPLPAR support SLB shadow buffer.
269	 */
270	addr = __pa(&slb_shadow[cpu]);
271	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
272		ret = register_slb_shadow(hwcpu, addr);
273		if (ret)
274			printk(KERN_ERR
275			       "WARNING: vpa_init: SLB shadow buffer "
276			       "registration for cpu %d (hw %d) of area %lx "
277			       "returns %ld\n", cpu, hwcpu, addr, ret);
278	}
279
280	/*
281	 * Register dispatch trace log, if one has been allocated.
282	 */
283	pp = &paca[cpu];
284	dtl = pp->dispatch_log;
285	if (dtl) {
286		pp->dtl_ridx = 0;
287		pp->dtl_curr = dtl;
288		lppaca_of(cpu).dtl_idx = 0;
289
290		/* hypervisor reads buffer length from this field */
291		dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
292		ret = register_dtl(hwcpu, __pa(dtl));
293		if (ret)
294			pr_warn("DTL registration failed for cpu %d (%ld)\n",
295				cpu, ret);
296		lppaca_of(cpu).dtl_enable_mask = 2;
297	}
298}
299
300static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
301 			      unsigned long va, unsigned long pa,
302 			      unsigned long rflags, unsigned long vflags,
303			      int psize, int ssize)
304{
305	unsigned long lpar_rc;
306	unsigned long flags;
307	unsigned long slot;
308	unsigned long hpte_v, hpte_r;
309
310	if (!(vflags & HPTE_V_BOLTED))
311		pr_devel("hpte_insert(group=%lx, va=%016lx, pa=%016lx, "
312			 "rflags=%lx, vflags=%lx, psize=%d)\n",
313			 hpte_group, va, pa, rflags, vflags, psize);
314
315	hpte_v = hpte_encode_v(va, psize, ssize) | vflags | HPTE_V_VALID;
316	hpte_r = hpte_encode_r(pa, psize) | rflags;
317
318	if (!(vflags & HPTE_V_BOLTED))
319		pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
320
321	/* Now fill in the actual HPTE */
322	/* Set CEC cookie to 0         */
323	/* Zero page = 0               */
324	/* I-cache Invalidate = 0      */
325	/* I-cache synchronize = 0     */
326	/* Exact = 0                   */
327	flags = 0;
328
329	/* Make pHyp happy */
330	if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
331		hpte_r &= ~_PAGE_COHERENT;
332
333	lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
334	if (unlikely(lpar_rc == H_PTEG_FULL)) {
335		if (!(vflags & HPTE_V_BOLTED))
336			pr_devel(" full\n");
337		return -1;
338	}
339
340	/*
341	 * Since we try and ioremap PHBs we don't own, the pte insert
342	 * will fail. However we must catch the failure in hash_page
343	 * or we will loop forever, so return -2 in this case.
344	 */
345	if (unlikely(lpar_rc != H_SUCCESS)) {
346		if (!(vflags & HPTE_V_BOLTED))
347			pr_devel(" lpar err %lu\n", lpar_rc);
348		return -2;
349	}
350	if (!(vflags & HPTE_V_BOLTED))
351		pr_devel(" -> slot: %lu\n", slot & 7);
352
353	/* Because of iSeries, we have to pass down the secondary
354	 * bucket bit here as well
355	 */
356	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
357}
358
359static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
360
361static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
362{
363	unsigned long slot_offset;
364	unsigned long lpar_rc;
365	int i;
366	unsigned long dummy1, dummy2;
367
368	/* pick a random slot to start at */
369	slot_offset = mftb() & 0x7;
370
371	for (i = 0; i < HPTES_PER_GROUP; i++) {
372
373		/* don't remove a bolted entry */
374		lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
375					   (0x1UL << 4), &dummy1, &dummy2);
376		if (lpar_rc == H_SUCCESS)
377			return i;
378		BUG_ON(lpar_rc != H_NOT_FOUND);
379
380		slot_offset++;
381		slot_offset &= 0x7;
382	}
383
384	return -1;
385}
386
387static void pSeries_lpar_hptab_clear(void)
388{
389	unsigned long size_bytes = 1UL << ppc64_pft_size;
390	unsigned long hpte_count = size_bytes >> 4;
391	struct {
392		unsigned long pteh;
393		unsigned long ptel;
394	} ptes[4];
395	long lpar_rc;
396	int i, j;
397
398	/* Read in batches of 4,
399	 * invalidate only valid entries not in the VRMA
400	 * hpte_count will be a multiple of 4
401         */
402	for (i = 0; i < hpte_count; i += 4) {
403		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
404		if (lpar_rc != H_SUCCESS)
405			continue;
406		for (j = 0; j < 4; j++){
407			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
408				HPTE_V_VRMA_MASK)
409				continue;
410			if (ptes[j].pteh & HPTE_V_VALID)
411				plpar_pte_remove_raw(0, i + j, 0,
412					&(ptes[j].pteh), &(ptes[j].ptel));
413		}
414	}
415}
416
417/*
418 * This computes the AVPN and B fields of the first dword of a HPTE,
419 * for use when we want to match an existing PTE.  The bottom 7 bits
420 * of the returned value are zero.
421 */
422static inline unsigned long hpte_encode_avpn(unsigned long va, int psize,
423					     int ssize)
424{
425	unsigned long v;
426
427	v = (va >> 23) & ~(mmu_psize_defs[psize].avpnm);
428	v <<= HPTE_V_AVPN_SHIFT;
429	v |= ((unsigned long) ssize) << HPTE_V_SSIZE_SHIFT;
430	return v;
431}
432
433/*
434 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
435 * the low 3 bits of flags happen to line up.  So no transform is needed.
436 * We can probably optimize here and assume the high bits of newpp are
437 * already zero.  For now I am paranoid.
438 */
439static long pSeries_lpar_hpte_updatepp(unsigned long slot,
440				       unsigned long newpp,
441				       unsigned long va,
442				       int psize, int ssize, int local)
443{
444	unsigned long lpar_rc;
445	unsigned long flags = (newpp & 7) | H_AVPN;
446	unsigned long want_v;
447
448	want_v = hpte_encode_avpn(va, psize, ssize);
449
450	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
451		 want_v, slot, flags, psize);
452
453	lpar_rc = plpar_pte_protect(flags, slot, want_v);
454
455	if (lpar_rc == H_NOT_FOUND) {
456		pr_devel("not found !\n");
457		return -1;
458	}
459
460	pr_devel("ok\n");
461
462	BUG_ON(lpar_rc != H_SUCCESS);
463
464	return 0;
465}
466
467static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
468{
469	unsigned long dword0;
470	unsigned long lpar_rc;
471	unsigned long dummy_word1;
472	unsigned long flags;
473
474	/* Read 1 pte at a time                        */
475	/* Do not need RPN to logical page translation */
476	/* No cross CEC PFT access                     */
477	flags = 0;
478
479	lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
480
481	BUG_ON(lpar_rc != H_SUCCESS);
482
483	return dword0;
484}
485
486static long pSeries_lpar_hpte_find(unsigned long va, int psize, int ssize)
487{
488	unsigned long hash;
489	unsigned long i;
490	long slot;
491	unsigned long want_v, hpte_v;
492
493	hash = hpt_hash(va, mmu_psize_defs[psize].shift, ssize);
494	want_v = hpte_encode_avpn(va, psize, ssize);
495
496	/* Bolted entries are always in the primary group */
497	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
498	for (i = 0; i < HPTES_PER_GROUP; i++) {
499		hpte_v = pSeries_lpar_hpte_getword0(slot);
500
501		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
502			/* HPTE matches */
503			return slot;
504		++slot;
505	}
506
507	return -1;
508}
509
510static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
511					     unsigned long ea,
512					     int psize, int ssize)
513{
514	unsigned long lpar_rc, slot, vsid, va, flags;
515
516	vsid = get_kernel_vsid(ea, ssize);
517	va = hpt_va(ea, vsid, ssize);
518
519	slot = pSeries_lpar_hpte_find(va, psize, ssize);
520	BUG_ON(slot == -1);
521
522	flags = newpp & 7;
523	lpar_rc = plpar_pte_protect(flags, slot, 0);
524
525	BUG_ON(lpar_rc != H_SUCCESS);
526}
527
528static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long va,
529					 int psize, int ssize, int local)
530{
531	unsigned long want_v;
532	unsigned long lpar_rc;
533	unsigned long dummy1, dummy2;
534
535	pr_devel("    inval : slot=%lx, va=%016lx, psize: %d, local: %d\n",
536		 slot, va, psize, local);
537
538	want_v = hpte_encode_avpn(va, psize, ssize);
539	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
540	if (lpar_rc == H_NOT_FOUND)
541		return;
542
543	BUG_ON(lpar_rc != H_SUCCESS);
544}
545
546static void pSeries_lpar_hpte_removebolted(unsigned long ea,
547					   int psize, int ssize)
548{
549	unsigned long slot, vsid, va;
550
551	vsid = get_kernel_vsid(ea, ssize);
552	va = hpt_va(ea, vsid, ssize);
553
554	slot = pSeries_lpar_hpte_find(va, psize, ssize);
555	BUG_ON(slot == -1);
556
557	pSeries_lpar_hpte_invalidate(slot, va, psize, ssize, 0);
558}
559
560/* Flag bits for H_BULK_REMOVE */
561#define HBR_REQUEST	0x4000000000000000UL
562#define HBR_RESPONSE	0x8000000000000000UL
563#define HBR_END		0xc000000000000000UL
564#define HBR_AVPN	0x0200000000000000UL
565#define HBR_ANDCOND	0x0100000000000000UL
566
567/*
568 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
569 * lock.
570 */
571static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
572{
573	unsigned long i, pix, rc;
574	unsigned long flags = 0;
575	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
576	int lock_tlbie = !cpu_has_feature(CPU_FTR_LOCKLESS_TLBIE);
577	unsigned long param[9];
578	unsigned long va;
579	unsigned long hash, index, shift, hidx, slot;
580	real_pte_t pte;
581	int psize, ssize;
582
583	if (lock_tlbie)
584		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
585
586	psize = batch->psize;
587	ssize = batch->ssize;
588	pix = 0;
589	for (i = 0; i < number; i++) {
590		va = batch->vaddr[i];
591		pte = batch->pte[i];
592		pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
593			hash = hpt_hash(va, shift, ssize);
594			hidx = __rpte_to_hidx(pte, index);
595			if (hidx & _PTEIDX_SECONDARY)
596				hash = ~hash;
597			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
598			slot += hidx & _PTEIDX_GROUP_IX;
599			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
600				pSeries_lpar_hpte_invalidate(slot, va, psize,
601							     ssize, local);
602			} else {
603				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
604				param[pix+1] = hpte_encode_avpn(va, psize,
605								ssize);
606				pix += 2;
607				if (pix == 8) {
608					rc = plpar_hcall9(H_BULK_REMOVE, param,
609						param[0], param[1], param[2],
610						param[3], param[4], param[5],
611						param[6], param[7]);
612					BUG_ON(rc != H_SUCCESS);
613					pix = 0;
614				}
615			}
616		} pte_iterate_hashed_end();
617	}
618	if (pix) {
619		param[pix] = HBR_END;
620		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
621				  param[2], param[3], param[4], param[5],
622				  param[6], param[7]);
623		BUG_ON(rc != H_SUCCESS);
624	}
625
626	if (lock_tlbie)
627		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
628}
629
630static int __init disable_bulk_remove(char *str)
631{
632	if (strcmp(str, "off") == 0 &&
633	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
634			printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
635			powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
636	}
637	return 1;
638}
639
640__setup("bulk_remove=", disable_bulk_remove);
641
642void __init hpte_init_lpar(void)
643{
644	ppc_md.hpte_invalidate	= pSeries_lpar_hpte_invalidate;
645	ppc_md.hpte_updatepp	= pSeries_lpar_hpte_updatepp;
646	ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
647	ppc_md.hpte_insert	= pSeries_lpar_hpte_insert;
648	ppc_md.hpte_remove	= pSeries_lpar_hpte_remove;
649	ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
650	ppc_md.flush_hash_range	= pSeries_lpar_flush_hash_range;
651	ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
652}
653
654#ifdef CONFIG_PPC_SMLPAR
655#define CMO_FREE_HINT_DEFAULT 1
656static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
657
658static int __init cmo_free_hint(char *str)
659{
660	char *parm;
661	parm = strstrip(str);
662
663	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
664		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
665		cmo_free_hint_flag = 0;
666		return 1;
667	}
668
669	cmo_free_hint_flag = 1;
670	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
671
672	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
673		return 1;
674
675	return 0;
676}
677
678__setup("cmo_free_hint=", cmo_free_hint);
679
680static void pSeries_set_page_state(struct page *page, int order,
681				   unsigned long state)
682{
683	int i, j;
684	unsigned long cmo_page_sz, addr;
685
686	cmo_page_sz = cmo_get_page_size();
687	addr = __pa((unsigned long)page_address(page));
688
689	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
690		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
691			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
692	}
693}
694
695void arch_free_page(struct page *page, int order)
696{
697	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
698		return;
699
700	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
701}
702EXPORT_SYMBOL(arch_free_page);
703
704#endif
705
706#ifdef CONFIG_TRACEPOINTS
707/*
708 * We optimise our hcall path by placing hcall_tracepoint_refcount
709 * directly in the TOC so we can check if the hcall tracepoints are
710 * enabled via a single load.
711 */
712
713/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
714extern long hcall_tracepoint_refcount;
715
716/*
717 * Since the tracing code might execute hcalls we need to guard against
718 * recursion. One example of this are spinlocks calling H_YIELD on
719 * shared processor partitions.
720 */
721static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
722
723void hcall_tracepoint_regfunc(void)
724{
725	hcall_tracepoint_refcount++;
726}
727
728void hcall_tracepoint_unregfunc(void)
729{
730	hcall_tracepoint_refcount--;
731}
732
733void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
734{
735	unsigned long flags;
736	unsigned int *depth;
737
738	local_irq_save(flags);
739
740	depth = &__get_cpu_var(hcall_trace_depth);
741
742	if (*depth)
743		goto out;
744
745	(*depth)++;
746	trace_hcall_entry(opcode, args);
747	(*depth)--;
748
749out:
750	local_irq_restore(flags);
751}
752
753void __trace_hcall_exit(long opcode, unsigned long retval,
754			unsigned long *retbuf)
755{
756	unsigned long flags;
757	unsigned int *depth;
758
759	local_irq_save(flags);
760
761	depth = &__get_cpu_var(hcall_trace_depth);
762
763	if (*depth)
764		goto out;
765
766	(*depth)++;
767	trace_hcall_exit(opcode, retval, retbuf);
768	(*depth)--;
769
770out:
771	local_irq_restore(flags);
772}
773#endif
774