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 */
57 EXPORT_SYMBOL(plpar_hcall);
58 EXPORT_SYMBOL(plpar_hcall9);
59 EXPORT_SYMBOL(plpar_hcall_norets);
60 
61 extern void pSeries_find_serial_port(void);
62 
63 void 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(&slb_shadow[cpu]);
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 
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)
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 &= ~_PAGE_COHERENT;
157 	if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
158 		flags |= H_COALESCE_CAND;
159 
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))
163 			pr_devel(" full\n");
164 		return -1;
165 	}
166 
167 	/*
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.
171 	 */
172 	if (unlikely(lpar_rc != H_SUCCESS)) {
173 		if (!(vflags & HPTE_V_BOLTED))
174 			pr_devel(" lpar err %ld\n", lpar_rc);
175 		return -2;
176 	}
177 	if (!(vflags & HPTE_V_BOLTED))
178 		pr_devel(" -> slot: %lu\n", slot & 7);
179 
180 	/* Because of iSeries, we have to pass down the secondary
181 	 * bucket bit here as well
182 	 */
183 	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
184 }
185 
186 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
187 
188 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
189 {
190 	unsigned long slot_offset;
191 	unsigned long lpar_rc;
192 	int i;
193 	unsigned long dummy1, dummy2;
194 
195 	/* pick a random slot to start at */
196 	slot_offset = mftb() & 0x7;
197 
198 	for (i = 0; i < HPTES_PER_GROUP; i++) {
199 
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)
204 			return i;
205 
206 		/*
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.
210 		 */
211 		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
212 
213 		slot_offset++;
214 		slot_offset &= 0x7;
215 	}
216 
217 	return -1;
218 }
219 
220 static void pSeries_lpar_hptab_clear(void)
221 {
222 	unsigned long size_bytes = 1UL << ppc64_pft_size;
223 	unsigned long hpte_count = size_bytes >> 4;
224 	struct {
225 		unsigned long pteh;
226 		unsigned long ptel;
227 	} ptes[4];
228 	long lpar_rc;
229 	unsigned long i, j;
230 
231 	/* Read in batches of 4,
232 	 * invalidate only valid entries not in the VRMA
233 	 * hpte_count will be a multiple of 4
234          */
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)
238 			continue;
239 		for (j = 0; j < 4; j++){
240 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
241 				HPTE_V_VRMA_MASK)
242 				continue;
243 			if (ptes[j].pteh & HPTE_V_VALID)
244 				plpar_pte_remove_raw(0, i + j, 0,
245 					&(ptes[j].pteh), &(ptes[j].ptel));
246 		}
247 	}
248 }
249 
250 /*
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.
255  */
256 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
257 				       unsigned long newpp,
258 				       unsigned long vpn,
259 				       int psize, int apsize,
260 				       int ssize, int local)
261 {
262 	unsigned long lpar_rc;
263 	unsigned long flags = (newpp & 7) | H_AVPN;
264 	unsigned long want_v;
265 
266 	want_v = hpte_encode_avpn(vpn, psize, ssize);
267 
268 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
269 		 want_v, slot, flags, psize);
270 
271 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
272 
273 	if (lpar_rc == H_NOT_FOUND) {
274 		pr_devel("not found !\n");
275 		return -1;
276 	}
277 
278 	pr_devel("ok\n");
279 
280 	BUG_ON(lpar_rc != H_SUCCESS);
281 
282 	return 0;
283 }
284 
285 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
286 {
287 	unsigned long dword0;
288 	unsigned long lpar_rc;
289 	unsigned long dummy_word1;
290 	unsigned long flags;
291 
292 	/* Read 1 pte at a time                        */
293 	/* Do not need RPN to logical page translation */
294 	/* No cross CEC PFT access                     */
295 	flags = 0;
296 
297 	lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
298 
299 	BUG_ON(lpar_rc != H_SUCCESS);
300 
301 	return dword0;
302 }
303 
304 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
305 {
306 	unsigned long hash;
307 	unsigned long i;
308 	long slot;
309 	unsigned long want_v, hpte_v;
310 
311 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
312 	want_v = hpte_encode_avpn(vpn, psize, ssize);
313 
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);
318 
319 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
320 			/* HPTE matches */
321 			return slot;
322 		++slot;
323 	}
324 
325 	return -1;
326 }
327 
328 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
329 					     unsigned long ea,
330 					     int psize, int ssize)
331 {
332 	unsigned long vpn;
333 	unsigned long lpar_rc, slot, vsid, flags;
334 
335 	vsid = get_kernel_vsid(ea, ssize);
336 	vpn = hpt_vpn(ea, vsid, ssize);
337 
338 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
339 	BUG_ON(slot == -1);
340 
341 	flags = newpp & 7;
342 	lpar_rc = plpar_pte_protect(flags, slot, 0);
343 
344 	BUG_ON(lpar_rc != H_SUCCESS);
345 }
346 
347 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
348 					 int psize, int apsize,
349 					 int ssize, int local)
350 {
351 	unsigned long want_v;
352 	unsigned long lpar_rc;
353 	unsigned long dummy1, dummy2;
354 
355 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
356 		 slot, vpn, psize, local);
357 
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)
361 		return;
362 
363 	BUG_ON(lpar_rc != H_SUCCESS);
364 }
365 
366 /*
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.
369  */
370 #define PPC64_HUGE_HPTE_BATCH 12
371 
372 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
373 					     unsigned long *vpn, int count,
374 					     int psize, int ssize)
375 {
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);
380 
381 	if (lock_tlbie)
382 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
383 
384 	for (i = 0; i < count; i++) {
385 
386 		if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
387 			pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
388 						     ssize, 0);
389 		} else {
390 			param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
391 			param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
392 			pix += 2;
393 			if (pix == 8) {
394 				rc = plpar_hcall9(H_BULK_REMOVE, param,
395 						  param[0], param[1], param[2],
396 						  param[3], param[4], param[5],
397 						  param[6], param[7]);
398 				BUG_ON(rc != H_SUCCESS);
399 				pix = 0;
400 			}
401 		}
402 	}
403 	if (pix) {
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],
407 				  param[6], param[7]);
408 		BUG_ON(rc != H_SUCCESS);
409 	}
410 
411 	if (lock_tlbie)
412 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
413 }
414 
415 static void pSeries_lpar_hugepage_invalidate(struct mm_struct *mm,
416 				       unsigned char *hpte_slot_array,
417 				       unsigned long addr, int psize)
418 {
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;
425 
426 	shift = mmu_psize_defs[psize].shift;
427 	max_hpte_count = 1U << (PMD_SHIFT - shift);
428 
429 	for (i = 0; i < max_hpte_count; i++) {
430 		valid = hpte_valid(hpte_slot_array, i);
431 		if (!valid)
432 			continue;
433 		hidx =  hpte_hash_index(hpte_slot_array, i);
434 
435 		/* get the vpn */
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);
440 			WARN_ON(vsid == 0);
441 		} else {
442 			vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
443 			ssize = mmu_kernel_ssize;
444 		}
445 
446 		vpn = hpt_vpn(addr, vsid, ssize);
447 		hash = hpt_hash(vpn, shift, ssize);
448 		if (hidx & _PTEIDX_SECONDARY)
449 			hash = ~hash;
450 
451 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
452 		slot += hidx & _PTEIDX_GROUP_IX;
453 
454 		slot_array[index] = slot;
455 		vpn_array[index] = vpn;
456 		if (index == PPC64_HUGE_HPTE_BATCH - 1) {
457 			/*
458 			 * Now do a bluk invalidate
459 			 */
460 			__pSeries_lpar_hugepage_invalidate(slot_array,
461 							   vpn_array,
462 							   PPC64_HUGE_HPTE_BATCH,
463 							   psize, ssize);
464 			index = 0;
465 		} else
466 			index++;
467 	}
468 	if (index)
469 		__pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
470 						   index, psize, ssize);
471 }
472 
473 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
474 					   int psize, int ssize)
475 {
476 	unsigned long vpn;
477 	unsigned long slot, vsid;
478 
479 	vsid = get_kernel_vsid(ea, ssize);
480 	vpn = hpt_vpn(ea, vsid, ssize);
481 
482 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
483 	BUG_ON(slot == -1);
484 	/*
485 	 * lpar doesn't use the passed actual page size
486 	 */
487 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
488 }
489 
490 /*
491  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
492  * lock.
493  */
494 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
495 {
496 	unsigned long vpn;
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;
503 	real_pte_t pte;
504 	int psize, ssize;
505 
506 	if (lock_tlbie)
507 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
508 
509 	psize = batch->psize;
510 	ssize = batch->ssize;
511 	pix = 0;
512 	for (i = 0; i < number; i++) {
513 		vpn = batch->vpn[i];
514 		pte = batch->pte[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)
519 				hash = ~hash;
520 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
521 			slot += hidx & _PTEIDX_GROUP_IX;
522 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
523 				/*
524 				 * lpar doesn't use the passed actual page size
525 				 */
526 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
527 							     0, ssize, local);
528 			} else {
529 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
530 				param[pix+1] = hpte_encode_avpn(vpn, psize,
531 								ssize);
532 				pix += 2;
533 				if (pix == 8) {
534 					rc = plpar_hcall9(H_BULK_REMOVE, param,
535 						param[0], param[1], param[2],
536 						param[3], param[4], param[5],
537 						param[6], param[7]);
538 					BUG_ON(rc != H_SUCCESS);
539 					pix = 0;
540 				}
541 			}
542 		} pte_iterate_hashed_end();
543 	}
544 	if (pix) {
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],
548 				  param[6], param[7]);
549 		BUG_ON(rc != H_SUCCESS);
550 	}
551 
552 	if (lock_tlbie)
553 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
554 }
555 
556 static int __init disable_bulk_remove(char *str)
557 {
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;
562 	}
563 	return 1;
564 }
565 
566 __setup("bulk_remove=", disable_bulk_remove);
567 
568 void __init hpte_init_lpar(void)
569 {
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;
579 }
580 
581 #ifdef CONFIG_PPC_SMLPAR
582 #define CMO_FREE_HINT_DEFAULT 1
583 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
584 
585 static int __init cmo_free_hint(char *str)
586 {
587 	char *parm;
588 	parm = strstrip(str);
589 
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;
593 		return 1;
594 	}
595 
596 	cmo_free_hint_flag = 1;
597 	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
598 
599 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
600 		return 1;
601 
602 	return 0;
603 }
604 
605 __setup("cmo_free_hint=", cmo_free_hint);
606 
607 static void pSeries_set_page_state(struct page *page, int order,
608 				   unsigned long state)
609 {
610 	int i, j;
611 	unsigned long cmo_page_sz, addr;
612 
613 	cmo_page_sz = cmo_get_page_size();
614 	addr = __pa((unsigned long)page_address(page));
615 
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);
619 	}
620 }
621 
622 void arch_free_page(struct page *page, int order)
623 {
624 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
625 		return;
626 
627 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
628 }
629 EXPORT_SYMBOL(arch_free_page);
630 
631 #endif
632 
633 #ifdef CONFIG_TRACEPOINTS
634 /*
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.
638  */
639 
640 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
641 extern long hcall_tracepoint_refcount;
642 
643 /*
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.
647  */
648 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
649 
650 void hcall_tracepoint_regfunc(void)
651 {
652 	hcall_tracepoint_refcount++;
653 }
654 
655 void hcall_tracepoint_unregfunc(void)
656 {
657 	hcall_tracepoint_refcount--;
658 }
659 
660 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
661 {
662 	unsigned long flags;
663 	unsigned int *depth;
664 
665 	/*
666 	 * We cannot call tracepoints inside RCU idle regions which
667 	 * means we must not trace H_CEDE.
668 	 */
669 	if (opcode == H_CEDE)
670 		return;
671 
672 	local_irq_save(flags);
673 
674 	depth = &__get_cpu_var(hcall_trace_depth);
675 
676 	if (*depth)
677 		goto out;
678 
679 	(*depth)++;
680 	preempt_disable();
681 	trace_hcall_entry(opcode, args);
682 	(*depth)--;
683 
684 out:
685 	local_irq_restore(flags);
686 }
687 
688 void __trace_hcall_exit(long opcode, unsigned long retval,
689 			unsigned long *retbuf)
690 {
691 	unsigned long flags;
692 	unsigned int *depth;
693 
694 	if (opcode == H_CEDE)
695 		return;
696 
697 	local_irq_save(flags);
698 
699 	depth = &__get_cpu_var(hcall_trace_depth);
700 
701 	if (*depth)
702 		goto out;
703 
704 	(*depth)++;
705 	trace_hcall_exit(opcode, retval, retbuf);
706 	preempt_enable();
707 	(*depth)--;
708 
709 out:
710 	local_irq_restore(flags);
711 }
712 #endif
713 
714 /**
715  * h_get_mpp
716  * H_GET_MPP hcall returns info in 7 parms
717  */
718 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
719 {
720 	int rc;
721 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
722 
723 	rc = plpar_hcall9(H_GET_MPP, retbuf);
724 
725 	mpp_data->entitled_mem = retbuf[0];
726 	mpp_data->mapped_mem = retbuf[1];
727 
728 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
729 	mpp_data->pool_num = retbuf[2] & 0xffff;
730 
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;
734 
735 	mpp_data->pool_size = retbuf[4];
736 	mpp_data->loan_request = retbuf[5];
737 	mpp_data->backing_mem = retbuf[6];
738 
739 	return rc;
740 }
741 EXPORT_SYMBOL(h_get_mpp);
742 
743 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
744 {
745 	int rc;
746 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
747 
748 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
749 
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];
754 
755 	return rc;
756 }
757