xref: /linux/arch/powerpc/platforms/pseries/lpar.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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 <linux/jump_label.h>
30 #include <linux/delay.h>
31 #include <linux/stop_machine.h>
32 #include <asm/processor.h>
33 #include <asm/mmu.h>
34 #include <asm/page.h>
35 #include <asm/pgtable.h>
36 #include <asm/machdep.h>
37 #include <asm/mmu_context.h>
38 #include <asm/iommu.h>
39 #include <asm/tlbflush.h>
40 #include <asm/tlb.h>
41 #include <asm/prom.h>
42 #include <asm/cputable.h>
43 #include <asm/udbg.h>
44 #include <asm/smp.h>
45 #include <asm/trace.h>
46 #include <asm/firmware.h>
47 #include <asm/plpar_wrappers.h>
48 #include <asm/kexec.h>
49 #include <asm/fadump.h>
50 #include <asm/asm-prototypes.h>
51 
52 #include "pseries.h"
53 
54 /* Flag bits for H_BULK_REMOVE */
55 #define HBR_REQUEST	0x4000000000000000UL
56 #define HBR_RESPONSE	0x8000000000000000UL
57 #define HBR_END		0xc000000000000000UL
58 #define HBR_AVPN	0x0200000000000000UL
59 #define HBR_ANDCOND	0x0100000000000000UL
60 
61 
62 /* in hvCall.S */
63 EXPORT_SYMBOL(plpar_hcall);
64 EXPORT_SYMBOL(plpar_hcall9);
65 EXPORT_SYMBOL(plpar_hcall_norets);
66 
67 void vpa_init(int cpu)
68 {
69 	int hwcpu = get_hard_smp_processor_id(cpu);
70 	unsigned long addr;
71 	long ret;
72 	struct paca_struct *pp;
73 	struct dtl_entry *dtl;
74 
75 	/*
76 	 * The spec says it "may be problematic" if CPU x registers the VPA of
77 	 * CPU y. We should never do that, but wail if we ever do.
78 	 */
79 	WARN_ON(cpu != smp_processor_id());
80 
81 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
82 		lppaca_of(cpu).vmxregs_in_use = 1;
83 
84 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
85 		lppaca_of(cpu).ebb_regs_in_use = 1;
86 
87 	addr = __pa(&lppaca_of(cpu));
88 	ret = register_vpa(hwcpu, addr);
89 
90 	if (ret) {
91 		pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
92 		       "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
93 		return;
94 	}
95 
96 #ifdef CONFIG_PPC_STD_MMU_64
97 	/*
98 	 * PAPR says this feature is SLB-Buffer but firmware never
99 	 * reports that.  All SPLPAR support SLB shadow buffer.
100 	 */
101 	if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
102 		addr = __pa(paca[cpu].slb_shadow_ptr);
103 		ret = register_slb_shadow(hwcpu, addr);
104 		if (ret)
105 			pr_err("WARNING: SLB shadow buffer registration for "
106 			       "cpu %d (hw %d) of area %lx failed with %ld\n",
107 			       cpu, hwcpu, addr, ret);
108 	}
109 #endif /* CONFIG_PPC_STD_MMU_64 */
110 
111 	/*
112 	 * Register dispatch trace log, if one has been allocated.
113 	 */
114 	pp = &paca[cpu];
115 	dtl = pp->dispatch_log;
116 	if (dtl) {
117 		pp->dtl_ridx = 0;
118 		pp->dtl_curr = dtl;
119 		lppaca_of(cpu).dtl_idx = 0;
120 
121 		/* hypervisor reads buffer length from this field */
122 		dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
123 		ret = register_dtl(hwcpu, __pa(dtl));
124 		if (ret)
125 			pr_err("WARNING: DTL registration of cpu %d (hw %d) "
126 			       "failed with %ld\n", smp_processor_id(),
127 			       hwcpu, ret);
128 		lppaca_of(cpu).dtl_enable_mask = 2;
129 	}
130 }
131 
132 #ifdef CONFIG_PPC_STD_MMU_64
133 
134 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
135 				     unsigned long vpn, unsigned long pa,
136 				     unsigned long rflags, unsigned long vflags,
137 				     int psize, int apsize, int ssize)
138 {
139 	unsigned long lpar_rc;
140 	unsigned long flags;
141 	unsigned long slot;
142 	unsigned long hpte_v, hpte_r;
143 
144 	if (!(vflags & HPTE_V_BOLTED))
145 		pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
146 			 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
147 			 hpte_group, vpn,  pa, rflags, vflags, psize);
148 
149 	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
150 	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
151 
152 	if (!(vflags & HPTE_V_BOLTED))
153 		pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
154 
155 	/* Now fill in the actual HPTE */
156 	/* Set CEC cookie to 0         */
157 	/* Zero page = 0               */
158 	/* I-cache Invalidate = 0      */
159 	/* I-cache synchronize = 0     */
160 	/* Exact = 0                   */
161 	flags = 0;
162 
163 	if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
164 		flags |= H_COALESCE_CAND;
165 
166 	lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
167 	if (unlikely(lpar_rc == H_PTEG_FULL)) {
168 		if (!(vflags & HPTE_V_BOLTED))
169 			pr_devel(" full\n");
170 		return -1;
171 	}
172 
173 	/*
174 	 * Since we try and ioremap PHBs we don't own, the pte insert
175 	 * will fail. However we must catch the failure in hash_page
176 	 * or we will loop forever, so return -2 in this case.
177 	 */
178 	if (unlikely(lpar_rc != H_SUCCESS)) {
179 		if (!(vflags & HPTE_V_BOLTED))
180 			pr_devel(" lpar err %ld\n", lpar_rc);
181 		return -2;
182 	}
183 	if (!(vflags & HPTE_V_BOLTED))
184 		pr_devel(" -> slot: %lu\n", slot & 7);
185 
186 	/* Because of iSeries, we have to pass down the secondary
187 	 * bucket bit here as well
188 	 */
189 	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
190 }
191 
192 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
193 
194 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
195 {
196 	unsigned long slot_offset;
197 	unsigned long lpar_rc;
198 	int i;
199 	unsigned long dummy1, dummy2;
200 
201 	/* pick a random slot to start at */
202 	slot_offset = mftb() & 0x7;
203 
204 	for (i = 0; i < HPTES_PER_GROUP; i++) {
205 
206 		/* don't remove a bolted entry */
207 		lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
208 					   (0x1UL << 4), &dummy1, &dummy2);
209 		if (lpar_rc == H_SUCCESS)
210 			return i;
211 
212 		/*
213 		 * The test for adjunct partition is performed before the
214 		 * ANDCOND test.  H_RESOURCE may be returned, so we need to
215 		 * check for that as well.
216 		 */
217 		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
218 
219 		slot_offset++;
220 		slot_offset &= 0x7;
221 	}
222 
223 	return -1;
224 }
225 
226 static void manual_hpte_clear_all(void)
227 {
228 	unsigned long size_bytes = 1UL << ppc64_pft_size;
229 	unsigned long hpte_count = size_bytes >> 4;
230 	struct {
231 		unsigned long pteh;
232 		unsigned long ptel;
233 	} ptes[4];
234 	long lpar_rc;
235 	unsigned long i, j;
236 
237 	/* Read in batches of 4,
238 	 * invalidate only valid entries not in the VRMA
239 	 * hpte_count will be a multiple of 4
240          */
241 	for (i = 0; i < hpte_count; i += 4) {
242 		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
243 		if (lpar_rc != H_SUCCESS)
244 			continue;
245 		for (j = 0; j < 4; j++){
246 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
247 				HPTE_V_VRMA_MASK)
248 				continue;
249 			if (ptes[j].pteh & HPTE_V_VALID)
250 				plpar_pte_remove_raw(0, i + j, 0,
251 					&(ptes[j].pteh), &(ptes[j].ptel));
252 		}
253 	}
254 }
255 
256 static int hcall_hpte_clear_all(void)
257 {
258 	int rc;
259 
260 	do {
261 		rc = plpar_hcall_norets(H_CLEAR_HPT);
262 	} while (rc == H_CONTINUE);
263 
264 	return rc;
265 }
266 
267 static void pseries_hpte_clear_all(void)
268 {
269 	int rc;
270 
271 	rc = hcall_hpte_clear_all();
272 	if (rc != H_SUCCESS)
273 		manual_hpte_clear_all();
274 
275 #ifdef __LITTLE_ENDIAN__
276 	/*
277 	 * Reset exceptions to big endian.
278 	 *
279 	 * FIXME this is a hack for kexec, we need to reset the exception
280 	 * endian before starting the new kernel and this is a convenient place
281 	 * to do it.
282 	 *
283 	 * This is also called on boot when a fadump happens. In that case we
284 	 * must not change the exception endian mode.
285 	 */
286 	if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
287 		pseries_big_endian_exceptions();
288 #endif
289 }
290 
291 /*
292  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
293  * the low 3 bits of flags happen to line up.  So no transform is needed.
294  * We can probably optimize here and assume the high bits of newpp are
295  * already zero.  For now I am paranoid.
296  */
297 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
298 				       unsigned long newpp,
299 				       unsigned long vpn,
300 				       int psize, int apsize,
301 				       int ssize, unsigned long inv_flags)
302 {
303 	unsigned long lpar_rc;
304 	unsigned long flags = (newpp & 7) | H_AVPN;
305 	unsigned long want_v;
306 
307 	want_v = hpte_encode_avpn(vpn, psize, ssize);
308 
309 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
310 		 want_v, slot, flags, psize);
311 
312 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
313 
314 	if (lpar_rc == H_NOT_FOUND) {
315 		pr_devel("not found !\n");
316 		return -1;
317 	}
318 
319 	pr_devel("ok\n");
320 
321 	BUG_ON(lpar_rc != H_SUCCESS);
322 
323 	return 0;
324 }
325 
326 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
327 {
328 	long lpar_rc;
329 	unsigned long i, j;
330 	struct {
331 		unsigned long pteh;
332 		unsigned long ptel;
333 	} ptes[4];
334 
335 	for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
336 
337 		lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
338 		if (lpar_rc != H_SUCCESS)
339 			continue;
340 
341 		for (j = 0; j < 4; j++) {
342 			if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
343 			    (ptes[j].pteh & HPTE_V_VALID))
344 				return i + j;
345 		}
346 	}
347 
348 	return -1;
349 }
350 
351 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
352 {
353 	long slot;
354 	unsigned long hash;
355 	unsigned long want_v;
356 	unsigned long hpte_group;
357 
358 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
359 	want_v = hpte_encode_avpn(vpn, psize, ssize);
360 
361 	/* Bolted entries are always in the primary group */
362 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
363 	slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
364 	if (slot < 0)
365 		return -1;
366 	return hpte_group + slot;
367 }
368 
369 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
370 					     unsigned long ea,
371 					     int psize, int ssize)
372 {
373 	unsigned long vpn;
374 	unsigned long lpar_rc, slot, vsid, flags;
375 
376 	vsid = get_kernel_vsid(ea, ssize);
377 	vpn = hpt_vpn(ea, vsid, ssize);
378 
379 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
380 	BUG_ON(slot == -1);
381 
382 	flags = newpp & 7;
383 	lpar_rc = plpar_pte_protect(flags, slot, 0);
384 
385 	BUG_ON(lpar_rc != H_SUCCESS);
386 }
387 
388 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
389 					 int psize, int apsize,
390 					 int ssize, int local)
391 {
392 	unsigned long want_v;
393 	unsigned long lpar_rc;
394 	unsigned long dummy1, dummy2;
395 
396 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
397 		 slot, vpn, psize, local);
398 
399 	want_v = hpte_encode_avpn(vpn, psize, ssize);
400 	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
401 	if (lpar_rc == H_NOT_FOUND)
402 		return;
403 
404 	BUG_ON(lpar_rc != H_SUCCESS);
405 }
406 
407 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
408 /*
409  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
410  * to make sure that we avoid bouncing the hypervisor tlbie lock.
411  */
412 #define PPC64_HUGE_HPTE_BATCH 12
413 
414 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
415 					     unsigned long *vpn, int count,
416 					     int psize, int ssize)
417 {
418 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
419 	int i = 0, pix = 0, rc;
420 	unsigned long flags = 0;
421 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
422 
423 	if (lock_tlbie)
424 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
425 
426 	for (i = 0; i < count; i++) {
427 
428 		if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
429 			pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
430 						     ssize, 0);
431 		} else {
432 			param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
433 			param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
434 			pix += 2;
435 			if (pix == 8) {
436 				rc = plpar_hcall9(H_BULK_REMOVE, param,
437 						  param[0], param[1], param[2],
438 						  param[3], param[4], param[5],
439 						  param[6], param[7]);
440 				BUG_ON(rc != H_SUCCESS);
441 				pix = 0;
442 			}
443 		}
444 	}
445 	if (pix) {
446 		param[pix] = HBR_END;
447 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
448 				  param[2], param[3], param[4], param[5],
449 				  param[6], param[7]);
450 		BUG_ON(rc != H_SUCCESS);
451 	}
452 
453 	if (lock_tlbie)
454 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
455 }
456 
457 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
458 					     unsigned long addr,
459 					     unsigned char *hpte_slot_array,
460 					     int psize, int ssize, int local)
461 {
462 	int i, index = 0;
463 	unsigned long s_addr = addr;
464 	unsigned int max_hpte_count, valid;
465 	unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
466 	unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
467 	unsigned long shift, hidx, vpn = 0, hash, slot;
468 
469 	shift = mmu_psize_defs[psize].shift;
470 	max_hpte_count = 1U << (PMD_SHIFT - shift);
471 
472 	for (i = 0; i < max_hpte_count; i++) {
473 		valid = hpte_valid(hpte_slot_array, i);
474 		if (!valid)
475 			continue;
476 		hidx =  hpte_hash_index(hpte_slot_array, i);
477 
478 		/* get the vpn */
479 		addr = s_addr + (i * (1ul << shift));
480 		vpn = hpt_vpn(addr, vsid, ssize);
481 		hash = hpt_hash(vpn, shift, ssize);
482 		if (hidx & _PTEIDX_SECONDARY)
483 			hash = ~hash;
484 
485 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
486 		slot += hidx & _PTEIDX_GROUP_IX;
487 
488 		slot_array[index] = slot;
489 		vpn_array[index] = vpn;
490 		if (index == PPC64_HUGE_HPTE_BATCH - 1) {
491 			/*
492 			 * Now do a bluk invalidate
493 			 */
494 			__pSeries_lpar_hugepage_invalidate(slot_array,
495 							   vpn_array,
496 							   PPC64_HUGE_HPTE_BATCH,
497 							   psize, ssize);
498 			index = 0;
499 		} else
500 			index++;
501 	}
502 	if (index)
503 		__pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
504 						   index, psize, ssize);
505 }
506 #else
507 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
508 					     unsigned long addr,
509 					     unsigned char *hpte_slot_array,
510 					     int psize, int ssize, int local)
511 {
512 	WARN(1, "%s called without THP support\n", __func__);
513 }
514 #endif
515 
516 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
517 					  int psize, int ssize)
518 {
519 	unsigned long vpn;
520 	unsigned long slot, vsid;
521 
522 	vsid = get_kernel_vsid(ea, ssize);
523 	vpn = hpt_vpn(ea, vsid, ssize);
524 
525 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
526 	if (slot == -1)
527 		return -ENOENT;
528 
529 	/*
530 	 * lpar doesn't use the passed actual page size
531 	 */
532 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
533 	return 0;
534 }
535 
536 /*
537  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
538  * lock.
539  */
540 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
541 {
542 	unsigned long vpn;
543 	unsigned long i, pix, rc;
544 	unsigned long flags = 0;
545 	struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
546 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
547 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
548 	unsigned long hash, index, shift, hidx, slot;
549 	real_pte_t pte;
550 	int psize, ssize;
551 
552 	if (lock_tlbie)
553 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
554 
555 	psize = batch->psize;
556 	ssize = batch->ssize;
557 	pix = 0;
558 	for (i = 0; i < number; i++) {
559 		vpn = batch->vpn[i];
560 		pte = batch->pte[i];
561 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
562 			hash = hpt_hash(vpn, shift, ssize);
563 			hidx = __rpte_to_hidx(pte, index);
564 			if (hidx & _PTEIDX_SECONDARY)
565 				hash = ~hash;
566 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
567 			slot += hidx & _PTEIDX_GROUP_IX;
568 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
569 				/*
570 				 * lpar doesn't use the passed actual page size
571 				 */
572 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
573 							     0, ssize, local);
574 			} else {
575 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
576 				param[pix+1] = hpte_encode_avpn(vpn, psize,
577 								ssize);
578 				pix += 2;
579 				if (pix == 8) {
580 					rc = plpar_hcall9(H_BULK_REMOVE, param,
581 						param[0], param[1], param[2],
582 						param[3], param[4], param[5],
583 						param[6], param[7]);
584 					BUG_ON(rc != H_SUCCESS);
585 					pix = 0;
586 				}
587 			}
588 		} pte_iterate_hashed_end();
589 	}
590 	if (pix) {
591 		param[pix] = HBR_END;
592 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
593 				  param[2], param[3], param[4], param[5],
594 				  param[6], param[7]);
595 		BUG_ON(rc != H_SUCCESS);
596 	}
597 
598 	if (lock_tlbie)
599 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
600 }
601 
602 static int __init disable_bulk_remove(char *str)
603 {
604 	if (strcmp(str, "off") == 0 &&
605 	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
606 			printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
607 			powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
608 	}
609 	return 1;
610 }
611 
612 __setup("bulk_remove=", disable_bulk_remove);
613 
614 #define HPT_RESIZE_TIMEOUT	10000 /* ms */
615 
616 struct hpt_resize_state {
617 	unsigned long shift;
618 	int commit_rc;
619 };
620 
621 static int pseries_lpar_resize_hpt_commit(void *data)
622 {
623 	struct hpt_resize_state *state = data;
624 
625 	state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
626 	if (state->commit_rc != H_SUCCESS)
627 		return -EIO;
628 
629 	/* Hypervisor has transitioned the HTAB, update our globals */
630 	ppc64_pft_size = state->shift;
631 	htab_size_bytes = 1UL << ppc64_pft_size;
632 	htab_hash_mask = (htab_size_bytes >> 7) - 1;
633 
634 	return 0;
635 }
636 
637 /* Must be called in user context */
638 static int pseries_lpar_resize_hpt(unsigned long shift)
639 {
640 	struct hpt_resize_state state = {
641 		.shift = shift,
642 		.commit_rc = H_FUNCTION,
643 	};
644 	unsigned int delay, total_delay = 0;
645 	int rc;
646 	ktime_t t0, t1, t2;
647 
648 	might_sleep();
649 
650 	if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
651 		return -ENODEV;
652 
653 	printk(KERN_INFO "lpar: Attempting to resize HPT to shift %lu\n",
654 	       shift);
655 
656 	t0 = ktime_get();
657 
658 	rc = plpar_resize_hpt_prepare(0, shift);
659 	while (H_IS_LONG_BUSY(rc)) {
660 		delay = get_longbusy_msecs(rc);
661 		total_delay += delay;
662 		if (total_delay > HPT_RESIZE_TIMEOUT) {
663 			/* prepare with shift==0 cancels an in-progress resize */
664 			rc = plpar_resize_hpt_prepare(0, 0);
665 			if (rc != H_SUCCESS)
666 				printk(KERN_WARNING
667 				       "lpar: Unexpected error %d cancelling timed out HPT resize\n",
668 				       rc);
669 			return -ETIMEDOUT;
670 		}
671 		msleep(delay);
672 		rc = plpar_resize_hpt_prepare(0, shift);
673 	};
674 
675 	switch (rc) {
676 	case H_SUCCESS:
677 		/* Continue on */
678 		break;
679 
680 	case H_PARAMETER:
681 		return -EINVAL;
682 	case H_RESOURCE:
683 		return -EPERM;
684 	default:
685 		printk(KERN_WARNING
686 		       "lpar: Unexpected error %d from H_RESIZE_HPT_PREPARE\n",
687 		       rc);
688 		return -EIO;
689 	}
690 
691 	t1 = ktime_get();
692 
693 	rc = stop_machine(pseries_lpar_resize_hpt_commit, &state, NULL);
694 
695 	t2 = ktime_get();
696 
697 	if (rc != 0) {
698 		switch (state.commit_rc) {
699 		case H_PTEG_FULL:
700 			printk(KERN_WARNING
701 			       "lpar: Hash collision while resizing HPT\n");
702 			return -ENOSPC;
703 
704 		default:
705 			printk(KERN_WARNING
706 			       "lpar: Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
707 			       state.commit_rc);
708 			return -EIO;
709 		};
710 	}
711 
712 	printk(KERN_INFO
713 	       "lpar: HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
714 	       shift, (long long) ktime_ms_delta(t1, t0),
715 	       (long long) ktime_ms_delta(t2, t1));
716 
717 	return 0;
718 }
719 
720 /* Actually only used for radix, so far */
721 static int pseries_lpar_register_process_table(unsigned long base,
722 			unsigned long page_size, unsigned long table_size)
723 {
724 	long rc;
725 	unsigned long flags = PROC_TABLE_NEW;
726 
727 	if (radix_enabled())
728 		flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
729 	for (;;) {
730 		rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
731 					page_size, table_size);
732 		if (!H_IS_LONG_BUSY(rc))
733 			break;
734 		mdelay(get_longbusy_msecs(rc));
735 	}
736 	if (rc != H_SUCCESS) {
737 		pr_err("Failed to register process table (rc=%ld)\n", rc);
738 		BUG();
739 	}
740 	return rc;
741 }
742 
743 void __init hpte_init_pseries(void)
744 {
745 	mmu_hash_ops.hpte_invalidate	 = pSeries_lpar_hpte_invalidate;
746 	mmu_hash_ops.hpte_updatepp	 = pSeries_lpar_hpte_updatepp;
747 	mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
748 	mmu_hash_ops.hpte_insert	 = pSeries_lpar_hpte_insert;
749 	mmu_hash_ops.hpte_remove	 = pSeries_lpar_hpte_remove;
750 	mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
751 	mmu_hash_ops.flush_hash_range	 = pSeries_lpar_flush_hash_range;
752 	mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
753 	mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
754 
755 	if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
756 		mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
757 }
758 
759 void radix_init_pseries(void)
760 {
761 	pr_info("Using radix MMU under hypervisor\n");
762 	register_process_table = pseries_lpar_register_process_table;
763 }
764 
765 #ifdef CONFIG_PPC_SMLPAR
766 #define CMO_FREE_HINT_DEFAULT 1
767 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
768 
769 static int __init cmo_free_hint(char *str)
770 {
771 	char *parm;
772 	parm = strstrip(str);
773 
774 	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
775 		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
776 		cmo_free_hint_flag = 0;
777 		return 1;
778 	}
779 
780 	cmo_free_hint_flag = 1;
781 	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
782 
783 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
784 		return 1;
785 
786 	return 0;
787 }
788 
789 __setup("cmo_free_hint=", cmo_free_hint);
790 
791 static void pSeries_set_page_state(struct page *page, int order,
792 				   unsigned long state)
793 {
794 	int i, j;
795 	unsigned long cmo_page_sz, addr;
796 
797 	cmo_page_sz = cmo_get_page_size();
798 	addr = __pa((unsigned long)page_address(page));
799 
800 	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
801 		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
802 			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
803 	}
804 }
805 
806 void arch_free_page(struct page *page, int order)
807 {
808 	if (radix_enabled())
809 		return;
810 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
811 		return;
812 
813 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
814 }
815 EXPORT_SYMBOL(arch_free_page);
816 
817 #endif /* CONFIG_PPC_SMLPAR */
818 #endif /* CONFIG_PPC_STD_MMU_64 */
819 
820 #ifdef CONFIG_TRACEPOINTS
821 #ifdef HAVE_JUMP_LABEL
822 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
823 
824 int hcall_tracepoint_regfunc(void)
825 {
826 	static_key_slow_inc(&hcall_tracepoint_key);
827 	return 0;
828 }
829 
830 void hcall_tracepoint_unregfunc(void)
831 {
832 	static_key_slow_dec(&hcall_tracepoint_key);
833 }
834 #else
835 /*
836  * We optimise our hcall path by placing hcall_tracepoint_refcount
837  * directly in the TOC so we can check if the hcall tracepoints are
838  * enabled via a single load.
839  */
840 
841 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
842 extern long hcall_tracepoint_refcount;
843 
844 int hcall_tracepoint_regfunc(void)
845 {
846 	hcall_tracepoint_refcount++;
847 	return 0;
848 }
849 
850 void hcall_tracepoint_unregfunc(void)
851 {
852 	hcall_tracepoint_refcount--;
853 }
854 #endif
855 
856 /*
857  * Since the tracing code might execute hcalls we need to guard against
858  * recursion. One example of this are spinlocks calling H_YIELD on
859  * shared processor partitions.
860  */
861 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
862 
863 
864 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
865 {
866 	unsigned long flags;
867 	unsigned int *depth;
868 
869 	/*
870 	 * We cannot call tracepoints inside RCU idle regions which
871 	 * means we must not trace H_CEDE.
872 	 */
873 	if (opcode == H_CEDE)
874 		return;
875 
876 	local_irq_save(flags);
877 
878 	depth = this_cpu_ptr(&hcall_trace_depth);
879 
880 	if (*depth)
881 		goto out;
882 
883 	(*depth)++;
884 	preempt_disable();
885 	trace_hcall_entry(opcode, args);
886 	(*depth)--;
887 
888 out:
889 	local_irq_restore(flags);
890 }
891 
892 void __trace_hcall_exit(long opcode, unsigned long retval,
893 			unsigned long *retbuf)
894 {
895 	unsigned long flags;
896 	unsigned int *depth;
897 
898 	if (opcode == H_CEDE)
899 		return;
900 
901 	local_irq_save(flags);
902 
903 	depth = this_cpu_ptr(&hcall_trace_depth);
904 
905 	if (*depth)
906 		goto out;
907 
908 	(*depth)++;
909 	trace_hcall_exit(opcode, retval, retbuf);
910 	preempt_enable();
911 	(*depth)--;
912 
913 out:
914 	local_irq_restore(flags);
915 }
916 #endif
917 
918 /**
919  * h_get_mpp
920  * H_GET_MPP hcall returns info in 7 parms
921  */
922 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
923 {
924 	int rc;
925 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
926 
927 	rc = plpar_hcall9(H_GET_MPP, retbuf);
928 
929 	mpp_data->entitled_mem = retbuf[0];
930 	mpp_data->mapped_mem = retbuf[1];
931 
932 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
933 	mpp_data->pool_num = retbuf[2] & 0xffff;
934 
935 	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
936 	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
937 	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
938 
939 	mpp_data->pool_size = retbuf[4];
940 	mpp_data->loan_request = retbuf[5];
941 	mpp_data->backing_mem = retbuf[6];
942 
943 	return rc;
944 }
945 EXPORT_SYMBOL(h_get_mpp);
946 
947 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
948 {
949 	int rc;
950 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
951 
952 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
953 
954 	mpp_x_data->coalesced_bytes = retbuf[0];
955 	mpp_x_data->pool_coalesced_bytes = retbuf[1];
956 	mpp_x_data->pool_purr_cycles = retbuf[2];
957 	mpp_x_data->pool_spurr_cycles = retbuf[3];
958 
959 	return rc;
960 }
961