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 #define pr_fmt(fmt) "lpar: " fmt
25 
26 #include <linux/kernel.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/console.h>
29 #include <linux/export.h>
30 #include <linux/jump_label.h>
31 #include <linux/delay.h>
32 #include <linux/stop_machine.h>
33 #include <asm/processor.h>
34 #include <asm/mmu.h>
35 #include <asm/page.h>
36 #include <asm/pgtable.h>
37 #include <asm/machdep.h>
38 #include <asm/mmu_context.h>
39 #include <asm/iommu.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_BOOK3S_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_ptrs[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_BOOK3S_64 */
110 
111 	/*
112 	 * Register dispatch trace log, if one has been allocated.
113 	 */
114 	pp = paca_ptrs[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_BOOK3S_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 		pr_devel("Hash table group is full\n");
169 		return -1;
170 	}
171 
172 	/*
173 	 * Since we try and ioremap PHBs we don't own, the pte insert
174 	 * will fail. However we must catch the failure in hash_page
175 	 * or we will loop forever, so return -2 in this case.
176 	 */
177 	if (unlikely(lpar_rc != H_SUCCESS)) {
178 		pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
179 		return -2;
180 	}
181 	if (!(vflags & HPTE_V_BOLTED))
182 		pr_devel(" -> slot: %lu\n", slot & 7);
183 
184 	/* Because of iSeries, we have to pass down the secondary
185 	 * bucket bit here as well
186 	 */
187 	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
188 }
189 
190 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
191 
192 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
193 {
194 	unsigned long slot_offset;
195 	unsigned long lpar_rc;
196 	int i;
197 	unsigned long dummy1, dummy2;
198 
199 	/* pick a random slot to start at */
200 	slot_offset = mftb() & 0x7;
201 
202 	for (i = 0; i < HPTES_PER_GROUP; i++) {
203 
204 		/* don't remove a bolted entry */
205 		lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
206 					   (0x1UL << 4), &dummy1, &dummy2);
207 		if (lpar_rc == H_SUCCESS)
208 			return i;
209 
210 		/*
211 		 * The test for adjunct partition is performed before the
212 		 * ANDCOND test.  H_RESOURCE may be returned, so we need to
213 		 * check for that as well.
214 		 */
215 		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
216 
217 		slot_offset++;
218 		slot_offset &= 0x7;
219 	}
220 
221 	return -1;
222 }
223 
224 static void manual_hpte_clear_all(void)
225 {
226 	unsigned long size_bytes = 1UL << ppc64_pft_size;
227 	unsigned long hpte_count = size_bytes >> 4;
228 	struct {
229 		unsigned long pteh;
230 		unsigned long ptel;
231 	} ptes[4];
232 	long lpar_rc;
233 	unsigned long i, j;
234 
235 	/* Read in batches of 4,
236 	 * invalidate only valid entries not in the VRMA
237 	 * hpte_count will be a multiple of 4
238          */
239 	for (i = 0; i < hpte_count; i += 4) {
240 		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
241 		if (lpar_rc != H_SUCCESS) {
242 			pr_info("Failed to read hash page table at %ld err %ld\n",
243 				i, lpar_rc);
244 			continue;
245 		}
246 		for (j = 0; j < 4; j++){
247 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
248 				HPTE_V_VRMA_MASK)
249 				continue;
250 			if (ptes[j].pteh & HPTE_V_VALID)
251 				plpar_pte_remove_raw(0, i + j, 0,
252 					&(ptes[j].pteh), &(ptes[j].ptel));
253 		}
254 	}
255 }
256 
257 static int hcall_hpte_clear_all(void)
258 {
259 	int rc;
260 
261 	do {
262 		rc = plpar_hcall_norets(H_CLEAR_HPT);
263 	} while (rc == H_CONTINUE);
264 
265 	return rc;
266 }
267 
268 static void pseries_hpte_clear_all(void)
269 {
270 	int rc;
271 
272 	rc = hcall_hpte_clear_all();
273 	if (rc != H_SUCCESS)
274 		manual_hpte_clear_all();
275 
276 #ifdef __LITTLE_ENDIAN__
277 	/*
278 	 * Reset exceptions to big endian.
279 	 *
280 	 * FIXME this is a hack for kexec, we need to reset the exception
281 	 * endian before starting the new kernel and this is a convenient place
282 	 * to do it.
283 	 *
284 	 * This is also called on boot when a fadump happens. In that case we
285 	 * must not change the exception endian mode.
286 	 */
287 	if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
288 		pseries_big_endian_exceptions();
289 #endif
290 }
291 
292 /*
293  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
294  * the low 3 bits of flags happen to line up.  So no transform is needed.
295  * We can probably optimize here and assume the high bits of newpp are
296  * already zero.  For now I am paranoid.
297  */
298 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
299 				       unsigned long newpp,
300 				       unsigned long vpn,
301 				       int psize, int apsize,
302 				       int ssize, unsigned long inv_flags)
303 {
304 	unsigned long lpar_rc;
305 	unsigned long flags;
306 	unsigned long want_v;
307 
308 	want_v = hpte_encode_avpn(vpn, psize, ssize);
309 
310 	flags = (newpp & 7) | H_AVPN;
311 	if (mmu_has_feature(MMU_FTR_KERNEL_RO))
312 		/* Move pp0 into bit 8 (IBM 55) */
313 		flags |= (newpp & HPTE_R_PP0) >> 55;
314 
315 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
316 		 want_v, slot, flags, psize);
317 
318 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
319 
320 	if (lpar_rc == H_NOT_FOUND) {
321 		pr_devel("not found !\n");
322 		return -1;
323 	}
324 
325 	pr_devel("ok\n");
326 
327 	BUG_ON(lpar_rc != H_SUCCESS);
328 
329 	return 0;
330 }
331 
332 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
333 {
334 	long lpar_rc;
335 	unsigned long i, j;
336 	struct {
337 		unsigned long pteh;
338 		unsigned long ptel;
339 	} ptes[4];
340 
341 	for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
342 
343 		lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
344 		if (lpar_rc != H_SUCCESS) {
345 			pr_info("Failed to read hash page table at %ld err %ld\n",
346 				hpte_group, lpar_rc);
347 			continue;
348 		}
349 
350 		for (j = 0; j < 4; j++) {
351 			if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
352 			    (ptes[j].pteh & HPTE_V_VALID))
353 				return i + j;
354 		}
355 	}
356 
357 	return -1;
358 }
359 
360 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
361 {
362 	long slot;
363 	unsigned long hash;
364 	unsigned long want_v;
365 	unsigned long hpte_group;
366 
367 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
368 	want_v = hpte_encode_avpn(vpn, psize, ssize);
369 
370 	/* Bolted entries are always in the primary group */
371 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
372 	slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
373 	if (slot < 0)
374 		return -1;
375 	return hpte_group + slot;
376 }
377 
378 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
379 					     unsigned long ea,
380 					     int psize, int ssize)
381 {
382 	unsigned long vpn;
383 	unsigned long lpar_rc, slot, vsid, flags;
384 
385 	vsid = get_kernel_vsid(ea, ssize);
386 	vpn = hpt_vpn(ea, vsid, ssize);
387 
388 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
389 	BUG_ON(slot == -1);
390 
391 	flags = newpp & 7;
392 	if (mmu_has_feature(MMU_FTR_KERNEL_RO))
393 		/* Move pp0 into bit 8 (IBM 55) */
394 		flags |= (newpp & HPTE_R_PP0) >> 55;
395 
396 	lpar_rc = plpar_pte_protect(flags, slot, 0);
397 
398 	BUG_ON(lpar_rc != H_SUCCESS);
399 }
400 
401 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
402 					 int psize, int apsize,
403 					 int ssize, int local)
404 {
405 	unsigned long want_v;
406 	unsigned long lpar_rc;
407 	unsigned long dummy1, dummy2;
408 
409 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
410 		 slot, vpn, psize, local);
411 
412 	want_v = hpte_encode_avpn(vpn, psize, ssize);
413 	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
414 	if (lpar_rc == H_NOT_FOUND)
415 		return;
416 
417 	BUG_ON(lpar_rc != H_SUCCESS);
418 }
419 
420 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
421 /*
422  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
423  * to make sure that we avoid bouncing the hypervisor tlbie lock.
424  */
425 #define PPC64_HUGE_HPTE_BATCH 12
426 
427 static void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
428 					     unsigned long *vpn, int count,
429 					     int psize, int ssize)
430 {
431 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
432 	int i = 0, pix = 0, rc;
433 	unsigned long flags = 0;
434 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
435 
436 	if (lock_tlbie)
437 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
438 
439 	for (i = 0; i < count; i++) {
440 
441 		if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
442 			pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
443 						     ssize, 0);
444 		} else {
445 			param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
446 			param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
447 			pix += 2;
448 			if (pix == 8) {
449 				rc = plpar_hcall9(H_BULK_REMOVE, param,
450 						  param[0], param[1], param[2],
451 						  param[3], param[4], param[5],
452 						  param[6], param[7]);
453 				BUG_ON(rc != H_SUCCESS);
454 				pix = 0;
455 			}
456 		}
457 	}
458 	if (pix) {
459 		param[pix] = HBR_END;
460 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
461 				  param[2], param[3], param[4], param[5],
462 				  param[6], param[7]);
463 		BUG_ON(rc != H_SUCCESS);
464 	}
465 
466 	if (lock_tlbie)
467 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
468 }
469 
470 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
471 					     unsigned long addr,
472 					     unsigned char *hpte_slot_array,
473 					     int psize, int ssize, int local)
474 {
475 	int i, index = 0;
476 	unsigned long s_addr = addr;
477 	unsigned int max_hpte_count, valid;
478 	unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
479 	unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
480 	unsigned long shift, hidx, vpn = 0, hash, slot;
481 
482 	shift = mmu_psize_defs[psize].shift;
483 	max_hpte_count = 1U << (PMD_SHIFT - shift);
484 
485 	for (i = 0; i < max_hpte_count; i++) {
486 		valid = hpte_valid(hpte_slot_array, i);
487 		if (!valid)
488 			continue;
489 		hidx =  hpte_hash_index(hpte_slot_array, i);
490 
491 		/* get the vpn */
492 		addr = s_addr + (i * (1ul << shift));
493 		vpn = hpt_vpn(addr, vsid, ssize);
494 		hash = hpt_hash(vpn, shift, ssize);
495 		if (hidx & _PTEIDX_SECONDARY)
496 			hash = ~hash;
497 
498 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
499 		slot += hidx & _PTEIDX_GROUP_IX;
500 
501 		slot_array[index] = slot;
502 		vpn_array[index] = vpn;
503 		if (index == PPC64_HUGE_HPTE_BATCH - 1) {
504 			/*
505 			 * Now do a bluk invalidate
506 			 */
507 			__pSeries_lpar_hugepage_invalidate(slot_array,
508 							   vpn_array,
509 							   PPC64_HUGE_HPTE_BATCH,
510 							   psize, ssize);
511 			index = 0;
512 		} else
513 			index++;
514 	}
515 	if (index)
516 		__pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
517 						   index, psize, ssize);
518 }
519 #else
520 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
521 					     unsigned long addr,
522 					     unsigned char *hpte_slot_array,
523 					     int psize, int ssize, int local)
524 {
525 	WARN(1, "%s called without THP support\n", __func__);
526 }
527 #endif
528 
529 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
530 					  int psize, int ssize)
531 {
532 	unsigned long vpn;
533 	unsigned long slot, vsid;
534 
535 	vsid = get_kernel_vsid(ea, ssize);
536 	vpn = hpt_vpn(ea, vsid, ssize);
537 
538 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
539 	if (slot == -1)
540 		return -ENOENT;
541 
542 	/*
543 	 * lpar doesn't use the passed actual page size
544 	 */
545 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
546 	return 0;
547 }
548 
549 /*
550  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
551  * lock.
552  */
553 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
554 {
555 	unsigned long vpn;
556 	unsigned long i, pix, rc;
557 	unsigned long flags = 0;
558 	struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
559 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
560 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
561 	unsigned long hash, index, shift, hidx, slot;
562 	real_pte_t pte;
563 	int psize, ssize;
564 
565 	if (lock_tlbie)
566 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
567 
568 	psize = batch->psize;
569 	ssize = batch->ssize;
570 	pix = 0;
571 	for (i = 0; i < number; i++) {
572 		vpn = batch->vpn[i];
573 		pte = batch->pte[i];
574 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
575 			hash = hpt_hash(vpn, shift, ssize);
576 			hidx = __rpte_to_hidx(pte, index);
577 			if (hidx & _PTEIDX_SECONDARY)
578 				hash = ~hash;
579 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
580 			slot += hidx & _PTEIDX_GROUP_IX;
581 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
582 				/*
583 				 * lpar doesn't use the passed actual page size
584 				 */
585 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
586 							     0, ssize, local);
587 			} else {
588 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
589 				param[pix+1] = hpte_encode_avpn(vpn, psize,
590 								ssize);
591 				pix += 2;
592 				if (pix == 8) {
593 					rc = plpar_hcall9(H_BULK_REMOVE, param,
594 						param[0], param[1], param[2],
595 						param[3], param[4], param[5],
596 						param[6], param[7]);
597 					BUG_ON(rc != H_SUCCESS);
598 					pix = 0;
599 				}
600 			}
601 		} pte_iterate_hashed_end();
602 	}
603 	if (pix) {
604 		param[pix] = HBR_END;
605 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
606 				  param[2], param[3], param[4], param[5],
607 				  param[6], param[7]);
608 		BUG_ON(rc != H_SUCCESS);
609 	}
610 
611 	if (lock_tlbie)
612 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
613 }
614 
615 static int __init disable_bulk_remove(char *str)
616 {
617 	if (strcmp(str, "off") == 0 &&
618 	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
619 		pr_info("Disabling BULK_REMOVE firmware feature");
620 		powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
621 	}
622 	return 1;
623 }
624 
625 __setup("bulk_remove=", disable_bulk_remove);
626 
627 #define HPT_RESIZE_TIMEOUT	10000 /* ms */
628 
629 struct hpt_resize_state {
630 	unsigned long shift;
631 	int commit_rc;
632 };
633 
634 static int pseries_lpar_resize_hpt_commit(void *data)
635 {
636 	struct hpt_resize_state *state = data;
637 
638 	state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
639 	if (state->commit_rc != H_SUCCESS)
640 		return -EIO;
641 
642 	/* Hypervisor has transitioned the HTAB, update our globals */
643 	ppc64_pft_size = state->shift;
644 	htab_size_bytes = 1UL << ppc64_pft_size;
645 	htab_hash_mask = (htab_size_bytes >> 7) - 1;
646 
647 	return 0;
648 }
649 
650 /* Must be called in user context */
651 static int pseries_lpar_resize_hpt(unsigned long shift)
652 {
653 	struct hpt_resize_state state = {
654 		.shift = shift,
655 		.commit_rc = H_FUNCTION,
656 	};
657 	unsigned int delay, total_delay = 0;
658 	int rc;
659 	ktime_t t0, t1, t2;
660 
661 	might_sleep();
662 
663 	if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
664 		return -ENODEV;
665 
666 	pr_info("Attempting to resize HPT to shift %lu\n", shift);
667 
668 	t0 = ktime_get();
669 
670 	rc = plpar_resize_hpt_prepare(0, shift);
671 	while (H_IS_LONG_BUSY(rc)) {
672 		delay = get_longbusy_msecs(rc);
673 		total_delay += delay;
674 		if (total_delay > HPT_RESIZE_TIMEOUT) {
675 			/* prepare with shift==0 cancels an in-progress resize */
676 			rc = plpar_resize_hpt_prepare(0, 0);
677 			if (rc != H_SUCCESS)
678 				pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
679 				       rc);
680 			return -ETIMEDOUT;
681 		}
682 		msleep(delay);
683 		rc = plpar_resize_hpt_prepare(0, shift);
684 	};
685 
686 	switch (rc) {
687 	case H_SUCCESS:
688 		/* Continue on */
689 		break;
690 
691 	case H_PARAMETER:
692 		return -EINVAL;
693 	case H_RESOURCE:
694 		return -EPERM;
695 	default:
696 		pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
697 		return -EIO;
698 	}
699 
700 	t1 = ktime_get();
701 
702 	rc = stop_machine(pseries_lpar_resize_hpt_commit, &state, NULL);
703 
704 	t2 = ktime_get();
705 
706 	if (rc != 0) {
707 		switch (state.commit_rc) {
708 		case H_PTEG_FULL:
709 			pr_warn("Hash collision while resizing HPT\n");
710 			return -ENOSPC;
711 
712 		default:
713 			pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
714 				state.commit_rc);
715 			return -EIO;
716 		};
717 	}
718 
719 	pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
720 		shift, (long long) ktime_ms_delta(t1, t0),
721 		(long long) ktime_ms_delta(t2, t1));
722 
723 	return 0;
724 }
725 
726 static int pseries_lpar_register_process_table(unsigned long base,
727 			unsigned long page_size, unsigned long table_size)
728 {
729 	long rc;
730 	unsigned long flags = 0;
731 
732 	if (table_size)
733 		flags |= PROC_TABLE_NEW;
734 	if (radix_enabled())
735 		flags |= PROC_TABLE_RADIX | PROC_TABLE_GTSE;
736 	else
737 		flags |= PROC_TABLE_HPT_SLB;
738 	for (;;) {
739 		rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
740 					page_size, table_size);
741 		if (!H_IS_LONG_BUSY(rc))
742 			break;
743 		mdelay(get_longbusy_msecs(rc));
744 	}
745 	if (rc != H_SUCCESS) {
746 		pr_err("Failed to register process table (rc=%ld)\n", rc);
747 		BUG();
748 	}
749 	return rc;
750 }
751 
752 void __init hpte_init_pseries(void)
753 {
754 	mmu_hash_ops.hpte_invalidate	 = pSeries_lpar_hpte_invalidate;
755 	mmu_hash_ops.hpte_updatepp	 = pSeries_lpar_hpte_updatepp;
756 	mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
757 	mmu_hash_ops.hpte_insert	 = pSeries_lpar_hpte_insert;
758 	mmu_hash_ops.hpte_remove	 = pSeries_lpar_hpte_remove;
759 	mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
760 	mmu_hash_ops.flush_hash_range	 = pSeries_lpar_flush_hash_range;
761 	mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
762 	mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
763 	register_process_table		 = pseries_lpar_register_process_table;
764 
765 	if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
766 		mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
767 }
768 
769 void radix_init_pseries(void)
770 {
771 	pr_info("Using radix MMU under hypervisor\n");
772 	register_process_table = pseries_lpar_register_process_table;
773 }
774 
775 #ifdef CONFIG_PPC_SMLPAR
776 #define CMO_FREE_HINT_DEFAULT 1
777 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
778 
779 static int __init cmo_free_hint(char *str)
780 {
781 	char *parm;
782 	parm = strstrip(str);
783 
784 	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
785 		pr_info("%s: CMO free page hinting is not active.\n", __func__);
786 		cmo_free_hint_flag = 0;
787 		return 1;
788 	}
789 
790 	cmo_free_hint_flag = 1;
791 	pr_info("%s: CMO free page hinting is active.\n", __func__);
792 
793 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
794 		return 1;
795 
796 	return 0;
797 }
798 
799 __setup("cmo_free_hint=", cmo_free_hint);
800 
801 static void pSeries_set_page_state(struct page *page, int order,
802 				   unsigned long state)
803 {
804 	int i, j;
805 	unsigned long cmo_page_sz, addr;
806 
807 	cmo_page_sz = cmo_get_page_size();
808 	addr = __pa((unsigned long)page_address(page));
809 
810 	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
811 		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
812 			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
813 	}
814 }
815 
816 void arch_free_page(struct page *page, int order)
817 {
818 	if (radix_enabled())
819 		return;
820 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
821 		return;
822 
823 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
824 }
825 EXPORT_SYMBOL(arch_free_page);
826 
827 #endif /* CONFIG_PPC_SMLPAR */
828 #endif /* CONFIG_PPC_BOOK3S_64 */
829 
830 #ifdef CONFIG_TRACEPOINTS
831 #ifdef HAVE_JUMP_LABEL
832 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
833 
834 int hcall_tracepoint_regfunc(void)
835 {
836 	static_key_slow_inc(&hcall_tracepoint_key);
837 	return 0;
838 }
839 
840 void hcall_tracepoint_unregfunc(void)
841 {
842 	static_key_slow_dec(&hcall_tracepoint_key);
843 }
844 #else
845 /*
846  * We optimise our hcall path by placing hcall_tracepoint_refcount
847  * directly in the TOC so we can check if the hcall tracepoints are
848  * enabled via a single load.
849  */
850 
851 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
852 extern long hcall_tracepoint_refcount;
853 
854 int hcall_tracepoint_regfunc(void)
855 {
856 	hcall_tracepoint_refcount++;
857 	return 0;
858 }
859 
860 void hcall_tracepoint_unregfunc(void)
861 {
862 	hcall_tracepoint_refcount--;
863 }
864 #endif
865 
866 /*
867  * Since the tracing code might execute hcalls we need to guard against
868  * recursion. One example of this are spinlocks calling H_YIELD on
869  * shared processor partitions.
870  */
871 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
872 
873 
874 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
875 {
876 	unsigned long flags;
877 	unsigned int *depth;
878 
879 	/*
880 	 * We cannot call tracepoints inside RCU idle regions which
881 	 * means we must not trace H_CEDE.
882 	 */
883 	if (opcode == H_CEDE)
884 		return;
885 
886 	local_irq_save(flags);
887 
888 	depth = this_cpu_ptr(&hcall_trace_depth);
889 
890 	if (*depth)
891 		goto out;
892 
893 	(*depth)++;
894 	preempt_disable();
895 	trace_hcall_entry(opcode, args);
896 	(*depth)--;
897 
898 out:
899 	local_irq_restore(flags);
900 }
901 
902 void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
903 {
904 	unsigned long flags;
905 	unsigned int *depth;
906 
907 	if (opcode == H_CEDE)
908 		return;
909 
910 	local_irq_save(flags);
911 
912 	depth = this_cpu_ptr(&hcall_trace_depth);
913 
914 	if (*depth)
915 		goto out;
916 
917 	(*depth)++;
918 	trace_hcall_exit(opcode, retval, retbuf);
919 	preempt_enable();
920 	(*depth)--;
921 
922 out:
923 	local_irq_restore(flags);
924 }
925 #endif
926 
927 /**
928  * h_get_mpp
929  * H_GET_MPP hcall returns info in 7 parms
930  */
931 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
932 {
933 	int rc;
934 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
935 
936 	rc = plpar_hcall9(H_GET_MPP, retbuf);
937 
938 	mpp_data->entitled_mem = retbuf[0];
939 	mpp_data->mapped_mem = retbuf[1];
940 
941 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
942 	mpp_data->pool_num = retbuf[2] & 0xffff;
943 
944 	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
945 	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
946 	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
947 
948 	mpp_data->pool_size = retbuf[4];
949 	mpp_data->loan_request = retbuf[5];
950 	mpp_data->backing_mem = retbuf[6];
951 
952 	return rc;
953 }
954 EXPORT_SYMBOL(h_get_mpp);
955 
956 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
957 {
958 	int rc;
959 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
960 
961 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
962 
963 	mpp_x_data->coalesced_bytes = retbuf[0];
964 	mpp_x_data->pool_coalesced_bytes = retbuf[1];
965 	mpp_x_data->pool_purr_cycles = retbuf[2];
966 	mpp_x_data->pool_spurr_cycles = retbuf[3];
967 
968 	return rc;
969 }
970 
971 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
972 {
973 	unsigned long protovsid;
974 	unsigned long va_bits = VA_BITS;
975 	unsigned long modinv, vsid_modulus;
976 	unsigned long max_mod_inv, tmp_modinv;
977 
978 	if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
979 		va_bits = 65;
980 
981 	if (ssize == MMU_SEGSIZE_256M) {
982 		modinv = VSID_MULINV_256M;
983 		vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
984 	} else {
985 		modinv = VSID_MULINV_1T;
986 		vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
987 	}
988 
989 	/*
990 	 * vsid outside our range.
991 	 */
992 	if (vsid >= vsid_modulus)
993 		return 0;
994 
995 	/*
996 	 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
997 	 * and vsid = (protovsid * x) % vsid_modulus, then we say:
998 	 *   protovsid = (vsid * modinv) % vsid_modulus
999 	 */
1000 
1001 	/* Check if (vsid * modinv) overflow (63 bits) */
1002 	max_mod_inv = 0x7fffffffffffffffull / vsid;
1003 	if (modinv < max_mod_inv)
1004 		return (vsid * modinv) % vsid_modulus;
1005 
1006 	tmp_modinv = modinv/max_mod_inv;
1007 	modinv %= max_mod_inv;
1008 
1009 	protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1010 	protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1011 
1012 	return protovsid;
1013 }
1014 
1015 static int __init reserve_vrma_context_id(void)
1016 {
1017 	unsigned long protovsid;
1018 
1019 	/*
1020 	 * Reserve context ids which map to reserved virtual addresses. For now
1021 	 * we only reserve the context id which maps to the VRMA VSID. We ignore
1022 	 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1023 	 * enable adjunct support via the "ibm,client-architecture-support"
1024 	 * interface.
1025 	 */
1026 	protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1027 	hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1028 	return 0;
1029 }
1030 machine_device_initcall(pseries, reserve_vrma_context_id);
1031