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 
45 #include "plpar_wrappers.h"
46 #include "pseries.h"
47 
48 
49 /* in hvCall.S */
50 EXPORT_SYMBOL(plpar_hcall);
51 EXPORT_SYMBOL(plpar_hcall9);
52 EXPORT_SYMBOL(plpar_hcall_norets);
53 
54 extern void pSeries_find_serial_port(void);
55 
56 void vpa_init(int cpu)
57 {
58 	int hwcpu = get_hard_smp_processor_id(cpu);
59 	unsigned long addr;
60 	long ret;
61 	struct paca_struct *pp;
62 	struct dtl_entry *dtl;
63 
64 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
65 		lppaca_of(cpu).vmxregs_in_use = 1;
66 
67 	addr = __pa(&lppaca_of(cpu));
68 	ret = register_vpa(hwcpu, addr);
69 
70 	if (ret) {
71 		pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
72 		       "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
73 		return;
74 	}
75 	/*
76 	 * PAPR says this feature is SLB-Buffer but firmware never
77 	 * reports that.  All SPLPAR support SLB shadow buffer.
78 	 */
79 	addr = __pa(&slb_shadow[cpu]);
80 	if (firmware_has_feature(FW_FEATURE_SPLPAR)) {
81 		ret = register_slb_shadow(hwcpu, addr);
82 		if (ret)
83 			pr_err("WARNING: SLB shadow buffer registration for "
84 			       "cpu %d (hw %d) of area %lx failed with %ld\n",
85 			       cpu, hwcpu, addr, ret);
86 	}
87 
88 	/*
89 	 * Register dispatch trace log, if one has been allocated.
90 	 */
91 	pp = &paca[cpu];
92 	dtl = pp->dispatch_log;
93 	if (dtl) {
94 		pp->dtl_ridx = 0;
95 		pp->dtl_curr = dtl;
96 		lppaca_of(cpu).dtl_idx = 0;
97 
98 		/* hypervisor reads buffer length from this field */
99 		dtl->enqueue_to_dispatch_time = DISPATCH_LOG_BYTES;
100 		ret = register_dtl(hwcpu, __pa(dtl));
101 		if (ret)
102 			pr_err("WARNING: DTL registration of cpu %d (hw %d) "
103 			       "failed with %ld\n", smp_processor_id(),
104 			       hwcpu, ret);
105 		lppaca_of(cpu).dtl_enable_mask = 2;
106 	}
107 }
108 
109 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
110 				     unsigned long vpn, unsigned long pa,
111 				     unsigned long rflags, unsigned long vflags,
112 				     int psize, int apsize, int ssize)
113 {
114 	unsigned long lpar_rc;
115 	unsigned long flags;
116 	unsigned long slot;
117 	unsigned long hpte_v, hpte_r;
118 
119 	if (!(vflags & HPTE_V_BOLTED))
120 		pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
121 			 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
122 			 hpte_group, vpn,  pa, rflags, vflags, psize);
123 
124 	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
125 	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
126 
127 	if (!(vflags & HPTE_V_BOLTED))
128 		pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
129 
130 	/* Now fill in the actual HPTE */
131 	/* Set CEC cookie to 0         */
132 	/* Zero page = 0               */
133 	/* I-cache Invalidate = 0      */
134 	/* I-cache synchronize = 0     */
135 	/* Exact = 0                   */
136 	flags = 0;
137 
138 	/* Make pHyp happy */
139 	if ((rflags & _PAGE_NO_CACHE) & !(rflags & _PAGE_WRITETHRU))
140 		hpte_r &= ~_PAGE_COHERENT;
141 	if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
142 		flags |= H_COALESCE_CAND;
143 
144 	lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
145 	if (unlikely(lpar_rc == H_PTEG_FULL)) {
146 		if (!(vflags & HPTE_V_BOLTED))
147 			pr_devel(" full\n");
148 		return -1;
149 	}
150 
151 	/*
152 	 * Since we try and ioremap PHBs we don't own, the pte insert
153 	 * will fail. However we must catch the failure in hash_page
154 	 * or we will loop forever, so return -2 in this case.
155 	 */
156 	if (unlikely(lpar_rc != H_SUCCESS)) {
157 		if (!(vflags & HPTE_V_BOLTED))
158 			pr_devel(" lpar err %ld\n", lpar_rc);
159 		return -2;
160 	}
161 	if (!(vflags & HPTE_V_BOLTED))
162 		pr_devel(" -> slot: %lu\n", slot & 7);
163 
164 	/* Because of iSeries, we have to pass down the secondary
165 	 * bucket bit here as well
166 	 */
167 	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
168 }
169 
170 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
171 
172 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
173 {
174 	unsigned long slot_offset;
175 	unsigned long lpar_rc;
176 	int i;
177 	unsigned long dummy1, dummy2;
178 
179 	/* pick a random slot to start at */
180 	slot_offset = mftb() & 0x7;
181 
182 	for (i = 0; i < HPTES_PER_GROUP; i++) {
183 
184 		/* don't remove a bolted entry */
185 		lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
186 					   (0x1UL << 4), &dummy1, &dummy2);
187 		if (lpar_rc == H_SUCCESS)
188 			return i;
189 
190 		/*
191 		 * The test for adjunct partition is performed before the
192 		 * ANDCOND test.  H_RESOURCE may be returned, so we need to
193 		 * check for that as well.
194 		 */
195 		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
196 
197 		slot_offset++;
198 		slot_offset &= 0x7;
199 	}
200 
201 	return -1;
202 }
203 
204 static void pSeries_lpar_hptab_clear(void)
205 {
206 	unsigned long size_bytes = 1UL << ppc64_pft_size;
207 	unsigned long hpte_count = size_bytes >> 4;
208 	struct {
209 		unsigned long pteh;
210 		unsigned long ptel;
211 	} ptes[4];
212 	long lpar_rc;
213 	unsigned long i, j;
214 
215 	/* Read in batches of 4,
216 	 * invalidate only valid entries not in the VRMA
217 	 * hpte_count will be a multiple of 4
218          */
219 	for (i = 0; i < hpte_count; i += 4) {
220 		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
221 		if (lpar_rc != H_SUCCESS)
222 			continue;
223 		for (j = 0; j < 4; j++){
224 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
225 				HPTE_V_VRMA_MASK)
226 				continue;
227 			if (ptes[j].pteh & HPTE_V_VALID)
228 				plpar_pte_remove_raw(0, i + j, 0,
229 					&(ptes[j].pteh), &(ptes[j].ptel));
230 		}
231 	}
232 }
233 
234 /*
235  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
236  * the low 3 bits of flags happen to line up.  So no transform is needed.
237  * We can probably optimize here and assume the high bits of newpp are
238  * already zero.  For now I am paranoid.
239  */
240 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
241 				       unsigned long newpp,
242 				       unsigned long vpn,
243 				       int psize, int ssize, int local)
244 {
245 	unsigned long lpar_rc;
246 	unsigned long flags = (newpp & 7) | H_AVPN;
247 	unsigned long want_v;
248 
249 	want_v = hpte_encode_avpn(vpn, psize, ssize);
250 
251 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
252 		 want_v, slot, flags, psize);
253 
254 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
255 
256 	if (lpar_rc == H_NOT_FOUND) {
257 		pr_devel("not found !\n");
258 		return -1;
259 	}
260 
261 	pr_devel("ok\n");
262 
263 	BUG_ON(lpar_rc != H_SUCCESS);
264 
265 	return 0;
266 }
267 
268 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
269 {
270 	unsigned long dword0;
271 	unsigned long lpar_rc;
272 	unsigned long dummy_word1;
273 	unsigned long flags;
274 
275 	/* Read 1 pte at a time                        */
276 	/* Do not need RPN to logical page translation */
277 	/* No cross CEC PFT access                     */
278 	flags = 0;
279 
280 	lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
281 
282 	BUG_ON(lpar_rc != H_SUCCESS);
283 
284 	return dword0;
285 }
286 
287 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
288 {
289 	unsigned long hash;
290 	unsigned long i;
291 	long slot;
292 	unsigned long want_v, hpte_v;
293 
294 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
295 	want_v = hpte_encode_avpn(vpn, psize, ssize);
296 
297 	/* Bolted entries are always in the primary group */
298 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
299 	for (i = 0; i < HPTES_PER_GROUP; i++) {
300 		hpte_v = pSeries_lpar_hpte_getword0(slot);
301 
302 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
303 			/* HPTE matches */
304 			return slot;
305 		++slot;
306 	}
307 
308 	return -1;
309 }
310 
311 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
312 					     unsigned long ea,
313 					     int psize, int ssize)
314 {
315 	unsigned long vpn;
316 	unsigned long lpar_rc, slot, vsid, flags;
317 
318 	vsid = get_kernel_vsid(ea, ssize);
319 	vpn = hpt_vpn(ea, vsid, ssize);
320 
321 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
322 	BUG_ON(slot == -1);
323 
324 	flags = newpp & 7;
325 	lpar_rc = plpar_pte_protect(flags, slot, 0);
326 
327 	BUG_ON(lpar_rc != H_SUCCESS);
328 }
329 
330 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
331 					 int psize, int ssize, int local)
332 {
333 	unsigned long want_v;
334 	unsigned long lpar_rc;
335 	unsigned long dummy1, dummy2;
336 
337 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
338 		 slot, vpn, psize, local);
339 
340 	want_v = hpte_encode_avpn(vpn, psize, ssize);
341 	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
342 	if (lpar_rc == H_NOT_FOUND)
343 		return;
344 
345 	BUG_ON(lpar_rc != H_SUCCESS);
346 }
347 
348 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
349 					   int psize, int ssize)
350 {
351 	unsigned long vpn;
352 	unsigned long slot, vsid;
353 
354 	vsid = get_kernel_vsid(ea, ssize);
355 	vpn = hpt_vpn(ea, vsid, ssize);
356 
357 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
358 	BUG_ON(slot == -1);
359 
360 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, ssize, 0);
361 }
362 
363 /* Flag bits for H_BULK_REMOVE */
364 #define HBR_REQUEST	0x4000000000000000UL
365 #define HBR_RESPONSE	0x8000000000000000UL
366 #define HBR_END		0xc000000000000000UL
367 #define HBR_AVPN	0x0200000000000000UL
368 #define HBR_ANDCOND	0x0100000000000000UL
369 
370 /*
371  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
372  * lock.
373  */
374 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
375 {
376 	unsigned long vpn;
377 	unsigned long i, pix, rc;
378 	unsigned long flags = 0;
379 	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
380 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
381 	unsigned long param[9];
382 	unsigned long hash, index, shift, hidx, slot;
383 	real_pte_t pte;
384 	int psize, ssize;
385 
386 	if (lock_tlbie)
387 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
388 
389 	psize = batch->psize;
390 	ssize = batch->ssize;
391 	pix = 0;
392 	for (i = 0; i < number; i++) {
393 		vpn = batch->vpn[i];
394 		pte = batch->pte[i];
395 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
396 			hash = hpt_hash(vpn, shift, ssize);
397 			hidx = __rpte_to_hidx(pte, index);
398 			if (hidx & _PTEIDX_SECONDARY)
399 				hash = ~hash;
400 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
401 			slot += hidx & _PTEIDX_GROUP_IX;
402 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
403 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
404 							     ssize, local);
405 			} else {
406 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
407 				param[pix+1] = hpte_encode_avpn(vpn, psize,
408 								ssize);
409 				pix += 2;
410 				if (pix == 8) {
411 					rc = plpar_hcall9(H_BULK_REMOVE, param,
412 						param[0], param[1], param[2],
413 						param[3], param[4], param[5],
414 						param[6], param[7]);
415 					BUG_ON(rc != H_SUCCESS);
416 					pix = 0;
417 				}
418 			}
419 		} pte_iterate_hashed_end();
420 	}
421 	if (pix) {
422 		param[pix] = HBR_END;
423 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
424 				  param[2], param[3], param[4], param[5],
425 				  param[6], param[7]);
426 		BUG_ON(rc != H_SUCCESS);
427 	}
428 
429 	if (lock_tlbie)
430 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
431 }
432 
433 static int __init disable_bulk_remove(char *str)
434 {
435 	if (strcmp(str, "off") == 0 &&
436 	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
437 			printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
438 			powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
439 	}
440 	return 1;
441 }
442 
443 __setup("bulk_remove=", disable_bulk_remove);
444 
445 void __init hpte_init_lpar(void)
446 {
447 	ppc_md.hpte_invalidate	= pSeries_lpar_hpte_invalidate;
448 	ppc_md.hpte_updatepp	= pSeries_lpar_hpte_updatepp;
449 	ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
450 	ppc_md.hpte_insert	= pSeries_lpar_hpte_insert;
451 	ppc_md.hpte_remove	= pSeries_lpar_hpte_remove;
452 	ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
453 	ppc_md.flush_hash_range	= pSeries_lpar_flush_hash_range;
454 	ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
455 }
456 
457 #ifdef CONFIG_PPC_SMLPAR
458 #define CMO_FREE_HINT_DEFAULT 1
459 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
460 
461 static int __init cmo_free_hint(char *str)
462 {
463 	char *parm;
464 	parm = strstrip(str);
465 
466 	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
467 		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
468 		cmo_free_hint_flag = 0;
469 		return 1;
470 	}
471 
472 	cmo_free_hint_flag = 1;
473 	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
474 
475 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
476 		return 1;
477 
478 	return 0;
479 }
480 
481 __setup("cmo_free_hint=", cmo_free_hint);
482 
483 static void pSeries_set_page_state(struct page *page, int order,
484 				   unsigned long state)
485 {
486 	int i, j;
487 	unsigned long cmo_page_sz, addr;
488 
489 	cmo_page_sz = cmo_get_page_size();
490 	addr = __pa((unsigned long)page_address(page));
491 
492 	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
493 		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
494 			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
495 	}
496 }
497 
498 void arch_free_page(struct page *page, int order)
499 {
500 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
501 		return;
502 
503 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
504 }
505 EXPORT_SYMBOL(arch_free_page);
506 
507 #endif
508 
509 #ifdef CONFIG_TRACEPOINTS
510 /*
511  * We optimise our hcall path by placing hcall_tracepoint_refcount
512  * directly in the TOC so we can check if the hcall tracepoints are
513  * enabled via a single load.
514  */
515 
516 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
517 extern long hcall_tracepoint_refcount;
518 
519 /*
520  * Since the tracing code might execute hcalls we need to guard against
521  * recursion. One example of this are spinlocks calling H_YIELD on
522  * shared processor partitions.
523  */
524 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
525 
526 void hcall_tracepoint_regfunc(void)
527 {
528 	hcall_tracepoint_refcount++;
529 }
530 
531 void hcall_tracepoint_unregfunc(void)
532 {
533 	hcall_tracepoint_refcount--;
534 }
535 
536 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
537 {
538 	unsigned long flags;
539 	unsigned int *depth;
540 
541 	/*
542 	 * We cannot call tracepoints inside RCU idle regions which
543 	 * means we must not trace H_CEDE.
544 	 */
545 	if (opcode == H_CEDE)
546 		return;
547 
548 	local_irq_save(flags);
549 
550 	depth = &__get_cpu_var(hcall_trace_depth);
551 
552 	if (*depth)
553 		goto out;
554 
555 	(*depth)++;
556 	preempt_disable();
557 	trace_hcall_entry(opcode, args);
558 	(*depth)--;
559 
560 out:
561 	local_irq_restore(flags);
562 }
563 
564 void __trace_hcall_exit(long opcode, unsigned long retval,
565 			unsigned long *retbuf)
566 {
567 	unsigned long flags;
568 	unsigned int *depth;
569 
570 	if (opcode == H_CEDE)
571 		return;
572 
573 	local_irq_save(flags);
574 
575 	depth = &__get_cpu_var(hcall_trace_depth);
576 
577 	if (*depth)
578 		goto out;
579 
580 	(*depth)++;
581 	trace_hcall_exit(opcode, retval, retbuf);
582 	preempt_enable();
583 	(*depth)--;
584 
585 out:
586 	local_irq_restore(flags);
587 }
588 #endif
589 
590 /**
591  * h_get_mpp
592  * H_GET_MPP hcall returns info in 7 parms
593  */
594 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
595 {
596 	int rc;
597 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
598 
599 	rc = plpar_hcall9(H_GET_MPP, retbuf);
600 
601 	mpp_data->entitled_mem = retbuf[0];
602 	mpp_data->mapped_mem = retbuf[1];
603 
604 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
605 	mpp_data->pool_num = retbuf[2] & 0xffff;
606 
607 	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
608 	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
609 	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
610 
611 	mpp_data->pool_size = retbuf[4];
612 	mpp_data->loan_request = retbuf[5];
613 	mpp_data->backing_mem = retbuf[6];
614 
615 	return rc;
616 }
617 EXPORT_SYMBOL(h_get_mpp);
618 
619 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
620 {
621 	int rc;
622 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
623 
624 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
625 
626 	mpp_x_data->coalesced_bytes = retbuf[0];
627 	mpp_x_data->pool_coalesced_bytes = retbuf[1];
628 	mpp_x_data->pool_purr_cycles = retbuf[2];
629 	mpp_x_data->pool_spurr_cycles = retbuf[3];
630 
631 	return rc;
632 }
633