xref: /linux/arch/powerpc/platforms/pseries/lpar.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 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, ssize) | vflags | HPTE_V_VALID;
125 	hpte_r = hpte_encode_r(pa, psize) | 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 %lu\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 		BUG_ON(lpar_rc != H_NOT_FOUND);
190 
191 		slot_offset++;
192 		slot_offset &= 0x7;
193 	}
194 
195 	return -1;
196 }
197 
198 static void pSeries_lpar_hptab_clear(void)
199 {
200 	unsigned long size_bytes = 1UL << ppc64_pft_size;
201 	unsigned long hpte_count = size_bytes >> 4;
202 	struct {
203 		unsigned long pteh;
204 		unsigned long ptel;
205 	} ptes[4];
206 	long lpar_rc;
207 	unsigned long i, j;
208 
209 	/* Read in batches of 4,
210 	 * invalidate only valid entries not in the VRMA
211 	 * hpte_count will be a multiple of 4
212          */
213 	for (i = 0; i < hpte_count; i += 4) {
214 		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
215 		if (lpar_rc != H_SUCCESS)
216 			continue;
217 		for (j = 0; j < 4; j++){
218 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
219 				HPTE_V_VRMA_MASK)
220 				continue;
221 			if (ptes[j].pteh & HPTE_V_VALID)
222 				plpar_pte_remove_raw(0, i + j, 0,
223 					&(ptes[j].pteh), &(ptes[j].ptel));
224 		}
225 	}
226 }
227 
228 /*
229  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
230  * the low 3 bits of flags happen to line up.  So no transform is needed.
231  * We can probably optimize here and assume the high bits of newpp are
232  * already zero.  For now I am paranoid.
233  */
234 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
235 				       unsigned long newpp,
236 				       unsigned long vpn,
237 				       int psize, int ssize, int local)
238 {
239 	unsigned long lpar_rc;
240 	unsigned long flags = (newpp & 7) | H_AVPN;
241 	unsigned long want_v;
242 
243 	want_v = hpte_encode_avpn(vpn, psize, ssize);
244 
245 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
246 		 want_v, slot, flags, psize);
247 
248 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
249 
250 	if (lpar_rc == H_NOT_FOUND) {
251 		pr_devel("not found !\n");
252 		return -1;
253 	}
254 
255 	pr_devel("ok\n");
256 
257 	BUG_ON(lpar_rc != H_SUCCESS);
258 
259 	return 0;
260 }
261 
262 static unsigned long pSeries_lpar_hpte_getword0(unsigned long slot)
263 {
264 	unsigned long dword0;
265 	unsigned long lpar_rc;
266 	unsigned long dummy_word1;
267 	unsigned long flags;
268 
269 	/* Read 1 pte at a time                        */
270 	/* Do not need RPN to logical page translation */
271 	/* No cross CEC PFT access                     */
272 	flags = 0;
273 
274 	lpar_rc = plpar_pte_read(flags, slot, &dword0, &dummy_word1);
275 
276 	BUG_ON(lpar_rc != H_SUCCESS);
277 
278 	return dword0;
279 }
280 
281 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
282 {
283 	unsigned long hash;
284 	unsigned long i;
285 	long slot;
286 	unsigned long want_v, hpte_v;
287 
288 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
289 	want_v = hpte_encode_avpn(vpn, psize, ssize);
290 
291 	/* Bolted entries are always in the primary group */
292 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
293 	for (i = 0; i < HPTES_PER_GROUP; i++) {
294 		hpte_v = pSeries_lpar_hpte_getword0(slot);
295 
296 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
297 			/* HPTE matches */
298 			return slot;
299 		++slot;
300 	}
301 
302 	return -1;
303 }
304 
305 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
306 					     unsigned long ea,
307 					     int psize, int ssize)
308 {
309 	unsigned long vpn;
310 	unsigned long lpar_rc, slot, vsid, flags;
311 
312 	vsid = get_kernel_vsid(ea, ssize);
313 	vpn = hpt_vpn(ea, vsid, ssize);
314 
315 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
316 	BUG_ON(slot == -1);
317 
318 	flags = newpp & 7;
319 	lpar_rc = plpar_pte_protect(flags, slot, 0);
320 
321 	BUG_ON(lpar_rc != H_SUCCESS);
322 }
323 
324 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
325 					 int psize, int ssize, int local)
326 {
327 	unsigned long want_v;
328 	unsigned long lpar_rc;
329 	unsigned long dummy1, dummy2;
330 
331 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
332 		 slot, vpn, psize, local);
333 
334 	want_v = hpte_encode_avpn(vpn, psize, ssize);
335 	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
336 	if (lpar_rc == H_NOT_FOUND)
337 		return;
338 
339 	BUG_ON(lpar_rc != H_SUCCESS);
340 }
341 
342 static void pSeries_lpar_hpte_removebolted(unsigned long ea,
343 					   int psize, int ssize)
344 {
345 	unsigned long vpn;
346 	unsigned long slot, vsid;
347 
348 	vsid = get_kernel_vsid(ea, ssize);
349 	vpn = hpt_vpn(ea, vsid, ssize);
350 
351 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
352 	BUG_ON(slot == -1);
353 
354 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, ssize, 0);
355 }
356 
357 /* Flag bits for H_BULK_REMOVE */
358 #define HBR_REQUEST	0x4000000000000000UL
359 #define HBR_RESPONSE	0x8000000000000000UL
360 #define HBR_END		0xc000000000000000UL
361 #define HBR_AVPN	0x0200000000000000UL
362 #define HBR_ANDCOND	0x0100000000000000UL
363 
364 /*
365  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
366  * lock.
367  */
368 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
369 {
370 	unsigned long vpn;
371 	unsigned long i, pix, rc;
372 	unsigned long flags = 0;
373 	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
374 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
375 	unsigned long param[9];
376 	unsigned long hash, index, shift, hidx, slot;
377 	real_pte_t pte;
378 	int psize, ssize;
379 
380 	if (lock_tlbie)
381 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
382 
383 	psize = batch->psize;
384 	ssize = batch->ssize;
385 	pix = 0;
386 	for (i = 0; i < number; i++) {
387 		vpn = batch->vpn[i];
388 		pte = batch->pte[i];
389 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
390 			hash = hpt_hash(vpn, shift, ssize);
391 			hidx = __rpte_to_hidx(pte, index);
392 			if (hidx & _PTEIDX_SECONDARY)
393 				hash = ~hash;
394 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
395 			slot += hidx & _PTEIDX_GROUP_IX;
396 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
397 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
398 							     ssize, local);
399 			} else {
400 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
401 				param[pix+1] = hpte_encode_avpn(vpn, psize,
402 								ssize);
403 				pix += 2;
404 				if (pix == 8) {
405 					rc = plpar_hcall9(H_BULK_REMOVE, param,
406 						param[0], param[1], param[2],
407 						param[3], param[4], param[5],
408 						param[6], param[7]);
409 					BUG_ON(rc != H_SUCCESS);
410 					pix = 0;
411 				}
412 			}
413 		} pte_iterate_hashed_end();
414 	}
415 	if (pix) {
416 		param[pix] = HBR_END;
417 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
418 				  param[2], param[3], param[4], param[5],
419 				  param[6], param[7]);
420 		BUG_ON(rc != H_SUCCESS);
421 	}
422 
423 	if (lock_tlbie)
424 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
425 }
426 
427 static int __init disable_bulk_remove(char *str)
428 {
429 	if (strcmp(str, "off") == 0 &&
430 	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
431 			printk(KERN_INFO "Disabling BULK_REMOVE firmware feature");
432 			powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
433 	}
434 	return 1;
435 }
436 
437 __setup("bulk_remove=", disable_bulk_remove);
438 
439 void __init hpte_init_lpar(void)
440 {
441 	ppc_md.hpte_invalidate	= pSeries_lpar_hpte_invalidate;
442 	ppc_md.hpte_updatepp	= pSeries_lpar_hpte_updatepp;
443 	ppc_md.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
444 	ppc_md.hpte_insert	= pSeries_lpar_hpte_insert;
445 	ppc_md.hpte_remove	= pSeries_lpar_hpte_remove;
446 	ppc_md.hpte_removebolted = pSeries_lpar_hpte_removebolted;
447 	ppc_md.flush_hash_range	= pSeries_lpar_flush_hash_range;
448 	ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
449 }
450 
451 #ifdef CONFIG_PPC_SMLPAR
452 #define CMO_FREE_HINT_DEFAULT 1
453 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
454 
455 static int __init cmo_free_hint(char *str)
456 {
457 	char *parm;
458 	parm = strstrip(str);
459 
460 	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
461 		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
462 		cmo_free_hint_flag = 0;
463 		return 1;
464 	}
465 
466 	cmo_free_hint_flag = 1;
467 	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
468 
469 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
470 		return 1;
471 
472 	return 0;
473 }
474 
475 __setup("cmo_free_hint=", cmo_free_hint);
476 
477 static void pSeries_set_page_state(struct page *page, int order,
478 				   unsigned long state)
479 {
480 	int i, j;
481 	unsigned long cmo_page_sz, addr;
482 
483 	cmo_page_sz = cmo_get_page_size();
484 	addr = __pa((unsigned long)page_address(page));
485 
486 	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
487 		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
488 			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
489 	}
490 }
491 
492 void arch_free_page(struct page *page, int order)
493 {
494 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
495 		return;
496 
497 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
498 }
499 EXPORT_SYMBOL(arch_free_page);
500 
501 #endif
502 
503 #ifdef CONFIG_TRACEPOINTS
504 /*
505  * We optimise our hcall path by placing hcall_tracepoint_refcount
506  * directly in the TOC so we can check if the hcall tracepoints are
507  * enabled via a single load.
508  */
509 
510 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
511 extern long hcall_tracepoint_refcount;
512 
513 /*
514  * Since the tracing code might execute hcalls we need to guard against
515  * recursion. One example of this are spinlocks calling H_YIELD on
516  * shared processor partitions.
517  */
518 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
519 
520 void hcall_tracepoint_regfunc(void)
521 {
522 	hcall_tracepoint_refcount++;
523 }
524 
525 void hcall_tracepoint_unregfunc(void)
526 {
527 	hcall_tracepoint_refcount--;
528 }
529 
530 void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
531 {
532 	unsigned long flags;
533 	unsigned int *depth;
534 
535 	/*
536 	 * We cannot call tracepoints inside RCU idle regions which
537 	 * means we must not trace H_CEDE.
538 	 */
539 	if (opcode == H_CEDE)
540 		return;
541 
542 	local_irq_save(flags);
543 
544 	depth = &__get_cpu_var(hcall_trace_depth);
545 
546 	if (*depth)
547 		goto out;
548 
549 	(*depth)++;
550 	preempt_disable();
551 	trace_hcall_entry(opcode, args);
552 	(*depth)--;
553 
554 out:
555 	local_irq_restore(flags);
556 }
557 
558 void __trace_hcall_exit(long opcode, unsigned long retval,
559 			unsigned long *retbuf)
560 {
561 	unsigned long flags;
562 	unsigned int *depth;
563 
564 	if (opcode == H_CEDE)
565 		return;
566 
567 	local_irq_save(flags);
568 
569 	depth = &__get_cpu_var(hcall_trace_depth);
570 
571 	if (*depth)
572 		goto out;
573 
574 	(*depth)++;
575 	trace_hcall_exit(opcode, retval, retbuf);
576 	preempt_enable();
577 	(*depth)--;
578 
579 out:
580 	local_irq_restore(flags);
581 }
582 #endif
583 
584 /**
585  * h_get_mpp
586  * H_GET_MPP hcall returns info in 7 parms
587  */
588 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
589 {
590 	int rc;
591 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
592 
593 	rc = plpar_hcall9(H_GET_MPP, retbuf);
594 
595 	mpp_data->entitled_mem = retbuf[0];
596 	mpp_data->mapped_mem = retbuf[1];
597 
598 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
599 	mpp_data->pool_num = retbuf[2] & 0xffff;
600 
601 	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
602 	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
603 	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffff;
604 
605 	mpp_data->pool_size = retbuf[4];
606 	mpp_data->loan_request = retbuf[5];
607 	mpp_data->backing_mem = retbuf[6];
608 
609 	return rc;
610 }
611 EXPORT_SYMBOL(h_get_mpp);
612 
613 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
614 {
615 	int rc;
616 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
617 
618 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
619 
620 	mpp_x_data->coalesced_bytes = retbuf[0];
621 	mpp_x_data->pool_coalesced_bytes = retbuf[1];
622 	mpp_x_data->pool_purr_cycles = retbuf[2];
623 	mpp_x_data->pool_spurr_cycles = retbuf[3];
624 
625 	return rc;
626 }
627