1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Low-level SPU handling
4 *
5 * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
6 *
7 * Author: Arnd Bergmann <arndb@de.ibm.com>
8 */
9
10#undef DEBUG
11
12#include <linux/interrupt.h>
13#include <linux/list.h>
14#include <linux/init.h>
15#include <linux/ptrace.h>
16#include <linux/slab.h>
17#include <linux/wait.h>
18#include <linux/mm.h>
19#include <linux/io.h>
20#include <linux/mutex.h>
21#include <linux/linux_logo.h>
22#include <linux/syscore_ops.h>
23#include <asm/spu.h>
24#include <asm/spu_priv1.h>
25#include <asm/spu_csa.h>
26#include <asm/xmon.h>
27#include <asm/prom.h>
28#include <asm/kexec.h>
29
30const struct spu_management_ops *spu_management_ops;
31EXPORT_SYMBOL_GPL(spu_management_ops);
32
33const struct spu_priv1_ops *spu_priv1_ops;
34EXPORT_SYMBOL_GPL(spu_priv1_ops);
35
36struct cbe_spu_info cbe_spu_info[MAX_NUMNODES];
37EXPORT_SYMBOL_GPL(cbe_spu_info);
38
39/*
40 * The spufs fault-handling code needs to call force_sig_fault to raise signals
41 * on DMA errors. Export it here to avoid general kernel-wide access to this
42 * function
43 */
44EXPORT_SYMBOL_GPL(force_sig_fault);
45
46/*
47 * Protects cbe_spu_info and spu->number.
48 */
49static DEFINE_SPINLOCK(spu_lock);
50
51/*
52 * List of all spus in the system.
53 *
54 * This list is iterated by callers from irq context and callers that
55 * want to sleep.  Thus modifications need to be done with both
56 * spu_full_list_lock and spu_full_list_mutex held, while iterating
57 * through it requires either of these locks.
58 *
59 * In addition spu_full_list_lock protects all assignments to
60 * spu->mm.
61 */
62static LIST_HEAD(spu_full_list);
63static DEFINE_SPINLOCK(spu_full_list_lock);
64static DEFINE_MUTEX(spu_full_list_mutex);
65
66void spu_invalidate_slbs(struct spu *spu)
67{
68	struct spu_priv2 __iomem *priv2 = spu->priv2;
69	unsigned long flags;
70
71	spin_lock_irqsave(&spu->register_lock, flags);
72	if (spu_mfc_sr1_get(spu) & MFC_STATE1_RELOCATE_MASK)
73		out_be64(&priv2->slb_invalidate_all_W, 0UL);
74	spin_unlock_irqrestore(&spu->register_lock, flags);
75}
76EXPORT_SYMBOL_GPL(spu_invalidate_slbs);
77
78/* This is called by the MM core when a segment size is changed, to
79 * request a flush of all the SPEs using a given mm
80 */
81void spu_flush_all_slbs(struct mm_struct *mm)
82{
83	struct spu *spu;
84	unsigned long flags;
85
86	spin_lock_irqsave(&spu_full_list_lock, flags);
87	list_for_each_entry(spu, &spu_full_list, full_list) {
88		if (spu->mm == mm)
89			spu_invalidate_slbs(spu);
90	}
91	spin_unlock_irqrestore(&spu_full_list_lock, flags);
92}
93
94/* The hack below stinks... try to do something better one of
95 * these days... Does it even work properly with NR_CPUS == 1 ?
96 */
97static inline void mm_needs_global_tlbie(struct mm_struct *mm)
98{
99	int nr = (NR_CPUS > 1) ? NR_CPUS : NR_CPUS + 1;
100
101	/* Global TLBIE broadcast required with SPEs. */
102	bitmap_fill(cpumask_bits(mm_cpumask(mm)), nr);
103}
104
105void spu_associate_mm(struct spu *spu, struct mm_struct *mm)
106{
107	unsigned long flags;
108
109	spin_lock_irqsave(&spu_full_list_lock, flags);
110	spu->mm = mm;
111	spin_unlock_irqrestore(&spu_full_list_lock, flags);
112	if (mm)
113		mm_needs_global_tlbie(mm);
114}
115EXPORT_SYMBOL_GPL(spu_associate_mm);
116
117int spu_64k_pages_available(void)
118{
119	return mmu_psize_defs[MMU_PAGE_64K].shift != 0;
120}
121EXPORT_SYMBOL_GPL(spu_64k_pages_available);
122
123static void spu_restart_dma(struct spu *spu)
124{
125	struct spu_priv2 __iomem *priv2 = spu->priv2;
126
127	if (!test_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags))
128		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
129	else {
130		set_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
131		mb();
132	}
133}
134
135static inline void spu_load_slb(struct spu *spu, int slbe, struct copro_slb *slb)
136{
137	struct spu_priv2 __iomem *priv2 = spu->priv2;
138
139	pr_debug("%s: adding SLB[%d] 0x%016llx 0x%016llx\n",
140			__func__, slbe, slb->vsid, slb->esid);
141
142	out_be64(&priv2->slb_index_W, slbe);
143	/* set invalid before writing vsid */
144	out_be64(&priv2->slb_esid_RW, 0);
145	/* now it's safe to write the vsid */
146	out_be64(&priv2->slb_vsid_RW, slb->vsid);
147	/* setting the new esid makes the entry valid again */
148	out_be64(&priv2->slb_esid_RW, slb->esid);
149}
150
151static int __spu_trap_data_seg(struct spu *spu, unsigned long ea)
152{
153	struct copro_slb slb;
154	int ret;
155
156	ret = copro_calculate_slb(spu->mm, ea, &slb);
157	if (ret)
158		return ret;
159
160	spu_load_slb(spu, spu->slb_replace, &slb);
161
162	spu->slb_replace++;
163	if (spu->slb_replace >= 8)
164		spu->slb_replace = 0;
165
166	spu_restart_dma(spu);
167	spu->stats.slb_flt++;
168	return 0;
169}
170
171extern int hash_page(unsigned long ea, unsigned long access,
172		     unsigned long trap, unsigned long dsisr); //XXX
173static int __spu_trap_data_map(struct spu *spu, unsigned long ea, u64 dsisr)
174{
175	int ret;
176
177	pr_debug("%s, %llx, %lx\n", __func__, dsisr, ea);
178
179	/*
180	 * Handle kernel space hash faults immediately. User hash
181	 * faults need to be deferred to process context.
182	 */
183	if ((dsisr & MFC_DSISR_PTE_NOT_FOUND) &&
184	    (get_region_id(ea) != USER_REGION_ID)) {
185
186		spin_unlock(&spu->register_lock);
187		ret = hash_page(ea,
188				_PAGE_PRESENT | _PAGE_READ | _PAGE_PRIVILEGED,
189				0x300, dsisr);
190		spin_lock(&spu->register_lock);
191
192		if (!ret) {
193			spu_restart_dma(spu);
194			return 0;
195		}
196	}
197
198	spu->class_1_dar = ea;
199	spu->class_1_dsisr = dsisr;
200
201	spu->stop_callback(spu, 1);
202
203	spu->class_1_dar = 0;
204	spu->class_1_dsisr = 0;
205
206	return 0;
207}
208
209static void __spu_kernel_slb(void *addr, struct copro_slb *slb)
210{
211	unsigned long ea = (unsigned long)addr;
212	u64 llp;
213
214	if (get_region_id(ea) == LINEAR_MAP_REGION_ID)
215		llp = mmu_psize_defs[mmu_linear_psize].sllp;
216	else
217		llp = mmu_psize_defs[mmu_virtual_psize].sllp;
218
219	slb->vsid = (get_kernel_vsid(ea, MMU_SEGSIZE_256M) << SLB_VSID_SHIFT) |
220		SLB_VSID_KERNEL | llp;
221	slb->esid = (ea & ESID_MASK) | SLB_ESID_V;
222}
223
224/**
225 * Given an array of @nr_slbs SLB entries, @slbs, return non-zero if the
226 * address @new_addr is present.
227 */
228static inline int __slb_present(struct copro_slb *slbs, int nr_slbs,
229		void *new_addr)
230{
231	unsigned long ea = (unsigned long)new_addr;
232	int i;
233
234	for (i = 0; i < nr_slbs; i++)
235		if (!((slbs[i].esid ^ ea) & ESID_MASK))
236			return 1;
237
238	return 0;
239}
240
241/**
242 * Setup the SPU kernel SLBs, in preparation for a context save/restore. We
243 * need to map both the context save area, and the save/restore code.
244 *
245 * Because the lscsa and code may cross segment boundaries, we check to see
246 * if mappings are required for the start and end of each range. We currently
247 * assume that the mappings are smaller that one segment - if not, something
248 * is seriously wrong.
249 */
250void spu_setup_kernel_slbs(struct spu *spu, struct spu_lscsa *lscsa,
251		void *code, int code_size)
252{
253	struct copro_slb slbs[4];
254	int i, nr_slbs = 0;
255	/* start and end addresses of both mappings */
256	void *addrs[] = {
257		lscsa, (void *)lscsa + sizeof(*lscsa) - 1,
258		code, code + code_size - 1
259	};
260
261	/* check the set of addresses, and create a new entry in the slbs array
262	 * if there isn't already a SLB for that address */
263	for (i = 0; i < ARRAY_SIZE(addrs); i++) {
264		if (__slb_present(slbs, nr_slbs, addrs[i]))
265			continue;
266
267		__spu_kernel_slb(addrs[i], &slbs[nr_slbs]);
268		nr_slbs++;
269	}
270
271	spin_lock_irq(&spu->register_lock);
272	/* Add the set of SLBs */
273	for (i = 0; i < nr_slbs; i++)
274		spu_load_slb(spu, i, &slbs[i]);
275	spin_unlock_irq(&spu->register_lock);
276}
277EXPORT_SYMBOL_GPL(spu_setup_kernel_slbs);
278
279static irqreturn_t
280spu_irq_class_0(int irq, void *data)
281{
282	struct spu *spu;
283	unsigned long stat, mask;
284
285	spu = data;
286
287	spin_lock(&spu->register_lock);
288	mask = spu_int_mask_get(spu, 0);
289	stat = spu_int_stat_get(spu, 0) & mask;
290
291	spu->class_0_pending |= stat;
292	spu->class_0_dar = spu_mfc_dar_get(spu);
293	spu->stop_callback(spu, 0);
294	spu->class_0_pending = 0;
295	spu->class_0_dar = 0;
296
297	spu_int_stat_clear(spu, 0, stat);
298	spin_unlock(&spu->register_lock);
299
300	return IRQ_HANDLED;
301}
302
303static irqreturn_t
304spu_irq_class_1(int irq, void *data)
305{
306	struct spu *spu;
307	unsigned long stat, mask, dar, dsisr;
308
309	spu = data;
310
311	/* atomically read & clear class1 status. */
312	spin_lock(&spu->register_lock);
313	mask  = spu_int_mask_get(spu, 1);
314	stat  = spu_int_stat_get(spu, 1) & mask;
315	dar   = spu_mfc_dar_get(spu);
316	dsisr = spu_mfc_dsisr_get(spu);
317	if (stat & CLASS1_STORAGE_FAULT_INTR)
318		spu_mfc_dsisr_set(spu, 0ul);
319	spu_int_stat_clear(spu, 1, stat);
320
321	pr_debug("%s: %lx %lx %lx %lx\n", __func__, mask, stat,
322			dar, dsisr);
323
324	if (stat & CLASS1_SEGMENT_FAULT_INTR)
325		__spu_trap_data_seg(spu, dar);
326
327	if (stat & CLASS1_STORAGE_FAULT_INTR)
328		__spu_trap_data_map(spu, dar, dsisr);
329
330	if (stat & CLASS1_LS_COMPARE_SUSPEND_ON_GET_INTR)
331		;
332
333	if (stat & CLASS1_LS_COMPARE_SUSPEND_ON_PUT_INTR)
334		;
335
336	spu->class_1_dsisr = 0;
337	spu->class_1_dar = 0;
338
339	spin_unlock(&spu->register_lock);
340
341	return stat ? IRQ_HANDLED : IRQ_NONE;
342}
343
344static irqreturn_t
345spu_irq_class_2(int irq, void *data)
346{
347	struct spu *spu;
348	unsigned long stat;
349	unsigned long mask;
350	const int mailbox_intrs =
351		CLASS2_MAILBOX_THRESHOLD_INTR | CLASS2_MAILBOX_INTR;
352
353	spu = data;
354	spin_lock(&spu->register_lock);
355	stat = spu_int_stat_get(spu, 2);
356	mask = spu_int_mask_get(spu, 2);
357	/* ignore interrupts we're not waiting for */
358	stat &= mask;
359	/* mailbox interrupts are level triggered. mask them now before
360	 * acknowledging */
361	if (stat & mailbox_intrs)
362		spu_int_mask_and(spu, 2, ~(stat & mailbox_intrs));
363	/* acknowledge all interrupts before the callbacks */
364	spu_int_stat_clear(spu, 2, stat);
365
366	pr_debug("class 2 interrupt %d, %lx, %lx\n", irq, stat, mask);
367
368	if (stat & CLASS2_MAILBOX_INTR)
369		spu->ibox_callback(spu);
370
371	if (stat & CLASS2_SPU_STOP_INTR)
372		spu->stop_callback(spu, 2);
373
374	if (stat & CLASS2_SPU_HALT_INTR)
375		spu->stop_callback(spu, 2);
376
377	if (stat & CLASS2_SPU_DMA_TAG_GROUP_COMPLETE_INTR)
378		spu->mfc_callback(spu);
379
380	if (stat & CLASS2_MAILBOX_THRESHOLD_INTR)
381		spu->wbox_callback(spu);
382
383	spu->stats.class2_intr++;
384
385	spin_unlock(&spu->register_lock);
386
387	return stat ? IRQ_HANDLED : IRQ_NONE;
388}
389
390static int spu_request_irqs(struct spu *spu)
391{
392	int ret = 0;
393
394	if (spu->irqs[0]) {
395		snprintf(spu->irq_c0, sizeof (spu->irq_c0), "spe%02d.0",
396			 spu->number);
397		ret = request_irq(spu->irqs[0], spu_irq_class_0,
398				  0, spu->irq_c0, spu);
399		if (ret)
400			goto bail0;
401	}
402	if (spu->irqs[1]) {
403		snprintf(spu->irq_c1, sizeof (spu->irq_c1), "spe%02d.1",
404			 spu->number);
405		ret = request_irq(spu->irqs[1], spu_irq_class_1,
406				  0, spu->irq_c1, spu);
407		if (ret)
408			goto bail1;
409	}
410	if (spu->irqs[2]) {
411		snprintf(spu->irq_c2, sizeof (spu->irq_c2), "spe%02d.2",
412			 spu->number);
413		ret = request_irq(spu->irqs[2], spu_irq_class_2,
414				  0, spu->irq_c2, spu);
415		if (ret)
416			goto bail2;
417	}
418	return 0;
419
420bail2:
421	if (spu->irqs[1])
422		free_irq(spu->irqs[1], spu);
423bail1:
424	if (spu->irqs[0])
425		free_irq(spu->irqs[0], spu);
426bail0:
427	return ret;
428}
429
430static void spu_free_irqs(struct spu *spu)
431{
432	if (spu->irqs[0])
433		free_irq(spu->irqs[0], spu);
434	if (spu->irqs[1])
435		free_irq(spu->irqs[1], spu);
436	if (spu->irqs[2])
437		free_irq(spu->irqs[2], spu);
438}
439
440void spu_init_channels(struct spu *spu)
441{
442	static const struct {
443		 unsigned channel;
444		 unsigned count;
445	} zero_list[] = {
446		{ 0x00, 1, }, { 0x01, 1, }, { 0x03, 1, }, { 0x04, 1, },
447		{ 0x18, 1, }, { 0x19, 1, }, { 0x1b, 1, }, { 0x1d, 1, },
448	}, count_list[] = {
449		{ 0x00, 0, }, { 0x03, 0, }, { 0x04, 0, }, { 0x15, 16, },
450		{ 0x17, 1, }, { 0x18, 0, }, { 0x19, 0, }, { 0x1b, 0, },
451		{ 0x1c, 1, }, { 0x1d, 0, }, { 0x1e, 1, },
452	};
453	struct spu_priv2 __iomem *priv2;
454	int i;
455
456	priv2 = spu->priv2;
457
458	/* initialize all channel data to zero */
459	for (i = 0; i < ARRAY_SIZE(zero_list); i++) {
460		int count;
461
462		out_be64(&priv2->spu_chnlcntptr_RW, zero_list[i].channel);
463		for (count = 0; count < zero_list[i].count; count++)
464			out_be64(&priv2->spu_chnldata_RW, 0);
465	}
466
467	/* initialize channel counts to meaningful values */
468	for (i = 0; i < ARRAY_SIZE(count_list); i++) {
469		out_be64(&priv2->spu_chnlcntptr_RW, count_list[i].channel);
470		out_be64(&priv2->spu_chnlcnt_RW, count_list[i].count);
471	}
472}
473EXPORT_SYMBOL_GPL(spu_init_channels);
474
475static struct bus_type spu_subsys = {
476	.name = "spu",
477	.dev_name = "spu",
478};
479
480int spu_add_dev_attr(struct device_attribute *attr)
481{
482	struct spu *spu;
483
484	mutex_lock(&spu_full_list_mutex);
485	list_for_each_entry(spu, &spu_full_list, full_list)
486		device_create_file(&spu->dev, attr);
487	mutex_unlock(&spu_full_list_mutex);
488
489	return 0;
490}
491EXPORT_SYMBOL_GPL(spu_add_dev_attr);
492
493int spu_add_dev_attr_group(struct attribute_group *attrs)
494{
495	struct spu *spu;
496	int rc = 0;
497
498	mutex_lock(&spu_full_list_mutex);
499	list_for_each_entry(spu, &spu_full_list, full_list) {
500		rc = sysfs_create_group(&spu->dev.kobj, attrs);
501
502		/* we're in trouble here, but try unwinding anyway */
503		if (rc) {
504			printk(KERN_ERR "%s: can't create sysfs group '%s'\n",
505					__func__, attrs->name);
506
507			list_for_each_entry_continue_reverse(spu,
508					&spu_full_list, full_list)
509				sysfs_remove_group(&spu->dev.kobj, attrs);
510			break;
511		}
512	}
513
514	mutex_unlock(&spu_full_list_mutex);
515
516	return rc;
517}
518EXPORT_SYMBOL_GPL(spu_add_dev_attr_group);
519
520
521void spu_remove_dev_attr(struct device_attribute *attr)
522{
523	struct spu *spu;
524
525	mutex_lock(&spu_full_list_mutex);
526	list_for_each_entry(spu, &spu_full_list, full_list)
527		device_remove_file(&spu->dev, attr);
528	mutex_unlock(&spu_full_list_mutex);
529}
530EXPORT_SYMBOL_GPL(spu_remove_dev_attr);
531
532void spu_remove_dev_attr_group(struct attribute_group *attrs)
533{
534	struct spu *spu;
535
536	mutex_lock(&spu_full_list_mutex);
537	list_for_each_entry(spu, &spu_full_list, full_list)
538		sysfs_remove_group(&spu->dev.kobj, attrs);
539	mutex_unlock(&spu_full_list_mutex);
540}
541EXPORT_SYMBOL_GPL(spu_remove_dev_attr_group);
542
543static int spu_create_dev(struct spu *spu)
544{
545	int ret;
546
547	spu->dev.id = spu->number;
548	spu->dev.bus = &spu_subsys;
549	ret = device_register(&spu->dev);
550	if (ret) {
551		printk(KERN_ERR "Can't register SPU %d with sysfs\n",
552				spu->number);
553		return ret;
554	}
555
556	sysfs_add_device_to_node(&spu->dev, spu->node);
557
558	return 0;
559}
560
561static int __init create_spu(void *data)
562{
563	struct spu *spu;
564	int ret;
565	static int number;
566	unsigned long flags;
567
568	ret = -ENOMEM;
569	spu = kzalloc(sizeof (*spu), GFP_KERNEL);
570	if (!spu)
571		goto out;
572
573	spu->alloc_state = SPU_FREE;
574
575	spin_lock_init(&spu->register_lock);
576	spin_lock(&spu_lock);
577	spu->number = number++;
578	spin_unlock(&spu_lock);
579
580	ret = spu_create_spu(spu, data);
581
582	if (ret)
583		goto out_free;
584
585	spu_mfc_sdr_setup(spu);
586	spu_mfc_sr1_set(spu, 0x33);
587	ret = spu_request_irqs(spu);
588	if (ret)
589		goto out_destroy;
590
591	ret = spu_create_dev(spu);
592	if (ret)
593		goto out_free_irqs;
594
595	mutex_lock(&cbe_spu_info[spu->node].list_mutex);
596	list_add(&spu->cbe_list, &cbe_spu_info[spu->node].spus);
597	cbe_spu_info[spu->node].n_spus++;
598	mutex_unlock(&cbe_spu_info[spu->node].list_mutex);
599
600	mutex_lock(&spu_full_list_mutex);
601	spin_lock_irqsave(&spu_full_list_lock, flags);
602	list_add(&spu->full_list, &spu_full_list);
603	spin_unlock_irqrestore(&spu_full_list_lock, flags);
604	mutex_unlock(&spu_full_list_mutex);
605
606	spu->stats.util_state = SPU_UTIL_IDLE_LOADED;
607	spu->stats.tstamp = ktime_get_ns();
608
609	INIT_LIST_HEAD(&spu->aff_list);
610
611	goto out;
612
613out_free_irqs:
614	spu_free_irqs(spu);
615out_destroy:
616	spu_destroy_spu(spu);
617out_free:
618	kfree(spu);
619out:
620	return ret;
621}
622
623static const char *spu_state_names[] = {
624	"user", "system", "iowait", "idle"
625};
626
627static unsigned long long spu_acct_time(struct spu *spu,
628		enum spu_utilization_state state)
629{
630	unsigned long long time = spu->stats.times[state];
631
632	/*
633	 * If the spu is idle or the context is stopped, utilization
634	 * statistics are not updated.  Apply the time delta from the
635	 * last recorded state of the spu.
636	 */
637	if (spu->stats.util_state == state)
638		time += ktime_get_ns() - spu->stats.tstamp;
639
640	return time / NSEC_PER_MSEC;
641}
642
643
644static ssize_t spu_stat_show(struct device *dev,
645				struct device_attribute *attr, char *buf)
646{
647	struct spu *spu = container_of(dev, struct spu, dev);
648
649	return sprintf(buf, "%s %llu %llu %llu %llu "
650		      "%llu %llu %llu %llu %llu %llu %llu %llu\n",
651		spu_state_names[spu->stats.util_state],
652		spu_acct_time(spu, SPU_UTIL_USER),
653		spu_acct_time(spu, SPU_UTIL_SYSTEM),
654		spu_acct_time(spu, SPU_UTIL_IOWAIT),
655		spu_acct_time(spu, SPU_UTIL_IDLE_LOADED),
656		spu->stats.vol_ctx_switch,
657		spu->stats.invol_ctx_switch,
658		spu->stats.slb_flt,
659		spu->stats.hash_flt,
660		spu->stats.min_flt,
661		spu->stats.maj_flt,
662		spu->stats.class2_intr,
663		spu->stats.libassist);
664}
665
666static DEVICE_ATTR(stat, 0444, spu_stat_show, NULL);
667
668#ifdef CONFIG_KEXEC_CORE
669
670struct crash_spu_info {
671	struct spu *spu;
672	u32 saved_spu_runcntl_RW;
673	u32 saved_spu_status_R;
674	u32 saved_spu_npc_RW;
675	u64 saved_mfc_sr1_RW;
676	u64 saved_mfc_dar;
677	u64 saved_mfc_dsisr;
678};
679
680#define CRASH_NUM_SPUS	16	/* Enough for current hardware */
681static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
682
683static void crash_kexec_stop_spus(void)
684{
685	struct spu *spu;
686	int i;
687	u64 tmp;
688
689	for (i = 0; i < CRASH_NUM_SPUS; i++) {
690		if (!crash_spu_info[i].spu)
691			continue;
692
693		spu = crash_spu_info[i].spu;
694
695		crash_spu_info[i].saved_spu_runcntl_RW =
696			in_be32(&spu->problem->spu_runcntl_RW);
697		crash_spu_info[i].saved_spu_status_R =
698			in_be32(&spu->problem->spu_status_R);
699		crash_spu_info[i].saved_spu_npc_RW =
700			in_be32(&spu->problem->spu_npc_RW);
701
702		crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
703		crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
704		tmp = spu_mfc_sr1_get(spu);
705		crash_spu_info[i].saved_mfc_sr1_RW = tmp;
706
707		tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
708		spu_mfc_sr1_set(spu, tmp);
709
710		__delay(200);
711	}
712}
713
714static void crash_register_spus(struct list_head *list)
715{
716	struct spu *spu;
717	int ret;
718
719	list_for_each_entry(spu, list, full_list) {
720		if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
721			continue;
722
723		crash_spu_info[spu->number].spu = spu;
724	}
725
726	ret = crash_shutdown_register(&crash_kexec_stop_spus);
727	if (ret)
728		printk(KERN_ERR "Could not register SPU crash handler");
729}
730
731#else
732static inline void crash_register_spus(struct list_head *list)
733{
734}
735#endif
736
737static void spu_shutdown(void)
738{
739	struct spu *spu;
740
741	mutex_lock(&spu_full_list_mutex);
742	list_for_each_entry(spu, &spu_full_list, full_list) {
743		spu_free_irqs(spu);
744		spu_destroy_spu(spu);
745	}
746	mutex_unlock(&spu_full_list_mutex);
747}
748
749static struct syscore_ops spu_syscore_ops = {
750	.shutdown = spu_shutdown,
751};
752
753static int __init init_spu_base(void)
754{
755	int i, ret = 0;
756
757	for (i = 0; i < MAX_NUMNODES; i++) {
758		mutex_init(&cbe_spu_info[i].list_mutex);
759		INIT_LIST_HEAD(&cbe_spu_info[i].spus);
760	}
761
762	if (!spu_management_ops)
763		goto out;
764
765	/* create system subsystem for spus */
766	ret = subsys_system_register(&spu_subsys, NULL);
767	if (ret)
768		goto out;
769
770	ret = spu_enumerate_spus(create_spu);
771
772	if (ret < 0) {
773		printk(KERN_WARNING "%s: Error initializing spus\n",
774			__func__);
775		goto out_unregister_subsys;
776	}
777
778	if (ret > 0)
779		fb_append_extra_logo(&logo_spe_clut224, ret);
780
781	mutex_lock(&spu_full_list_mutex);
782	xmon_register_spus(&spu_full_list);
783	crash_register_spus(&spu_full_list);
784	mutex_unlock(&spu_full_list_mutex);
785	spu_add_dev_attr(&dev_attr_stat);
786	register_syscore_ops(&spu_syscore_ops);
787
788	spu_init_affinity();
789
790	return 0;
791
792 out_unregister_subsys:
793	bus_unregister(&spu_subsys);
794 out:
795	return ret;
796}
797device_initcall(init_spu_base);
798