xref: /linux/arch/s390/pci/pci.c (revision 7fc2cd2e4b398c57c9cf961cfea05eadbf34c05c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright IBM Corp. 2012
4  *
5  * Author(s):
6  *   Jan Glauber <jang@linux.vnet.ibm.com>
7  *
8  * The System z PCI code is a rewrite from a prototype by
9  * the following people (Kudoz!):
10  *   Alexander Schmidt
11  *   Christoph Raisch
12  *   Hannes Hering
13  *   Hoang-Nam Nguyen
14  *   Jan-Bernd Themann
15  *   Stefan Roscher
16  *   Thomas Klein
17  */
18 
19 #define pr_fmt(fmt) "zpci: " fmt
20 
21 #include <linux/kernel.h>
22 #include <linux/slab.h>
23 #include <linux/err.h>
24 #include <linux/export.h>
25 #include <linux/delay.h>
26 #include <linux/seq_file.h>
27 #include <linux/jump_label.h>
28 #include <linux/pci.h>
29 #include <linux/printk.h>
30 #include <linux/lockdep.h>
31 #include <linux/list_sort.h>
32 
33 #include <asm/machine.h>
34 #include <asm/isc.h>
35 #include <asm/airq.h>
36 #include <asm/facility.h>
37 #include <asm/pci_insn.h>
38 #include <asm/pci_clp.h>
39 #include <asm/pci_dma.h>
40 
41 #include "pci_bus.h"
42 #include "pci_iov.h"
43 
44 /* list of all detected zpci devices */
45 static LIST_HEAD(zpci_list);
46 static DEFINE_SPINLOCK(zpci_list_lock);
47 static DEFINE_MUTEX(zpci_add_remove_lock);
48 
49 static DECLARE_BITMAP(zpci_domain, ZPCI_DOMAIN_BITMAP_SIZE);
50 static DEFINE_SPINLOCK(zpci_domain_lock);
51 
52 #define ZPCI_IOMAP_ENTRIES						\
53 	min(((unsigned long) ZPCI_NR_DEVICES * PCI_STD_NUM_BARS / 2),	\
54 	    ZPCI_IOMAP_MAX_ENTRIES)
55 
56 unsigned int s390_pci_no_rid;
57 
58 static DEFINE_SPINLOCK(zpci_iomap_lock);
59 static unsigned long *zpci_iomap_bitmap;
60 struct zpci_iomap_entry *zpci_iomap_start;
61 EXPORT_SYMBOL_GPL(zpci_iomap_start);
62 
63 DEFINE_STATIC_KEY_FALSE(have_mio);
64 
65 static struct kmem_cache *zdev_fmb_cache;
66 
67 /* AEN structures that must be preserved over KVM module re-insertion */
68 union zpci_sic_iib *zpci_aipb;
69 EXPORT_SYMBOL_GPL(zpci_aipb);
70 struct airq_iv *zpci_aif_sbv;
71 EXPORT_SYMBOL_GPL(zpci_aif_sbv);
72 
73 void zpci_zdev_put(struct zpci_dev *zdev)
74 {
75 	if (!zdev)
76 		return;
77 	mutex_lock(&zpci_add_remove_lock);
78 	kref_put_lock(&zdev->kref, zpci_release_device, &zpci_list_lock);
79 	mutex_unlock(&zpci_add_remove_lock);
80 }
81 
82 struct zpci_dev *get_zdev_by_fid(u32 fid)
83 {
84 	struct zpci_dev *tmp, *zdev = NULL;
85 
86 	spin_lock(&zpci_list_lock);
87 	list_for_each_entry(tmp, &zpci_list, entry) {
88 		if (tmp->fid == fid) {
89 			zdev = tmp;
90 			zpci_zdev_get(zdev);
91 			break;
92 		}
93 	}
94 	spin_unlock(&zpci_list_lock);
95 	return zdev;
96 }
97 
98 void zpci_remove_reserved_devices(void)
99 {
100 	struct zpci_dev *tmp, *zdev;
101 	enum zpci_state state;
102 	LIST_HEAD(remove);
103 
104 	spin_lock(&zpci_list_lock);
105 	list_for_each_entry_safe(zdev, tmp, &zpci_list, entry) {
106 		if (zdev->state == ZPCI_FN_STATE_STANDBY &&
107 		    !clp_get_state(zdev->fid, &state) &&
108 		    state == ZPCI_FN_STATE_RESERVED)
109 			list_move_tail(&zdev->entry, &remove);
110 	}
111 	spin_unlock(&zpci_list_lock);
112 
113 	list_for_each_entry_safe(zdev, tmp, &remove, entry)
114 		zpci_device_reserved(zdev);
115 }
116 
117 int pci_domain_nr(struct pci_bus *bus)
118 {
119 	return ((struct zpci_bus *) bus->sysdata)->domain_nr;
120 }
121 EXPORT_SYMBOL_GPL(pci_domain_nr);
122 
123 int pci_proc_domain(struct pci_bus *bus)
124 {
125 	return pci_domain_nr(bus);
126 }
127 EXPORT_SYMBOL_GPL(pci_proc_domain);
128 
129 /* Modify PCI: Register I/O address translation parameters */
130 int zpci_register_ioat(struct zpci_dev *zdev, u8 dmaas,
131 		       u64 base, u64 limit, u64 iota, u8 *status)
132 {
133 	u64 req = ZPCI_CREATE_REQ(zdev->fh, dmaas, ZPCI_MOD_FC_REG_IOAT);
134 	struct zpci_fib fib = {0};
135 	u8 cc;
136 
137 	fib.pba = base;
138 	/* Work around off by one in ISM virt device */
139 	if (zdev->pft == PCI_FUNC_TYPE_ISM && limit > base)
140 		fib.pal = limit + (1 << 12);
141 	else
142 		fib.pal = limit;
143 	fib.iota = iota;
144 	fib.gd = zdev->gisa;
145 	cc = zpci_mod_fc(req, &fib, status);
146 	if (cc)
147 		zpci_dbg(3, "reg ioat fid:%x, cc:%d, status:%d\n", zdev->fid, cc, *status);
148 	return cc;
149 }
150 EXPORT_SYMBOL_GPL(zpci_register_ioat);
151 
152 /* Modify PCI: Unregister I/O address translation parameters */
153 int zpci_unregister_ioat(struct zpci_dev *zdev, u8 dmaas)
154 {
155 	u64 req = ZPCI_CREATE_REQ(zdev->fh, dmaas, ZPCI_MOD_FC_DEREG_IOAT);
156 	struct zpci_fib fib = {0};
157 	u8 cc, status;
158 
159 	fib.gd = zdev->gisa;
160 
161 	cc = zpci_mod_fc(req, &fib, &status);
162 	if (cc)
163 		zpci_dbg(3, "unreg ioat fid:%x, cc:%d, status:%d\n", zdev->fid, cc, status);
164 	return cc;
165 }
166 
167 /* Modify PCI: Set PCI function measurement parameters */
168 int zpci_fmb_enable_device(struct zpci_dev *zdev)
169 {
170 	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_SET_MEASURE);
171 	struct zpci_iommu_ctrs *ctrs;
172 	struct zpci_fib fib = {0};
173 	unsigned long flags;
174 	u8 cc, status;
175 
176 	if (zdev->fmb || sizeof(*zdev->fmb) < zdev->fmb_length)
177 		return -EINVAL;
178 
179 	zdev->fmb = kmem_cache_zalloc(zdev_fmb_cache, GFP_KERNEL);
180 	if (!zdev->fmb)
181 		return -ENOMEM;
182 	WARN_ON((u64) zdev->fmb & 0xf);
183 
184 	/* reset software counters */
185 	spin_lock_irqsave(&zdev->dom_lock, flags);
186 	ctrs = zpci_get_iommu_ctrs(zdev);
187 	if (ctrs) {
188 		atomic64_set(&ctrs->mapped_pages, 0);
189 		atomic64_set(&ctrs->unmapped_pages, 0);
190 		atomic64_set(&ctrs->global_rpcits, 0);
191 		atomic64_set(&ctrs->sync_map_rpcits, 0);
192 		atomic64_set(&ctrs->sync_rpcits, 0);
193 	}
194 	spin_unlock_irqrestore(&zdev->dom_lock, flags);
195 
196 
197 	fib.fmb_addr = virt_to_phys(zdev->fmb);
198 	fib.gd = zdev->gisa;
199 	cc = zpci_mod_fc(req, &fib, &status);
200 	if (cc) {
201 		kmem_cache_free(zdev_fmb_cache, zdev->fmb);
202 		zdev->fmb = NULL;
203 	}
204 	return cc ? -EIO : 0;
205 }
206 
207 /* Modify PCI: Disable PCI function measurement */
208 int zpci_fmb_disable_device(struct zpci_dev *zdev)
209 {
210 	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_SET_MEASURE);
211 	struct zpci_fib fib = {0};
212 	u8 cc, status;
213 
214 	if (!zdev->fmb)
215 		return -EINVAL;
216 
217 	fib.gd = zdev->gisa;
218 
219 	/* Function measurement is disabled if fmb address is zero */
220 	cc = zpci_mod_fc(req, &fib, &status);
221 	if (cc == 3) /* Function already gone. */
222 		cc = 0;
223 
224 	if (!cc) {
225 		kmem_cache_free(zdev_fmb_cache, zdev->fmb);
226 		zdev->fmb = NULL;
227 	}
228 	return cc ? -EIO : 0;
229 }
230 
231 static int zpci_cfg_load(struct zpci_dev *zdev, int offset, u32 *val, u8 len)
232 {
233 	u64 req = ZPCI_CREATE_REQ(zdev->fh, ZPCI_PCIAS_CFGSPC, len);
234 	u64 data;
235 	int rc;
236 
237 	rc = __zpci_load(&data, req, offset);
238 	if (!rc) {
239 		data = le64_to_cpu((__force __le64) data);
240 		data >>= (8 - len) * 8;
241 		*val = (u32) data;
242 	} else
243 		*val = 0xffffffff;
244 	return rc;
245 }
246 
247 static int zpci_cfg_store(struct zpci_dev *zdev, int offset, u32 val, u8 len)
248 {
249 	u64 req = ZPCI_CREATE_REQ(zdev->fh, ZPCI_PCIAS_CFGSPC, len);
250 	u64 data = val;
251 	int rc;
252 
253 	data <<= (8 - len) * 8;
254 	data = (__force u64) cpu_to_le64(data);
255 	rc = __zpci_store(data, req, offset);
256 	return rc;
257 }
258 
259 resource_size_t pcibios_align_resource(void *data, const struct resource *res,
260 				       resource_size_t size,
261 				       resource_size_t align)
262 {
263 	return 0;
264 }
265 
266 void __iomem *ioremap_prot(phys_addr_t phys_addr, size_t size,
267 			   pgprot_t prot)
268 {
269 	/*
270 	 * When PCI MIO instructions are unavailable the "physical" address
271 	 * encodes a hint for accessing the PCI memory space it represents.
272 	 * Just pass it unchanged such that ioread/iowrite can decode it.
273 	 */
274 	if (!static_branch_unlikely(&have_mio))
275 		return (void __iomem *)phys_addr;
276 
277 	return generic_ioremap_prot(phys_addr, size, prot);
278 }
279 EXPORT_SYMBOL(ioremap_prot);
280 
281 void iounmap(volatile void __iomem *addr)
282 {
283 	if (static_branch_likely(&have_mio))
284 		generic_iounmap(addr);
285 }
286 EXPORT_SYMBOL(iounmap);
287 
288 /* Create a virtual mapping cookie for a PCI BAR */
289 static void __iomem *pci_iomap_range_fh(struct pci_dev *pdev, int bar,
290 					unsigned long offset, unsigned long max)
291 {
292 	struct zpci_dev *zdev =	to_zpci(pdev);
293 	int idx;
294 
295 	idx = zdev->bars[bar].map_idx;
296 	spin_lock(&zpci_iomap_lock);
297 	/* Detect overrun */
298 	WARN_ON(!++zpci_iomap_start[idx].count);
299 	zpci_iomap_start[idx].fh = zdev->fh;
300 	zpci_iomap_start[idx].bar = bar;
301 	spin_unlock(&zpci_iomap_lock);
302 
303 	return (void __iomem *) ZPCI_ADDR(idx) + offset;
304 }
305 
306 static void __iomem *pci_iomap_range_mio(struct pci_dev *pdev, int bar,
307 					 unsigned long offset,
308 					 unsigned long max)
309 {
310 	unsigned long barsize = pci_resource_len(pdev, bar);
311 	struct zpci_dev *zdev = to_zpci(pdev);
312 	void __iomem *iova;
313 
314 	iova = ioremap((unsigned long) zdev->bars[bar].mio_wt, barsize);
315 	return iova ? iova + offset : iova;
316 }
317 
318 void __iomem *pci_iomap_range(struct pci_dev *pdev, int bar,
319 			      unsigned long offset, unsigned long max)
320 {
321 	if (bar >= PCI_STD_NUM_BARS || !pci_resource_len(pdev, bar))
322 		return NULL;
323 
324 	if (static_branch_likely(&have_mio))
325 		return pci_iomap_range_mio(pdev, bar, offset, max);
326 	else
327 		return pci_iomap_range_fh(pdev, bar, offset, max);
328 }
329 EXPORT_SYMBOL(pci_iomap_range);
330 
331 void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long maxlen)
332 {
333 	return pci_iomap_range(dev, bar, 0, maxlen);
334 }
335 EXPORT_SYMBOL(pci_iomap);
336 
337 static void __iomem *pci_iomap_wc_range_mio(struct pci_dev *pdev, int bar,
338 					    unsigned long offset, unsigned long max)
339 {
340 	unsigned long barsize = pci_resource_len(pdev, bar);
341 	struct zpci_dev *zdev = to_zpci(pdev);
342 	void __iomem *iova;
343 
344 	iova = ioremap((unsigned long) zdev->bars[bar].mio_wb, barsize);
345 	return iova ? iova + offset : iova;
346 }
347 
348 void __iomem *pci_iomap_wc_range(struct pci_dev *pdev, int bar,
349 				 unsigned long offset, unsigned long max)
350 {
351 	if (bar >= PCI_STD_NUM_BARS || !pci_resource_len(pdev, bar))
352 		return NULL;
353 
354 	if (static_branch_likely(&have_mio))
355 		return pci_iomap_wc_range_mio(pdev, bar, offset, max);
356 	else
357 		return pci_iomap_range_fh(pdev, bar, offset, max);
358 }
359 EXPORT_SYMBOL(pci_iomap_wc_range);
360 
361 void __iomem *pci_iomap_wc(struct pci_dev *dev, int bar, unsigned long maxlen)
362 {
363 	return pci_iomap_wc_range(dev, bar, 0, maxlen);
364 }
365 EXPORT_SYMBOL(pci_iomap_wc);
366 
367 static void pci_iounmap_fh(struct pci_dev *pdev, void __iomem *addr)
368 {
369 	unsigned int idx = ZPCI_IDX(addr);
370 
371 	spin_lock(&zpci_iomap_lock);
372 	/* Detect underrun */
373 	WARN_ON(!zpci_iomap_start[idx].count);
374 	if (!--zpci_iomap_start[idx].count) {
375 		zpci_iomap_start[idx].fh = 0;
376 		zpci_iomap_start[idx].bar = 0;
377 	}
378 	spin_unlock(&zpci_iomap_lock);
379 }
380 
381 static void pci_iounmap_mio(struct pci_dev *pdev, void __iomem *addr)
382 {
383 	iounmap(addr);
384 }
385 
386 void pci_iounmap(struct pci_dev *pdev, void __iomem *addr)
387 {
388 	if (static_branch_likely(&have_mio))
389 		pci_iounmap_mio(pdev, addr);
390 	else
391 		pci_iounmap_fh(pdev, addr);
392 }
393 EXPORT_SYMBOL(pci_iounmap);
394 
395 static int pci_read(struct pci_bus *bus, unsigned int devfn, int where,
396 		    int size, u32 *val)
397 {
398 	struct zpci_dev *zdev = zdev_from_bus(bus, devfn);
399 
400 	return (zdev) ? zpci_cfg_load(zdev, where, val, size) : -ENODEV;
401 }
402 
403 static int pci_write(struct pci_bus *bus, unsigned int devfn, int where,
404 		     int size, u32 val)
405 {
406 	struct zpci_dev *zdev = zdev_from_bus(bus, devfn);
407 
408 	return (zdev) ? zpci_cfg_store(zdev, where, val, size) : -ENODEV;
409 }
410 
411 static struct pci_ops pci_root_ops = {
412 	.read = pci_read,
413 	.write = pci_write,
414 };
415 
416 static void zpci_map_resources(struct pci_dev *pdev)
417 {
418 	struct zpci_dev *zdev = to_zpci(pdev);
419 	resource_size_t len;
420 	int i;
421 
422 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
423 		len = pci_resource_len(pdev, i);
424 		if (!len)
425 			continue;
426 
427 		if (zpci_use_mio(zdev))
428 			pdev->resource[i].start =
429 				(resource_size_t __force) zdev->bars[i].mio_wt;
430 		else
431 			pdev->resource[i].start = (resource_size_t __force)
432 				pci_iomap_range_fh(pdev, i, 0, 0);
433 		pdev->resource[i].end = pdev->resource[i].start + len - 1;
434 	}
435 
436 	zpci_iov_map_resources(pdev);
437 }
438 
439 static void zpci_unmap_resources(struct pci_dev *pdev)
440 {
441 	struct zpci_dev *zdev = to_zpci(pdev);
442 	resource_size_t len;
443 	int i;
444 
445 	if (zpci_use_mio(zdev))
446 		return;
447 
448 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
449 		len = pci_resource_len(pdev, i);
450 		if (!len)
451 			continue;
452 		pci_iounmap_fh(pdev, (void __iomem __force *)
453 			       pdev->resource[i].start);
454 	}
455 }
456 
457 static int zpci_alloc_iomap(struct zpci_dev *zdev)
458 {
459 	unsigned long entry;
460 
461 	spin_lock(&zpci_iomap_lock);
462 	entry = find_first_zero_bit(zpci_iomap_bitmap, ZPCI_IOMAP_ENTRIES);
463 	if (entry == ZPCI_IOMAP_ENTRIES) {
464 		spin_unlock(&zpci_iomap_lock);
465 		return -ENOSPC;
466 	}
467 	set_bit(entry, zpci_iomap_bitmap);
468 	spin_unlock(&zpci_iomap_lock);
469 	return entry;
470 }
471 
472 static void zpci_free_iomap(struct zpci_dev *zdev, int entry)
473 {
474 	spin_lock(&zpci_iomap_lock);
475 	memset(&zpci_iomap_start[entry], 0, sizeof(struct zpci_iomap_entry));
476 	clear_bit(entry, zpci_iomap_bitmap);
477 	spin_unlock(&zpci_iomap_lock);
478 }
479 
480 static void zpci_do_update_iomap_fh(struct zpci_dev *zdev, u32 fh)
481 {
482 	int bar, idx;
483 
484 	spin_lock(&zpci_iomap_lock);
485 	for (bar = 0; bar < PCI_STD_NUM_BARS; bar++) {
486 		if (!zdev->bars[bar].size)
487 			continue;
488 		idx = zdev->bars[bar].map_idx;
489 		if (!zpci_iomap_start[idx].count)
490 			continue;
491 		WRITE_ONCE(zpci_iomap_start[idx].fh, zdev->fh);
492 	}
493 	spin_unlock(&zpci_iomap_lock);
494 }
495 
496 void zpci_update_fh(struct zpci_dev *zdev, u32 fh)
497 {
498 	if (!fh || zdev->fh == fh)
499 		return;
500 
501 	zdev->fh = fh;
502 	if (zpci_use_mio(zdev))
503 		return;
504 	if (zdev->has_resources && zdev_enabled(zdev))
505 		zpci_do_update_iomap_fh(zdev, fh);
506 }
507 
508 static struct resource *__alloc_res(struct zpci_dev *zdev, unsigned long start,
509 				    unsigned long size, unsigned long flags)
510 {
511 	struct resource *r;
512 
513 	r = kzalloc(sizeof(*r), GFP_KERNEL);
514 	if (!r)
515 		return NULL;
516 
517 	r->start = start;
518 	r->end = r->start + size - 1;
519 	r->flags = flags;
520 	r->name = zdev->res_name;
521 
522 	if (request_resource(&iomem_resource, r)) {
523 		kfree(r);
524 		return NULL;
525 	}
526 	return r;
527 }
528 
529 int zpci_setup_bus_resources(struct zpci_dev *zdev)
530 {
531 	unsigned long addr, size, flags;
532 	struct resource *res;
533 	int i, entry;
534 
535 	snprintf(zdev->res_name, sizeof(zdev->res_name),
536 		 "PCI Bus %04x:%02x", zdev->uid, ZPCI_BUS_NR);
537 
538 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
539 		if (!zdev->bars[i].size)
540 			continue;
541 		entry = zpci_alloc_iomap(zdev);
542 		if (entry < 0)
543 			return entry;
544 		zdev->bars[i].map_idx = entry;
545 
546 		/* only MMIO is supported */
547 		flags = IORESOURCE_MEM;
548 		if (zdev->bars[i].val & 8)
549 			flags |= IORESOURCE_PREFETCH;
550 		if (zdev->bars[i].val & 4)
551 			flags |= IORESOURCE_MEM_64;
552 
553 		if (zpci_use_mio(zdev))
554 			addr = (unsigned long) zdev->bars[i].mio_wt;
555 		else
556 			addr = ZPCI_ADDR(entry);
557 		size = 1UL << zdev->bars[i].size;
558 
559 		res = __alloc_res(zdev, addr, size, flags);
560 		if (!res) {
561 			zpci_free_iomap(zdev, entry);
562 			return -ENOMEM;
563 		}
564 		zdev->bars[i].res = res;
565 	}
566 	zdev->has_resources = 1;
567 
568 	return 0;
569 }
570 
571 static void zpci_cleanup_bus_resources(struct zpci_dev *zdev)
572 {
573 	struct resource *res;
574 	int i;
575 
576 	pci_lock_rescan_remove();
577 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
578 		res = zdev->bars[i].res;
579 		if (!res)
580 			continue;
581 
582 		release_resource(res);
583 		pci_bus_remove_resource(zdev->zbus->bus, res);
584 		zpci_free_iomap(zdev, zdev->bars[i].map_idx);
585 		zdev->bars[i].res = NULL;
586 		kfree(res);
587 	}
588 	zdev->has_resources = 0;
589 	pci_unlock_rescan_remove();
590 }
591 
592 int pcibios_device_add(struct pci_dev *pdev)
593 {
594 	struct zpci_dev *zdev = to_zpci(pdev);
595 	struct resource *res;
596 	int i;
597 
598 	/* The pdev has a reference to the zdev via its bus */
599 	zpci_zdev_get(zdev);
600 	if (pdev->is_physfn)
601 		pdev->no_vf_scan = 1;
602 
603 	zpci_map_resources(pdev);
604 
605 	for (i = 0; i < PCI_STD_NUM_BARS; i++) {
606 		res = &pdev->resource[i];
607 		if (res->parent || !res->flags)
608 			continue;
609 		pci_claim_resource(pdev, i);
610 	}
611 
612 	return 0;
613 }
614 
615 void pcibios_release_device(struct pci_dev *pdev)
616 {
617 	struct zpci_dev *zdev = to_zpci(pdev);
618 
619 	zpci_unmap_resources(pdev);
620 	zpci_zdev_put(zdev);
621 }
622 
623 int pcibios_enable_device(struct pci_dev *pdev, int mask)
624 {
625 	struct zpci_dev *zdev = to_zpci(pdev);
626 
627 	zpci_debug_init_device(zdev, dev_name(&pdev->dev));
628 	zpci_fmb_enable_device(zdev);
629 
630 	return pci_enable_resources(pdev, mask);
631 }
632 
633 void pcibios_disable_device(struct pci_dev *pdev)
634 {
635 	struct zpci_dev *zdev = to_zpci(pdev);
636 
637 	zpci_fmb_disable_device(zdev);
638 	zpci_debug_exit_device(zdev);
639 }
640 
641 static int __zpci_register_domain(int domain)
642 {
643 	spin_lock(&zpci_domain_lock);
644 	if (test_bit(domain, zpci_domain)) {
645 		spin_unlock(&zpci_domain_lock);
646 		pr_err("Domain %04x is already assigned\n", domain);
647 		return -EEXIST;
648 	}
649 	set_bit(domain, zpci_domain);
650 	spin_unlock(&zpci_domain_lock);
651 	return domain;
652 }
653 
654 static int __zpci_alloc_domain(void)
655 {
656 	int domain;
657 
658 	spin_lock(&zpci_domain_lock);
659 	/*
660 	 * We can always auto allocate domains below ZPCI_NR_DEVICES.
661 	 * There is either a free domain or we have reached the maximum in
662 	 * which case we would have bailed earlier.
663 	 */
664 	domain = find_first_zero_bit(zpci_domain, ZPCI_NR_DEVICES);
665 	set_bit(domain, zpci_domain);
666 	spin_unlock(&zpci_domain_lock);
667 	return domain;
668 }
669 
670 int zpci_alloc_domain(int domain)
671 {
672 	if (zpci_unique_uid) {
673 		if (domain)
674 			return __zpci_register_domain(domain);
675 		pr_warn("UID checking was active but no UID is provided: switching to automatic domain allocation\n");
676 		update_uid_checking(false);
677 	}
678 	return __zpci_alloc_domain();
679 }
680 
681 void zpci_free_domain(int domain)
682 {
683 	spin_lock(&zpci_domain_lock);
684 	clear_bit(domain, zpci_domain);
685 	spin_unlock(&zpci_domain_lock);
686 }
687 
688 
689 int zpci_enable_device(struct zpci_dev *zdev)
690 {
691 	u32 fh = zdev->fh;
692 	int rc = 0;
693 
694 	if (clp_enable_fh(zdev, &fh, ZPCI_NR_DMA_SPACES))
695 		rc = -EIO;
696 	else
697 		zpci_update_fh(zdev, fh);
698 	return rc;
699 }
700 EXPORT_SYMBOL_GPL(zpci_enable_device);
701 
702 int zpci_reenable_device(struct zpci_dev *zdev)
703 {
704 	u8 status;
705 	int rc;
706 
707 	rc = zpci_enable_device(zdev);
708 	if (rc)
709 		return rc;
710 
711 	rc = zpci_iommu_register_ioat(zdev, &status);
712 	if (rc)
713 		zpci_disable_device(zdev);
714 
715 	return rc;
716 }
717 EXPORT_SYMBOL_GPL(zpci_reenable_device);
718 
719 int zpci_disable_device(struct zpci_dev *zdev)
720 {
721 	u32 fh = zdev->fh;
722 	int cc, rc = 0;
723 
724 	cc = clp_disable_fh(zdev, &fh);
725 	if (!cc) {
726 		zpci_update_fh(zdev, fh);
727 	} else if (cc == CLP_RC_SETPCIFN_ALRDY) {
728 		pr_info("Disabling PCI function %08x had no effect as it was already disabled\n",
729 			zdev->fid);
730 		/* Function is already disabled - update handle */
731 		rc = clp_refresh_fh(zdev->fid, &fh);
732 		if (!rc) {
733 			zpci_update_fh(zdev, fh);
734 			rc = -EINVAL;
735 		}
736 	} else {
737 		rc = -EIO;
738 	}
739 	return rc;
740 }
741 EXPORT_SYMBOL_GPL(zpci_disable_device);
742 
743 /**
744  * zpci_hot_reset_device - perform a reset of the given zPCI function
745  * @zdev: the slot which should be reset
746  *
747  * Performs a low level reset of the zPCI function. The reset is low level in
748  * the sense that the zPCI function can be reset without detaching it from the
749  * common PCI subsystem. The reset may be performed while under control of
750  * either DMA or IOMMU APIs in which case the existing DMA/IOMMU translation
751  * table is reinstated at the end of the reset.
752  *
753  * After the reset the functions internal state is reset to an initial state
754  * equivalent to its state during boot when first probing a driver.
755  * Consequently after reset the PCI function requires re-initialization via the
756  * common PCI code including re-enabling IRQs via pci_alloc_irq_vectors()
757  * and enabling the function via e.g. pci_enable_device_flags(). The caller
758  * must guard against concurrent reset attempts.
759  *
760  * In most cases this function should not be called directly but through
761  * pci_reset_function() or pci_reset_bus() which handle the save/restore and
762  * locking - asserted by lockdep.
763  *
764  * Return: 0 on success and an error value otherwise
765  */
766 int zpci_hot_reset_device(struct zpci_dev *zdev)
767 {
768 	int rc;
769 
770 	lockdep_assert_held(&zdev->state_lock);
771 	zpci_dbg(3, "rst fid:%x, fh:%x\n", zdev->fid, zdev->fh);
772 	if (zdev_enabled(zdev)) {
773 		/* Disables device access, DMAs and IRQs (reset state) */
774 		rc = zpci_disable_device(zdev);
775 		/*
776 		 * Due to a z/VM vs LPAR inconsistency in the error state the
777 		 * FH may indicate an enabled device but disable says the
778 		 * device is already disabled don't treat it as an error here.
779 		 */
780 		if (rc == -EINVAL)
781 			rc = 0;
782 		if (rc)
783 			return rc;
784 	}
785 
786 	rc = zpci_reenable_device(zdev);
787 
788 	return rc;
789 }
790 
791 /**
792  * zpci_create_device() - Create a new zpci_dev and add it to the zbus
793  * @fid: Function ID of the device to be created
794  * @fh: Current Function Handle of the device to be created
795  * @state: Initial state after creation either Standby or Configured
796  *
797  * Allocates a new struct zpci_dev and queries the platform for its details.
798  * If successful the device can subsequently be added to the zPCI subsystem
799  * using zpci_add_device().
800  *
801  * Returns: the zdev on success or an error pointer otherwise
802  */
803 struct zpci_dev *zpci_create_device(u32 fid, u32 fh, enum zpci_state state)
804 {
805 	struct zpci_dev *zdev;
806 	int rc;
807 
808 	zdev = kzalloc(sizeof(*zdev), GFP_KERNEL);
809 	if (!zdev)
810 		return ERR_PTR(-ENOMEM);
811 
812 	/* FID and Function Handle are the static/dynamic identifiers */
813 	zdev->fid = fid;
814 	zdev->fh = fh;
815 
816 	/* Query function properties and update zdev */
817 	rc = clp_query_pci_fn(zdev);
818 	if (rc)
819 		goto error;
820 	zdev->state =  state;
821 
822 	mutex_init(&zdev->state_lock);
823 	mutex_init(&zdev->fmb_lock);
824 	mutex_init(&zdev->kzdev_lock);
825 
826 	return zdev;
827 
828 error:
829 	zpci_dbg(0, "crt fid:%x, rc:%d\n", fid, rc);
830 	kfree(zdev);
831 	return ERR_PTR(rc);
832 }
833 
834 /**
835  * zpci_add_device() - Add a previously created zPCI device to the zPCI subsystem
836  * @zdev: The zPCI device to be added
837  *
838  * A struct zpci_dev is added to the zPCI subsystem and to a virtual PCI bus creating
839  * a new one as necessary. A hotplug slot is created and events start to be handled.
840  * If successful from this point on zpci_zdev_get() and zpci_zdev_put() must be used.
841  * If adding the struct zpci_dev fails the device was not added and should be freed.
842  *
843  * Return: 0 on success, or an error code otherwise
844  */
845 int zpci_add_device(struct zpci_dev *zdev)
846 {
847 	int rc;
848 
849 	mutex_lock(&zpci_add_remove_lock);
850 	zpci_dbg(1, "add fid:%x, fh:%x, c:%d\n", zdev->fid, zdev->fh, zdev->state);
851 	rc = zpci_init_iommu(zdev);
852 	if (rc)
853 		goto error;
854 
855 	rc = zpci_bus_device_register(zdev, &pci_root_ops);
856 	if (rc)
857 		goto error_destroy_iommu;
858 
859 	kref_init(&zdev->kref);
860 	spin_lock(&zpci_list_lock);
861 	list_add_tail(&zdev->entry, &zpci_list);
862 	spin_unlock(&zpci_list_lock);
863 	mutex_unlock(&zpci_add_remove_lock);
864 	return 0;
865 
866 error_destroy_iommu:
867 	zpci_destroy_iommu(zdev);
868 error:
869 	zpci_dbg(0, "add fid:%x, rc:%d\n", zdev->fid, rc);
870 	mutex_unlock(&zpci_add_remove_lock);
871 	return rc;
872 }
873 
874 bool zpci_is_device_configured(struct zpci_dev *zdev)
875 {
876 	enum zpci_state state = zdev->state;
877 
878 	return state != ZPCI_FN_STATE_RESERVED &&
879 		state != ZPCI_FN_STATE_STANDBY;
880 }
881 
882 /**
883  * zpci_scan_configured_device() - Scan a freshly configured zpci_dev
884  * @zdev: The zpci_dev to be configured
885  * @fh: The general function handle supplied by the platform
886  *
887  * Given a device in the configuration state Configured, enables, scans and
888  * adds it to the common code PCI subsystem if possible. If any failure occurs,
889  * the zpci_dev is left disabled.
890  *
891  * Return: 0 on success, or an error code otherwise
892  */
893 int zpci_scan_configured_device(struct zpci_dev *zdev, u32 fh)
894 {
895 	zpci_update_fh(zdev, fh);
896 	return zpci_bus_scan_device(zdev);
897 }
898 
899 /**
900  * zpci_deconfigure_device() - Deconfigure a zpci_dev
901  * @zdev: The zpci_dev to configure
902  *
903  * Deconfigure a zPCI function that is currently configured and possibly known
904  * to the common code PCI subsystem.
905  * If any failure occurs the device is left as is.
906  *
907  * Return: 0 on success, or an error code otherwise
908  */
909 int zpci_deconfigure_device(struct zpci_dev *zdev)
910 {
911 	int rc;
912 
913 	lockdep_assert_held(&zdev->state_lock);
914 	if (zdev->state != ZPCI_FN_STATE_CONFIGURED)
915 		return 0;
916 
917 	if (zdev->zbus->bus)
918 		zpci_bus_remove_device(zdev, false);
919 
920 	if (zdev_enabled(zdev)) {
921 		rc = zpci_disable_device(zdev);
922 		if (rc)
923 			return rc;
924 	}
925 
926 	rc = sclp_pci_deconfigure(zdev->fid);
927 	zpci_dbg(3, "deconf fid:%x, rc:%d\n", zdev->fid, rc);
928 	if (rc)
929 		return rc;
930 	zdev->state = ZPCI_FN_STATE_STANDBY;
931 
932 	return 0;
933 }
934 
935 /**
936  * zpci_device_reserved() - Mark device as reserved
937  * @zdev: the zpci_dev that was reserved
938  *
939  * Handle the case that a given zPCI function was reserved by another system.
940  */
941 void zpci_device_reserved(struct zpci_dev *zdev)
942 {
943 	lockdep_assert_held(&zdev->state_lock);
944 	/* We may declare the device reserved multiple times */
945 	if (zdev->state == ZPCI_FN_STATE_RESERVED)
946 		return;
947 	zdev->state = ZPCI_FN_STATE_RESERVED;
948 	zpci_dbg(3, "rsv fid:%x\n", zdev->fid);
949 	/*
950 	 * The underlying device is gone. Allow the zdev to be freed
951 	 * as soon as all other references are gone by accounting for
952 	 * the removal as a dropped reference.
953 	 */
954 	zpci_zdev_put(zdev);
955 }
956 
957 void zpci_release_device(struct kref *kref)
958 {
959 	struct zpci_dev *zdev = container_of(kref, struct zpci_dev, kref);
960 
961 	lockdep_assert_held(&zpci_add_remove_lock);
962 	WARN_ON(zdev->state != ZPCI_FN_STATE_RESERVED);
963 	/*
964 	 * We already hold zpci_list_lock thanks to kref_put_lock().
965 	 * This makes sure no new reference can be taken from the list.
966 	 */
967 	list_del(&zdev->entry);
968 	spin_unlock(&zpci_list_lock);
969 
970 	if (zdev->has_hp_slot)
971 		zpci_exit_slot(zdev);
972 
973 	if (zdev->has_resources)
974 		zpci_cleanup_bus_resources(zdev);
975 
976 	zpci_bus_device_unregister(zdev);
977 	zpci_destroy_iommu(zdev);
978 	zpci_dbg(3, "rem fid:%x\n", zdev->fid);
979 	kfree_rcu(zdev, rcu);
980 }
981 
982 int zpci_report_error(struct pci_dev *pdev,
983 		      struct zpci_report_error_header *report)
984 {
985 	struct zpci_dev *zdev = to_zpci(pdev);
986 
987 	return sclp_pci_report(report, zdev->fh, zdev->fid);
988 }
989 EXPORT_SYMBOL(zpci_report_error);
990 
991 /**
992  * zpci_clear_error_state() - Clears the zPCI error state of the device
993  * @zdev: The zdev for which the zPCI error state should be reset
994  *
995  * Clear the zPCI error state of the device. If clearing the zPCI error state
996  * fails the device is left in the error state. In this case it may make sense
997  * to call zpci_io_perm_failure() on the associated pdev if it exists.
998  *
999  * Returns: 0 on success, -EIO otherwise
1000  */
1001 int zpci_clear_error_state(struct zpci_dev *zdev)
1002 {
1003 	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_ERROR);
1004 	struct zpci_fib fib = {0};
1005 	u8 status;
1006 	int cc;
1007 
1008 	cc = zpci_mod_fc(req, &fib, &status);
1009 	if (cc) {
1010 		zpci_dbg(3, "ces fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status);
1011 		return -EIO;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 /**
1018  * zpci_reset_load_store_blocked() - Re-enables L/S from error state
1019  * @zdev: The zdev for which to unblock load/store access
1020  *
1021  * Re-enables load/store access for a PCI function in the error state while
1022  * keeping DMA blocked. In this state drivers can poke MMIO space to determine
1023  * if error recovery is possible while catching any rogue DMA access from the
1024  * device.
1025  *
1026  * Returns: 0 on success, -EIO otherwise
1027  */
1028 int zpci_reset_load_store_blocked(struct zpci_dev *zdev)
1029 {
1030 	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_BLOCK);
1031 	struct zpci_fib fib = {0};
1032 	u8 status;
1033 	int cc;
1034 
1035 	cc = zpci_mod_fc(req, &fib, &status);
1036 	if (cc) {
1037 		zpci_dbg(3, "rls fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status);
1038 		return -EIO;
1039 	}
1040 
1041 	return 0;
1042 }
1043 
1044 static int zpci_mem_init(void)
1045 {
1046 	BUILD_BUG_ON(!is_power_of_2(__alignof__(struct zpci_fmb)) ||
1047 		     __alignof__(struct zpci_fmb) < sizeof(struct zpci_fmb));
1048 
1049 	zdev_fmb_cache = kmem_cache_create("PCI_FMB_cache", sizeof(struct zpci_fmb),
1050 					   __alignof__(struct zpci_fmb), 0, NULL);
1051 	if (!zdev_fmb_cache)
1052 		goto error_fmb;
1053 
1054 	zpci_iomap_start = kcalloc(ZPCI_IOMAP_ENTRIES,
1055 				   sizeof(*zpci_iomap_start), GFP_KERNEL);
1056 	if (!zpci_iomap_start)
1057 		goto error_iomap;
1058 
1059 	zpci_iomap_bitmap = kcalloc(BITS_TO_LONGS(ZPCI_IOMAP_ENTRIES),
1060 				    sizeof(*zpci_iomap_bitmap), GFP_KERNEL);
1061 	if (!zpci_iomap_bitmap)
1062 		goto error_iomap_bitmap;
1063 
1064 	if (static_branch_likely(&have_mio))
1065 		clp_setup_writeback_mio();
1066 
1067 	return 0;
1068 error_iomap_bitmap:
1069 	kfree(zpci_iomap_start);
1070 error_iomap:
1071 	kmem_cache_destroy(zdev_fmb_cache);
1072 error_fmb:
1073 	return -ENOMEM;
1074 }
1075 
1076 static void zpci_mem_exit(void)
1077 {
1078 	kfree(zpci_iomap_bitmap);
1079 	kfree(zpci_iomap_start);
1080 	kmem_cache_destroy(zdev_fmb_cache);
1081 }
1082 
1083 static unsigned int s390_pci_probe __initdata = 1;
1084 unsigned int s390_pci_force_floating __initdata;
1085 static unsigned int s390_pci_initialized;
1086 
1087 char * __init pcibios_setup(char *str)
1088 {
1089 	if (!strcmp(str, "off")) {
1090 		s390_pci_probe = 0;
1091 		return NULL;
1092 	}
1093 	if (!strcmp(str, "nomio")) {
1094 		clear_machine_feature(MFEATURE_PCI_MIO);
1095 		return NULL;
1096 	}
1097 	if (!strcmp(str, "force_floating")) {
1098 		s390_pci_force_floating = 1;
1099 		return NULL;
1100 	}
1101 	if (!strcmp(str, "norid")) {
1102 		s390_pci_no_rid = 1;
1103 		return NULL;
1104 	}
1105 	return str;
1106 }
1107 
1108 bool zpci_is_enabled(void)
1109 {
1110 	return s390_pci_initialized;
1111 }
1112 
1113 static int zpci_cmp_rid(void *priv, const struct list_head *a,
1114 			const struct list_head *b)
1115 {
1116 	struct zpci_dev *za = container_of(a, struct zpci_dev, entry);
1117 	struct zpci_dev *zb = container_of(b, struct zpci_dev, entry);
1118 
1119 	/*
1120 	 * PCI functions without RID available maintain original order
1121 	 * between themselves but sort before those with RID.
1122 	 */
1123 	if (za->rid == zb->rid)
1124 		return za->rid_available > zb->rid_available;
1125 	/*
1126 	 * PCI functions with RID sort by RID ascending.
1127 	 */
1128 	return za->rid > zb->rid;
1129 }
1130 
1131 static void zpci_add_devices(struct list_head *scan_list)
1132 {
1133 	struct zpci_dev *zdev, *tmp;
1134 
1135 	list_sort(NULL, scan_list, &zpci_cmp_rid);
1136 	list_for_each_entry_safe(zdev, tmp, scan_list, entry) {
1137 		list_del_init(&zdev->entry);
1138 		if (zpci_add_device(zdev))
1139 			kfree(zdev);
1140 	}
1141 }
1142 
1143 int zpci_scan_devices(void)
1144 {
1145 	LIST_HEAD(scan_list);
1146 	int rc;
1147 
1148 	rc = clp_scan_pci_devices(&scan_list);
1149 	if (rc)
1150 		return rc;
1151 
1152 	zpci_add_devices(&scan_list);
1153 	zpci_bus_scan_busses();
1154 	return 0;
1155 }
1156 
1157 static int __init pci_base_init(void)
1158 {
1159 	int rc;
1160 
1161 	if (!s390_pci_probe)
1162 		return 0;
1163 
1164 	if (!test_facility(69) || !test_facility(71)) {
1165 		pr_info("PCI is not supported because CPU facilities 69 or 71 are not available\n");
1166 		return 0;
1167 	}
1168 
1169 	if (test_machine_feature(MFEATURE_PCI_MIO)) {
1170 		static_branch_enable(&have_mio);
1171 		system_ctl_set_bit(2, CR2_MIO_ADDRESSING_BIT);
1172 	}
1173 
1174 	rc = zpci_debug_init();
1175 	if (rc)
1176 		goto out;
1177 
1178 	rc = zpci_mem_init();
1179 	if (rc)
1180 		goto out_mem;
1181 
1182 	rc = zpci_irq_init();
1183 	if (rc)
1184 		goto out_irq;
1185 
1186 	rc = zpci_scan_devices();
1187 	if (rc)
1188 		goto out_find;
1189 
1190 	rc = zpci_fw_sysfs_init();
1191 	if (rc)
1192 		goto out_find;
1193 
1194 	s390_pci_initialized = 1;
1195 	return 0;
1196 
1197 out_find:
1198 	zpci_irq_exit();
1199 out_irq:
1200 	zpci_mem_exit();
1201 out_mem:
1202 	zpci_debug_exit();
1203 out:
1204 	return rc;
1205 }
1206 subsys_initcall_sync(pci_base_init);
1207