xref: /linux/arch/powerpc/platforms/pseries/msi.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  * Copyright 2006 Jake Moilanen <moilanen@austin.ibm.com>, IBM Corp.
3  * Copyright 2006-2007 Michael Ellerman, IBM Corp.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License
7  * as published by the Free Software Foundation; version 2 of the
8  * License.
9  *
10  */
11 
12 #include <linux/device.h>
13 #include <linux/irq.h>
14 #include <linux/msi.h>
15 
16 #include <asm/rtas.h>
17 #include <asm/hw_irq.h>
18 #include <asm/ppc-pci.h>
19 
20 static int query_token, change_token;
21 
22 #define RTAS_QUERY_FN		0
23 #define RTAS_CHANGE_FN		1
24 #define RTAS_RESET_FN		2
25 #define RTAS_CHANGE_MSI_FN	3
26 #define RTAS_CHANGE_MSIX_FN	4
27 
28 static struct pci_dn *get_pdn(struct pci_dev *pdev)
29 {
30 	struct device_node *dn;
31 	struct pci_dn *pdn;
32 
33 	dn = pci_device_to_OF_node(pdev);
34 	if (!dn) {
35 		dev_dbg(&pdev->dev, "rtas_msi: No OF device node\n");
36 		return NULL;
37 	}
38 
39 	pdn = PCI_DN(dn);
40 	if (!pdn) {
41 		dev_dbg(&pdev->dev, "rtas_msi: No PCI DN\n");
42 		return NULL;
43 	}
44 
45 	return pdn;
46 }
47 
48 /* RTAS Helpers */
49 
50 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
51 {
52 	u32 addr, seq_num, rtas_ret[3];
53 	unsigned long buid;
54 	int rc;
55 
56 	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
57 	buid = pdn->phb->buid;
58 
59 	seq_num = 1;
60 	do {
61 		if (func == RTAS_CHANGE_MSI_FN || func == RTAS_CHANGE_MSIX_FN)
62 			rc = rtas_call(change_token, 6, 4, rtas_ret, addr,
63 					BUID_HI(buid), BUID_LO(buid),
64 					func, num_irqs, seq_num);
65 		else
66 			rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
67 					BUID_HI(buid), BUID_LO(buid),
68 					func, num_irqs, seq_num);
69 
70 		seq_num = rtas_ret[1];
71 	} while (rtas_busy_delay(rc));
72 
73 	/*
74 	 * If the RTAS call succeeded, return the number of irqs allocated.
75 	 * If not, make sure we return a negative error code.
76 	 */
77 	if (rc == 0)
78 		rc = rtas_ret[0];
79 	else if (rc > 0)
80 		rc = -rc;
81 
82 	pr_debug("rtas_msi: ibm,change_msi(func=%d,num=%d), got %d rc = %d\n",
83 		 func, num_irqs, rtas_ret[0], rc);
84 
85 	return rc;
86 }
87 
88 static void rtas_disable_msi(struct pci_dev *pdev)
89 {
90 	struct pci_dn *pdn;
91 
92 	pdn = get_pdn(pdev);
93 	if (!pdn)
94 		return;
95 
96 	/*
97 	 * disabling MSI with the explicit interface also disables MSI-X
98 	 */
99 	if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
100 		/*
101 		 * may have failed because explicit interface is not
102 		 * present
103 		 */
104 		if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
105 			pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
106 		}
107 	}
108 }
109 
110 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
111 {
112 	u32 addr, rtas_ret[2];
113 	unsigned long buid;
114 	int rc;
115 
116 	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
117 	buid = pdn->phb->buid;
118 
119 	do {
120 		rc = rtas_call(query_token, 4, 3, rtas_ret, addr,
121 			       BUID_HI(buid), BUID_LO(buid), offset);
122 	} while (rtas_busy_delay(rc));
123 
124 	if (rc) {
125 		pr_debug("rtas_msi: error (%d) querying source number\n", rc);
126 		return rc;
127 	}
128 
129 	return rtas_ret[0];
130 }
131 
132 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
133 {
134 	struct msi_desc *entry;
135 
136 	list_for_each_entry(entry, &pdev->msi_list, list) {
137 		if (entry->irq == NO_IRQ)
138 			continue;
139 
140 		irq_set_msi_desc(entry->irq, NULL);
141 		irq_dispose_mapping(entry->irq);
142 	}
143 
144 	rtas_disable_msi(pdev);
145 }
146 
147 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
148 {
149 	struct device_node *dn;
150 	struct pci_dn *pdn;
151 	const u32 *req_msi;
152 
153 	pdn = get_pdn(pdev);
154 	if (!pdn)
155 		return -ENODEV;
156 
157 	dn = pdn->node;
158 
159 	req_msi = of_get_property(dn, prop_name, NULL);
160 	if (!req_msi) {
161 		pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name);
162 		return -ENOENT;
163 	}
164 
165 	if (*req_msi < nvec) {
166 		pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
167 
168 		if (*req_msi == 0) /* Be paranoid */
169 			return -ENOSPC;
170 
171 		return *req_msi;
172 	}
173 
174 	return 0;
175 }
176 
177 static int check_req_msi(struct pci_dev *pdev, int nvec)
178 {
179 	return check_req(pdev, nvec, "ibm,req#msi");
180 }
181 
182 static int check_req_msix(struct pci_dev *pdev, int nvec)
183 {
184 	return check_req(pdev, nvec, "ibm,req#msi-x");
185 }
186 
187 /* Quota calculation */
188 
189 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
190 {
191 	struct device_node *dn;
192 	const u32 *p;
193 
194 	dn = of_node_get(pci_device_to_OF_node(dev));
195 	while (dn) {
196 		p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
197 		if (p) {
198 			pr_debug("rtas_msi: found prop on dn %s\n",
199 				dn->full_name);
200 			*total = *p;
201 			return dn;
202 		}
203 
204 		dn = of_get_next_parent(dn);
205 	}
206 
207 	return NULL;
208 }
209 
210 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
211 {
212 	struct device_node *dn;
213 	struct eeh_dev *edev;
214 
215 	/* Found our PE and assume 8 at that point. */
216 
217 	dn = pci_device_to_OF_node(dev);
218 	if (!dn)
219 		return NULL;
220 
221 	/* Get the top level device in the PE */
222 	edev = of_node_to_eeh_dev(dn);
223 	if (edev->pe)
224 		edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
225 	dn = eeh_dev_to_of_node(edev);
226 	if (!dn)
227 		return NULL;
228 
229 	/* We actually want the parent */
230 	dn = of_get_parent(dn);
231 	if (!dn)
232 		return NULL;
233 
234 	/* Hardcode of 8 for old firmwares */
235 	*total = 8;
236 	pr_debug("rtas_msi: using PE dn %s\n", dn->full_name);
237 
238 	return dn;
239 }
240 
241 struct msi_counts {
242 	struct device_node *requestor;
243 	int num_devices;
244 	int request;
245 	int quota;
246 	int spare;
247 	int over_quota;
248 };
249 
250 static void *count_non_bridge_devices(struct device_node *dn, void *data)
251 {
252 	struct msi_counts *counts = data;
253 	const u32 *p;
254 	u32 class;
255 
256 	pr_debug("rtas_msi: counting %s\n", dn->full_name);
257 
258 	p = of_get_property(dn, "class-code", NULL);
259 	class = p ? *p : 0;
260 
261 	if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
262 		counts->num_devices++;
263 
264 	return NULL;
265 }
266 
267 static void *count_spare_msis(struct device_node *dn, void *data)
268 {
269 	struct msi_counts *counts = data;
270 	const u32 *p;
271 	int req;
272 
273 	if (dn == counts->requestor)
274 		req = counts->request;
275 	else {
276 		/* We don't know if a driver will try to use MSI or MSI-X,
277 		 * so we just have to punt and use the larger of the two. */
278 		req = 0;
279 		p = of_get_property(dn, "ibm,req#msi", NULL);
280 		if (p)
281 			req = *p;
282 
283 		p = of_get_property(dn, "ibm,req#msi-x", NULL);
284 		if (p)
285 			req = max(req, (int)*p);
286 	}
287 
288 	if (req < counts->quota)
289 		counts->spare += counts->quota - req;
290 	else if (req > counts->quota)
291 		counts->over_quota++;
292 
293 	return NULL;
294 }
295 
296 static int msi_quota_for_device(struct pci_dev *dev, int request)
297 {
298 	struct device_node *pe_dn;
299 	struct msi_counts counts;
300 	int total;
301 
302 	pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
303 		  request);
304 
305 	pe_dn = find_pe_total_msi(dev, &total);
306 	if (!pe_dn)
307 		pe_dn = find_pe_dn(dev, &total);
308 
309 	if (!pe_dn) {
310 		pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
311 		goto out;
312 	}
313 
314 	pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name);
315 
316 	memset(&counts, 0, sizeof(struct msi_counts));
317 
318 	/* Work out how many devices we have below this PE */
319 	traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts);
320 
321 	if (counts.num_devices == 0) {
322 		pr_err("rtas_msi: found 0 devices under PE for %s\n",
323 			pci_name(dev));
324 		goto out;
325 	}
326 
327 	counts.quota = total / counts.num_devices;
328 	if (request <= counts.quota)
329 		goto out;
330 
331 	/* else, we have some more calculating to do */
332 	counts.requestor = pci_device_to_OF_node(dev);
333 	counts.request = request;
334 	traverse_pci_devices(pe_dn, count_spare_msis, &counts);
335 
336 	/* If the quota isn't an integer multiple of the total, we can
337 	 * use the remainder as spare MSIs for anyone that wants them. */
338 	counts.spare += total % counts.num_devices;
339 
340 	/* Divide any spare by the number of over-quota requestors */
341 	if (counts.over_quota)
342 		counts.quota += counts.spare / counts.over_quota;
343 
344 	/* And finally clamp the request to the possibly adjusted quota */
345 	request = min(counts.quota, request);
346 
347 	pr_debug("rtas_msi: request clamped to quota %d\n", request);
348 out:
349 	of_node_put(pe_dn);
350 
351 	return request;
352 }
353 
354 static int rtas_msi_check_device(struct pci_dev *pdev, int nvec, int type)
355 {
356 	int quota, rc;
357 
358 	if (type == PCI_CAP_ID_MSIX)
359 		rc = check_req_msix(pdev, nvec);
360 	else
361 		rc = check_req_msi(pdev, nvec);
362 
363 	if (rc)
364 		return rc;
365 
366 	quota = msi_quota_for_device(pdev, nvec);
367 
368 	if (quota && quota < nvec)
369 		return quota;
370 
371 	return 0;
372 }
373 
374 static int check_msix_entries(struct pci_dev *pdev)
375 {
376 	struct msi_desc *entry;
377 	int expected;
378 
379 	/* There's no way for us to express to firmware that we want
380 	 * a discontiguous, or non-zero based, range of MSI-X entries.
381 	 * So we must reject such requests. */
382 
383 	expected = 0;
384 	list_for_each_entry(entry, &pdev->msi_list, list) {
385 		if (entry->msi_attrib.entry_nr != expected) {
386 			pr_debug("rtas_msi: bad MSI-X entries.\n");
387 			return -EINVAL;
388 		}
389 		expected++;
390 	}
391 
392 	return 0;
393 }
394 
395 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
396 {
397 	struct pci_dn *pdn;
398 	int hwirq, virq, i, rc;
399 	struct msi_desc *entry;
400 	struct msi_msg msg;
401 	int nvec = nvec_in;
402 
403 	pdn = get_pdn(pdev);
404 	if (!pdn)
405 		return -ENODEV;
406 
407 	if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
408 		return -EINVAL;
409 
410 	/*
411 	 * Firmware currently refuse any non power of two allocation
412 	 * so we round up if the quota will allow it.
413 	 */
414 	if (type == PCI_CAP_ID_MSIX) {
415 		int m = roundup_pow_of_two(nvec);
416 		int quota = msi_quota_for_device(pdev, m);
417 
418 		if (quota >= m)
419 			nvec = m;
420 	}
421 
422 	/*
423 	 * Try the new more explicit firmware interface, if that fails fall
424 	 * back to the old interface. The old interface is known to never
425 	 * return MSI-Xs.
426 	 */
427 again:
428 	if (type == PCI_CAP_ID_MSI) {
429 		rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
430 
431 		if (rc < 0) {
432 			pr_debug("rtas_msi: trying the old firmware call.\n");
433 			rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
434 		}
435 	} else
436 		rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
437 
438 	if (rc != nvec) {
439 		if (nvec != nvec_in) {
440 			nvec = nvec_in;
441 			goto again;
442 		}
443 		pr_debug("rtas_msi: rtas_change_msi() failed\n");
444 		return rc;
445 	}
446 
447 	i = 0;
448 	list_for_each_entry(entry, &pdev->msi_list, list) {
449 		hwirq = rtas_query_irq_number(pdn, i++);
450 		if (hwirq < 0) {
451 			pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
452 			return hwirq;
453 		}
454 
455 		virq = irq_create_mapping(NULL, hwirq);
456 
457 		if (virq == NO_IRQ) {
458 			pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
459 			return -ENOSPC;
460 		}
461 
462 		dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
463 		irq_set_msi_desc(virq, entry);
464 
465 		/* Read config space back so we can restore after reset */
466 		read_msi_msg(virq, &msg);
467 		entry->msg = msg;
468 	}
469 
470 	return 0;
471 }
472 
473 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
474 {
475 	/* No LSI -> leave MSIs (if any) configured */
476 	if (pdev->irq == NO_IRQ) {
477 		dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
478 		return;
479 	}
480 
481 	/* No MSI -> MSIs can't have been assigned by fw, leave LSI */
482 	if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
483 		dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
484 		return;
485 	}
486 
487 	dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
488 	rtas_disable_msi(pdev);
489 }
490 
491 static int rtas_msi_init(void)
492 {
493 	query_token  = rtas_token("ibm,query-interrupt-source-number");
494 	change_token = rtas_token("ibm,change-msi");
495 
496 	if ((query_token == RTAS_UNKNOWN_SERVICE) ||
497 			(change_token == RTAS_UNKNOWN_SERVICE)) {
498 		pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
499 		return -1;
500 	}
501 
502 	pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
503 
504 	WARN_ON(ppc_md.setup_msi_irqs);
505 	ppc_md.setup_msi_irqs = rtas_setup_msi_irqs;
506 	ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs;
507 	ppc_md.msi_check_device = rtas_msi_check_device;
508 
509 	WARN_ON(ppc_md.pci_irq_fixup);
510 	ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
511 
512 	return 0;
513 }
514 arch_initcall(rtas_msi_init);
515