1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright 2006 Jake Moilanen <moilanen@austin.ibm.com>, IBM Corp.
4  * Copyright 2006-2007 Michael Ellerman, IBM Corp.
5  */
6 
7 #include <linux/device.h>
8 #include <linux/irq.h>
9 #include <linux/msi.h>
10 
11 #include <asm/rtas.h>
12 #include <asm/hw_irq.h>
13 #include <asm/ppc-pci.h>
14 #include <asm/machdep.h>
15 
16 #include "pseries.h"
17 
18 static int query_token, change_token;
19 
20 #define RTAS_QUERY_FN		0
21 #define RTAS_CHANGE_FN		1
22 #define RTAS_RESET_FN		2
23 #define RTAS_CHANGE_MSI_FN	3
24 #define RTAS_CHANGE_MSIX_FN	4
25 #define RTAS_CHANGE_32MSI_FN	5
26 
27 /* RTAS Helpers */
28 
29 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
30 {
31 	u32 addr, seq_num, rtas_ret[3];
32 	unsigned long buid;
33 	int rc;
34 
35 	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
36 	buid = pdn->phb->buid;
37 
38 	seq_num = 1;
39 	do {
40 		if (func == RTAS_CHANGE_MSI_FN || func == RTAS_CHANGE_MSIX_FN ||
41 		    func == RTAS_CHANGE_32MSI_FN)
42 			rc = rtas_call(change_token, 6, 4, rtas_ret, addr,
43 					BUID_HI(buid), BUID_LO(buid),
44 					func, num_irqs, seq_num);
45 		else
46 			rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
47 					BUID_HI(buid), BUID_LO(buid),
48 					func, num_irqs, seq_num);
49 
50 		seq_num = rtas_ret[1];
51 	} while (rtas_busy_delay(rc));
52 
53 	/*
54 	 * If the RTAS call succeeded, return the number of irqs allocated.
55 	 * If not, make sure we return a negative error code.
56 	 */
57 	if (rc == 0)
58 		rc = rtas_ret[0];
59 	else if (rc > 0)
60 		rc = -rc;
61 
62 	pr_debug("rtas_msi: ibm,change_msi(func=%d,num=%d), got %d rc = %d\n",
63 		 func, num_irqs, rtas_ret[0], rc);
64 
65 	return rc;
66 }
67 
68 static void rtas_disable_msi(struct pci_dev *pdev)
69 {
70 	struct pci_dn *pdn;
71 
72 	pdn = pci_get_pdn(pdev);
73 	if (!pdn)
74 		return;
75 
76 	/*
77 	 * disabling MSI with the explicit interface also disables MSI-X
78 	 */
79 	if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
80 		/*
81 		 * may have failed because explicit interface is not
82 		 * present
83 		 */
84 		if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
85 			pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
86 		}
87 	}
88 }
89 
90 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
91 {
92 	u32 addr, rtas_ret[2];
93 	unsigned long buid;
94 	int rc;
95 
96 	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
97 	buid = pdn->phb->buid;
98 
99 	do {
100 		rc = rtas_call(query_token, 4, 3, rtas_ret, addr,
101 			       BUID_HI(buid), BUID_LO(buid), offset);
102 	} while (rtas_busy_delay(rc));
103 
104 	if (rc) {
105 		pr_debug("rtas_msi: error (%d) querying source number\n", rc);
106 		return rc;
107 	}
108 
109 	return rtas_ret[0];
110 }
111 
112 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
113 {
114 	struct msi_desc *entry;
115 
116 	for_each_pci_msi_entry(entry, pdev) {
117 		if (!entry->irq)
118 			continue;
119 
120 		irq_set_msi_desc(entry->irq, NULL);
121 		irq_dispose_mapping(entry->irq);
122 	}
123 
124 	rtas_disable_msi(pdev);
125 }
126 
127 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
128 {
129 	struct device_node *dn;
130 	const __be32 *p;
131 	u32 req_msi;
132 
133 	dn = pci_device_to_OF_node(pdev);
134 
135 	p = of_get_property(dn, prop_name, NULL);
136 	if (!p) {
137 		pr_debug("rtas_msi: No %s on %pOF\n", prop_name, dn);
138 		return -ENOENT;
139 	}
140 
141 	req_msi = be32_to_cpup(p);
142 	if (req_msi < nvec) {
143 		pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
144 
145 		if (req_msi == 0) /* Be paranoid */
146 			return -ENOSPC;
147 
148 		return req_msi;
149 	}
150 
151 	return 0;
152 }
153 
154 static int check_req_msi(struct pci_dev *pdev, int nvec)
155 {
156 	return check_req(pdev, nvec, "ibm,req#msi");
157 }
158 
159 static int check_req_msix(struct pci_dev *pdev, int nvec)
160 {
161 	return check_req(pdev, nvec, "ibm,req#msi-x");
162 }
163 
164 /* Quota calculation */
165 
166 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
167 {
168 	struct device_node *dn;
169 	const __be32 *p;
170 
171 	dn = of_node_get(pci_device_to_OF_node(dev));
172 	while (dn) {
173 		p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
174 		if (p) {
175 			pr_debug("rtas_msi: found prop on dn %pOF\n",
176 				dn);
177 			*total = be32_to_cpup(p);
178 			return dn;
179 		}
180 
181 		dn = of_get_next_parent(dn);
182 	}
183 
184 	return NULL;
185 }
186 
187 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
188 {
189 	struct device_node *dn;
190 	struct eeh_dev *edev;
191 
192 	/* Found our PE and assume 8 at that point. */
193 
194 	dn = pci_device_to_OF_node(dev);
195 	if (!dn)
196 		return NULL;
197 
198 	/* Get the top level device in the PE */
199 	edev = pdn_to_eeh_dev(PCI_DN(dn));
200 	if (edev->pe)
201 		edev = list_first_entry(&edev->pe->edevs, struct eeh_dev,
202 					entry);
203 	dn = pci_device_to_OF_node(edev->pdev);
204 	if (!dn)
205 		return NULL;
206 
207 	/* We actually want the parent */
208 	dn = of_get_parent(dn);
209 	if (!dn)
210 		return NULL;
211 
212 	/* Hardcode of 8 for old firmwares */
213 	*total = 8;
214 	pr_debug("rtas_msi: using PE dn %pOF\n", dn);
215 
216 	return dn;
217 }
218 
219 struct msi_counts {
220 	struct device_node *requestor;
221 	int num_devices;
222 	int request;
223 	int quota;
224 	int spare;
225 	int over_quota;
226 };
227 
228 static void *count_non_bridge_devices(struct device_node *dn, void *data)
229 {
230 	struct msi_counts *counts = data;
231 	const __be32 *p;
232 	u32 class;
233 
234 	pr_debug("rtas_msi: counting %pOF\n", dn);
235 
236 	p = of_get_property(dn, "class-code", NULL);
237 	class = p ? be32_to_cpup(p) : 0;
238 
239 	if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
240 		counts->num_devices++;
241 
242 	return NULL;
243 }
244 
245 static void *count_spare_msis(struct device_node *dn, void *data)
246 {
247 	struct msi_counts *counts = data;
248 	const __be32 *p;
249 	int req;
250 
251 	if (dn == counts->requestor)
252 		req = counts->request;
253 	else {
254 		/* We don't know if a driver will try to use MSI or MSI-X,
255 		 * so we just have to punt and use the larger of the two. */
256 		req = 0;
257 		p = of_get_property(dn, "ibm,req#msi", NULL);
258 		if (p)
259 			req = be32_to_cpup(p);
260 
261 		p = of_get_property(dn, "ibm,req#msi-x", NULL);
262 		if (p)
263 			req = max(req, (int)be32_to_cpup(p));
264 	}
265 
266 	if (req < counts->quota)
267 		counts->spare += counts->quota - req;
268 	else if (req > counts->quota)
269 		counts->over_quota++;
270 
271 	return NULL;
272 }
273 
274 static int msi_quota_for_device(struct pci_dev *dev, int request)
275 {
276 	struct device_node *pe_dn;
277 	struct msi_counts counts;
278 	int total;
279 
280 	pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
281 		  request);
282 
283 	pe_dn = find_pe_total_msi(dev, &total);
284 	if (!pe_dn)
285 		pe_dn = find_pe_dn(dev, &total);
286 
287 	if (!pe_dn) {
288 		pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
289 		goto out;
290 	}
291 
292 	pr_debug("rtas_msi: found PE %pOF\n", pe_dn);
293 
294 	memset(&counts, 0, sizeof(struct msi_counts));
295 
296 	/* Work out how many devices we have below this PE */
297 	pci_traverse_device_nodes(pe_dn, count_non_bridge_devices, &counts);
298 
299 	if (counts.num_devices == 0) {
300 		pr_err("rtas_msi: found 0 devices under PE for %s\n",
301 			pci_name(dev));
302 		goto out;
303 	}
304 
305 	counts.quota = total / counts.num_devices;
306 	if (request <= counts.quota)
307 		goto out;
308 
309 	/* else, we have some more calculating to do */
310 	counts.requestor = pci_device_to_OF_node(dev);
311 	counts.request = request;
312 	pci_traverse_device_nodes(pe_dn, count_spare_msis, &counts);
313 
314 	/* If the quota isn't an integer multiple of the total, we can
315 	 * use the remainder as spare MSIs for anyone that wants them. */
316 	counts.spare += total % counts.num_devices;
317 
318 	/* Divide any spare by the number of over-quota requestors */
319 	if (counts.over_quota)
320 		counts.quota += counts.spare / counts.over_quota;
321 
322 	/* And finally clamp the request to the possibly adjusted quota */
323 	request = min(counts.quota, request);
324 
325 	pr_debug("rtas_msi: request clamped to quota %d\n", request);
326 out:
327 	of_node_put(pe_dn);
328 
329 	return request;
330 }
331 
332 static int check_msix_entries(struct pci_dev *pdev)
333 {
334 	struct msi_desc *entry;
335 	int expected;
336 
337 	/* There's no way for us to express to firmware that we want
338 	 * a discontiguous, or non-zero based, range of MSI-X entries.
339 	 * So we must reject such requests. */
340 
341 	expected = 0;
342 	for_each_pci_msi_entry(entry, pdev) {
343 		if (entry->msi_attrib.entry_nr != expected) {
344 			pr_debug("rtas_msi: bad MSI-X entries.\n");
345 			return -EINVAL;
346 		}
347 		expected++;
348 	}
349 
350 	return 0;
351 }
352 
353 static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
354 {
355 	u32 addr_hi, addr_lo;
356 
357 	/*
358 	 * We should only get in here for IODA1 configs. This is based on the
359 	 * fact that we using RTAS for MSIs, we don't have the 32 bit MSI RTAS
360 	 * support, and we are in a PCIe Gen2 slot.
361 	 */
362 	dev_info(&pdev->dev,
363 		 "rtas_msi: No 32 bit MSI firmware support, forcing 32 bit MSI\n");
364 	pci_read_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, &addr_hi);
365 	addr_lo = 0xffff0000 | ((addr_hi >> (48 - 32)) << 4);
366 	pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_LO, addr_lo);
367 	pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, 0);
368 }
369 
370 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
371 {
372 	struct pci_dn *pdn;
373 	int hwirq, virq, i, quota, rc;
374 	struct msi_desc *entry;
375 	struct msi_msg msg;
376 	int nvec = nvec_in;
377 	int use_32bit_msi_hack = 0;
378 
379 	if (type == PCI_CAP_ID_MSIX)
380 		rc = check_req_msix(pdev, nvec);
381 	else
382 		rc = check_req_msi(pdev, nvec);
383 
384 	if (rc)
385 		return rc;
386 
387 	quota = msi_quota_for_device(pdev, nvec);
388 
389 	if (quota && quota < nvec)
390 		return quota;
391 
392 	if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
393 		return -EINVAL;
394 
395 	/*
396 	 * Firmware currently refuse any non power of two allocation
397 	 * so we round up if the quota will allow it.
398 	 */
399 	if (type == PCI_CAP_ID_MSIX) {
400 		int m = roundup_pow_of_two(nvec);
401 		quota = msi_quota_for_device(pdev, m);
402 
403 		if (quota >= m)
404 			nvec = m;
405 	}
406 
407 	pdn = pci_get_pdn(pdev);
408 
409 	/*
410 	 * Try the new more explicit firmware interface, if that fails fall
411 	 * back to the old interface. The old interface is known to never
412 	 * return MSI-Xs.
413 	 */
414 again:
415 	if (type == PCI_CAP_ID_MSI) {
416 		if (pdev->no_64bit_msi) {
417 			rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
418 			if (rc < 0) {
419 				/*
420 				 * We only want to run the 32 bit MSI hack below if
421 				 * the max bus speed is Gen2 speed
422 				 */
423 				if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
424 					return rc;
425 
426 				use_32bit_msi_hack = 1;
427 			}
428 		} else
429 			rc = -1;
430 
431 		if (rc < 0)
432 			rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
433 
434 		if (rc < 0) {
435 			pr_debug("rtas_msi: trying the old firmware call.\n");
436 			rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
437 		}
438 
439 		if (use_32bit_msi_hack && rc > 0)
440 			rtas_hack_32bit_msi_gen2(pdev);
441 	} else
442 		rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
443 
444 	if (rc != nvec) {
445 		if (nvec != nvec_in) {
446 			nvec = nvec_in;
447 			goto again;
448 		}
449 		pr_debug("rtas_msi: rtas_change_msi() failed\n");
450 		return rc;
451 	}
452 
453 	i = 0;
454 	for_each_pci_msi_entry(entry, pdev) {
455 		hwirq = rtas_query_irq_number(pdn, i++);
456 		if (hwirq < 0) {
457 			pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
458 			return hwirq;
459 		}
460 
461 		virq = irq_create_mapping_affinity(NULL, hwirq,
462 						   entry->affinity);
463 
464 		if (!virq) {
465 			pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
466 			return -ENOSPC;
467 		}
468 
469 		dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
470 		irq_set_msi_desc(virq, entry);
471 
472 		/* Read config space back so we can restore after reset */
473 		__pci_read_msi_msg(entry, &msg);
474 		entry->msg = msg;
475 	}
476 
477 	return 0;
478 }
479 
480 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
481 {
482 	/* No LSI -> leave MSIs (if any) configured */
483 	if (!pdev->irq) {
484 		dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
485 		return;
486 	}
487 
488 	/* No MSI -> MSIs can't have been assigned by fw, leave LSI */
489 	if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
490 		dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
491 		return;
492 	}
493 
494 	dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
495 	rtas_disable_msi(pdev);
496 }
497 
498 static int rtas_msi_init(void)
499 {
500 	struct pci_controller *phb;
501 
502 	query_token  = rtas_token("ibm,query-interrupt-source-number");
503 	change_token = rtas_token("ibm,change-msi");
504 
505 	if ((query_token == RTAS_UNKNOWN_SERVICE) ||
506 			(change_token == RTAS_UNKNOWN_SERVICE)) {
507 		pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
508 		return -1;
509 	}
510 
511 	pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
512 
513 	WARN_ON(pseries_pci_controller_ops.setup_msi_irqs);
514 	pseries_pci_controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
515 	pseries_pci_controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
516 
517 	list_for_each_entry(phb, &hose_list, list_node) {
518 		WARN_ON(phb->controller_ops.setup_msi_irqs);
519 		phb->controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
520 		phb->controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
521 	}
522 
523 	WARN_ON(ppc_md.pci_irq_fixup);
524 	ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
525 
526 	return 0;
527 }
528 machine_arch_initcall(pseries, rtas_msi_init);
529