Commit | Line | Data |
---|---|---|
85f2bf9f ME |
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 | ||
d385366a | 73 | /* |
6071ed04 ME |
74 | * If the RTAS call succeeded, return the number of irqs allocated. |
75 | * If not, make sure we return a negative error code. | |
d385366a | 76 | */ |
6071ed04 ME |
77 | if (rc == 0) |
78 | rc = rtas_ret[0]; | |
79 | else if (rc > 0) | |
80 | rc = -rc; | |
85f2bf9f | 81 | |
d385366a ME |
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); | |
85f2bf9f ME |
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 | ||
964a2996 NA |
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 | } | |
85f2bf9f ME |
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 | ||
ec775d0e | 140 | irq_set_msi_desc(entry->irq, NULL); |
85f2bf9f ME |
141 | irq_dispose_mapping(entry->irq); |
142 | } | |
143 | ||
144 | rtas_disable_msi(pdev); | |
145 | } | |
146 | ||
3a51c0cb | 147 | static int check_req(struct pci_dev *pdev, int nvec, char *prop_name) |
85f2bf9f ME |
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 | ||
3a51c0cb | 159 | req_msi = of_get_property(dn, prop_name, NULL); |
85f2bf9f | 160 | if (!req_msi) { |
3a51c0cb | 161 | pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name); |
85f2bf9f ME |
162 | return -ENOENT; |
163 | } | |
164 | ||
165 | if (*req_msi < nvec) { | |
3a51c0cb | 166 | pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec); |
d523cc37 ME |
167 | |
168 | if (*req_msi == 0) /* Be paranoid */ | |
169 | return -ENOSPC; | |
170 | ||
171 | return *req_msi; | |
85f2bf9f ME |
172 | } |
173 | ||
174 | return 0; | |
175 | } | |
176 | ||
3a51c0cb ME |
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 | ||
448e2ca0 ME |
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 | ||
214 | /* Found our PE and assume 8 at that point. */ | |
215 | ||
216 | dn = pci_device_to_OF_node(dev); | |
217 | if (!dn) | |
218 | return NULL; | |
219 | ||
cce4b2d2 | 220 | dn = eeh_find_device_pe(dn); |
448e2ca0 ME |
221 | if (!dn) |
222 | return NULL; | |
223 | ||
224 | /* We actually want the parent */ | |
225 | dn = of_get_parent(dn); | |
226 | if (!dn) | |
227 | return NULL; | |
228 | ||
229 | /* Hardcode of 8 for old firmwares */ | |
230 | *total = 8; | |
231 | pr_debug("rtas_msi: using PE dn %s\n", dn->full_name); | |
232 | ||
233 | return dn; | |
234 | } | |
235 | ||
236 | struct msi_counts { | |
237 | struct device_node *requestor; | |
238 | int num_devices; | |
239 | int request; | |
240 | int quota; | |
241 | int spare; | |
242 | int over_quota; | |
243 | }; | |
244 | ||
245 | static void *count_non_bridge_devices(struct device_node *dn, void *data) | |
246 | { | |
247 | struct msi_counts *counts = data; | |
248 | const u32 *p; | |
249 | u32 class; | |
250 | ||
251 | pr_debug("rtas_msi: counting %s\n", dn->full_name); | |
252 | ||
253 | p = of_get_property(dn, "class-code", NULL); | |
254 | class = p ? *p : 0; | |
255 | ||
256 | if ((class >> 8) != PCI_CLASS_BRIDGE_PCI) | |
257 | counts->num_devices++; | |
258 | ||
259 | return NULL; | |
260 | } | |
261 | ||
262 | static void *count_spare_msis(struct device_node *dn, void *data) | |
263 | { | |
264 | struct msi_counts *counts = data; | |
265 | const u32 *p; | |
266 | int req; | |
267 | ||
268 | if (dn == counts->requestor) | |
269 | req = counts->request; | |
270 | else { | |
271 | /* We don't know if a driver will try to use MSI or MSI-X, | |
272 | * so we just have to punt and use the larger of the two. */ | |
273 | req = 0; | |
274 | p = of_get_property(dn, "ibm,req#msi", NULL); | |
275 | if (p) | |
276 | req = *p; | |
277 | ||
278 | p = of_get_property(dn, "ibm,req#msi-x", NULL); | |
279 | if (p) | |
280 | req = max(req, (int)*p); | |
281 | } | |
282 | ||
283 | if (req < counts->quota) | |
284 | counts->spare += counts->quota - req; | |
285 | else if (req > counts->quota) | |
286 | counts->over_quota++; | |
287 | ||
288 | return NULL; | |
289 | } | |
290 | ||
291 | static int msi_quota_for_device(struct pci_dev *dev, int request) | |
292 | { | |
293 | struct device_node *pe_dn; | |
294 | struct msi_counts counts; | |
295 | int total; | |
296 | ||
297 | pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev), | |
298 | request); | |
299 | ||
300 | pe_dn = find_pe_total_msi(dev, &total); | |
301 | if (!pe_dn) | |
302 | pe_dn = find_pe_dn(dev, &total); | |
303 | ||
304 | if (!pe_dn) { | |
305 | pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev)); | |
306 | goto out; | |
307 | } | |
308 | ||
309 | pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name); | |
310 | ||
311 | memset(&counts, 0, sizeof(struct msi_counts)); | |
312 | ||
313 | /* Work out how many devices we have below this PE */ | |
314 | traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts); | |
315 | ||
316 | if (counts.num_devices == 0) { | |
317 | pr_err("rtas_msi: found 0 devices under PE for %s\n", | |
318 | pci_name(dev)); | |
319 | goto out; | |
320 | } | |
321 | ||
322 | counts.quota = total / counts.num_devices; | |
323 | if (request <= counts.quota) | |
324 | goto out; | |
325 | ||
326 | /* else, we have some more calculating to do */ | |
327 | counts.requestor = pci_device_to_OF_node(dev); | |
328 | counts.request = request; | |
329 | traverse_pci_devices(pe_dn, count_spare_msis, &counts); | |
330 | ||
331 | /* If the quota isn't an integer multiple of the total, we can | |
332 | * use the remainder as spare MSIs for anyone that wants them. */ | |
333 | counts.spare += total % counts.num_devices; | |
334 | ||
335 | /* Divide any spare by the number of over-quota requestors */ | |
336 | if (counts.over_quota) | |
337 | counts.quota += counts.spare / counts.over_quota; | |
338 | ||
339 | /* And finally clamp the request to the possibly adjusted quota */ | |
340 | request = min(counts.quota, request); | |
341 | ||
342 | pr_debug("rtas_msi: request clamped to quota %d\n", request); | |
343 | out: | |
344 | of_node_put(pe_dn); | |
345 | ||
346 | return request; | |
347 | } | |
348 | ||
85f2bf9f ME |
349 | static int rtas_msi_check_device(struct pci_dev *pdev, int nvec, int type) |
350 | { | |
448e2ca0 ME |
351 | int quota, rc; |
352 | ||
85f2bf9f | 353 | if (type == PCI_CAP_ID_MSIX) |
448e2ca0 ME |
354 | rc = check_req_msix(pdev, nvec); |
355 | else | |
356 | rc = check_req_msi(pdev, nvec); | |
357 | ||
358 | if (rc) | |
359 | return rc; | |
85f2bf9f | 360 | |
448e2ca0 ME |
361 | quota = msi_quota_for_device(pdev, nvec); |
362 | ||
363 | if (quota && quota < nvec) | |
364 | return quota; | |
365 | ||
366 | return 0; | |
85f2bf9f ME |
367 | } |
368 | ||
94afa5a5 ME |
369 | static int check_msix_entries(struct pci_dev *pdev) |
370 | { | |
371 | struct msi_desc *entry; | |
372 | int expected; | |
373 | ||
374 | /* There's no way for us to express to firmware that we want | |
375 | * a discontiguous, or non-zero based, range of MSI-X entries. | |
376 | * So we must reject such requests. */ | |
377 | ||
378 | expected = 0; | |
379 | list_for_each_entry(entry, &pdev->msi_list, list) { | |
380 | if (entry->msi_attrib.entry_nr != expected) { | |
381 | pr_debug("rtas_msi: bad MSI-X entries.\n"); | |
382 | return -EINVAL; | |
383 | } | |
384 | expected++; | |
385 | } | |
386 | ||
387 | return 0; | |
388 | } | |
389 | ||
752f5216 | 390 | static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type) |
85f2bf9f ME |
391 | { |
392 | struct pci_dn *pdn; | |
393 | int hwirq, virq, i, rc; | |
394 | struct msi_desc *entry; | |
1db3e890 | 395 | struct msi_msg msg; |
752f5216 | 396 | int nvec = nvec_in; |
85f2bf9f ME |
397 | |
398 | pdn = get_pdn(pdev); | |
399 | if (!pdn) | |
400 | return -ENODEV; | |
401 | ||
94afa5a5 ME |
402 | if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev)) |
403 | return -EINVAL; | |
404 | ||
752f5216 AB |
405 | /* |
406 | * Firmware currently refuse any non power of two allocation | |
407 | * so we round up if the quota will allow it. | |
408 | */ | |
409 | if (type == PCI_CAP_ID_MSIX) { | |
410 | int m = roundup_pow_of_two(nvec); | |
411 | int quota = msi_quota_for_device(pdev, m); | |
412 | ||
413 | if (quota >= m) | |
414 | nvec = m; | |
415 | } | |
416 | ||
85f2bf9f ME |
417 | /* |
418 | * Try the new more explicit firmware interface, if that fails fall | |
419 | * back to the old interface. The old interface is known to never | |
420 | * return MSI-Xs. | |
421 | */ | |
752f5216 | 422 | again: |
85f2bf9f ME |
423 | if (type == PCI_CAP_ID_MSI) { |
424 | rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec); | |
425 | ||
6071ed04 | 426 | if (rc < 0) { |
85f2bf9f ME |
427 | pr_debug("rtas_msi: trying the old firmware call.\n"); |
428 | rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec); | |
429 | } | |
430 | } else | |
431 | rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec); | |
432 | ||
6071ed04 | 433 | if (rc != nvec) { |
752f5216 AB |
434 | if (nvec != nvec_in) { |
435 | nvec = nvec_in; | |
436 | goto again; | |
437 | } | |
85f2bf9f | 438 | pr_debug("rtas_msi: rtas_change_msi() failed\n"); |
fcbe8090 | 439 | return rc; |
85f2bf9f ME |
440 | } |
441 | ||
442 | i = 0; | |
443 | list_for_each_entry(entry, &pdev->msi_list, list) { | |
e27ed698 | 444 | hwirq = rtas_query_irq_number(pdn, i++); |
85f2bf9f | 445 | if (hwirq < 0) { |
85f2bf9f | 446 | pr_debug("rtas_msi: error (%d) getting hwirq\n", rc); |
fcbe8090 | 447 | return hwirq; |
85f2bf9f ME |
448 | } |
449 | ||
450 | virq = irq_create_mapping(NULL, hwirq); | |
451 | ||
452 | if (virq == NO_IRQ) { | |
453 | pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq); | |
fcbe8090 | 454 | return -ENOSPC; |
85f2bf9f ME |
455 | } |
456 | ||
457 | dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq); | |
ec775d0e | 458 | irq_set_msi_desc(virq, entry); |
1db3e890 ME |
459 | |
460 | /* Read config space back so we can restore after reset */ | |
461 | read_msi_msg(virq, &msg); | |
462 | entry->msg = msg; | |
85f2bf9f ME |
463 | } |
464 | ||
465 | return 0; | |
85f2bf9f ME |
466 | } |
467 | ||
468 | static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev) | |
469 | { | |
470 | /* No LSI -> leave MSIs (if any) configured */ | |
471 | if (pdev->irq == NO_IRQ) { | |
472 | dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n"); | |
473 | return; | |
474 | } | |
475 | ||
476 | /* No MSI -> MSIs can't have been assigned by fw, leave LSI */ | |
649781f8 ME |
477 | if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) { |
478 | dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n"); | |
85f2bf9f ME |
479 | return; |
480 | } | |
481 | ||
482 | dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n"); | |
483 | rtas_disable_msi(pdev); | |
484 | } | |
485 | ||
486 | static int rtas_msi_init(void) | |
487 | { | |
488 | query_token = rtas_token("ibm,query-interrupt-source-number"); | |
489 | change_token = rtas_token("ibm,change-msi"); | |
490 | ||
491 | if ((query_token == RTAS_UNKNOWN_SERVICE) || | |
492 | (change_token == RTAS_UNKNOWN_SERVICE)) { | |
493 | pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n"); | |
494 | return -1; | |
495 | } | |
496 | ||
497 | pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n"); | |
498 | ||
499 | WARN_ON(ppc_md.setup_msi_irqs); | |
500 | ppc_md.setup_msi_irqs = rtas_setup_msi_irqs; | |
501 | ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs; | |
502 | ppc_md.msi_check_device = rtas_msi_check_device; | |
503 | ||
504 | WARN_ON(ppc_md.pci_irq_fixup); | |
505 | ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup; | |
506 | ||
507 | return 0; | |
508 | } | |
509 | arch_initcall(rtas_msi_init); |