2 * Copyright 2006 Jake Moilanen <moilanen@austin.ibm.com>, IBM Corp.
3 * Copyright 2006-2007 Michael Ellerman, IBM Corp.
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
12 #include <linux/device.h>
13 #include <linux/irq.h>
14 #include <linux/msi.h>
17 #include <asm/hw_irq.h>
18 #include <asm/ppc-pci.h>
20 static int query_token, change_token;
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
28 static struct pci_dn *get_pdn(struct pci_dev *pdev)
30 struct device_node *dn;
33 dn = pci_device_to_OF_node(pdev);
35 dev_dbg(&pdev->dev, "rtas_msi: No OF device node\n");
41 dev_dbg(&pdev->dev, "rtas_msi: No PCI DN\n");
50 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
52 u32 addr, seq_num, rtas_ret[3];
56 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
57 buid = pdn->phb->buid;
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);
66 rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
67 BUID_HI(buid), BUID_LO(buid),
68 func, num_irqs, seq_num);
70 seq_num = rtas_ret[1];
71 } while (rtas_busy_delay(rc));
74 * If the RTAS call succeeded, return the number of irqs allocated.
75 * If not, make sure we return a negative error code.
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);
88 static void rtas_disable_msi(struct pci_dev *pdev)
97 * disabling MSI with the explicit interface also disables MSI-X
99 if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
101 * may have failed because explicit interface is not
104 if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
105 pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
110 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
112 u32 addr, rtas_ret[2];
116 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
117 buid = pdn->phb->buid;
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));
125 pr_debug("rtas_msi: error (%d) querying source number\n", rc);
132 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
134 struct msi_desc *entry;
136 list_for_each_entry(entry, &pdev->msi_list, list) {
137 if (entry->irq == NO_IRQ)
140 irq_set_msi_desc(entry->irq, NULL);
141 irq_dispose_mapping(entry->irq);
144 rtas_disable_msi(pdev);
147 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
149 struct device_node *dn;
159 req_msi = of_get_property(dn, prop_name, NULL);
161 pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name);
165 if (*req_msi < nvec) {
166 pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
168 if (*req_msi == 0) /* Be paranoid */
177 static int check_req_msi(struct pci_dev *pdev, int nvec)
179 return check_req(pdev, nvec, "ibm,req#msi");
182 static int check_req_msix(struct pci_dev *pdev, int nvec)
184 return check_req(pdev, nvec, "ibm,req#msi-x");
187 /* Quota calculation */
189 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
191 struct device_node *dn;
194 dn = of_node_get(pci_device_to_OF_node(dev));
196 p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
198 pr_debug("rtas_msi: found prop on dn %s\n",
204 dn = of_get_next_parent(dn);
210 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
212 struct device_node *dn;
213 struct eeh_dev *edev;
215 /* Found our PE and assume 8 at that point. */
217 dn = pci_device_to_OF_node(dev);
221 /* Get the top level device in the PE */
222 edev = of_node_to_eeh_dev(dn);
223 edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
224 dn = eeh_dev_to_of_node(edev);
228 /* We actually want the parent */
229 dn = of_get_parent(dn);
233 /* Hardcode of 8 for old firmwares */
235 pr_debug("rtas_msi: using PE dn %s\n", dn->full_name);
241 struct device_node *requestor;
249 static void *count_non_bridge_devices(struct device_node *dn, void *data)
251 struct msi_counts *counts = data;
255 pr_debug("rtas_msi: counting %s\n", dn->full_name);
257 p = of_get_property(dn, "class-code", NULL);
260 if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
261 counts->num_devices++;
266 static void *count_spare_msis(struct device_node *dn, void *data)
268 struct msi_counts *counts = data;
272 if (dn == counts->requestor)
273 req = counts->request;
275 /* We don't know if a driver will try to use MSI or MSI-X,
276 * so we just have to punt and use the larger of the two. */
278 p = of_get_property(dn, "ibm,req#msi", NULL);
282 p = of_get_property(dn, "ibm,req#msi-x", NULL);
284 req = max(req, (int)*p);
287 if (req < counts->quota)
288 counts->spare += counts->quota - req;
289 else if (req > counts->quota)
290 counts->over_quota++;
295 static int msi_quota_for_device(struct pci_dev *dev, int request)
297 struct device_node *pe_dn;
298 struct msi_counts counts;
301 pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
304 pe_dn = find_pe_total_msi(dev, &total);
306 pe_dn = find_pe_dn(dev, &total);
309 pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
313 pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name);
315 memset(&counts, 0, sizeof(struct msi_counts));
317 /* Work out how many devices we have below this PE */
318 traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts);
320 if (counts.num_devices == 0) {
321 pr_err("rtas_msi: found 0 devices under PE for %s\n",
326 counts.quota = total / counts.num_devices;
327 if (request <= counts.quota)
330 /* else, we have some more calculating to do */
331 counts.requestor = pci_device_to_OF_node(dev);
332 counts.request = request;
333 traverse_pci_devices(pe_dn, count_spare_msis, &counts);
335 /* If the quota isn't an integer multiple of the total, we can
336 * use the remainder as spare MSIs for anyone that wants them. */
337 counts.spare += total % counts.num_devices;
339 /* Divide any spare by the number of over-quota requestors */
340 if (counts.over_quota)
341 counts.quota += counts.spare / counts.over_quota;
343 /* And finally clamp the request to the possibly adjusted quota */
344 request = min(counts.quota, request);
346 pr_debug("rtas_msi: request clamped to quota %d\n", request);
353 static int rtas_msi_check_device(struct pci_dev *pdev, int nvec, int type)
357 if (type == PCI_CAP_ID_MSIX)
358 rc = check_req_msix(pdev, nvec);
360 rc = check_req_msi(pdev, nvec);
365 quota = msi_quota_for_device(pdev, nvec);
367 if (quota && quota < nvec)
373 static int check_msix_entries(struct pci_dev *pdev)
375 struct msi_desc *entry;
378 /* There's no way for us to express to firmware that we want
379 * a discontiguous, or non-zero based, range of MSI-X entries.
380 * So we must reject such requests. */
383 list_for_each_entry(entry, &pdev->msi_list, list) {
384 if (entry->msi_attrib.entry_nr != expected) {
385 pr_debug("rtas_msi: bad MSI-X entries.\n");
394 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
397 int hwirq, virq, i, rc;
398 struct msi_desc *entry;
406 if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
410 * Firmware currently refuse any non power of two allocation
411 * so we round up if the quota will allow it.
413 if (type == PCI_CAP_ID_MSIX) {
414 int m = roundup_pow_of_two(nvec);
415 int quota = msi_quota_for_device(pdev, m);
422 * Try the new more explicit firmware interface, if that fails fall
423 * back to the old interface. The old interface is known to never
427 if (type == PCI_CAP_ID_MSI) {
428 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
431 pr_debug("rtas_msi: trying the old firmware call.\n");
432 rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
435 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
438 if (nvec != nvec_in) {
442 pr_debug("rtas_msi: rtas_change_msi() failed\n");
447 list_for_each_entry(entry, &pdev->msi_list, list) {
448 hwirq = rtas_query_irq_number(pdn, i++);
450 pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
454 virq = irq_create_mapping(NULL, hwirq);
456 if (virq == NO_IRQ) {
457 pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
461 dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
462 irq_set_msi_desc(virq, entry);
464 /* Read config space back so we can restore after reset */
465 read_msi_msg(virq, &msg);
472 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
474 /* No LSI -> leave MSIs (if any) configured */
475 if (pdev->irq == NO_IRQ) {
476 dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
480 /* No MSI -> MSIs can't have been assigned by fw, leave LSI */
481 if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
482 dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
486 dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
487 rtas_disable_msi(pdev);
490 static int rtas_msi_init(void)
492 query_token = rtas_token("ibm,query-interrupt-source-number");
493 change_token = rtas_token("ibm,change-msi");
495 if ((query_token == RTAS_UNKNOWN_SERVICE) ||
496 (change_token == RTAS_UNKNOWN_SERVICE)) {
497 pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
501 pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
503 WARN_ON(ppc_md.setup_msi_irqs);
504 ppc_md.setup_msi_irqs = rtas_setup_msi_irqs;
505 ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs;
506 ppc_md.msi_check_device = rtas_msi_check_device;
508 WARN_ON(ppc_md.pci_irq_fixup);
509 ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
513 arch_initcall(rtas_msi_init);