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
27 #define RTAS_CHANGE_32MSI_FN 5
31 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
33 u32 addr, seq_num, rtas_ret[3];
37 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
38 buid = pdn->phb->buid;
42 if (func == RTAS_CHANGE_MSI_FN || func == RTAS_CHANGE_MSIX_FN ||
43 func == RTAS_CHANGE_32MSI_FN)
44 rc = rtas_call(change_token, 6, 4, rtas_ret, addr,
45 BUID_HI(buid), BUID_LO(buid),
46 func, num_irqs, seq_num);
48 rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
49 BUID_HI(buid), BUID_LO(buid),
50 func, num_irqs, seq_num);
52 seq_num = rtas_ret[1];
53 } while (rtas_busy_delay(rc));
56 * If the RTAS call succeeded, return the number of irqs allocated.
57 * If not, make sure we return a negative error code.
64 pr_debug("rtas_msi: ibm,change_msi(func=%d,num=%d), got %d rc = %d\n",
65 func, num_irqs, rtas_ret[0], rc);
70 static void rtas_disable_msi(struct pci_dev *pdev)
74 pdn = pci_get_pdn(pdev);
79 * disabling MSI with the explicit interface also disables MSI-X
81 if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
83 * may have failed because explicit interface is not
86 if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
87 pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
92 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
94 u32 addr, rtas_ret[2];
98 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
99 buid = pdn->phb->buid;
102 rc = rtas_call(query_token, 4, 3, rtas_ret, addr,
103 BUID_HI(buid), BUID_LO(buid), offset);
104 } while (rtas_busy_delay(rc));
107 pr_debug("rtas_msi: error (%d) querying source number\n", rc);
114 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
116 struct msi_desc *entry;
118 list_for_each_entry(entry, &pdev->msi_list, list) {
119 if (entry->irq == NO_IRQ)
122 irq_set_msi_desc(entry->irq, NULL);
123 irq_dispose_mapping(entry->irq);
126 rtas_disable_msi(pdev);
129 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
131 struct device_node *dn;
135 pdn = pci_get_pdn(pdev);
141 req_msi = of_get_property(dn, prop_name, NULL);
143 pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name);
147 if (*req_msi < nvec) {
148 pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
150 if (*req_msi == 0) /* Be paranoid */
159 static int check_req_msi(struct pci_dev *pdev, int nvec)
161 return check_req(pdev, nvec, "ibm,req#msi");
164 static int check_req_msix(struct pci_dev *pdev, int nvec)
166 return check_req(pdev, nvec, "ibm,req#msi-x");
169 /* Quota calculation */
171 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
173 struct device_node *dn;
176 dn = of_node_get(pci_device_to_OF_node(dev));
178 p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
180 pr_debug("rtas_msi: found prop on dn %s\n",
186 dn = of_get_next_parent(dn);
192 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
194 struct device_node *dn;
195 struct eeh_dev *edev;
197 /* Found our PE and assume 8 at that point. */
199 dn = pci_device_to_OF_node(dev);
203 /* Get the top level device in the PE */
204 edev = of_node_to_eeh_dev(dn);
206 edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
207 dn = eeh_dev_to_of_node(edev);
211 /* We actually want the parent */
212 dn = of_get_parent(dn);
216 /* Hardcode of 8 for old firmwares */
218 pr_debug("rtas_msi: using PE dn %s\n", dn->full_name);
224 struct device_node *requestor;
232 static void *count_non_bridge_devices(struct device_node *dn, void *data)
234 struct msi_counts *counts = data;
238 pr_debug("rtas_msi: counting %s\n", dn->full_name);
240 p = of_get_property(dn, "class-code", NULL);
243 if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
244 counts->num_devices++;
249 static void *count_spare_msis(struct device_node *dn, void *data)
251 struct msi_counts *counts = data;
255 if (dn == counts->requestor)
256 req = counts->request;
258 /* We don't know if a driver will try to use MSI or MSI-X,
259 * so we just have to punt and use the larger of the two. */
261 p = of_get_property(dn, "ibm,req#msi", NULL);
265 p = of_get_property(dn, "ibm,req#msi-x", NULL);
267 req = max(req, (int)*p);
270 if (req < counts->quota)
271 counts->spare += counts->quota - req;
272 else if (req > counts->quota)
273 counts->over_quota++;
278 static int msi_quota_for_device(struct pci_dev *dev, int request)
280 struct device_node *pe_dn;
281 struct msi_counts counts;
284 pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
287 pe_dn = find_pe_total_msi(dev, &total);
289 pe_dn = find_pe_dn(dev, &total);
292 pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
296 pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name);
298 memset(&counts, 0, sizeof(struct msi_counts));
300 /* Work out how many devices we have below this PE */
301 traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts);
303 if (counts.num_devices == 0) {
304 pr_err("rtas_msi: found 0 devices under PE for %s\n",
309 counts.quota = total / counts.num_devices;
310 if (request <= counts.quota)
313 /* else, we have some more calculating to do */
314 counts.requestor = pci_device_to_OF_node(dev);
315 counts.request = request;
316 traverse_pci_devices(pe_dn, count_spare_msis, &counts);
318 /* If the quota isn't an integer multiple of the total, we can
319 * use the remainder as spare MSIs for anyone that wants them. */
320 counts.spare += total % counts.num_devices;
322 /* Divide any spare by the number of over-quota requestors */
323 if (counts.over_quota)
324 counts.quota += counts.spare / counts.over_quota;
326 /* And finally clamp the request to the possibly adjusted quota */
327 request = min(counts.quota, request);
329 pr_debug("rtas_msi: request clamped to quota %d\n", request);
336 static int rtas_msi_check_device(struct pci_dev *pdev, int nvec, int type)
340 if (type == PCI_CAP_ID_MSIX)
341 rc = check_req_msix(pdev, nvec);
343 rc = check_req_msi(pdev, nvec);
348 quota = msi_quota_for_device(pdev, nvec);
350 if (quota && quota < nvec)
356 static int check_msix_entries(struct pci_dev *pdev)
358 struct msi_desc *entry;
361 /* There's no way for us to express to firmware that we want
362 * a discontiguous, or non-zero based, range of MSI-X entries.
363 * So we must reject such requests. */
366 list_for_each_entry(entry, &pdev->msi_list, list) {
367 if (entry->msi_attrib.entry_nr != expected) {
368 pr_debug("rtas_msi: bad MSI-X entries.\n");
377 static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
379 u32 addr_hi, addr_lo;
382 * We should only get in here for IODA1 configs. This is based on the
383 * fact that we using RTAS for MSIs, we don't have the 32 bit MSI RTAS
384 * support, and we are in a PCIe Gen2 slot.
387 "rtas_msi: No 32 bit MSI firmware support, forcing 32 bit MSI\n");
388 pci_read_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, &addr_hi);
389 addr_lo = 0xffff0000 | ((addr_hi >> (48 - 32)) << 4);
390 pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_LO, addr_lo);
391 pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, 0);
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;
401 int use_32bit_msi_hack = 0;
403 pdn = pci_get_pdn(pdev);
407 if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
411 * Firmware currently refuse any non power of two allocation
412 * so we round up if the quota will allow it.
414 if (type == PCI_CAP_ID_MSIX) {
415 int m = roundup_pow_of_two(nvec);
416 int quota = msi_quota_for_device(pdev, m);
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
428 if (type == PCI_CAP_ID_MSI) {
429 if (pdn->force_32bit_msi) {
430 rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
433 * We only want to run the 32 bit MSI hack below if
434 * the max bus speed is Gen2 speed
436 if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
439 use_32bit_msi_hack = 1;
445 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
448 pr_debug("rtas_msi: trying the old firmware call.\n");
449 rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
452 if (use_32bit_msi_hack && rc > 0)
453 rtas_hack_32bit_msi_gen2(pdev);
455 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
458 if (nvec != nvec_in) {
462 pr_debug("rtas_msi: rtas_change_msi() failed\n");
467 list_for_each_entry(entry, &pdev->msi_list, list) {
468 hwirq = rtas_query_irq_number(pdn, i++);
470 pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
474 virq = irq_create_mapping(NULL, hwirq);
476 if (virq == NO_IRQ) {
477 pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
481 dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
482 irq_set_msi_desc(virq, entry);
484 /* Read config space back so we can restore after reset */
485 read_msi_msg(virq, &msg);
492 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
494 /* No LSI -> leave MSIs (if any) configured */
495 if (pdev->irq == NO_IRQ) {
496 dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
500 /* No MSI -> MSIs can't have been assigned by fw, leave LSI */
501 if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
502 dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
506 dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
507 rtas_disable_msi(pdev);
510 static int rtas_msi_init(void)
512 query_token = rtas_token("ibm,query-interrupt-source-number");
513 change_token = rtas_token("ibm,change-msi");
515 if ((query_token == RTAS_UNKNOWN_SERVICE) ||
516 (change_token == RTAS_UNKNOWN_SERVICE)) {
517 pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
521 pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
523 WARN_ON(ppc_md.setup_msi_irqs);
524 ppc_md.setup_msi_irqs = rtas_setup_msi_irqs;
525 ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs;
526 ppc_md.msi_check_device = rtas_msi_check_device;
528 WARN_ON(ppc_md.pci_irq_fixup);
529 ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
533 arch_initcall(rtas_msi_init);