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>
19 #include <asm/machdep.h>
21 static int query_token, change_token;
23 #define RTAS_QUERY_FN 0
24 #define RTAS_CHANGE_FN 1
25 #define RTAS_RESET_FN 2
26 #define RTAS_CHANGE_MSI_FN 3
27 #define RTAS_CHANGE_MSIX_FN 4
28 #define RTAS_CHANGE_32MSI_FN 5
32 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
34 u32 addr, seq_num, rtas_ret[3];
38 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
39 buid = pdn->phb->buid;
43 if (func == RTAS_CHANGE_MSI_FN || func == RTAS_CHANGE_MSIX_FN ||
44 func == RTAS_CHANGE_32MSI_FN)
45 rc = rtas_call(change_token, 6, 4, rtas_ret, addr,
46 BUID_HI(buid), BUID_LO(buid),
47 func, num_irqs, seq_num);
49 rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
50 BUID_HI(buid), BUID_LO(buid),
51 func, num_irqs, seq_num);
53 seq_num = rtas_ret[1];
54 } while (rtas_busy_delay(rc));
57 * If the RTAS call succeeded, return the number of irqs allocated.
58 * If not, make sure we return a negative error code.
65 pr_debug("rtas_msi: ibm,change_msi(func=%d,num=%d), got %d rc = %d\n",
66 func, num_irqs, rtas_ret[0], rc);
71 static void rtas_disable_msi(struct pci_dev *pdev)
75 pdn = pci_get_pdn(pdev);
80 * disabling MSI with the explicit interface also disables MSI-X
82 if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
84 * may have failed because explicit interface is not
87 if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
88 pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
93 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
95 u32 addr, rtas_ret[2];
99 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
100 buid = pdn->phb->buid;
103 rc = rtas_call(query_token, 4, 3, rtas_ret, addr,
104 BUID_HI(buid), BUID_LO(buid), offset);
105 } while (rtas_busy_delay(rc));
108 pr_debug("rtas_msi: error (%d) querying source number\n", rc);
115 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
117 struct msi_desc *entry;
119 list_for_each_entry(entry, &pdev->msi_list, list) {
120 if (entry->irq == NO_IRQ)
123 irq_set_msi_desc(entry->irq, NULL);
124 irq_dispose_mapping(entry->irq);
127 rtas_disable_msi(pdev);
130 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
132 struct device_node *dn;
137 pdn = pci_get_pdn(pdev);
143 p = of_get_property(dn, prop_name, NULL);
145 pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name);
149 req_msi = be32_to_cpup(p);
150 if (req_msi < nvec) {
151 pr_debug("rtas_msi: %s requests < %d MSIs\n", prop_name, nvec);
153 if (req_msi == 0) /* Be paranoid */
162 static int check_req_msi(struct pci_dev *pdev, int nvec)
164 return check_req(pdev, nvec, "ibm,req#msi");
167 static int check_req_msix(struct pci_dev *pdev, int nvec)
169 return check_req(pdev, nvec, "ibm,req#msi-x");
172 /* Quota calculation */
174 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
176 struct device_node *dn;
179 dn = of_node_get(pci_device_to_OF_node(dev));
181 p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
183 pr_debug("rtas_msi: found prop on dn %s\n",
185 *total = be32_to_cpup(p);
189 dn = of_get_next_parent(dn);
195 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
197 struct device_node *dn;
198 struct eeh_dev *edev;
200 /* Found our PE and assume 8 at that point. */
202 dn = pci_device_to_OF_node(dev);
206 /* Get the top level device in the PE */
207 edev = of_node_to_eeh_dev(dn);
209 edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
210 dn = eeh_dev_to_of_node(edev);
214 /* We actually want the parent */
215 dn = of_get_parent(dn);
219 /* Hardcode of 8 for old firmwares */
221 pr_debug("rtas_msi: using PE dn %s\n", dn->full_name);
227 struct device_node *requestor;
235 static void *count_non_bridge_devices(struct device_node *dn, void *data)
237 struct msi_counts *counts = data;
241 pr_debug("rtas_msi: counting %s\n", dn->full_name);
243 p = of_get_property(dn, "class-code", NULL);
244 class = p ? be32_to_cpup(p) : 0;
246 if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
247 counts->num_devices++;
252 static void *count_spare_msis(struct device_node *dn, void *data)
254 struct msi_counts *counts = data;
258 if (dn == counts->requestor)
259 req = counts->request;
261 /* We don't know if a driver will try to use MSI or MSI-X,
262 * so we just have to punt and use the larger of the two. */
264 p = of_get_property(dn, "ibm,req#msi", NULL);
266 req = be32_to_cpup(p);
268 p = of_get_property(dn, "ibm,req#msi-x", NULL);
270 req = max(req, (int)be32_to_cpup(p));
273 if (req < counts->quota)
274 counts->spare += counts->quota - req;
275 else if (req > counts->quota)
276 counts->over_quota++;
281 static int msi_quota_for_device(struct pci_dev *dev, int request)
283 struct device_node *pe_dn;
284 struct msi_counts counts;
287 pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
290 pe_dn = find_pe_total_msi(dev, &total);
292 pe_dn = find_pe_dn(dev, &total);
295 pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
299 pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name);
301 memset(&counts, 0, sizeof(struct msi_counts));
303 /* Work out how many devices we have below this PE */
304 traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts);
306 if (counts.num_devices == 0) {
307 pr_err("rtas_msi: found 0 devices under PE for %s\n",
312 counts.quota = total / counts.num_devices;
313 if (request <= counts.quota)
316 /* else, we have some more calculating to do */
317 counts.requestor = pci_device_to_OF_node(dev);
318 counts.request = request;
319 traverse_pci_devices(pe_dn, count_spare_msis, &counts);
321 /* If the quota isn't an integer multiple of the total, we can
322 * use the remainder as spare MSIs for anyone that wants them. */
323 counts.spare += total % counts.num_devices;
325 /* Divide any spare by the number of over-quota requestors */
326 if (counts.over_quota)
327 counts.quota += counts.spare / counts.over_quota;
329 /* And finally clamp the request to the possibly adjusted quota */
330 request = min(counts.quota, request);
332 pr_debug("rtas_msi: request clamped to quota %d\n", request);
339 static int check_msix_entries(struct pci_dev *pdev)
341 struct msi_desc *entry;
344 /* There's no way for us to express to firmware that we want
345 * a discontiguous, or non-zero based, range of MSI-X entries.
346 * So we must reject such requests. */
349 list_for_each_entry(entry, &pdev->msi_list, list) {
350 if (entry->msi_attrib.entry_nr != expected) {
351 pr_debug("rtas_msi: bad MSI-X entries.\n");
360 static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
362 u32 addr_hi, addr_lo;
365 * We should only get in here for IODA1 configs. This is based on the
366 * fact that we using RTAS for MSIs, we don't have the 32 bit MSI RTAS
367 * support, and we are in a PCIe Gen2 slot.
370 "rtas_msi: No 32 bit MSI firmware support, forcing 32 bit MSI\n");
371 pci_read_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, &addr_hi);
372 addr_lo = 0xffff0000 | ((addr_hi >> (48 - 32)) << 4);
373 pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_LO, addr_lo);
374 pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, 0);
377 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
380 int hwirq, virq, i, quota, rc;
381 struct msi_desc *entry;
384 int use_32bit_msi_hack = 0;
386 if (type == PCI_CAP_ID_MSIX)
387 rc = check_req_msix(pdev, nvec);
389 rc = check_req_msi(pdev, nvec);
394 quota = msi_quota_for_device(pdev, nvec);
396 if (quota && quota < nvec)
399 if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
403 * Firmware currently refuse any non power of two allocation
404 * so we round up if the quota will allow it.
406 if (type == PCI_CAP_ID_MSIX) {
407 int m = roundup_pow_of_two(nvec);
408 quota = msi_quota_for_device(pdev, m);
414 pdn = pci_get_pdn(pdev);
417 * Try the new more explicit firmware interface, if that fails fall
418 * back to the old interface. The old interface is known to never
422 if (type == PCI_CAP_ID_MSI) {
423 if (pdev->no_64bit_msi) {
424 rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
427 * We only want to run the 32 bit MSI hack below if
428 * the max bus speed is Gen2 speed
430 if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
433 use_32bit_msi_hack = 1;
439 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
442 pr_debug("rtas_msi: trying the old firmware call.\n");
443 rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
446 if (use_32bit_msi_hack && rc > 0)
447 rtas_hack_32bit_msi_gen2(pdev);
449 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
452 if (nvec != nvec_in) {
456 pr_debug("rtas_msi: rtas_change_msi() failed\n");
461 list_for_each_entry(entry, &pdev->msi_list, list) {
462 hwirq = rtas_query_irq_number(pdn, i++);
464 pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
468 virq = irq_create_mapping(NULL, hwirq);
470 if (virq == NO_IRQ) {
471 pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
475 dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
476 irq_set_msi_desc(virq, entry);
478 /* Read config space back so we can restore after reset */
479 __read_msi_msg(entry, &msg);
486 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
488 /* No LSI -> leave MSIs (if any) configured */
489 if (pdev->irq == NO_IRQ) {
490 dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
494 /* No MSI -> MSIs can't have been assigned by fw, leave LSI */
495 if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
496 dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
500 dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
501 rtas_disable_msi(pdev);
504 static int rtas_msi_init(void)
506 query_token = rtas_token("ibm,query-interrupt-source-number");
507 change_token = rtas_token("ibm,change-msi");
509 if ((query_token == RTAS_UNKNOWN_SERVICE) ||
510 (change_token == RTAS_UNKNOWN_SERVICE)) {
511 pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
515 pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
517 WARN_ON(ppc_md.setup_msi_irqs);
518 ppc_md.setup_msi_irqs = rtas_setup_msi_irqs;
519 ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs;
521 WARN_ON(ppc_md.pci_irq_fixup);
522 ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
526 machine_arch_initcall(pseries, rtas_msi_init);