1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2006 Jake Moilanen <moilanen@austin.ibm.com>, IBM Corp.
4 * Copyright 2006-2007 Michael Ellerman, IBM Corp.
7 #include <linux/device.h>
12 #include <asm/hw_irq.h>
13 #include <asm/ppc-pci.h>
14 #include <asm/machdep.h>
18 static int query_token, change_token;
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
29 static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
31 u32 addr, seq_num, rtas_ret[3];
35 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
36 buid = pdn->phb->buid;
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);
46 rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
47 BUID_HI(buid), BUID_LO(buid),
48 func, num_irqs, seq_num);
50 seq_num = rtas_ret[1];
51 } while (rtas_busy_delay(rc));
54 * If the RTAS call succeeded, return the number of irqs allocated.
55 * If not, make sure we return a negative error code.
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);
68 static void rtas_disable_msi(struct pci_dev *pdev)
72 pdn = pci_get_pdn(pdev);
77 * disabling MSI with the explicit interface also disables MSI-X
79 if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
81 * may have failed because explicit interface is not
84 if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
85 pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
90 static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
92 u32 addr, rtas_ret[2];
96 addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
97 buid = pdn->phb->buid;
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));
105 pr_debug("rtas_msi: error (%d) querying source number\n", rc);
112 static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
114 struct msi_desc *entry;
116 for_each_pci_msi_entry(entry, pdev) {
120 irq_set_msi_desc(entry->irq, NULL);
121 irq_dispose_mapping(entry->irq);
124 rtas_disable_msi(pdev);
127 static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
129 struct device_node *dn;
133 dn = pci_device_to_OF_node(pdev);
135 p = of_get_property(dn, prop_name, NULL);
137 pr_debug("rtas_msi: No %s on %pOF\n", prop_name, dn);
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);
145 if (req_msi == 0) /* Be paranoid */
154 static int check_req_msi(struct pci_dev *pdev, int nvec)
156 return check_req(pdev, nvec, "ibm,req#msi");
159 static int check_req_msix(struct pci_dev *pdev, int nvec)
161 return check_req(pdev, nvec, "ibm,req#msi-x");
164 /* Quota calculation */
166 static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
168 struct device_node *dn;
171 dn = of_node_get(pci_device_to_OF_node(dev));
173 p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
175 pr_debug("rtas_msi: found prop on dn %pOF\n",
177 *total = be32_to_cpup(p);
181 dn = of_get_next_parent(dn);
187 static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
189 struct device_node *dn;
190 struct eeh_dev *edev;
192 /* Found our PE and assume 8 at that point. */
194 dn = pci_device_to_OF_node(dev);
198 /* Get the top level device in the PE */
199 edev = pdn_to_eeh_dev(PCI_DN(dn));
201 edev = list_first_entry(&edev->pe->edevs, struct eeh_dev,
203 dn = pci_device_to_OF_node(edev->pdev);
207 /* We actually want the parent */
208 dn = of_get_parent(dn);
212 /* Hardcode of 8 for old firmwares */
214 pr_debug("rtas_msi: using PE dn %pOF\n", dn);
220 struct device_node *requestor;
228 static void *count_non_bridge_devices(struct device_node *dn, void *data)
230 struct msi_counts *counts = data;
234 pr_debug("rtas_msi: counting %pOF\n", dn);
236 p = of_get_property(dn, "class-code", NULL);
237 class = p ? be32_to_cpup(p) : 0;
239 if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
240 counts->num_devices++;
245 static void *count_spare_msis(struct device_node *dn, void *data)
247 struct msi_counts *counts = data;
251 if (dn == counts->requestor)
252 req = counts->request;
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. */
257 p = of_get_property(dn, "ibm,req#msi", NULL);
259 req = be32_to_cpup(p);
261 p = of_get_property(dn, "ibm,req#msi-x", NULL);
263 req = max(req, (int)be32_to_cpup(p));
266 if (req < counts->quota)
267 counts->spare += counts->quota - req;
268 else if (req > counts->quota)
269 counts->over_quota++;
274 static int msi_quota_for_device(struct pci_dev *dev, int request)
276 struct device_node *pe_dn;
277 struct msi_counts counts;
280 pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
283 pe_dn = find_pe_total_msi(dev, &total);
285 pe_dn = find_pe_dn(dev, &total);
288 pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
292 pr_debug("rtas_msi: found PE %pOF\n", pe_dn);
294 memset(&counts, 0, sizeof(struct msi_counts));
296 /* Work out how many devices we have below this PE */
297 pci_traverse_device_nodes(pe_dn, count_non_bridge_devices, &counts);
299 if (counts.num_devices == 0) {
300 pr_err("rtas_msi: found 0 devices under PE for %s\n",
305 counts.quota = total / counts.num_devices;
306 if (request <= counts.quota)
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);
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;
318 /* Divide any spare by the number of over-quota requestors */
319 if (counts.over_quota)
320 counts.quota += counts.spare / counts.over_quota;
322 /* And finally clamp the request to the possibly adjusted quota */
323 request = min(counts.quota, request);
325 pr_debug("rtas_msi: request clamped to quota %d\n", request);
332 static int check_msix_entries(struct pci_dev *pdev)
334 struct msi_desc *entry;
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. */
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");
353 static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
355 u32 addr_hi, addr_lo;
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.
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);
370 static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
373 int hwirq, virq, i, quota, rc;
374 struct msi_desc *entry;
377 int use_32bit_msi_hack = 0;
379 if (type == PCI_CAP_ID_MSIX)
380 rc = check_req_msix(pdev, nvec);
382 rc = check_req_msi(pdev, nvec);
387 quota = msi_quota_for_device(pdev, nvec);
389 if (quota && quota < nvec)
392 if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
396 * Firmware currently refuse any non power of two allocation
397 * so we round up if the quota will allow it.
399 if (type == PCI_CAP_ID_MSIX) {
400 int m = roundup_pow_of_two(nvec);
401 quota = msi_quota_for_device(pdev, m);
407 pdn = pci_get_pdn(pdev);
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
415 if (type == PCI_CAP_ID_MSI) {
416 if (pdev->no_64bit_msi) {
417 rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
420 * We only want to run the 32 bit MSI hack below if
421 * the max bus speed is Gen2 speed
423 if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
426 use_32bit_msi_hack = 1;
432 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
435 pr_debug("rtas_msi: trying the old firmware call.\n");
436 rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
439 if (use_32bit_msi_hack && rc > 0)
440 rtas_hack_32bit_msi_gen2(pdev);
442 rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
445 if (nvec != nvec_in) {
449 pr_debug("rtas_msi: rtas_change_msi() failed\n");
454 for_each_pci_msi_entry(entry, pdev) {
455 hwirq = rtas_query_irq_number(pdn, i++);
457 pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
461 virq = irq_create_mapping(NULL, hwirq);
464 pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
468 dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
469 irq_set_msi_desc(virq, entry);
471 /* Read config space back so we can restore after reset */
472 __pci_read_msi_msg(entry, &msg);
479 static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
481 /* No LSI -> leave MSIs (if any) configured */
483 dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
487 /* No MSI -> MSIs can't have been assigned by fw, leave LSI */
488 if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
489 dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
493 dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
494 rtas_disable_msi(pdev);
497 static int rtas_msi_init(void)
499 struct pci_controller *phb;
501 query_token = rtas_token("ibm,query-interrupt-source-number");
502 change_token = rtas_token("ibm,change-msi");
504 if ((query_token == RTAS_UNKNOWN_SERVICE) ||
505 (change_token == RTAS_UNKNOWN_SERVICE)) {
506 pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
510 pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
512 WARN_ON(pseries_pci_controller_ops.setup_msi_irqs);
513 pseries_pci_controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
514 pseries_pci_controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
516 list_for_each_entry(phb, &hose_list, list_node) {
517 WARN_ON(phb->controller_ops.setup_msi_irqs);
518 phb->controller_ops.setup_msi_irqs = rtas_setup_msi_irqs;
519 phb->controller_ops.teardown_msi_irqs = rtas_teardown_msi_irqs;
522 WARN_ON(ppc_md.pci_irq_fixup);
523 ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
527 machine_arch_initcall(pseries, rtas_msi_init);