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#include <asm/machdep.h>
20
21static int query_token, change_token;
22
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
29
30/* RTAS Helpers */
31
32static int rtas_change_msi(struct pci_dn *pdn, u32 func, u32 num_irqs)
33{
34	u32 addr, seq_num, rtas_ret[3];
35	unsigned long buid;
36	int rc;
37
38	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
39	buid = pdn->phb->buid;
40
41	seq_num = 1;
42	do {
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);
48		else
49			rc = rtas_call(change_token, 6, 3, rtas_ret, addr,
50					BUID_HI(buid), BUID_LO(buid),
51					func, num_irqs, seq_num);
52
53		seq_num = rtas_ret[1];
54	} while (rtas_busy_delay(rc));
55
56	/*
57	 * If the RTAS call succeeded, return the number of irqs allocated.
58	 * If not, make sure we return a negative error code.
59	 */
60	if (rc == 0)
61		rc = rtas_ret[0];
62	else if (rc > 0)
63		rc = -rc;
64
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);
67
68	return rc;
69}
70
71static void rtas_disable_msi(struct pci_dev *pdev)
72{
73	struct pci_dn *pdn;
74
75	pdn = pci_get_pdn(pdev);
76	if (!pdn)
77		return;
78
79	/*
80	 * disabling MSI with the explicit interface also disables MSI-X
81	 */
82	if (rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, 0) != 0) {
83		/*
84		 * may have failed because explicit interface is not
85		 * present
86		 */
87		if (rtas_change_msi(pdn, RTAS_CHANGE_FN, 0) != 0) {
88			pr_debug("rtas_msi: Setting MSIs to 0 failed!\n");
89		}
90	}
91}
92
93static int rtas_query_irq_number(struct pci_dn *pdn, int offset)
94{
95	u32 addr, rtas_ret[2];
96	unsigned long buid;
97	int rc;
98
99	addr = rtas_config_addr(pdn->busno, pdn->devfn, 0);
100	buid = pdn->phb->buid;
101
102	do {
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));
106
107	if (rc) {
108		pr_debug("rtas_msi: error (%d) querying source number\n", rc);
109		return rc;
110	}
111
112	return rtas_ret[0];
113}
114
115static void rtas_teardown_msi_irqs(struct pci_dev *pdev)
116{
117	struct msi_desc *entry;
118
119	list_for_each_entry(entry, &pdev->msi_list, list) {
120		if (entry->irq == NO_IRQ)
121			continue;
122
123		irq_set_msi_desc(entry->irq, NULL);
124		irq_dispose_mapping(entry->irq);
125	}
126
127	rtas_disable_msi(pdev);
128}
129
130static int check_req(struct pci_dev *pdev, int nvec, char *prop_name)
131{
132	struct device_node *dn;
133	struct pci_dn *pdn;
134	const __be32 *p;
135	u32 req_msi;
136
137	pdn = pci_get_pdn(pdev);
138	if (!pdn)
139		return -ENODEV;
140
141	dn = pdn->node;
142
143	p = of_get_property(dn, prop_name, NULL);
144	if (!p) {
145		pr_debug("rtas_msi: No %s on %s\n", prop_name, dn->full_name);
146		return -ENOENT;
147	}
148
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);
152
153		if (req_msi == 0) /* Be paranoid */
154			return -ENOSPC;
155
156		return req_msi;
157	}
158
159	return 0;
160}
161
162static int check_req_msi(struct pci_dev *pdev, int nvec)
163{
164	return check_req(pdev, nvec, "ibm,req#msi");
165}
166
167static int check_req_msix(struct pci_dev *pdev, int nvec)
168{
169	return check_req(pdev, nvec, "ibm,req#msi-x");
170}
171
172/* Quota calculation */
173
174static struct device_node *find_pe_total_msi(struct pci_dev *dev, int *total)
175{
176	struct device_node *dn;
177	const __be32 *p;
178
179	dn = of_node_get(pci_device_to_OF_node(dev));
180	while (dn) {
181		p = of_get_property(dn, "ibm,pe-total-#msi", NULL);
182		if (p) {
183			pr_debug("rtas_msi: found prop on dn %s\n",
184				dn->full_name);
185			*total = be32_to_cpup(p);
186			return dn;
187		}
188
189		dn = of_get_next_parent(dn);
190	}
191
192	return NULL;
193}
194
195static struct device_node *find_pe_dn(struct pci_dev *dev, int *total)
196{
197	struct device_node *dn;
198	struct pci_dn *pdn;
199	struct eeh_dev *edev;
200
201	/* Found our PE and assume 8 at that point. */
202
203	dn = pci_device_to_OF_node(dev);
204	if (!dn)
205		return NULL;
206
207	/* Get the top level device in the PE */
208	edev = pdn_to_eeh_dev(PCI_DN(dn));
209	if (edev->pe)
210		edev = list_first_entry(&edev->pe->edevs, struct eeh_dev, list);
211	pdn = eeh_dev_to_pdn(edev);
212	dn = pdn ? pdn->node : NULL;
213	if (!dn)
214		return NULL;
215
216	/* We actually want the parent */
217	dn = of_get_parent(dn);
218	if (!dn)
219		return NULL;
220
221	/* Hardcode of 8 for old firmwares */
222	*total = 8;
223	pr_debug("rtas_msi: using PE dn %s\n", dn->full_name);
224
225	return dn;
226}
227
228struct msi_counts {
229	struct device_node *requestor;
230	int num_devices;
231	int request;
232	int quota;
233	int spare;
234	int over_quota;
235};
236
237static void *count_non_bridge_devices(struct device_node *dn, void *data)
238{
239	struct msi_counts *counts = data;
240	const __be32 *p;
241	u32 class;
242
243	pr_debug("rtas_msi: counting %s\n", dn->full_name);
244
245	p = of_get_property(dn, "class-code", NULL);
246	class = p ? be32_to_cpup(p) : 0;
247
248	if ((class >> 8) != PCI_CLASS_BRIDGE_PCI)
249		counts->num_devices++;
250
251	return NULL;
252}
253
254static void *count_spare_msis(struct device_node *dn, void *data)
255{
256	struct msi_counts *counts = data;
257	const __be32 *p;
258	int req;
259
260	if (dn == counts->requestor)
261		req = counts->request;
262	else {
263		/* We don't know if a driver will try to use MSI or MSI-X,
264		 * so we just have to punt and use the larger of the two. */
265		req = 0;
266		p = of_get_property(dn, "ibm,req#msi", NULL);
267		if (p)
268			req = be32_to_cpup(p);
269
270		p = of_get_property(dn, "ibm,req#msi-x", NULL);
271		if (p)
272			req = max(req, (int)be32_to_cpup(p));
273	}
274
275	if (req < counts->quota)
276		counts->spare += counts->quota - req;
277	else if (req > counts->quota)
278		counts->over_quota++;
279
280	return NULL;
281}
282
283static int msi_quota_for_device(struct pci_dev *dev, int request)
284{
285	struct device_node *pe_dn;
286	struct msi_counts counts;
287	int total;
288
289	pr_debug("rtas_msi: calc quota for %s, request %d\n", pci_name(dev),
290		  request);
291
292	pe_dn = find_pe_total_msi(dev, &total);
293	if (!pe_dn)
294		pe_dn = find_pe_dn(dev, &total);
295
296	if (!pe_dn) {
297		pr_err("rtas_msi: couldn't find PE for %s\n", pci_name(dev));
298		goto out;
299	}
300
301	pr_debug("rtas_msi: found PE %s\n", pe_dn->full_name);
302
303	memset(&counts, 0, sizeof(struct msi_counts));
304
305	/* Work out how many devices we have below this PE */
306	traverse_pci_devices(pe_dn, count_non_bridge_devices, &counts);
307
308	if (counts.num_devices == 0) {
309		pr_err("rtas_msi: found 0 devices under PE for %s\n",
310			pci_name(dev));
311		goto out;
312	}
313
314	counts.quota = total / counts.num_devices;
315	if (request <= counts.quota)
316		goto out;
317
318	/* else, we have some more calculating to do */
319	counts.requestor = pci_device_to_OF_node(dev);
320	counts.request = request;
321	traverse_pci_devices(pe_dn, count_spare_msis, &counts);
322
323	/* If the quota isn't an integer multiple of the total, we can
324	 * use the remainder as spare MSIs for anyone that wants them. */
325	counts.spare += total % counts.num_devices;
326
327	/* Divide any spare by the number of over-quota requestors */
328	if (counts.over_quota)
329		counts.quota += counts.spare / counts.over_quota;
330
331	/* And finally clamp the request to the possibly adjusted quota */
332	request = min(counts.quota, request);
333
334	pr_debug("rtas_msi: request clamped to quota %d\n", request);
335out:
336	of_node_put(pe_dn);
337
338	return request;
339}
340
341static int check_msix_entries(struct pci_dev *pdev)
342{
343	struct msi_desc *entry;
344	int expected;
345
346	/* There's no way for us to express to firmware that we want
347	 * a discontiguous, or non-zero based, range of MSI-X entries.
348	 * So we must reject such requests. */
349
350	expected = 0;
351	list_for_each_entry(entry, &pdev->msi_list, list) {
352		if (entry->msi_attrib.entry_nr != expected) {
353			pr_debug("rtas_msi: bad MSI-X entries.\n");
354			return -EINVAL;
355		}
356		expected++;
357	}
358
359	return 0;
360}
361
362static void rtas_hack_32bit_msi_gen2(struct pci_dev *pdev)
363{
364	u32 addr_hi, addr_lo;
365
366	/*
367	 * We should only get in here for IODA1 configs. This is based on the
368	 * fact that we using RTAS for MSIs, we don't have the 32 bit MSI RTAS
369	 * support, and we are in a PCIe Gen2 slot.
370	 */
371	dev_info(&pdev->dev,
372		 "rtas_msi: No 32 bit MSI firmware support, forcing 32 bit MSI\n");
373	pci_read_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, &addr_hi);
374	addr_lo = 0xffff0000 | ((addr_hi >> (48 - 32)) << 4);
375	pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_LO, addr_lo);
376	pci_write_config_dword(pdev, pdev->msi_cap + PCI_MSI_ADDRESS_HI, 0);
377}
378
379static int rtas_setup_msi_irqs(struct pci_dev *pdev, int nvec_in, int type)
380{
381	struct pci_dn *pdn;
382	int hwirq, virq, i, quota, rc;
383	struct msi_desc *entry;
384	struct msi_msg msg;
385	int nvec = nvec_in;
386	int use_32bit_msi_hack = 0;
387
388	if (type == PCI_CAP_ID_MSIX)
389		rc = check_req_msix(pdev, nvec);
390	else
391		rc = check_req_msi(pdev, nvec);
392
393	if (rc)
394		return rc;
395
396	quota = msi_quota_for_device(pdev, nvec);
397
398	if (quota && quota < nvec)
399		return quota;
400
401	if (type == PCI_CAP_ID_MSIX && check_msix_entries(pdev))
402		return -EINVAL;
403
404	/*
405	 * Firmware currently refuse any non power of two allocation
406	 * so we round up if the quota will allow it.
407	 */
408	if (type == PCI_CAP_ID_MSIX) {
409		int m = roundup_pow_of_two(nvec);
410		quota = msi_quota_for_device(pdev, m);
411
412		if (quota >= m)
413			nvec = m;
414	}
415
416	pdn = pci_get_pdn(pdev);
417
418	/*
419	 * Try the new more explicit firmware interface, if that fails fall
420	 * back to the old interface. The old interface is known to never
421	 * return MSI-Xs.
422	 */
423again:
424	if (type == PCI_CAP_ID_MSI) {
425		if (pdev->no_64bit_msi) {
426			rc = rtas_change_msi(pdn, RTAS_CHANGE_32MSI_FN, nvec);
427			if (rc < 0) {
428				/*
429				 * We only want to run the 32 bit MSI hack below if
430				 * the max bus speed is Gen2 speed
431				 */
432				if (pdev->bus->max_bus_speed != PCIE_SPEED_5_0GT)
433					return rc;
434
435				use_32bit_msi_hack = 1;
436			}
437		} else
438			rc = -1;
439
440		if (rc < 0)
441			rc = rtas_change_msi(pdn, RTAS_CHANGE_MSI_FN, nvec);
442
443		if (rc < 0) {
444			pr_debug("rtas_msi: trying the old firmware call.\n");
445			rc = rtas_change_msi(pdn, RTAS_CHANGE_FN, nvec);
446		}
447
448		if (use_32bit_msi_hack && rc > 0)
449			rtas_hack_32bit_msi_gen2(pdev);
450	} else
451		rc = rtas_change_msi(pdn, RTAS_CHANGE_MSIX_FN, nvec);
452
453	if (rc != nvec) {
454		if (nvec != nvec_in) {
455			nvec = nvec_in;
456			goto again;
457		}
458		pr_debug("rtas_msi: rtas_change_msi() failed\n");
459		return rc;
460	}
461
462	i = 0;
463	list_for_each_entry(entry, &pdev->msi_list, list) {
464		hwirq = rtas_query_irq_number(pdn, i++);
465		if (hwirq < 0) {
466			pr_debug("rtas_msi: error (%d) getting hwirq\n", rc);
467			return hwirq;
468		}
469
470		virq = irq_create_mapping(NULL, hwirq);
471
472		if (virq == NO_IRQ) {
473			pr_debug("rtas_msi: Failed mapping hwirq %d\n", hwirq);
474			return -ENOSPC;
475		}
476
477		dev_dbg(&pdev->dev, "rtas_msi: allocated virq %d\n", virq);
478		irq_set_msi_desc(virq, entry);
479
480		/* Read config space back so we can restore after reset */
481		__pci_read_msi_msg(entry, &msg);
482		entry->msg = msg;
483	}
484
485	return 0;
486}
487
488static void rtas_msi_pci_irq_fixup(struct pci_dev *pdev)
489{
490	/* No LSI -> leave MSIs (if any) configured */
491	if (pdev->irq == NO_IRQ) {
492		dev_dbg(&pdev->dev, "rtas_msi: no LSI, nothing to do.\n");
493		return;
494	}
495
496	/* No MSI -> MSIs can't have been assigned by fw, leave LSI */
497	if (check_req_msi(pdev, 1) && check_req_msix(pdev, 1)) {
498		dev_dbg(&pdev->dev, "rtas_msi: no req#msi/x, nothing to do.\n");
499		return;
500	}
501
502	dev_dbg(&pdev->dev, "rtas_msi: disabling existing MSI.\n");
503	rtas_disable_msi(pdev);
504}
505
506static int rtas_msi_init(void)
507{
508	query_token  = rtas_token("ibm,query-interrupt-source-number");
509	change_token = rtas_token("ibm,change-msi");
510
511	if ((query_token == RTAS_UNKNOWN_SERVICE) ||
512			(change_token == RTAS_UNKNOWN_SERVICE)) {
513		pr_debug("rtas_msi: no RTAS tokens, no MSI support.\n");
514		return -1;
515	}
516
517	pr_debug("rtas_msi: Registering RTAS MSI callbacks.\n");
518
519	WARN_ON(ppc_md.setup_msi_irqs);
520	ppc_md.setup_msi_irqs = rtas_setup_msi_irqs;
521	ppc_md.teardown_msi_irqs = rtas_teardown_msi_irqs;
522
523	WARN_ON(ppc_md.pci_irq_fixup);
524	ppc_md.pci_irq_fixup = rtas_msi_pci_irq_fixup;
525
526	return 0;
527}
528machine_arch_initcall(pseries, rtas_msi_init);
529