1/*
2 * Setup routines for AGP 3.5 compliant bridges.
3 */
4
5#include <linux/list.h>
6#include <linux/pci.h>
7#include <linux/agp_backend.h>
8#include <linux/module.h>
9#include <linux/slab.h>
10
11#include "agp.h"
12
13/* Generic AGP 3.5 enabling routines */
14
15struct agp_3_5_dev {
16	struct list_head list;
17	u8 capndx;
18	u32 maxbw;
19	struct pci_dev *dev;
20};
21
22static void agp_3_5_dev_list_insert(struct list_head *head, struct list_head *new)
23{
24	struct agp_3_5_dev *cur, *n = list_entry(new, struct agp_3_5_dev, list);
25	struct list_head *pos;
26
27	list_for_each(pos, head) {
28		cur = list_entry(pos, struct agp_3_5_dev, list);
29		if (cur->maxbw > n->maxbw)
30			break;
31	}
32	list_add_tail(new, pos);
33}
34
35static void agp_3_5_dev_list_sort(struct agp_3_5_dev *list, unsigned int ndevs)
36{
37	struct agp_3_5_dev *cur;
38	struct pci_dev *dev;
39	struct list_head *pos, *tmp, *head = &list->list, *start = head->next;
40	u32 nistat;
41
42	INIT_LIST_HEAD(head);
43
44	for (pos=start; pos!=head; ) {
45		cur = list_entry(pos, struct agp_3_5_dev, list);
46		dev = cur->dev;
47
48		pci_read_config_dword(dev, cur->capndx+AGPNISTAT, &nistat);
49		cur->maxbw = (nistat >> 16) & 0xff;
50
51		tmp = pos;
52		pos = pos->next;
53		agp_3_5_dev_list_insert(head, tmp);
54	}
55}
56
57/*
58 * Initialize all isochronous transfer parameters for an AGP 3.0
59 * node (i.e. a host bridge in combination with the adapters
60 * lying behind it...)
61 */
62
63static int agp_3_5_isochronous_node_enable(struct agp_bridge_data *bridge,
64		struct agp_3_5_dev *dev_list, unsigned int ndevs)
65{
66	/*
67	 * Convenience structure to make the calculations clearer
68	 * here.  The field names come straight from the AGP 3.0 spec.
69	 */
70	struct isoch_data {
71		u32 maxbw;
72		u32 n;
73		u32 y;
74		u32 l;
75		u32 rq;
76		struct agp_3_5_dev *dev;
77	};
78
79	struct pci_dev *td = bridge->dev, *dev;
80	struct list_head *head = &dev_list->list, *pos;
81	struct agp_3_5_dev *cur;
82	struct isoch_data *master, target;
83	unsigned int cdev = 0;
84	u32 mnistat, tnistat, tstatus, mcmd;
85	u16 tnicmd, mnicmd;
86	u8 mcapndx;
87	u32 tot_bw = 0, tot_n = 0, tot_rq = 0, y_max, rq_isoch, rq_async;
88	u32 step, rem, rem_isoch, rem_async;
89	int ret = 0;
90
91	/*
92	 * We'll work with an array of isoch_data's (one for each
93	 * device in dev_list) throughout this function.
94	 */
95	if ((master = kmalloc(ndevs * sizeof(*master), GFP_KERNEL)) == NULL) {
96		ret = -ENOMEM;
97		goto get_out;
98	}
99
100	/*
101	 * Sort the device list by maxbw.  We need to do this because the
102	 * spec suggests that the devices with the smallest requirements
103	 * have their resources allocated first, with all remaining resources
104	 * falling to the device with the largest requirement.
105	 *
106	 * We don't exactly do this, we divide target resources by ndevs
107	 * and split them amongst the AGP 3.0 devices.  The remainder of such
108	 * division operations are dropped on the last device, sort of like
109	 * the spec mentions it should be done.
110	 *
111	 * We can't do this sort when we initially construct the dev_list
112	 * because we don't know until this function whether isochronous
113	 * transfers are enabled and consequently whether maxbw will mean
114	 * anything.
115	 */
116	agp_3_5_dev_list_sort(dev_list, ndevs);
117
118	pci_read_config_dword(td, bridge->capndx+AGPNISTAT, &tnistat);
119	pci_read_config_dword(td, bridge->capndx+AGPSTAT, &tstatus);
120
121	/* Extract power-on defaults from the target */
122	target.maxbw = (tnistat >> 16) & 0xff;
123	target.n     = (tnistat >> 8)  & 0xff;
124	target.y     = (tnistat >> 6)  & 0x3;
125	target.l     = (tnistat >> 3)  & 0x7;
126	target.rq    = (tstatus >> 24) & 0xff;
127
128	y_max = target.y;
129
130	/*
131	 * Extract power-on defaults for each device in dev_list.  Along
132	 * the way, calculate the total isochronous bandwidth required
133	 * by these devices and the largest requested payload size.
134	 */
135	list_for_each(pos, head) {
136		cur = list_entry(pos, struct agp_3_5_dev, list);
137		dev = cur->dev;
138
139		mcapndx = cur->capndx;
140
141		pci_read_config_dword(dev, cur->capndx+AGPNISTAT, &mnistat);
142
143		master[cdev].maxbw = (mnistat >> 16) & 0xff;
144		master[cdev].n     = (mnistat >> 8)  & 0xff;
145		master[cdev].y     = (mnistat >> 6)  & 0x3;
146		master[cdev].dev   = cur;
147
148		tot_bw += master[cdev].maxbw;
149		y_max = max(y_max, master[cdev].y);
150
151		cdev++;
152	}
153
154	/* Check if this configuration has any chance of working */
155	if (tot_bw > target.maxbw) {
156		dev_err(&td->dev, "isochronous bandwidth required "
157			"by AGP 3.0 devices exceeds that which is supported by "
158			"the AGP 3.0 bridge!\n");
159		ret = -ENODEV;
160		goto free_and_exit;
161	}
162
163	target.y = y_max;
164
165	/*
166	 * Write the calculated payload size into the target's NICMD
167	 * register.  Doing this directly effects the ISOCH_N value
168	 * in the target's NISTAT register, so we need to do this now
169	 * to get an accurate value for ISOCH_N later.
170	 */
171	pci_read_config_word(td, bridge->capndx+AGPNICMD, &tnicmd);
172	tnicmd &= ~(0x3 << 6);
173	tnicmd |= target.y << 6;
174	pci_write_config_word(td, bridge->capndx+AGPNICMD, tnicmd);
175
176	/* Reread the target's ISOCH_N */
177	pci_read_config_dword(td, bridge->capndx+AGPNISTAT, &tnistat);
178	target.n = (tnistat >> 8) & 0xff;
179
180	/* Calculate the minimum ISOCH_N needed by each master */
181	for (cdev=0; cdev<ndevs; cdev++) {
182		master[cdev].y = target.y;
183		master[cdev].n = master[cdev].maxbw / (master[cdev].y + 1);
184
185		tot_n += master[cdev].n;
186	}
187
188	/* Exit if the minimal ISOCH_N allocation among the masters is more
189	 * than the target can handle. */
190	if (tot_n > target.n) {
191		dev_err(&td->dev, "number of isochronous "
192			"transactions per period required by AGP 3.0 devices "
193			"exceeds that which is supported by the AGP 3.0 "
194			"bridge!\n");
195		ret = -ENODEV;
196		goto free_and_exit;
197	}
198
199	/* Calculate left over ISOCH_N capability in the target.  We'll give
200	 * this to the hungriest device (as per the spec) */
201	rem  = target.n - tot_n;
202
203	/*
204	 * Calculate the minimum isochronous RQ depth needed by each master.
205	 * Along the way, distribute the extra ISOCH_N capability calculated
206	 * above.
207	 */
208	for (cdev=0; cdev<ndevs; cdev++) {
209		/*
210		 * This is a little subtle.  If ISOCH_Y > 64B, then ISOCH_Y
211		 * byte isochronous writes will be broken into 64B pieces.
212		 * This means we need to budget more RQ depth to account for
213		 * these kind of writes (each isochronous write is actually
214		 * many writes on the AGP bus).
215		 */
216		master[cdev].rq = master[cdev].n;
217		if (master[cdev].y > 0x1)
218			master[cdev].rq *= (1 << (master[cdev].y - 1));
219
220		tot_rq += master[cdev].rq;
221	}
222	master[ndevs-1].n += rem;
223
224	/* Figure the number of isochronous and asynchronous RQ slots the
225	 * target is providing. */
226	rq_isoch = (target.y > 0x1) ? target.n * (1 << (target.y - 1)) : target.n;
227	rq_async = target.rq - rq_isoch;
228
229	/* Exit if the minimal RQ needs of the masters exceeds what the target
230	 * can provide. */
231	if (tot_rq > rq_isoch) {
232		dev_err(&td->dev, "number of request queue slots "
233			"required by the isochronous bandwidth requested by "
234			"AGP 3.0 devices exceeds the number provided by the "
235			"AGP 3.0 bridge!\n");
236		ret = -ENODEV;
237		goto free_and_exit;
238	}
239
240	/* Calculate asynchronous RQ capability in the target (per master) as
241	 * well as the total number of leftover isochronous RQ slots. */
242	step      = rq_async / ndevs;
243	rem_async = step + (rq_async % ndevs);
244	rem_isoch = rq_isoch - tot_rq;
245
246	/* Distribute the extra RQ slots calculated above and write our
247	 * isochronous settings out to the actual devices. */
248	for (cdev=0; cdev<ndevs; cdev++) {
249		cur = master[cdev].dev;
250		dev = cur->dev;
251
252		mcapndx = cur->capndx;
253
254		master[cdev].rq += (cdev == ndevs - 1)
255		              ? (rem_async + rem_isoch) : step;
256
257		pci_read_config_word(dev, cur->capndx+AGPNICMD, &mnicmd);
258		pci_read_config_dword(dev, cur->capndx+AGPCMD, &mcmd);
259
260		mnicmd &= ~(0xff << 8);
261		mnicmd &= ~(0x3  << 6);
262		mcmd   &= ~(0xff << 24);
263
264		mnicmd |= master[cdev].n  << 8;
265		mnicmd |= master[cdev].y  << 6;
266		mcmd   |= master[cdev].rq << 24;
267
268		pci_write_config_dword(dev, cur->capndx+AGPCMD, mcmd);
269		pci_write_config_word(dev, cur->capndx+AGPNICMD, mnicmd);
270	}
271
272free_and_exit:
273	kfree(master);
274
275get_out:
276	return ret;
277}
278
279/*
280 * This function basically allocates request queue slots among the
281 * AGP 3.0 systems in nonisochronous nodes.  The algorithm is
282 * pretty stupid, divide the total number of RQ slots provided by the
283 * target by ndevs.  Distribute this many slots to each AGP 3.0 device,
284 * giving any left over slots to the last device in dev_list.
285 */
286static void agp_3_5_nonisochronous_node_enable(struct agp_bridge_data *bridge,
287		struct agp_3_5_dev *dev_list, unsigned int ndevs)
288{
289	struct agp_3_5_dev *cur;
290	struct list_head *head = &dev_list->list, *pos;
291	u32 tstatus, mcmd;
292	u32 trq, mrq, rem;
293	unsigned int cdev = 0;
294
295	pci_read_config_dword(bridge->dev, bridge->capndx+AGPSTAT, &tstatus);
296
297	trq = (tstatus >> 24) & 0xff;
298	mrq = trq / ndevs;
299
300	rem = mrq + (trq % ndevs);
301
302	for (pos=head->next; cdev<ndevs; cdev++, pos=pos->next) {
303		cur = list_entry(pos, struct agp_3_5_dev, list);
304
305		pci_read_config_dword(cur->dev, cur->capndx+AGPCMD, &mcmd);
306		mcmd &= ~(0xff << 24);
307		mcmd |= ((cdev == ndevs - 1) ? rem : mrq) << 24;
308		pci_write_config_dword(cur->dev, cur->capndx+AGPCMD, mcmd);
309	}
310}
311
312/*
313 * Fully configure and enable an AGP 3.0 host bridge and all the devices
314 * lying behind it.
315 */
316int agp_3_5_enable(struct agp_bridge_data *bridge)
317{
318	struct pci_dev *td = bridge->dev, *dev = NULL;
319	u8 mcapndx;
320	u32 isoch, arqsz;
321	u32 tstatus, mstatus, ncapid;
322	u32 mmajor;
323	u16 mpstat;
324	struct agp_3_5_dev *dev_list, *cur;
325	struct list_head *head, *pos;
326	unsigned int ndevs = 0;
327	int ret = 0;
328
329	/* Extract some power-on defaults from the target */
330	pci_read_config_dword(td, bridge->capndx+AGPSTAT, &tstatus);
331	isoch     = (tstatus >> 17) & 0x1;
332	if (isoch == 0)	/* isoch xfers not available, bail out. */
333		return -ENODEV;
334
335	arqsz     = (tstatus >> 13) & 0x7;
336
337	/*
338	 * Allocate a head for our AGP 3.5 device list
339	 * (multiple AGP v3 devices are allowed behind a single bridge).
340	 */
341	if ((dev_list = kmalloc(sizeof(*dev_list), GFP_KERNEL)) == NULL) {
342		ret = -ENOMEM;
343		goto get_out;
344	}
345	head = &dev_list->list;
346	INIT_LIST_HEAD(head);
347
348	/* Find all AGP devices, and add them to dev_list. */
349	for_each_pci_dev(dev) {
350		mcapndx = pci_find_capability(dev, PCI_CAP_ID_AGP);
351		if (mcapndx == 0)
352			continue;
353
354		switch ((dev->class >>8) & 0xff00) {
355			case 0x0600:    /* Bridge */
356				/* Skip bridges. We should call this function for each one. */
357				continue;
358
359			case 0x0001:    /* Unclassified device */
360				/* Don't know what this is, but log it for investigation. */
361				if (mcapndx != 0) {
362					dev_info(&td->dev, "wacky, found unclassified AGP device %s [%04x/%04x]\n",
363						 pci_name(dev),
364						 dev->vendor, dev->device);
365				}
366				continue;
367
368			case 0x0300:    /* Display controller */
369			case 0x0400:    /* Multimedia controller */
370				if ((cur = kmalloc(sizeof(*cur), GFP_KERNEL)) == NULL) {
371					ret = -ENOMEM;
372					goto free_and_exit;
373				}
374				cur->dev = dev;
375
376				pos = &cur->list;
377				list_add(pos, head);
378				ndevs++;
379				continue;
380
381			default:
382				continue;
383		}
384	}
385
386	/*
387	 * Take an initial pass through the devices lying behind our host
388	 * bridge.  Make sure each one is actually an AGP 3.0 device, otherwise
389	 * exit with an error message.  Along the way store the AGP 3.0
390	 * cap_ptr for each device
391	 */
392	list_for_each(pos, head) {
393		cur = list_entry(pos, struct agp_3_5_dev, list);
394		dev = cur->dev;
395
396		pci_read_config_word(dev, PCI_STATUS, &mpstat);
397		if ((mpstat & PCI_STATUS_CAP_LIST) == 0)
398			continue;
399
400		pci_read_config_byte(dev, PCI_CAPABILITY_LIST, &mcapndx);
401		if (mcapndx != 0) {
402			do {
403				pci_read_config_dword(dev, mcapndx, &ncapid);
404				if ((ncapid & 0xff) != 2)
405					mcapndx = (ncapid >> 8) & 0xff;
406			}
407			while (((ncapid & 0xff) != 2) && (mcapndx != 0));
408		}
409
410		if (mcapndx == 0) {
411			dev_err(&td->dev, "woah!  Non-AGP device %s on "
412				"secondary bus of AGP 3.5 bridge!\n",
413				pci_name(dev));
414			ret = -ENODEV;
415			goto free_and_exit;
416		}
417
418		mmajor = (ncapid >> AGP_MAJOR_VERSION_SHIFT) & 0xf;
419		if (mmajor < 3) {
420			dev_err(&td->dev, "woah!  AGP 2.0 device %s on "
421				"secondary bus of AGP 3.5 bridge operating "
422				"with AGP 3.0 electricals!\n", pci_name(dev));
423			ret = -ENODEV;
424			goto free_and_exit;
425		}
426
427		cur->capndx = mcapndx;
428
429		pci_read_config_dword(dev, cur->capndx+AGPSTAT, &mstatus);
430
431		if (((mstatus >> 3) & 0x1) == 0) {
432			dev_err(&td->dev, "woah!  AGP 3.x device %s not "
433				"operating in AGP 3.x mode on secondary bus "
434				"of AGP 3.5 bridge operating with AGP 3.0 "
435				"electricals!\n", pci_name(dev));
436			ret = -ENODEV;
437			goto free_and_exit;
438		}
439	}
440
441	/*
442	 * Call functions to divide target resources amongst the AGP 3.0
443	 * masters.  This process is dramatically different depending on
444	 * whether isochronous transfers are supported.
445	 */
446	if (isoch) {
447		ret = agp_3_5_isochronous_node_enable(bridge, dev_list, ndevs);
448		if (ret) {
449			dev_info(&td->dev, "something bad happened setting "
450				 "up isochronous xfers; falling back to "
451				 "non-isochronous xfer mode\n");
452		} else {
453			goto free_and_exit;
454		}
455	}
456	agp_3_5_nonisochronous_node_enable(bridge, dev_list, ndevs);
457
458free_and_exit:
459	/* Be sure to free the dev_list */
460	for (pos=head->next; pos!=head; ) {
461		cur = list_entry(pos, struct agp_3_5_dev, list);
462
463		pos = pos->next;
464		kfree(cur);
465	}
466	kfree(dev_list);
467
468get_out:
469	return ret;
470}
471