]>
Commit | Line | Data |
---|---|---|
7bb206e3 | 1 | /* |
87427da5 | 2 | * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved. |
7bb206e3 BS |
3 | * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved. |
4 | * | |
5 | * This software is available to you under a choice of one of two | |
6 | * licenses. You may choose to be licensed under the terms of the GNU | |
7 | * General Public License (GPL) Version 2, available from the file | |
8 | * COPYING in the main directory of this source tree, or the | |
9 | * OpenIB.org BSD license below: | |
10 | * | |
11 | * Redistribution and use in source and binary forms, with or | |
12 | * without modification, are permitted provided that the following | |
13 | * conditions are met: | |
14 | * | |
15 | * - Redistributions of source code must retain the above | |
16 | * copyright notice, this list of conditions and the following | |
17 | * disclaimer. | |
18 | * | |
19 | * - Redistributions in binary form must reproduce the above | |
20 | * copyright notice, this list of conditions and the following | |
21 | * disclaimer in the documentation and/or other materials | |
22 | * provided with the distribution. | |
23 | * | |
24 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
25 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
26 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
27 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
28 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
29 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
30 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
31 | * SOFTWARE. | |
32 | */ | |
33 | ||
34 | #include <linux/spinlock.h> | |
35 | #include <linux/idr.h> | |
36 | #include <linux/pci.h> | |
9bec3992 | 37 | #include <linux/io.h> |
7bb206e3 BS |
38 | #include <linux/delay.h> |
39 | #include <linux/netdevice.h> | |
40 | #include <linux/vmalloc.h> | |
41 | ||
42 | #include "ipath_kernel.h" | |
b1c1b6a3 | 43 | #include "ipath_verbs.h" |
7bb206e3 BS |
44 | |
45 | static void ipath_update_pio_bufs(struct ipath_devdata *); | |
46 | ||
47 | const char *ipath_get_unit_name(int unit) | |
48 | { | |
49 | static char iname[16]; | |
50 | snprintf(iname, sizeof iname, "infinipath%u", unit); | |
51 | return iname; | |
52 | } | |
53 | ||
759d5768 | 54 | #define DRIVER_LOAD_MSG "QLogic " IPATH_DRV_NAME " loaded: " |
7bb206e3 BS |
55 | #define PFX IPATH_DRV_NAME ": " |
56 | ||
57 | /* | |
58 | * The size has to be longer than this string, so we can append | |
59 | * board/chip information to it in the init code. | |
60 | */ | |
b55f4f06 | 61 | const char ib_ipath_version[] = IPATH_IDSTR "\n"; |
7bb206e3 BS |
62 | |
63 | static struct idr unit_table; | |
64 | DEFINE_SPINLOCK(ipath_devs_lock); | |
65 | LIST_HEAD(ipath_dev_list); | |
66 | ||
0fd41363 | 67 | wait_queue_head_t ipath_state_wait; |
7bb206e3 BS |
68 | |
69 | unsigned ipath_debug = __IPATH_INFO; | |
70 | ||
71 | module_param_named(debug, ipath_debug, uint, S_IWUSR | S_IRUGO); | |
72 | MODULE_PARM_DESC(debug, "mask for debug prints"); | |
73 | EXPORT_SYMBOL_GPL(ipath_debug); | |
74 | ||
826d8010 DO |
75 | unsigned ipath_mtu4096 = 1; /* max 4KB IB mtu by default, if supported */ |
76 | module_param_named(mtu4096, ipath_mtu4096, uint, S_IRUGO); | |
77 | MODULE_PARM_DESC(mtu4096, "enable MTU of 4096 bytes, if supported"); | |
78 | ||
58411d1c JG |
79 | static unsigned ipath_hol_timeout_ms = 13000; |
80 | module_param_named(hol_timeout_ms, ipath_hol_timeout_ms, uint, S_IRUGO); | |
81 | MODULE_PARM_DESC(hol_timeout_ms, | |
82 | "duration of user app suspension after link failure"); | |
83 | ||
72708a0a DO |
84 | unsigned ipath_linkrecovery = 1; |
85 | module_param_named(linkrecovery, ipath_linkrecovery, uint, S_IWUSR | S_IRUGO); | |
86 | MODULE_PARM_DESC(linkrecovery, "enable workaround for link recovery issue"); | |
87 | ||
7bb206e3 | 88 | MODULE_LICENSE("GPL"); |
928e3e4b | 89 | MODULE_AUTHOR("QLogic <support@qlogic.com>"); |
759d5768 | 90 | MODULE_DESCRIPTION("QLogic InfiniPath driver"); |
7bb206e3 BS |
91 | |
92 | const char *ipath_ibcstatus_str[] = { | |
93 | "Disabled", | |
94 | "LinkUp", | |
95 | "PollActive", | |
96 | "PollQuiet", | |
97 | "SleepDelay", | |
98 | "SleepQuiet", | |
99 | "LState6", /* unused */ | |
100 | "LState7", /* unused */ | |
101 | "CfgDebounce", | |
102 | "CfgRcvfCfg", | |
103 | "CfgWaitRmt", | |
104 | "CfgIdle", | |
105 | "RecovRetrain", | |
106 | "LState0xD", /* unused */ | |
107 | "RecovWaitRmt", | |
108 | "RecovIdle", | |
109 | }; | |
110 | ||
7bb206e3 BS |
111 | static void __devexit ipath_remove_one(struct pci_dev *); |
112 | static int __devinit ipath_init_one(struct pci_dev *, | |
113 | const struct pci_device_id *); | |
114 | ||
115 | /* Only needed for registration, nothing else needs this info */ | |
116 | #define PCI_VENDOR_ID_PATHSCALE 0x1fc1 | |
117 | #define PCI_DEVICE_ID_INFINIPATH_HT 0xd | |
118 | #define PCI_DEVICE_ID_INFINIPATH_PE800 0x10 | |
119 | ||
3588423f AJ |
120 | /* Number of seconds before our card status check... */ |
121 | #define STATUS_TIMEOUT 60 | |
122 | ||
7bb206e3 | 123 | static const struct pci_device_id ipath_pci_tbl[] = { |
6f4bb3d8 RD |
124 | { PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_HT) }, |
125 | { PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_PE800) }, | |
126 | { 0, } | |
7bb206e3 BS |
127 | }; |
128 | ||
129 | MODULE_DEVICE_TABLE(pci, ipath_pci_tbl); | |
130 | ||
131 | static struct pci_driver ipath_driver = { | |
132 | .name = IPATH_DRV_NAME, | |
133 | .probe = ipath_init_one, | |
134 | .remove = __devexit_p(ipath_remove_one), | |
135 | .id_table = ipath_pci_tbl, | |
23b9c1ab GKH |
136 | .driver = { |
137 | .groups = ipath_driver_attr_groups, | |
138 | }, | |
7bb206e3 BS |
139 | }; |
140 | ||
3588423f AJ |
141 | static void ipath_check_status(struct work_struct *work) |
142 | { | |
143 | struct ipath_devdata *dd = container_of(work, struct ipath_devdata, | |
144 | status_work.work); | |
145 | ||
146 | /* | |
147 | * If we don't have any interrupts, let the user know and | |
148 | * don't bother checking again. | |
149 | */ | |
150 | if (dd->ipath_int_counter == 0) | |
151 | dev_err(&dd->pcidev->dev, "No interrupts detected.\n"); | |
152 | } | |
7bb206e3 BS |
153 | |
154 | static inline void read_bars(struct ipath_devdata *dd, struct pci_dev *dev, | |
155 | u32 *bar0, u32 *bar1) | |
156 | { | |
157 | int ret; | |
158 | ||
159 | ret = pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, bar0); | |
160 | if (ret) | |
161 | ipath_dev_err(dd, "failed to read bar0 before enable: " | |
162 | "error %d\n", -ret); | |
163 | ||
164 | ret = pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, bar1); | |
165 | if (ret) | |
166 | ipath_dev_err(dd, "failed to read bar1 before enable: " | |
167 | "error %d\n", -ret); | |
168 | ||
169 | ipath_dbg("Read bar0 %x bar1 %x\n", *bar0, *bar1); | |
170 | } | |
171 | ||
172 | static void ipath_free_devdata(struct pci_dev *pdev, | |
173 | struct ipath_devdata *dd) | |
174 | { | |
175 | unsigned long flags; | |
176 | ||
177 | pci_set_drvdata(pdev, NULL); | |
178 | ||
179 | if (dd->ipath_unit != -1) { | |
180 | spin_lock_irqsave(&ipath_devs_lock, flags); | |
181 | idr_remove(&unit_table, dd->ipath_unit); | |
182 | list_del(&dd->ipath_list); | |
183 | spin_unlock_irqrestore(&ipath_devs_lock, flags); | |
184 | } | |
06993ca6 | 185 | vfree(dd); |
7bb206e3 BS |
186 | } |
187 | ||
188 | static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev) | |
189 | { | |
190 | unsigned long flags; | |
191 | struct ipath_devdata *dd; | |
7bb206e3 BS |
192 | int ret; |
193 | ||
194 | if (!idr_pre_get(&unit_table, GFP_KERNEL)) { | |
195 | dd = ERR_PTR(-ENOMEM); | |
196 | goto bail; | |
197 | } | |
198 | ||
06993ca6 | 199 | dd = vmalloc(sizeof(*dd)); |
7bb206e3 BS |
200 | if (!dd) { |
201 | dd = ERR_PTR(-ENOMEM); | |
202 | goto bail; | |
203 | } | |
06993ca6 | 204 | memset(dd, 0, sizeof(*dd)); |
7bb206e3 BS |
205 | dd->ipath_unit = -1; |
206 | ||
207 | spin_lock_irqsave(&ipath_devs_lock, flags); | |
208 | ||
209 | ret = idr_get_new(&unit_table, dd, &dd->ipath_unit); | |
210 | if (ret < 0) { | |
211 | printk(KERN_ERR IPATH_DRV_NAME | |
212 | ": Could not allocate unit ID: error %d\n", -ret); | |
213 | ipath_free_devdata(pdev, dd); | |
214 | dd = ERR_PTR(ret); | |
215 | goto bail_unlock; | |
216 | } | |
217 | ||
218 | dd->pcidev = pdev; | |
219 | pci_set_drvdata(pdev, dd); | |
220 | ||
3588423f AJ |
221 | INIT_DELAYED_WORK(&dd->status_work, ipath_check_status); |
222 | ||
7bb206e3 BS |
223 | list_add(&dd->ipath_list, &ipath_dev_list); |
224 | ||
225 | bail_unlock: | |
226 | spin_unlock_irqrestore(&ipath_devs_lock, flags); | |
227 | ||
228 | bail: | |
229 | return dd; | |
230 | } | |
231 | ||
232 | static inline struct ipath_devdata *__ipath_lookup(int unit) | |
233 | { | |
234 | return idr_find(&unit_table, unit); | |
235 | } | |
236 | ||
237 | struct ipath_devdata *ipath_lookup(int unit) | |
238 | { | |
239 | struct ipath_devdata *dd; | |
240 | unsigned long flags; | |
241 | ||
242 | spin_lock_irqsave(&ipath_devs_lock, flags); | |
243 | dd = __ipath_lookup(unit); | |
244 | spin_unlock_irqrestore(&ipath_devs_lock, flags); | |
245 | ||
246 | return dd; | |
247 | } | |
248 | ||
6ef6aee2 | 249 | int ipath_count_units(int *npresentp, int *nupp, int *maxportsp) |
7bb206e3 BS |
250 | { |
251 | int nunits, npresent, nup; | |
252 | struct ipath_devdata *dd; | |
253 | unsigned long flags; | |
6ef6aee2 | 254 | int maxports; |
7bb206e3 BS |
255 | |
256 | nunits = npresent = nup = maxports = 0; | |
257 | ||
258 | spin_lock_irqsave(&ipath_devs_lock, flags); | |
259 | ||
260 | list_for_each_entry(dd, &ipath_dev_list, ipath_list) { | |
261 | nunits++; | |
262 | if ((dd->ipath_flags & IPATH_PRESENT) && dd->ipath_kregbase) | |
263 | npresent++; | |
264 | if (dd->ipath_lid && | |
265 | !(dd->ipath_flags & (IPATH_DISABLED | IPATH_LINKDOWN | |
266 | | IPATH_LINKUNK))) | |
267 | nup++; | |
268 | if (dd->ipath_cfgports > maxports) | |
269 | maxports = dd->ipath_cfgports; | |
270 | } | |
271 | ||
272 | spin_unlock_irqrestore(&ipath_devs_lock, flags); | |
273 | ||
274 | if (npresentp) | |
275 | *npresentp = npresent; | |
276 | if (nupp) | |
277 | *nupp = nup; | |
278 | if (maxportsp) | |
279 | *maxportsp = maxports; | |
280 | ||
281 | return nunits; | |
282 | } | |
283 | ||
7bb206e3 BS |
284 | /* |
285 | * These next two routines are placeholders in case we don't have per-arch | |
286 | * code for controlling write combining. If explicit control of write | |
287 | * combining is not available, performance will probably be awful. | |
288 | */ | |
289 | ||
290 | int __attribute__((weak)) ipath_enable_wc(struct ipath_devdata *dd) | |
291 | { | |
292 | return -EOPNOTSUPP; | |
293 | } | |
294 | ||
295 | void __attribute__((weak)) ipath_disable_wc(struct ipath_devdata *dd) | |
296 | { | |
297 | } | |
298 | ||
9bec3992 DO |
299 | /* |
300 | * Perform a PIO buffer bandwidth write test, to verify proper system | |
301 | * configuration. Even when all the setup calls work, occasionally | |
302 | * BIOS or other issues can prevent write combining from working, or | |
303 | * can cause other bandwidth problems to the chip. | |
304 | * | |
305 | * This test simply writes the same buffer over and over again, and | |
306 | * measures close to the peak bandwidth to the chip (not testing | |
307 | * data bandwidth to the wire). On chips that use an address-based | |
308 | * trigger to send packets to the wire, this is easy. On chips that | |
309 | * use a count to trigger, we want to make sure that the packet doesn't | |
310 | * go out on the wire, or trigger flow control checks. | |
311 | */ | |
312 | static void ipath_verify_pioperf(struct ipath_devdata *dd) | |
313 | { | |
314 | u32 pbnum, cnt, lcnt; | |
315 | u32 __iomem *piobuf; | |
316 | u32 *addr; | |
317 | u64 msecs, emsecs; | |
318 | ||
c4b4d16e | 319 | piobuf = ipath_getpiobuf(dd, 0, &pbnum); |
9bec3992 DO |
320 | if (!piobuf) { |
321 | dev_info(&dd->pcidev->dev, | |
322 | "No PIObufs for checking perf, skipping\n"); | |
323 | return; | |
324 | } | |
325 | ||
326 | /* | |
327 | * Enough to give us a reasonable test, less than piobuf size, and | |
328 | * likely multiple of store buffer length. | |
329 | */ | |
330 | cnt = 1024; | |
331 | ||
332 | addr = vmalloc(cnt); | |
333 | if (!addr) { | |
334 | dev_info(&dd->pcidev->dev, | |
335 | "Couldn't get memory for checking PIO perf," | |
336 | " skipping\n"); | |
337 | goto done; | |
338 | } | |
339 | ||
340 | preempt_disable(); /* we want reasonably accurate elapsed time */ | |
341 | msecs = 1 + jiffies_to_msecs(jiffies); | |
342 | for (lcnt = 0; lcnt < 10000U; lcnt++) { | |
343 | /* wait until we cross msec boundary */ | |
344 | if (jiffies_to_msecs(jiffies) >= msecs) | |
345 | break; | |
346 | udelay(1); | |
347 | } | |
348 | ||
6ac50727 DO |
349 | ipath_disable_armlaunch(dd); |
350 | ||
9bec3992 DO |
351 | writeq(0, piobuf); /* length 0, no dwords actually sent */ |
352 | ipath_flush_wc(); | |
353 | ||
354 | /* | |
355 | * this is only roughly accurate, since even with preempt we | |
356 | * still take interrupts that could take a while. Running for | |
357 | * >= 5 msec seems to get us "close enough" to accurate values | |
358 | */ | |
359 | msecs = jiffies_to_msecs(jiffies); | |
360 | for (emsecs = lcnt = 0; emsecs <= 5UL; lcnt++) { | |
361 | __iowrite32_copy(piobuf + 64, addr, cnt >> 2); | |
362 | emsecs = jiffies_to_msecs(jiffies) - msecs; | |
363 | } | |
364 | ||
365 | /* 1 GiB/sec, slightly over IB SDR line rate */ | |
366 | if (lcnt < (emsecs * 1024U)) | |
367 | ipath_dev_err(dd, | |
368 | "Performance problem: bandwidth to PIO buffers is " | |
369 | "only %u MiB/sec\n", | |
370 | lcnt / (u32) emsecs); | |
371 | else | |
372 | ipath_dbg("PIO buffer bandwidth %u MiB/sec is OK\n", | |
373 | lcnt / (u32) emsecs); | |
374 | ||
375 | preempt_enable(); | |
376 | ||
377 | vfree(addr); | |
378 | ||
379 | done: | |
380 | /* disarm piobuf, so it's available again */ | |
381 | ipath_disarm_piobufs(dd, pbnum, 1); | |
6ac50727 | 382 | ipath_enable_armlaunch(dd); |
9bec3992 DO |
383 | } |
384 | ||
7bb206e3 BS |
385 | static int __devinit ipath_init_one(struct pci_dev *pdev, |
386 | const struct pci_device_id *ent) | |
387 | { | |
388 | int ret, len, j; | |
389 | struct ipath_devdata *dd; | |
390 | unsigned long long addr; | |
391 | u32 bar0 = 0, bar1 = 0; | |
7bb206e3 | 392 | |
7bb206e3 BS |
393 | dd = ipath_alloc_devdata(pdev); |
394 | if (IS_ERR(dd)) { | |
395 | ret = PTR_ERR(dd); | |
396 | printk(KERN_ERR IPATH_DRV_NAME | |
397 | ": Could not allocate devdata: error %d\n", -ret); | |
f37bda92 | 398 | goto bail; |
7bb206e3 BS |
399 | } |
400 | ||
401 | ipath_cdbg(VERBOSE, "initializing unit #%u\n", dd->ipath_unit); | |
402 | ||
7bb206e3 BS |
403 | ret = pci_enable_device(pdev); |
404 | if (ret) { | |
405 | /* This can happen iff: | |
406 | * | |
407 | * We did a chip reset, and then failed to reprogram the | |
408 | * BAR, or the chip reset due to an internal error. We then | |
409 | * unloaded the driver and reloaded it. | |
410 | * | |
411 | * Both reset cases set the BAR back to initial state. For | |
412 | * the latter case, the AER sticky error bit at offset 0x718 | |
413 | * should be set, but the Linux kernel doesn't yet know | |
414 | * about that, it appears. If the original BAR was retained | |
415 | * in the kernel data structures, this may be OK. | |
416 | */ | |
417 | ipath_dev_err(dd, "enable unit %d failed: error %d\n", | |
418 | dd->ipath_unit, -ret); | |
419 | goto bail_devdata; | |
420 | } | |
421 | addr = pci_resource_start(pdev, 0); | |
422 | len = pci_resource_len(pdev, 0); | |
51f65ebc | 423 | ipath_cdbg(VERBOSE, "regbase (0) %llx len %d pdev->irq %d, vend %x/%x " |
7bb206e3 BS |
424 | "driver_data %lx\n", addr, len, pdev->irq, ent->vendor, |
425 | ent->device, ent->driver_data); | |
426 | ||
427 | read_bars(dd, pdev, &bar0, &bar1); | |
428 | ||
429 | if (!bar1 && !(bar0 & ~0xf)) { | |
430 | if (addr) { | |
431 | dev_info(&pdev->dev, "BAR is 0 (probable RESET), " | |
432 | "rewriting as %llx\n", addr); | |
433 | ret = pci_write_config_dword( | |
434 | pdev, PCI_BASE_ADDRESS_0, addr); | |
435 | if (ret) { | |
436 | ipath_dev_err(dd, "rewrite of BAR0 " | |
437 | "failed: err %d\n", -ret); | |
438 | goto bail_disable; | |
439 | } | |
440 | ret = pci_write_config_dword( | |
441 | pdev, PCI_BASE_ADDRESS_1, addr >> 32); | |
442 | if (ret) { | |
443 | ipath_dev_err(dd, "rewrite of BAR1 " | |
444 | "failed: err %d\n", -ret); | |
445 | goto bail_disable; | |
446 | } | |
447 | } else { | |
448 | ipath_dev_err(dd, "BAR is 0 (probable RESET), " | |
449 | "not usable until reboot\n"); | |
450 | ret = -ENODEV; | |
451 | goto bail_disable; | |
452 | } | |
453 | } | |
454 | ||
455 | ret = pci_request_regions(pdev, IPATH_DRV_NAME); | |
456 | if (ret) { | |
457 | dev_info(&pdev->dev, "pci_request_regions unit %u fails: " | |
458 | "err %d\n", dd->ipath_unit, -ret); | |
459 | goto bail_disable; | |
460 | } | |
461 | ||
462 | ret = pci_set_dma_mask(pdev, DMA_64BIT_MASK); | |
463 | if (ret) { | |
68dd43a1 BS |
464 | /* |
465 | * if the 64 bit setup fails, try 32 bit. Some systems | |
466 | * do not setup 64 bit maps on systems with 2GB or less | |
467 | * memory installed. | |
468 | */ | |
469 | ret = pci_set_dma_mask(pdev, DMA_32BIT_MASK); | |
470 | if (ret) { | |
b1d8865a BS |
471 | dev_info(&pdev->dev, |
472 | "Unable to set DMA mask for unit %u: %d\n", | |
473 | dd->ipath_unit, ret); | |
68dd43a1 BS |
474 | goto bail_regions; |
475 | } | |
b1d8865a | 476 | else { |
68dd43a1 | 477 | ipath_dbg("No 64bit DMA mask, used 32 bit mask\n"); |
b1d8865a BS |
478 | ret = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK); |
479 | if (ret) | |
480 | dev_info(&pdev->dev, | |
481 | "Unable to set DMA consistent mask " | |
482 | "for unit %u: %d\n", | |
483 | dd->ipath_unit, ret); | |
484 | ||
485 | } | |
486 | } | |
487 | else { | |
488 | ret = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK); | |
489 | if (ret) | |
490 | dev_info(&pdev->dev, | |
491 | "Unable to set DMA consistent mask " | |
492 | "for unit %u: %d\n", | |
493 | dd->ipath_unit, ret); | |
7bb206e3 BS |
494 | } |
495 | ||
496 | pci_set_master(pdev); | |
497 | ||
498 | /* | |
499 | * Save BARs to rewrite after device reset. Save all 64 bits of | |
500 | * BAR, just in case. | |
501 | */ | |
502 | dd->ipath_pcibar0 = addr; | |
503 | dd->ipath_pcibar1 = addr >> 32; | |
504 | dd->ipath_deviceid = ent->device; /* save for later use */ | |
505 | dd->ipath_vendorid = ent->vendor; | |
506 | ||
507 | /* setup the chip-specific functions, as early as possible. */ | |
508 | switch (ent->device) { | |
509 | case PCI_DEVICE_ID_INFINIPATH_HT: | |
820054b7 | 510 | #ifdef CONFIG_HT_IRQ |
525d0ca1 | 511 | ipath_init_iba6110_funcs(dd); |
7bb206e3 | 512 | break; |
820054b7 BS |
513 | #else |
514 | ipath_dev_err(dd, "QLogic HT device 0x%x cannot work if " | |
515 | "CONFIG_HT_IRQ is not enabled\n", ent->device); | |
516 | return -ENODEV; | |
e757bef2 | 517 | #endif |
7bb206e3 | 518 | case PCI_DEVICE_ID_INFINIPATH_PE800: |
820054b7 | 519 | #ifdef CONFIG_PCI_MSI |
525d0ca1 | 520 | ipath_init_iba6120_funcs(dd); |
7bb206e3 | 521 | break; |
820054b7 BS |
522 | #else |
523 | ipath_dev_err(dd, "QLogic PCIE device 0x%x cannot work if " | |
524 | "CONFIG_PCI_MSI is not enabled\n", ent->device); | |
525 | return -ENODEV; | |
e757bef2 | 526 | #endif |
7bb206e3 | 527 | default: |
759d5768 | 528 | ipath_dev_err(dd, "Found unknown QLogic deviceid 0x%x, " |
7bb206e3 BS |
529 | "failing\n", ent->device); |
530 | return -ENODEV; | |
531 | } | |
532 | ||
533 | for (j = 0; j < 6; j++) { | |
534 | if (!pdev->resource[j].start) | |
535 | continue; | |
e29419ff GKH |
536 | ipath_cdbg(VERBOSE, "BAR %d start %llx, end %llx, len %llx\n", |
537 | j, (unsigned long long)pdev->resource[j].start, | |
538 | (unsigned long long)pdev->resource[j].end, | |
539 | (unsigned long long)pci_resource_len(pdev, j)); | |
7bb206e3 BS |
540 | } |
541 | ||
542 | if (!addr) { | |
543 | ipath_dev_err(dd, "No valid address in BAR 0!\n"); | |
544 | ret = -ENODEV; | |
545 | goto bail_regions; | |
546 | } | |
547 | ||
44c10138 | 548 | dd->ipath_pcirev = pdev->revision; |
7bb206e3 | 549 | |
eb9dc6f4 BS |
550 | #if defined(__powerpc__) |
551 | /* There isn't a generic way to specify writethrough mappings */ | |
552 | dd->ipath_kregbase = __ioremap(addr, len, | |
553 | (_PAGE_NO_CACHE|_PAGE_WRITETHRU)); | |
554 | #else | |
7bb206e3 | 555 | dd->ipath_kregbase = ioremap_nocache(addr, len); |
eb9dc6f4 | 556 | #endif |
7bb206e3 BS |
557 | |
558 | if (!dd->ipath_kregbase) { | |
559 | ipath_dbg("Unable to map io addr %llx to kvirt, failing\n", | |
560 | addr); | |
561 | ret = -ENOMEM; | |
562 | goto bail_iounmap; | |
563 | } | |
564 | dd->ipath_kregend = (u64 __iomem *) | |
565 | ((void __iomem *)dd->ipath_kregbase + len); | |
566 | dd->ipath_physaddr = addr; /* used for io_remap, etc. */ | |
567 | /* for user mmap */ | |
b35f004d BS |
568 | ipath_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p\n", |
569 | addr, dd->ipath_kregbase); | |
7bb206e3 BS |
570 | |
571 | /* | |
572 | * clear ipath_flags here instead of in ipath_init_chip as it is set | |
573 | * by ipath_setup_htconfig. | |
574 | */ | |
575 | dd->ipath_flags = 0; | |
fba75200 BS |
576 | dd->ipath_lli_counter = 0; |
577 | dd->ipath_lli_errors = 0; | |
7bb206e3 BS |
578 | |
579 | if (dd->ipath_f_bus(dd, pdev)) | |
580 | ipath_dev_err(dd, "Failed to setup config space; " | |
581 | "continuing anyway\n"); | |
582 | ||
583 | /* | |
dace1453 | 584 | * set up our interrupt handler; IRQF_SHARED probably not needed, |
7bb206e3 BS |
585 | * since MSI interrupts shouldn't be shared but won't hurt for now. |
586 | * check 0 irq after we return from chip-specific bus setup, since | |
587 | * that can affect this due to setup | |
588 | */ | |
51f65ebc | 589 | if (!dd->ipath_irq) |
7bb206e3 BS |
590 | ipath_dev_err(dd, "irq is 0, BIOS error? Interrupts won't " |
591 | "work\n"); | |
592 | else { | |
51f65ebc | 593 | ret = request_irq(dd->ipath_irq, ipath_intr, IRQF_SHARED, |
7bb206e3 BS |
594 | IPATH_DRV_NAME, dd); |
595 | if (ret) { | |
596 | ipath_dev_err(dd, "Couldn't setup irq handler, " | |
51f65ebc | 597 | "irq=%d: %d\n", dd->ipath_irq, ret); |
7bb206e3 BS |
598 | goto bail_iounmap; |
599 | } | |
600 | } | |
601 | ||
602 | ret = ipath_init_chip(dd, 0); /* do the chip-specific init */ | |
603 | if (ret) | |
7b196e2f | 604 | goto bail_irqsetup; |
7bb206e3 BS |
605 | |
606 | ret = ipath_enable_wc(dd); | |
607 | ||
608 | if (ret) { | |
609 | ipath_dev_err(dd, "Write combining not enabled " | |
610 | "(err %d): performance may be poor\n", | |
611 | -ret); | |
612 | ret = 0; | |
613 | } | |
614 | ||
9bec3992 DO |
615 | ipath_verify_pioperf(dd); |
616 | ||
7bb206e3 BS |
617 | ipath_device_create_group(&pdev->dev, dd); |
618 | ipathfs_add_device(dd); | |
619 | ipath_user_add(dd); | |
a2acb2ff | 620 | ipath_diag_add(dd); |
b1c1b6a3 | 621 | ipath_register_ib_device(dd); |
7bb206e3 | 622 | |
3588423f AJ |
623 | /* Check that card status in STATUS_TIMEOUT seconds. */ |
624 | schedule_delayed_work(&dd->status_work, HZ * STATUS_TIMEOUT); | |
625 | ||
7bb206e3 BS |
626 | goto bail; |
627 | ||
7b196e2f | 628 | bail_irqsetup: |
2ba3f56e RC |
629 | if (pdev->irq) |
630 | free_irq(pdev->irq, dd); | |
7b196e2f | 631 | |
7bb206e3 BS |
632 | bail_iounmap: |
633 | iounmap((volatile void __iomem *) dd->ipath_kregbase); | |
634 | ||
635 | bail_regions: | |
636 | pci_release_regions(pdev); | |
637 | ||
638 | bail_disable: | |
639 | pci_disable_device(pdev); | |
640 | ||
641 | bail_devdata: | |
642 | ipath_free_devdata(pdev, dd); | |
643 | ||
7bb206e3 BS |
644 | bail: |
645 | return ret; | |
646 | } | |
647 | ||
7227aac4 | 648 | static void __devexit cleanup_device(struct ipath_devdata *dd) |
7bb206e3 | 649 | { |
7227aac4 | 650 | int port; |
7bb206e3 | 651 | |
7227aac4 BS |
652 | if (*dd->ipath_statusp & IPATH_STATUS_CHIP_PRESENT) { |
653 | /* can't do anything more with chip; needs re-init */ | |
654 | *dd->ipath_statusp &= ~IPATH_STATUS_CHIP_PRESENT; | |
655 | if (dd->ipath_kregbase) { | |
656 | /* | |
657 | * if we haven't already cleaned up before these are | |
658 | * to ensure any register reads/writes "fail" until | |
659 | * re-init | |
660 | */ | |
661 | dd->ipath_kregbase = NULL; | |
662 | dd->ipath_uregbase = 0; | |
663 | dd->ipath_sregbase = 0; | |
664 | dd->ipath_cregbase = 0; | |
665 | dd->ipath_kregsize = 0; | |
666 | } | |
667 | ipath_disable_wc(dd); | |
668 | } | |
c78f6415 | 669 | |
7227aac4 BS |
670 | if (dd->ipath_pioavailregs_dma) { |
671 | dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, | |
672 | (void *) dd->ipath_pioavailregs_dma, | |
673 | dd->ipath_pioavailregs_phys); | |
674 | dd->ipath_pioavailregs_dma = NULL; | |
675 | } | |
676 | if (dd->ipath_dummy_hdrq) { | |
677 | dma_free_coherent(&dd->pcidev->dev, | |
678 | dd->ipath_pd[0]->port_rcvhdrq_size, | |
679 | dd->ipath_dummy_hdrq, dd->ipath_dummy_hdrq_phys); | |
680 | dd->ipath_dummy_hdrq = NULL; | |
681 | } | |
682 | ||
683 | if (dd->ipath_pageshadow) { | |
684 | struct page **tmpp = dd->ipath_pageshadow; | |
685 | dma_addr_t *tmpd = dd->ipath_physshadow; | |
686 | int i, cnt = 0; | |
687 | ||
688 | ipath_cdbg(VERBOSE, "Unlocking any expTID pages still " | |
689 | "locked\n"); | |
690 | for (port = 0; port < dd->ipath_cfgports; port++) { | |
691 | int port_tidbase = port * dd->ipath_rcvtidcnt; | |
692 | int maxtid = port_tidbase + dd->ipath_rcvtidcnt; | |
693 | for (i = port_tidbase; i < maxtid; i++) { | |
694 | if (!tmpp[i]) | |
695 | continue; | |
696 | pci_unmap_page(dd->pcidev, tmpd[i], | |
697 | PAGE_SIZE, PCI_DMA_FROMDEVICE); | |
698 | ipath_release_user_pages(&tmpp[i], 1); | |
699 | tmpp[i] = NULL; | |
700 | cnt++; | |
701 | } | |
702 | } | |
703 | if (cnt) { | |
704 | ipath_stats.sps_pageunlocks += cnt; | |
705 | ipath_cdbg(VERBOSE, "There were still %u expTID " | |
706 | "entries locked\n", cnt); | |
707 | } | |
708 | if (ipath_stats.sps_pagelocks || | |
709 | ipath_stats.sps_pageunlocks) | |
710 | ipath_cdbg(VERBOSE, "%llu pages locked, %llu " | |
711 | "unlocked via ipath_m{un}lock\n", | |
712 | (unsigned long long) | |
713 | ipath_stats.sps_pagelocks, | |
714 | (unsigned long long) | |
715 | ipath_stats.sps_pageunlocks); | |
716 | ||
717 | ipath_cdbg(VERBOSE, "Free shadow page tid array at %p\n", | |
718 | dd->ipath_pageshadow); | |
9783ab40 | 719 | tmpp = dd->ipath_pageshadow; |
7227aac4 | 720 | dd->ipath_pageshadow = NULL; |
9783ab40 | 721 | vfree(tmpp); |
9355fb6a RC |
722 | |
723 | dd->ipath_egrtidbase = NULL; | |
c78f6415 BS |
724 | } |
725 | ||
7227aac4 BS |
726 | /* |
727 | * free any resources still in use (usually just kernel ports) | |
728 | * at unload; we do for portcnt, not cfgports, because cfgports | |
729 | * could have changed while we were loaded. | |
730 | */ | |
731 | for (port = 0; port < dd->ipath_portcnt; port++) { | |
732 | struct ipath_portdata *pd = dd->ipath_pd[port]; | |
733 | dd->ipath_pd[port] = NULL; | |
734 | ipath_free_pddata(dd, pd); | |
735 | } | |
736 | kfree(dd->ipath_pd); | |
737 | /* | |
738 | * debuggability, in case some cleanup path tries to use it | |
739 | * after this | |
740 | */ | |
741 | dd->ipath_pd = NULL; | |
742 | } | |
743 | ||
744 | static void __devexit ipath_remove_one(struct pci_dev *pdev) | |
745 | { | |
746 | struct ipath_devdata *dd = pci_get_drvdata(pdev); | |
747 | ||
748 | ipath_cdbg(VERBOSE, "removing, pdev=%p, dd=%p\n", pdev, dd); | |
749 | ||
53c1d2c9 BS |
750 | /* |
751 | * disable the IB link early, to be sure no new packets arrive, which | |
752 | * complicates the shutdown process | |
753 | */ | |
754 | ipath_shutdown_device(dd); | |
755 | ||
3588423f AJ |
756 | cancel_delayed_work(&dd->status_work); |
757 | flush_scheduled_work(); | |
758 | ||
7227aac4 BS |
759 | if (dd->verbs_dev) |
760 | ipath_unregister_ib_device(dd->verbs_dev); | |
761 | ||
a2acb2ff BS |
762 | ipath_diag_remove(dd); |
763 | ipath_user_remove(dd); | |
7bb206e3 BS |
764 | ipathfs_remove_device(dd); |
765 | ipath_device_remove_group(&pdev->dev, dd); | |
7227aac4 | 766 | |
7bb206e3 BS |
767 | ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, " |
768 | "unit %u\n", dd, (u32) dd->ipath_unit); | |
7227aac4 BS |
769 | |
770 | cleanup_device(dd); | |
771 | ||
772 | /* | |
773 | * turn off rcv, send, and interrupts for all ports, all drivers | |
774 | * should also hard reset the chip here? | |
775 | * free up port 0 (kernel) rcvhdr, egr bufs, and eventually tid bufs | |
776 | * for all versions of the driver, if they were allocated | |
777 | */ | |
51f65ebc BS |
778 | if (dd->ipath_irq) { |
779 | ipath_cdbg(VERBOSE, "unit %u free irq %d\n", | |
780 | dd->ipath_unit, dd->ipath_irq); | |
781 | dd->ipath_f_free_irq(dd); | |
7227aac4 BS |
782 | } else |
783 | ipath_dbg("irq is 0, not doing free_irq " | |
784 | "for unit %u\n", dd->ipath_unit); | |
785 | /* | |
786 | * we check for NULL here, because it's outside | |
787 | * the kregbase check, and we need to call it | |
788 | * after the free_irq. Thus it's possible that | |
789 | * the function pointers were never initialized. | |
790 | */ | |
791 | if (dd->ipath_f_cleanup) | |
792 | /* clean up chip-specific stuff */ | |
793 | dd->ipath_f_cleanup(dd); | |
794 | ||
795 | ipath_cdbg(VERBOSE, "Unmapping kregbase %p\n", dd->ipath_kregbase); | |
796 | iounmap((volatile void __iomem *) dd->ipath_kregbase); | |
7bb206e3 BS |
797 | pci_release_regions(pdev); |
798 | ipath_cdbg(VERBOSE, "calling pci_disable_device\n"); | |
799 | pci_disable_device(pdev); | |
800 | ||
801 | ipath_free_devdata(pdev, dd); | |
7bb206e3 BS |
802 | } |
803 | ||
804 | /* general driver use */ | |
805 | DEFINE_MUTEX(ipath_mutex); | |
806 | ||
807 | static DEFINE_SPINLOCK(ipath_pioavail_lock); | |
808 | ||
809 | /** | |
810 | * ipath_disarm_piobufs - cancel a range of PIO buffers | |
811 | * @dd: the infinipath device | |
812 | * @first: the first PIO buffer to cancel | |
813 | * @cnt: the number of PIO buffers to cancel | |
814 | * | |
815 | * cancel a range of PIO buffers, used when they might be armed, but | |
816 | * not triggered. Used at init to ensure buffer state, and also user | |
817 | * process close, in case it died while writing to a PIO buffer | |
818 | * Also after errors. | |
819 | */ | |
820 | void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first, | |
821 | unsigned cnt) | |
822 | { | |
823 | unsigned i, last = first + cnt; | |
e342c119 | 824 | unsigned long flags; |
7bb206e3 BS |
825 | |
826 | ipath_cdbg(PKT, "disarm %u PIObufs first=%u\n", cnt, first); | |
7bb206e3 | 827 | for (i = first; i < last; i++) { |
e342c119 JG |
828 | spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); |
829 | /* | |
830 | * The disarm-related bits are write-only, so it | |
831 | * is ok to OR them in with our copy of sendctrl | |
832 | * while we hold the lock. | |
833 | */ | |
7bb206e3 | 834 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, |
e342c119 JG |
835 | dd->ipath_sendctrl | INFINIPATH_S_DISARM | |
836 | (i << INFINIPATH_S_DISARMPIOBUF_SHIFT)); | |
837 | /* can't disarm bufs back-to-back per iba7220 spec */ | |
838 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); | |
839 | spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); | |
7bb206e3 | 840 | } |
c4b4d16e RC |
841 | /* on some older chips, update may not happen after cancel */ |
842 | ipath_force_pio_avail_update(dd); | |
7bb206e3 BS |
843 | } |
844 | ||
845 | /** | |
846 | * ipath_wait_linkstate - wait for an IB link state change to occur | |
847 | * @dd: the infinipath device | |
848 | * @state: the state to wait for | |
849 | * @msecs: the number of milliseconds to wait | |
850 | * | |
851 | * wait up to msecs milliseconds for IB link state change to occur for | |
852 | * now, take the easy polling route. Currently used only by | |
34b2aafe | 853 | * ipath_set_linkstate. Returns 0 if state reached, otherwise |
7bb206e3 BS |
854 | * -ETIMEDOUT state can have multiple states set, for any of several |
855 | * transitions. | |
856 | */ | |
140277e9 | 857 | int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs) |
7bb206e3 | 858 | { |
0fd41363 BS |
859 | dd->ipath_state_wanted = state; |
860 | wait_event_interruptible_timeout(ipath_state_wait, | |
7bb206e3 BS |
861 | (dd->ipath_flags & state), |
862 | msecs_to_jiffies(msecs)); | |
0fd41363 | 863 | dd->ipath_state_wanted = 0; |
7bb206e3 BS |
864 | |
865 | if (!(dd->ipath_flags & state)) { | |
866 | u64 val; | |
0fd41363 BS |
867 | ipath_cdbg(VERBOSE, "Didn't reach linkstate %s within %u" |
868 | " ms\n", | |
7bb206e3 BS |
869 | /* test INIT ahead of DOWN, both can be set */ |
870 | (state & IPATH_LINKINIT) ? "INIT" : | |
871 | ((state & IPATH_LINKDOWN) ? "DOWN" : | |
872 | ((state & IPATH_LINKARMED) ? "ARM" : "ACTIVE")), | |
873 | msecs); | |
874 | val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); | |
875 | ipath_cdbg(VERBOSE, "ibcc=%llx ibcstatus=%llx (%s)\n", | |
876 | (unsigned long long) ipath_read_kreg64( | |
877 | dd, dd->ipath_kregs->kr_ibcctrl), | |
878 | (unsigned long long) val, | |
879 | ipath_ibcstatus_str[val & 0xf]); | |
880 | } | |
881 | return (dd->ipath_flags & state) ? 0 : -ETIMEDOUT; | |
882 | } | |
883 | ||
8ec1077b BS |
884 | /* |
885 | * Decode the error status into strings, deciding whether to always | |
886 | * print * it or not depending on "normal packet errors" vs everything | |
887 | * else. Return 1 if "real" errors, otherwise 0 if only packet | |
888 | * errors, so caller can decide what to print with the string. | |
889 | */ | |
890 | int ipath_decode_err(char *buf, size_t blen, ipath_err_t err) | |
7bb206e3 | 891 | { |
8ec1077b | 892 | int iserr = 1; |
7bb206e3 | 893 | *buf = '\0'; |
8ec1077b BS |
894 | if (err & INFINIPATH_E_PKTERRS) { |
895 | if (!(err & ~INFINIPATH_E_PKTERRS)) | |
896 | iserr = 0; // if only packet errors. | |
897 | if (ipath_debug & __IPATH_ERRPKTDBG) { | |
898 | if (err & INFINIPATH_E_REBP) | |
899 | strlcat(buf, "EBP ", blen); | |
900 | if (err & INFINIPATH_E_RVCRC) | |
901 | strlcat(buf, "VCRC ", blen); | |
902 | if (err & INFINIPATH_E_RICRC) { | |
903 | strlcat(buf, "CRC ", blen); | |
904 | // clear for check below, so only once | |
905 | err &= INFINIPATH_E_RICRC; | |
906 | } | |
907 | if (err & INFINIPATH_E_RSHORTPKTLEN) | |
908 | strlcat(buf, "rshortpktlen ", blen); | |
909 | if (err & INFINIPATH_E_SDROPPEDDATAPKT) | |
910 | strlcat(buf, "sdroppeddatapkt ", blen); | |
911 | if (err & INFINIPATH_E_SPKTLEN) | |
912 | strlcat(buf, "spktlen ", blen); | |
913 | } | |
914 | if ((err & INFINIPATH_E_RICRC) && | |
915 | !(err&(INFINIPATH_E_RVCRC|INFINIPATH_E_REBP))) | |
916 | strlcat(buf, "CRC ", blen); | |
917 | if (!iserr) | |
918 | goto done; | |
919 | } | |
7bb206e3 BS |
920 | if (err & INFINIPATH_E_RHDRLEN) |
921 | strlcat(buf, "rhdrlen ", blen); | |
922 | if (err & INFINIPATH_E_RBADTID) | |
923 | strlcat(buf, "rbadtid ", blen); | |
924 | if (err & INFINIPATH_E_RBADVERSION) | |
925 | strlcat(buf, "rbadversion ", blen); | |
926 | if (err & INFINIPATH_E_RHDR) | |
927 | strlcat(buf, "rhdr ", blen); | |
928 | if (err & INFINIPATH_E_RLONGPKTLEN) | |
929 | strlcat(buf, "rlongpktlen ", blen); | |
7bb206e3 BS |
930 | if (err & INFINIPATH_E_RMAXPKTLEN) |
931 | strlcat(buf, "rmaxpktlen ", blen); | |
932 | if (err & INFINIPATH_E_RMINPKTLEN) | |
933 | strlcat(buf, "rminpktlen ", blen); | |
8ec1077b BS |
934 | if (err & INFINIPATH_E_SMINPKTLEN) |
935 | strlcat(buf, "sminpktlen ", blen); | |
7bb206e3 BS |
936 | if (err & INFINIPATH_E_RFORMATERR) |
937 | strlcat(buf, "rformaterr ", blen); | |
938 | if (err & INFINIPATH_E_RUNSUPVL) | |
939 | strlcat(buf, "runsupvl ", blen); | |
940 | if (err & INFINIPATH_E_RUNEXPCHAR) | |
941 | strlcat(buf, "runexpchar ", blen); | |
942 | if (err & INFINIPATH_E_RIBFLOW) | |
943 | strlcat(buf, "ribflow ", blen); | |
7bb206e3 BS |
944 | if (err & INFINIPATH_E_SUNDERRUN) |
945 | strlcat(buf, "sunderrun ", blen); | |
946 | if (err & INFINIPATH_E_SPIOARMLAUNCH) | |
947 | strlcat(buf, "spioarmlaunch ", blen); | |
948 | if (err & INFINIPATH_E_SUNEXPERRPKTNUM) | |
949 | strlcat(buf, "sunexperrpktnum ", blen); | |
7bb206e3 BS |
950 | if (err & INFINIPATH_E_SDROPPEDSMPPKT) |
951 | strlcat(buf, "sdroppedsmppkt ", blen); | |
952 | if (err & INFINIPATH_E_SMAXPKTLEN) | |
953 | strlcat(buf, "smaxpktlen ", blen); | |
7bb206e3 BS |
954 | if (err & INFINIPATH_E_SUNSUPVL) |
955 | strlcat(buf, "sunsupVL ", blen); | |
7bb206e3 BS |
956 | if (err & INFINIPATH_E_INVALIDADDR) |
957 | strlcat(buf, "invalidaddr ", blen); | |
7bb206e3 BS |
958 | if (err & INFINIPATH_E_RRCVEGRFULL) |
959 | strlcat(buf, "rcvegrfull ", blen); | |
960 | if (err & INFINIPATH_E_RRCVHDRFULL) | |
961 | strlcat(buf, "rcvhdrfull ", blen); | |
962 | if (err & INFINIPATH_E_IBSTATUSCHANGED) | |
963 | strlcat(buf, "ibcstatuschg ", blen); | |
964 | if (err & INFINIPATH_E_RIBLOSTLINK) | |
965 | strlcat(buf, "riblostlink ", blen); | |
966 | if (err & INFINIPATH_E_HARDWARE) | |
967 | strlcat(buf, "hardware ", blen); | |
968 | if (err & INFINIPATH_E_RESET) | |
969 | strlcat(buf, "reset ", blen); | |
8ec1077b BS |
970 | done: |
971 | return iserr; | |
7bb206e3 BS |
972 | } |
973 | ||
974 | /** | |
975 | * get_rhf_errstring - decode RHF errors | |
976 | * @err: the err number | |
977 | * @msg: the output buffer | |
978 | * @len: the length of the output buffer | |
979 | * | |
980 | * only used one place now, may want more later | |
981 | */ | |
982 | static void get_rhf_errstring(u32 err, char *msg, size_t len) | |
983 | { | |
984 | /* if no errors, and so don't need to check what's first */ | |
985 | *msg = '\0'; | |
986 | ||
987 | if (err & INFINIPATH_RHF_H_ICRCERR) | |
988 | strlcat(msg, "icrcerr ", len); | |
989 | if (err & INFINIPATH_RHF_H_VCRCERR) | |
990 | strlcat(msg, "vcrcerr ", len); | |
991 | if (err & INFINIPATH_RHF_H_PARITYERR) | |
992 | strlcat(msg, "parityerr ", len); | |
993 | if (err & INFINIPATH_RHF_H_LENERR) | |
994 | strlcat(msg, "lenerr ", len); | |
995 | if (err & INFINIPATH_RHF_H_MTUERR) | |
996 | strlcat(msg, "mtuerr ", len); | |
997 | if (err & INFINIPATH_RHF_H_IHDRERR) | |
998 | /* infinipath hdr checksum error */ | |
999 | strlcat(msg, "ipathhdrerr ", len); | |
1000 | if (err & INFINIPATH_RHF_H_TIDERR) | |
1001 | strlcat(msg, "tiderr ", len); | |
1002 | if (err & INFINIPATH_RHF_H_MKERR) | |
1003 | /* bad port, offset, etc. */ | |
1004 | strlcat(msg, "invalid ipathhdr ", len); | |
1005 | if (err & INFINIPATH_RHF_H_IBERR) | |
1006 | strlcat(msg, "iberr ", len); | |
1007 | if (err & INFINIPATH_RHF_L_SWA) | |
1008 | strlcat(msg, "swA ", len); | |
1009 | if (err & INFINIPATH_RHF_L_SWB) | |
1010 | strlcat(msg, "swB ", len); | |
1011 | } | |
1012 | ||
1013 | /** | |
1014 | * ipath_get_egrbuf - get an eager buffer | |
1015 | * @dd: the infinipath device | |
1016 | * @bufnum: the eager buffer to get | |
7bb206e3 BS |
1017 | * |
1018 | * must only be called if ipath_pd[port] is known to be allocated | |
1019 | */ | |
d65708f3 | 1020 | static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum) |
7bb206e3 | 1021 | { |
1fd3b40f BS |
1022 | return dd->ipath_port0_skbinfo ? |
1023 | (void *) dd->ipath_port0_skbinfo[bufnum].skb->data : NULL; | |
7bb206e3 BS |
1024 | } |
1025 | ||
1026 | /** | |
1027 | * ipath_alloc_skb - allocate an skb and buffer with possible constraints | |
1028 | * @dd: the infinipath device | |
1029 | * @gfp_mask: the sk_buff SFP mask | |
1030 | */ | |
1031 | struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd, | |
1032 | gfp_t gfp_mask) | |
1033 | { | |
1034 | struct sk_buff *skb; | |
1035 | u32 len; | |
1036 | ||
1037 | /* | |
1038 | * Only fully supported way to handle this is to allocate lots | |
1039 | * extra, align as needed, and then do skb_reserve(). That wastes | |
1040 | * a lot of memory... I'll have to hack this into infinipath_copy | |
1041 | * also. | |
1042 | */ | |
1043 | ||
1044 | /* | |
1fd3b40f BS |
1045 | * We need 2 extra bytes for ipath_ether data sent in the |
1046 | * key header. In order to keep everything dword aligned, | |
1047 | * we'll reserve 4 bytes. | |
7bb206e3 | 1048 | */ |
1fd3b40f BS |
1049 | len = dd->ipath_ibmaxlen + 4; |
1050 | ||
7bb206e3 | 1051 | if (dd->ipath_flags & IPATH_4BYTE_TID) { |
1fd3b40f | 1052 | /* We need a 2KB multiple alignment, and there is no way |
7bb206e3 BS |
1053 | * to do it except to allocate extra and then skb_reserve |
1054 | * enough to bring it up to the right alignment. | |
1055 | */ | |
1fd3b40f | 1056 | len += 2047; |
7bb206e3 | 1057 | } |
1fd3b40f | 1058 | |
7bb206e3 BS |
1059 | skb = __dev_alloc_skb(len, gfp_mask); |
1060 | if (!skb) { | |
1061 | ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n", | |
1062 | len); | |
1063 | goto bail; | |
1064 | } | |
1fd3b40f BS |
1065 | |
1066 | skb_reserve(skb, 4); | |
1067 | ||
7bb206e3 | 1068 | if (dd->ipath_flags & IPATH_4BYTE_TID) { |
1fd3b40f | 1069 | u32 una = (unsigned long)skb->data & 2047; |
7bb206e3 | 1070 | if (una) |
1fd3b40f BS |
1071 | skb_reserve(skb, 2048 - una); |
1072 | } | |
7bb206e3 BS |
1073 | |
1074 | bail: | |
1075 | return skb; | |
1076 | } | |
1077 | ||
3d37b9e2 RC |
1078 | static void ipath_rcv_hdrerr(struct ipath_devdata *dd, |
1079 | u32 eflags, | |
1080 | u32 l, | |
1081 | u32 etail, | |
9355fb6a RC |
1082 | __le32 *rhf_addr, |
1083 | struct ipath_message_header *hdr) | |
3d37b9e2 RC |
1084 | { |
1085 | char emsg[128]; | |
3d37b9e2 RC |
1086 | |
1087 | get_rhf_errstring(eflags, emsg, sizeof emsg); | |
3d37b9e2 RC |
1088 | ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u " |
1089 | "tlen=%x opcode=%x egridx=%x: %s\n", | |
1090 | eflags, l, | |
9355fb6a RC |
1091 | ipath_hdrget_rcv_type(rhf_addr), |
1092 | ipath_hdrget_length_in_bytes(rhf_addr), | |
3d37b9e2 RC |
1093 | be32_to_cpu(hdr->bth[0]) >> 24, |
1094 | etail, emsg); | |
1095 | ||
1096 | /* Count local link integrity errors. */ | |
1097 | if (eflags & (INFINIPATH_RHF_H_ICRCERR | INFINIPATH_RHF_H_VCRCERR)) { | |
1098 | u8 n = (dd->ipath_ibcctrl >> | |
1099 | INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) & | |
1100 | INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK; | |
1101 | ||
1102 | if (++dd->ipath_lli_counter > n) { | |
1103 | dd->ipath_lli_counter = 0; | |
1104 | dd->ipath_lli_errors++; | |
1105 | } | |
1106 | } | |
1107 | } | |
1108 | ||
7bb206e3 BS |
1109 | /* |
1110 | * ipath_kreceive - receive a packet | |
c59a80ac | 1111 | * @pd: the infinipath port |
7bb206e3 BS |
1112 | * |
1113 | * called from interrupt handler for errors or receive interrupt | |
1114 | */ | |
c59a80ac | 1115 | void ipath_kreceive(struct ipath_portdata *pd) |
7bb206e3 | 1116 | { |
c59a80ac | 1117 | struct ipath_devdata *dd = pd->port_dd; |
9355fb6a | 1118 | __le32 *rhf_addr; |
7bb206e3 BS |
1119 | void *ebuf; |
1120 | const u32 rsize = dd->ipath_rcvhdrentsize; /* words */ | |
1121 | const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize; /* words */ | |
1122 | u32 etail = -1, l, hdrqtail; | |
27b678dd | 1123 | struct ipath_message_header *hdr; |
9355fb6a | 1124 | u32 eflags, i, etype, tlen, pkttot = 0, updegr = 0, reloop = 0; |
7bb206e3 | 1125 | static u64 totcalls; /* stats, may eventually remove */ |
9355fb6a | 1126 | int last; |
7bb206e3 | 1127 | |
c59a80ac | 1128 | l = pd->port_head; |
9355fb6a RC |
1129 | rhf_addr = (__le32 *) pd->port_rcvhdrq + l + dd->ipath_rhf_offset; |
1130 | if (dd->ipath_flags & IPATH_NODMA_RTAIL) { | |
1131 | u32 seq = ipath_hdrget_seq(rhf_addr); | |
7bb206e3 | 1132 | |
9355fb6a RC |
1133 | if (seq != pd->port_seq_cnt) |
1134 | goto bail; | |
1135 | hdrqtail = 0; | |
1136 | } else { | |
1137 | hdrqtail = ipath_get_rcvhdrtail(pd); | |
1138 | if (l == hdrqtail) | |
1139 | goto bail; | |
1140 | smp_rmb(); | |
1141 | } | |
7bb206e3 | 1142 | |
9355fb6a RC |
1143 | reloop: |
1144 | for (last = 0, i = 1; !last; i++) { | |
1145 | hdr = dd->ipath_f_get_msgheader(dd, rhf_addr); | |
1146 | eflags = ipath_hdrget_err_flags(rhf_addr); | |
1147 | etype = ipath_hdrget_rcv_type(rhf_addr); | |
7bb206e3 | 1148 | /* total length */ |
9355fb6a | 1149 | tlen = ipath_hdrget_length_in_bytes(rhf_addr); |
7bb206e3 | 1150 | ebuf = NULL; |
9355fb6a RC |
1151 | if ((dd->ipath_flags & IPATH_NODMA_RTAIL) ? |
1152 | ipath_hdrget_use_egr_buf(rhf_addr) : | |
1153 | (etype != RCVHQ_RCV_TYPE_EXPECTED)) { | |
7bb206e3 | 1154 | /* |
9355fb6a | 1155 | * It turns out that the chip uses an eager buffer |
7bb206e3 BS |
1156 | * for all non-expected packets, whether it "needs" |
1157 | * one or not. So always get the index, but don't | |
1158 | * set ebuf (so we try to copy data) unless the | |
1159 | * length requires it. | |
1160 | */ | |
9355fb6a RC |
1161 | etail = ipath_hdrget_index(rhf_addr); |
1162 | updegr = 1; | |
7bb206e3 BS |
1163 | if (tlen > sizeof(*hdr) || |
1164 | etype == RCVHQ_RCV_TYPE_NON_KD) | |
d65708f3 | 1165 | ebuf = ipath_get_egrbuf(dd, etail); |
7bb206e3 BS |
1166 | } |
1167 | ||
1168 | /* | |
1169 | * both tiderr and ipathhdrerr are set for all plain IB | |
1170 | * packets; only ipathhdrerr should be set. | |
1171 | */ | |
1172 | ||
9355fb6a RC |
1173 | if (etype != RCVHQ_RCV_TYPE_NON_KD && |
1174 | etype != RCVHQ_RCV_TYPE_ERROR && | |
1175 | ipath_hdrget_ipath_ver(hdr->iph.ver_port_tid_offset) != | |
1176 | IPS_PROTO_VERSION) | |
7bb206e3 BS |
1177 | ipath_cdbg(PKT, "Bad InfiniPath protocol version " |
1178 | "%x\n", etype); | |
7bb206e3 | 1179 | |
3d37b9e2 | 1180 | if (unlikely(eflags)) |
9355fb6a | 1181 | ipath_rcv_hdrerr(dd, eflags, l, etail, rhf_addr, hdr); |
3d37b9e2 | 1182 | else if (etype == RCVHQ_RCV_TYPE_NON_KD) { |
9355fb6a | 1183 | ipath_ib_rcv(dd->verbs_dev, (u32 *)hdr, ebuf, tlen); |
34b2aafe BS |
1184 | if (dd->ipath_lli_counter) |
1185 | dd->ipath_lli_counter--; | |
9355fb6a RC |
1186 | } else if (etype == RCVHQ_RCV_TYPE_EAGER) { |
1187 | u8 opcode = be32_to_cpu(hdr->bth[0]) >> 24; | |
1188 | u32 qp = be32_to_cpu(hdr->bth[1]) & 0xffffff; | |
34b2aafe BS |
1189 | ipath_cdbg(PKT, "typ %x, opcode %x (eager, " |
1190 | "qp=%x), len %x; ignored\n", | |
9355fb6a | 1191 | etype, opcode, qp, tlen); |
7bb206e3 BS |
1192 | } |
1193 | else if (etype == RCVHQ_RCV_TYPE_EXPECTED) | |
1194 | ipath_dbg("Bug: Expected TID, opcode %x; ignored\n", | |
9355fb6a | 1195 | be32_to_cpu(hdr->bth[0]) >> 24); |
3d37b9e2 | 1196 | else { |
7bb206e3 | 1197 | /* |
9e2ef36b | 1198 | * error packet, type of error unknown. |
7bb206e3 BS |
1199 | * Probably type 3, but we don't know, so don't |
1200 | * even try to print the opcode, etc. | |
9355fb6a RC |
1201 | * Usually caused by a "bad packet", that has no |
1202 | * BTH, when the LRH says it should. | |
7bb206e3 | 1203 | */ |
9355fb6a RC |
1204 | ipath_cdbg(ERRPKT, "Error Pkt, but no eflags! egrbuf" |
1205 | " %x, len %x hdrq+%x rhf: %Lx\n", | |
1206 | etail, tlen, l, | |
1207 | le64_to_cpu(*(__le64 *) rhf_addr)); | |
1208 | if (ipath_debug & __IPATH_ERRPKTDBG) { | |
1209 | u32 j, *d, dw = rsize-2; | |
1210 | if (rsize > (tlen>>2)) | |
1211 | dw = tlen>>2; | |
1212 | d = (u32 *)hdr; | |
1213 | printk(KERN_DEBUG "EPkt rcvhdr(%x dw):\n", | |
1214 | dw); | |
1215 | for (j = 0; j < dw; j++) | |
1216 | printk(KERN_DEBUG "%8x%s", d[j], | |
1217 | (j%8) == 7 ? "\n" : " "); | |
1218 | printk(KERN_DEBUG ".\n"); | |
1219 | } | |
7bb206e3 BS |
1220 | } |
1221 | l += rsize; | |
1222 | if (l >= maxcnt) | |
1223 | l = 0; | |
9355fb6a RC |
1224 | rhf_addr = (__le32 *) pd->port_rcvhdrq + |
1225 | l + dd->ipath_rhf_offset; | |
1226 | if (dd->ipath_flags & IPATH_NODMA_RTAIL) { | |
1227 | u32 seq = ipath_hdrget_seq(rhf_addr); | |
1228 | ||
1229 | if (++pd->port_seq_cnt > 13) | |
1230 | pd->port_seq_cnt = 1; | |
1231 | if (seq != pd->port_seq_cnt) | |
1232 | last = 1; | |
1233 | } else if (l == hdrqtail) | |
1234 | last = 1; | |
7bb206e3 | 1235 | /* |
f5f99929 BS |
1236 | * update head regs on last packet, and every 16 packets. |
1237 | * Reduce bus traffic, while still trying to prevent | |
1238 | * rcvhdrq overflows, for when the queue is nearly full | |
7bb206e3 | 1239 | */ |
9355fb6a RC |
1240 | if (last || !(i & 0xf)) { |
1241 | u64 lval = l; | |
1242 | ||
1243 | /* request IBA6120 and 7220 interrupt only on last */ | |
1244 | if (last) | |
1245 | lval |= dd->ipath_rhdrhead_intr_off; | |
1246 | ipath_write_ureg(dd, ur_rcvhdrhead, lval, | |
1247 | pd->port_port); | |
f5f99929 | 1248 | if (updegr) { |
8c641d4b | 1249 | ipath_write_ureg(dd, ur_rcvegrindexhead, |
9355fb6a | 1250 | etail, pd->port_port); |
f5f99929 BS |
1251 | updegr = 0; |
1252 | } | |
1253 | } | |
7bb206e3 BS |
1254 | } |
1255 | ||
9355fb6a RC |
1256 | if (!dd->ipath_rhdrhead_intr_off && !reloop && |
1257 | !(dd->ipath_flags & IPATH_NODMA_RTAIL)) { | |
525d0ca1 | 1258 | /* IBA6110 workaround; we can have a race clearing chip |
57abad25 BS |
1259 | * interrupt with another interrupt about to be delivered, |
1260 | * and can clear it before it is delivered on the GPIO | |
1261 | * workaround. By doing the extra check here for the | |
1262 | * in-memory tail register updating while we were doing | |
1263 | * earlier packets, we "almost" guarantee we have covered | |
1264 | * that case. | |
1265 | */ | |
c59a80ac | 1266 | u32 hqtail = ipath_get_rcvhdrtail(pd); |
57abad25 BS |
1267 | if (hqtail != hdrqtail) { |
1268 | hdrqtail = hqtail; | |
1269 | reloop = 1; /* loop 1 extra time at most */ | |
1270 | goto reloop; | |
1271 | } | |
1272 | } | |
1273 | ||
7bb206e3 BS |
1274 | pkttot += i; |
1275 | ||
c59a80ac | 1276 | pd->port_head = l; |
7bb206e3 | 1277 | |
7bb206e3 BS |
1278 | if (pkttot > ipath_stats.sps_maxpkts_call) |
1279 | ipath_stats.sps_maxpkts_call = pkttot; | |
1280 | ipath_stats.sps_port0pkts += pkttot; | |
1281 | ipath_stats.sps_avgpkts_call = | |
1282 | ipath_stats.sps_port0pkts / ++totcalls; | |
1283 | ||
7bb206e3 BS |
1284 | bail:; |
1285 | } | |
1286 | ||
1287 | /** | |
1288 | * ipath_update_pio_bufs - update shadow copy of the PIO availability map | |
1289 | * @dd: the infinipath device | |
1290 | * | |
1291 | * called whenever our local copy indicates we have run out of send buffers | |
1292 | * NOTE: This can be called from interrupt context by some code | |
1293 | * and from non-interrupt context by ipath_getpiobuf(). | |
1294 | */ | |
1295 | ||
1296 | static void ipath_update_pio_bufs(struct ipath_devdata *dd) | |
1297 | { | |
1298 | unsigned long flags; | |
1299 | int i; | |
1300 | const unsigned piobregs = (unsigned)dd->ipath_pioavregs; | |
1301 | ||
1302 | /* If the generation (check) bits have changed, then we update the | |
1303 | * busy bit for the corresponding PIO buffer. This algorithm will | |
1304 | * modify positions to the value they already have in some cases | |
1305 | * (i.e., no change), but it's faster than changing only the bits | |
1306 | * that have changed. | |
1307 | * | |
1308 | * We would like to do this atomicly, to avoid spinlocks in the | |
1309 | * critical send path, but that's not really possible, given the | |
1310 | * type of changes, and that this routine could be called on | |
1311 | * multiple cpu's simultaneously, so we lock in this routine only, | |
1312 | * to avoid conflicting updates; all we change is the shadow, and | |
1313 | * it's a single 64 bit memory location, so by definition the update | |
1314 | * is atomic in terms of what other cpu's can see in testing the | |
1315 | * bits. The spin_lock overhead isn't too bad, since it only | |
1316 | * happens when all buffers are in use, so only cpu overhead, not | |
1317 | * latency or bandwidth is affected. | |
1318 | */ | |
7bb206e3 BS |
1319 | if (!dd->ipath_pioavailregs_dma) { |
1320 | ipath_dbg("Update shadow pioavail, but regs_dma NULL!\n"); | |
1321 | return; | |
1322 | } | |
1323 | if (ipath_debug & __IPATH_VERBDBG) { | |
1324 | /* only if packet debug and verbose */ | |
1325 | volatile __le64 *dma = dd->ipath_pioavailregs_dma; | |
1326 | unsigned long *shadow = dd->ipath_pioavailshadow; | |
1327 | ||
1328 | ipath_cdbg(PKT, "Refill avail, dma0=%llx shad0=%lx, " | |
1329 | "d1=%llx s1=%lx, d2=%llx s2=%lx, d3=%llx " | |
1330 | "s3=%lx\n", | |
1331 | (unsigned long long) le64_to_cpu(dma[0]), | |
1332 | shadow[0], | |
1333 | (unsigned long long) le64_to_cpu(dma[1]), | |
1334 | shadow[1], | |
1335 | (unsigned long long) le64_to_cpu(dma[2]), | |
1336 | shadow[2], | |
1337 | (unsigned long long) le64_to_cpu(dma[3]), | |
1338 | shadow[3]); | |
1339 | if (piobregs > 4) | |
1340 | ipath_cdbg( | |
1341 | PKT, "2nd group, dma4=%llx shad4=%lx, " | |
1342 | "d5=%llx s5=%lx, d6=%llx s6=%lx, " | |
1343 | "d7=%llx s7=%lx\n", | |
1344 | (unsigned long long) le64_to_cpu(dma[4]), | |
1345 | shadow[4], | |
1346 | (unsigned long long) le64_to_cpu(dma[5]), | |
1347 | shadow[5], | |
1348 | (unsigned long long) le64_to_cpu(dma[6]), | |
1349 | shadow[6], | |
1350 | (unsigned long long) le64_to_cpu(dma[7]), | |
1351 | shadow[7]); | |
1352 | } | |
1353 | spin_lock_irqsave(&ipath_pioavail_lock, flags); | |
1354 | for (i = 0; i < piobregs; i++) { | |
1355 | u64 pchbusy, pchg, piov, pnew; | |
1356 | /* | |
1357 | * Chip Errata: bug 6641; even and odd qwords>3 are swapped | |
1358 | */ | |
4ea61b54 RC |
1359 | if (i > 3 && (dd->ipath_flags & IPATH_SWAP_PIOBUFS)) |
1360 | piov = le64_to_cpu(dd->ipath_pioavailregs_dma[i ^ 1]); | |
1361 | else | |
7bb206e3 | 1362 | piov = le64_to_cpu(dd->ipath_pioavailregs_dma[i]); |
c4b4d16e | 1363 | pchg = dd->ipath_pioavailkernel[i] & |
7bb206e3 BS |
1364 | ~(dd->ipath_pioavailshadow[i] ^ piov); |
1365 | pchbusy = pchg << INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT; | |
1366 | if (pchg && (pchbusy & dd->ipath_pioavailshadow[i])) { | |
1367 | pnew = dd->ipath_pioavailshadow[i] & ~pchbusy; | |
1368 | pnew |= piov & pchbusy; | |
1369 | dd->ipath_pioavailshadow[i] = pnew; | |
1370 | } | |
1371 | } | |
1372 | spin_unlock_irqrestore(&ipath_pioavail_lock, flags); | |
1373 | } | |
1374 | ||
1375 | /** | |
1376 | * ipath_setrcvhdrsize - set the receive header size | |
1377 | * @dd: the infinipath device | |
1378 | * @rhdrsize: the receive header size | |
1379 | * | |
1380 | * called from user init code, and also layered driver init | |
1381 | */ | |
1382 | int ipath_setrcvhdrsize(struct ipath_devdata *dd, unsigned rhdrsize) | |
1383 | { | |
1384 | int ret = 0; | |
1385 | ||
1386 | if (dd->ipath_flags & IPATH_RCVHDRSZ_SET) { | |
1387 | if (dd->ipath_rcvhdrsize != rhdrsize) { | |
1388 | dev_info(&dd->pcidev->dev, | |
1389 | "Error: can't set protocol header " | |
1390 | "size %u, already %u\n", | |
1391 | rhdrsize, dd->ipath_rcvhdrsize); | |
1392 | ret = -EAGAIN; | |
1393 | } else | |
1394 | ipath_cdbg(VERBOSE, "Reuse same protocol header " | |
1395 | "size %u\n", dd->ipath_rcvhdrsize); | |
1396 | } else if (rhdrsize > (dd->ipath_rcvhdrentsize - | |
1397 | (sizeof(u64) / sizeof(u32)))) { | |
1398 | ipath_dbg("Error: can't set protocol header size %u " | |
1399 | "(> max %u)\n", rhdrsize, | |
1400 | dd->ipath_rcvhdrentsize - | |
1401 | (u32) (sizeof(u64) / sizeof(u32))); | |
1402 | ret = -EOVERFLOW; | |
1403 | } else { | |
1404 | dd->ipath_flags |= IPATH_RCVHDRSZ_SET; | |
1405 | dd->ipath_rcvhdrsize = rhdrsize; | |
1406 | ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvhdrsize, | |
1407 | dd->ipath_rcvhdrsize); | |
1408 | ipath_cdbg(VERBOSE, "Set protocol header size to %u\n", | |
1409 | dd->ipath_rcvhdrsize); | |
1410 | } | |
1411 | return ret; | |
1412 | } | |
1413 | ||
c4b4d16e RC |
1414 | /* |
1415 | * debugging code and stats updates if no pio buffers available. | |
1416 | */ | |
1417 | static noinline void no_pio_bufs(struct ipath_devdata *dd) | |
1418 | { | |
1419 | unsigned long *shadow = dd->ipath_pioavailshadow; | |
1420 | __le64 *dma = (__le64 *)dd->ipath_pioavailregs_dma; | |
1421 | ||
1422 | dd->ipath_upd_pio_shadow = 1; | |
1423 | ||
1424 | /* | |
1425 | * not atomic, but if we lose a stat count in a while, that's OK | |
1426 | */ | |
1427 | ipath_stats.sps_nopiobufs++; | |
1428 | if (!(++dd->ipath_consec_nopiobuf % 100000)) { | |
1429 | ipath_dbg("%u pio sends with no bufavail; dmacopy: " | |
1430 | "%llx %llx %llx %llx; shadow: %lx %lx %lx %lx\n", | |
1431 | dd->ipath_consec_nopiobuf, | |
1432 | (unsigned long long) le64_to_cpu(dma[0]), | |
1433 | (unsigned long long) le64_to_cpu(dma[1]), | |
1434 | (unsigned long long) le64_to_cpu(dma[2]), | |
1435 | (unsigned long long) le64_to_cpu(dma[3]), | |
1436 | shadow[0], shadow[1], shadow[2], shadow[3]); | |
1437 | /* | |
1438 | * 4 buffers per byte, 4 registers above, cover rest | |
1439 | * below | |
1440 | */ | |
1441 | if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) > | |
1442 | (sizeof(shadow[0]) * 4 * 4)) | |
1443 | ipath_dbg("2nd group: dmacopy: %llx %llx " | |
1444 | "%llx %llx; shadow: %lx %lx %lx %lx\n", | |
1445 | (unsigned long long)le64_to_cpu(dma[4]), | |
1446 | (unsigned long long)le64_to_cpu(dma[5]), | |
1447 | (unsigned long long)le64_to_cpu(dma[6]), | |
1448 | (unsigned long long)le64_to_cpu(dma[7]), | |
1449 | shadow[4], shadow[5], shadow[6], | |
1450 | shadow[7]); | |
1451 | } | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * common code for normal driver pio buffer allocation, and reserved | |
1456 | * allocation. | |
7bb206e3 BS |
1457 | * |
1458 | * do appropriate marking as busy, etc. | |
1459 | * returns buffer number if one found (>=0), negative number is error. | |
7bb206e3 | 1460 | */ |
c4b4d16e RC |
1461 | static u32 __iomem *ipath_getpiobuf_range(struct ipath_devdata *dd, |
1462 | u32 *pbufnum, u32 first, u32 last, u32 firsti) | |
7bb206e3 | 1463 | { |
c4b4d16e RC |
1464 | int i, j, updated = 0; |
1465 | unsigned piobcnt; | |
7bb206e3 BS |
1466 | unsigned long flags; |
1467 | unsigned long *shadow = dd->ipath_pioavailshadow; | |
1468 | u32 __iomem *buf; | |
1469 | ||
c4b4d16e | 1470 | piobcnt = last - first; |
7bb206e3 BS |
1471 | if (dd->ipath_upd_pio_shadow) { |
1472 | /* | |
1473 | * Minor optimization. If we had no buffers on last call, | |
1474 | * start out by doing the update; continue and do scan even | |
1475 | * if no buffers were updated, to be paranoid | |
1476 | */ | |
1477 | ipath_update_pio_bufs(dd); | |
c4b4d16e RC |
1478 | updated++; |
1479 | i = first; | |
7bb206e3 | 1480 | } else |
c4b4d16e | 1481 | i = firsti; |
7bb206e3 BS |
1482 | rescan: |
1483 | /* | |
1484 | * while test_and_set_bit() is atomic, we do that and then the | |
1485 | * change_bit(), and the pair is not. See if this is the cause | |
1486 | * of the remaining armlaunch errors. | |
1487 | */ | |
1488 | spin_lock_irqsave(&ipath_pioavail_lock, flags); | |
c4b4d16e RC |
1489 | for (j = 0; j < piobcnt; j++, i++) { |
1490 | if (i >= last) | |
1491 | i = first; | |
1492 | if (__test_and_set_bit((2 * i) + 1, shadow)) | |
7bb206e3 BS |
1493 | continue; |
1494 | /* flip generation bit */ | |
c4b4d16e | 1495 | __change_bit(2 * i, shadow); |
7bb206e3 BS |
1496 | break; |
1497 | } | |
1498 | spin_unlock_irqrestore(&ipath_pioavail_lock, flags); | |
1499 | ||
c4b4d16e | 1500 | if (j == piobcnt) { |
7bb206e3 | 1501 | if (!updated) { |
c4b4d16e RC |
1502 | /* |
1503 | * first time through; shadow exhausted, but may be | |
1504 | * buffers available, try an update and then rescan. | |
1505 | */ | |
7bb206e3 | 1506 | ipath_update_pio_bufs(dd); |
c4b4d16e RC |
1507 | updated++; |
1508 | i = first; | |
7bb206e3 | 1509 | goto rescan; |
c4b4d16e RC |
1510 | } else if (updated == 1 && piobcnt <= |
1511 | ((dd->ipath_sendctrl | |
1512 | >> INFINIPATH_S_UPDTHRESH_SHIFT) & | |
1513 | INFINIPATH_S_UPDTHRESH_MASK)) { | |
7bb206e3 | 1514 | /* |
c4b4d16e RC |
1515 | * for chips supporting and using the update |
1516 | * threshold we need to force an update of the | |
1517 | * in-memory copy if the count is less than the | |
1518 | * thershold, then check one more time. | |
7bb206e3 | 1519 | */ |
c4b4d16e RC |
1520 | ipath_force_pio_avail_update(dd); |
1521 | ipath_update_pio_bufs(dd); | |
1522 | updated++; | |
1523 | i = first; | |
1524 | goto rescan; | |
7bb206e3 | 1525 | } |
c4b4d16e RC |
1526 | |
1527 | no_pio_bufs(dd); | |
7bb206e3 | 1528 | buf = NULL; |
c4b4d16e RC |
1529 | } else { |
1530 | if (i < dd->ipath_piobcnt2k) | |
1531 | buf = (u32 __iomem *) (dd->ipath_pio2kbase + | |
1532 | i * dd->ipath_palign); | |
1533 | else | |
1534 | buf = (u32 __iomem *) | |
1535 | (dd->ipath_pio4kbase + | |
1536 | (i - dd->ipath_piobcnt2k) * dd->ipath_4kalign); | |
1537 | if (pbufnum) | |
1538 | *pbufnum = i; | |
7bb206e3 BS |
1539 | } |
1540 | ||
c4b4d16e RC |
1541 | return buf; |
1542 | } | |
7bb206e3 | 1543 | |
c4b4d16e RC |
1544 | /** |
1545 | * ipath_getpiobuf - find an available pio buffer | |
1546 | * @dd: the infinipath device | |
1547 | * @plen: the size of the PIO buffer needed in 32-bit words | |
1548 | * @pbufnum: the buffer number is placed here | |
1549 | */ | |
1550 | u32 __iomem *ipath_getpiobuf(struct ipath_devdata *dd, u32 plen, u32 *pbufnum) | |
1551 | { | |
1552 | u32 __iomem *buf; | |
1553 | u32 pnum, nbufs; | |
1554 | u32 first, lasti; | |
1555 | ||
1556 | if (plen + 1 >= IPATH_SMALLBUF_DWORDS) { | |
1557 | first = dd->ipath_piobcnt2k; | |
1558 | lasti = dd->ipath_lastpioindexl; | |
1559 | } else { | |
1560 | first = 0; | |
1561 | lasti = dd->ipath_lastpioindex; | |
1562 | } | |
1563 | nbufs = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k; | |
1564 | buf = ipath_getpiobuf_range(dd, &pnum, first, nbufs, lasti); | |
1565 | ||
1566 | if (buf) { | |
1567 | /* | |
1568 | * Set next starting place. It's just an optimization, | |
1569 | * it doesn't matter who wins on this, so no locking | |
1570 | */ | |
1571 | if (plen + 1 >= IPATH_SMALLBUF_DWORDS) | |
1572 | dd->ipath_lastpioindexl = pnum + 1; | |
1573 | else | |
1574 | dd->ipath_lastpioindex = pnum + 1; | |
1575 | if (dd->ipath_upd_pio_shadow) | |
1576 | dd->ipath_upd_pio_shadow = 0; | |
1577 | if (dd->ipath_consec_nopiobuf) | |
1578 | dd->ipath_consec_nopiobuf = 0; | |
1579 | ipath_cdbg(VERBOSE, "Return piobuf%u %uk @ %p\n", | |
1580 | pnum, (pnum < dd->ipath_piobcnt2k) ? 2 : 4, buf); | |
1581 | if (pbufnum) | |
1582 | *pbufnum = pnum; | |
1583 | ||
1584 | } | |
7bb206e3 BS |
1585 | return buf; |
1586 | } | |
1587 | ||
c4b4d16e RC |
1588 | /** |
1589 | * ipath_chg_pioavailkernel - change which send buffers are available for kernel | |
1590 | * @dd: the infinipath device | |
1591 | * @start: the starting send buffer number | |
1592 | * @len: the number of send buffers | |
1593 | * @avail: true if the buffers are available for kernel use, false otherwise | |
1594 | */ | |
1595 | void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start, | |
1596 | unsigned len, int avail) | |
1597 | { | |
1598 | unsigned long flags; | |
1599 | unsigned end; | |
1600 | ||
1601 | /* There are two bits per send buffer (busy and generation) */ | |
1602 | start *= 2; | |
1603 | len *= 2; | |
1604 | end = start + len; | |
1605 | ||
1606 | /* Set or clear the generation bits. */ | |
1607 | spin_lock_irqsave(&ipath_pioavail_lock, flags); | |
1608 | while (start < end) { | |
1609 | if (avail) { | |
1610 | __clear_bit(start + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT, | |
1611 | dd->ipath_pioavailshadow); | |
1612 | __set_bit(start, dd->ipath_pioavailkernel); | |
1613 | } else { | |
1614 | __set_bit(start + INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT, | |
1615 | dd->ipath_pioavailshadow); | |
1616 | __clear_bit(start, dd->ipath_pioavailkernel); | |
1617 | } | |
1618 | start += 2; | |
1619 | } | |
1620 | spin_unlock_irqrestore(&ipath_pioavail_lock, flags); | |
1621 | } | |
1622 | ||
7bb206e3 BS |
1623 | /** |
1624 | * ipath_create_rcvhdrq - create a receive header queue | |
1625 | * @dd: the infinipath device | |
1626 | * @pd: the port data | |
1627 | * | |
f37bda92 BS |
1628 | * this must be contiguous memory (from an i/o perspective), and must be |
1629 | * DMA'able (which means for some systems, it will go through an IOMMU, | |
1630 | * or be forced into a low address range). | |
7bb206e3 BS |
1631 | */ |
1632 | int ipath_create_rcvhdrq(struct ipath_devdata *dd, | |
1633 | struct ipath_portdata *pd) | |
1634 | { | |
f37bda92 | 1635 | int ret = 0; |
7bb206e3 | 1636 | |
7bb206e3 | 1637 | if (!pd->port_rcvhdrq) { |
f37bda92 | 1638 | dma_addr_t phys_hdrqtail; |
7bb206e3 | 1639 | gfp_t gfp_flags = GFP_USER | __GFP_COMP; |
f37bda92 BS |
1640 | int amt = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize * |
1641 | sizeof(u32), PAGE_SIZE); | |
7bb206e3 BS |
1642 | |
1643 | pd->port_rcvhdrq = dma_alloc_coherent( | |
1644 | &dd->pcidev->dev, amt, &pd->port_rcvhdrq_phys, | |
1645 | gfp_flags); | |
1646 | ||
1647 | if (!pd->port_rcvhdrq) { | |
1648 | ipath_dev_err(dd, "attempt to allocate %d bytes " | |
1649 | "for port %u rcvhdrq failed\n", | |
1650 | amt, pd->port_port); | |
1651 | ret = -ENOMEM; | |
1652 | goto bail; | |
1653 | } | |
9355fb6a RC |
1654 | |
1655 | if (!(dd->ipath_flags & IPATH_NODMA_RTAIL)) { | |
1656 | pd->port_rcvhdrtail_kvaddr = dma_alloc_coherent( | |
1657 | &dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail, | |
1658 | GFP_KERNEL); | |
1659 | if (!pd->port_rcvhdrtail_kvaddr) { | |
1660 | ipath_dev_err(dd, "attempt to allocate 1 page " | |
1661 | "for port %u rcvhdrqtailaddr " | |
1662 | "failed\n", pd->port_port); | |
1663 | ret = -ENOMEM; | |
1664 | dma_free_coherent(&dd->pcidev->dev, amt, | |
1665 | pd->port_rcvhdrq, | |
1666 | pd->port_rcvhdrq_phys); | |
1667 | pd->port_rcvhdrq = NULL; | |
1668 | goto bail; | |
1669 | } | |
1670 | pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail; | |
1671 | ipath_cdbg(VERBOSE, "port %d hdrtailaddr, %llx " | |
1672 | "physical\n", pd->port_port, | |
1673 | (unsigned long long) phys_hdrqtail); | |
f37bda92 | 1674 | } |
7bb206e3 BS |
1675 | |
1676 | pd->port_rcvhdrq_size = amt; | |
1677 | ||
1678 | ipath_cdbg(VERBOSE, "%d pages at %p (phys %lx) size=%lu " | |
1679 | "for port %u rcvhdr Q\n", | |
1680 | amt >> PAGE_SHIFT, pd->port_rcvhdrq, | |
1681 | (unsigned long) pd->port_rcvhdrq_phys, | |
1682 | (unsigned long) pd->port_rcvhdrq_size, | |
1683 | pd->port_port); | |
7bb206e3 | 1684 | } |
f37bda92 BS |
1685 | else |
1686 | ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; " | |
1687 | "hdrtailaddr@%p %llx physical\n", | |
1688 | pd->port_port, pd->port_rcvhdrq, | |
1fd3b40f BS |
1689 | (unsigned long long) pd->port_rcvhdrq_phys, |
1690 | pd->port_rcvhdrtail_kvaddr, (unsigned long long) | |
1691 | pd->port_rcvhdrqtailaddr_phys); | |
f37bda92 BS |
1692 | |
1693 | /* clear for security and sanity on each use */ | |
1694 | memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size); | |
c59a80ac RC |
1695 | if (pd->port_rcvhdrtail_kvaddr) |
1696 | memset(pd->port_rcvhdrtail_kvaddr, 0, PAGE_SIZE); | |
7bb206e3 BS |
1697 | |
1698 | /* | |
1699 | * tell chip each time we init it, even if we are re-using previous | |
f37bda92 | 1700 | * memory (we zero the register at process close) |
7bb206e3 | 1701 | */ |
f37bda92 BS |
1702 | ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdrtailaddr, |
1703 | pd->port_port, pd->port_rcvhdrqtailaddr_phys); | |
7bb206e3 BS |
1704 | ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr, |
1705 | pd->port_port, pd->port_rcvhdrq_phys); | |
1706 | ||
7bb206e3 BS |
1707 | bail: |
1708 | return ret; | |
1709 | } | |
1710 | ||
9380068f DO |
1711 | |
1712 | /* | |
1713 | * Flush all sends that might be in the ready to send state, as well as any | |
1714 | * that are in the process of being sent. Used whenever we need to be | |
1715 | * sure the send side is idle. Cleans up all buffer state by canceling | |
1716 | * all pio buffers, and issuing an abort, which cleans up anything in the | |
1717 | * launch fifo. The cancel is superfluous on some chip versions, but | |
1718 | * it's safer to always do it. | |
1719 | * PIOAvail bits are updated by the chip as if normal send had happened. | |
1720 | */ | |
3810f2a8 | 1721 | void ipath_cancel_sends(struct ipath_devdata *dd, int restore_sendctrl) |
9380068f DO |
1722 | { |
1723 | ipath_dbg("Cancelling all in-progress send buffers\n"); | |
2ba3f56e RC |
1724 | |
1725 | /* skip armlaunch errs for a while */ | |
1726 | dd->ipath_lastcancel = jiffies + HZ / 2; | |
1727 | ||
9380068f DO |
1728 | /* |
1729 | * the abort bit is auto-clearing. We read scratch to be sure | |
1730 | * that cancels and the abort have taken effect in the chip. | |
1731 | */ | |
1732 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, | |
1733 | INFINIPATH_S_ABORT); | |
1734 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); | |
1735 | ipath_disarm_piobufs(dd, 0, | |
1736 | (unsigned)(dd->ipath_piobcnt2k + dd->ipath_piobcnt4k)); | |
3810f2a8 DO |
1737 | if (restore_sendctrl) /* else done by caller later */ |
1738 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, | |
1739 | dd->ipath_sendctrl); | |
9380068f DO |
1740 | |
1741 | /* and again, be sure all have hit the chip */ | |
1742 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); | |
1743 | } | |
1744 | ||
c4b4d16e RC |
1745 | /* |
1746 | * Force an update of in-memory copy of the pioavail registers, when | |
1747 | * needed for any of a variety of reasons. We read the scratch register | |
1748 | * to make it highly likely that the update will have happened by the | |
1749 | * time we return. If already off (as in cancel_sends above), this | |
1750 | * routine is a nop, on the assumption that the caller will "do the | |
1751 | * right thing". | |
1752 | */ | |
1753 | void ipath_force_pio_avail_update(struct ipath_devdata *dd) | |
1754 | { | |
1755 | unsigned long flags; | |
1756 | ||
1757 | spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); | |
1758 | if (dd->ipath_sendctrl & INFINIPATH_S_PIOBUFAVAILUPD) { | |
1759 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, | |
1760 | dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD); | |
1761 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); | |
1762 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, | |
1763 | dd->ipath_sendctrl); | |
1764 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); | |
1765 | } | |
1766 | spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); | |
1767 | } | |
1768 | ||
4330e4da MA |
1769 | static void ipath_set_ib_lstate(struct ipath_devdata *dd, int linkcmd, |
1770 | int linitcmd) | |
7bb206e3 | 1771 | { |
4330e4da | 1772 | u64 mod_wd; |
7bb206e3 | 1773 | static const char *what[4] = { |
140277e9 RC |
1774 | [0] = "NOP", |
1775 | [INFINIPATH_IBCC_LINKCMD_DOWN] = "DOWN", | |
7bb206e3 BS |
1776 | [INFINIPATH_IBCC_LINKCMD_ARMED] = "ARMED", |
1777 | [INFINIPATH_IBCC_LINKCMD_ACTIVE] = "ACTIVE" | |
1778 | }; | |
f37bda92 | 1779 | |
4330e4da MA |
1780 | if (linitcmd == INFINIPATH_IBCC_LINKINITCMD_DISABLE) { |
1781 | /* | |
1782 | * If we are told to disable, note that so link-recovery | |
1783 | * code does not attempt to bring us back up. | |
1784 | */ | |
1785 | preempt_disable(); | |
1786 | dd->ipath_flags |= IPATH_IB_LINK_DISABLED; | |
1787 | preempt_enable(); | |
1788 | } else if (linitcmd) { | |
1789 | /* | |
1790 | * Any other linkinitcmd will lead to LINKDOWN and then | |
1791 | * to INIT (if all is well), so clear flag to let | |
1792 | * link-recovery code attempt to bring us back up. | |
1793 | */ | |
1794 | preempt_disable(); | |
1795 | dd->ipath_flags &= ~IPATH_IB_LINK_DISABLED; | |
1796 | preempt_enable(); | |
1797 | } | |
1798 | ||
1799 | mod_wd = (linkcmd << dd->ibcc_lc_shift) | | |
1800 | (linitcmd << INFINIPATH_IBCC_LINKINITCMD_SHIFT); | |
1801 | ipath_cdbg(VERBOSE, | |
1802 | "Moving unit %u to %s (initcmd=0x%x), current ltstate is %s\n", | |
1803 | dd->ipath_unit, what[linkcmd], linitcmd, | |
1804 | ipath_ibcstatus_str[ipath_ib_linktrstate(dd, | |
58411d1c | 1805 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus))]); |
7bb206e3 BS |
1806 | |
1807 | ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, | |
4330e4da MA |
1808 | dd->ipath_ibcctrl | mod_wd); |
1809 | /* read from chip so write is flushed */ | |
1810 | (void) ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); | |
7bb206e3 BS |
1811 | } |
1812 | ||
34b2aafe BS |
1813 | int ipath_set_linkstate(struct ipath_devdata *dd, u8 newstate) |
1814 | { | |
1815 | u32 lstate; | |
1816 | int ret; | |
1817 | ||
1818 | switch (newstate) { | |
140277e9 | 1819 | case IPATH_IB_LINKDOWN_ONLY: |
4330e4da | 1820 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, 0); |
140277e9 RC |
1821 | /* don't wait */ |
1822 | ret = 0; | |
1823 | goto bail; | |
1824 | ||
34b2aafe | 1825 | case IPATH_IB_LINKDOWN: |
4330e4da MA |
1826 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, |
1827 | INFINIPATH_IBCC_LINKINITCMD_POLL); | |
34b2aafe BS |
1828 | /* don't wait */ |
1829 | ret = 0; | |
1830 | goto bail; | |
1831 | ||
1832 | case IPATH_IB_LINKDOWN_SLEEP: | |
4330e4da MA |
1833 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, |
1834 | INFINIPATH_IBCC_LINKINITCMD_SLEEP); | |
34b2aafe BS |
1835 | /* don't wait */ |
1836 | ret = 0; | |
1837 | goto bail; | |
1838 | ||
1839 | case IPATH_IB_LINKDOWN_DISABLE: | |
4330e4da MA |
1840 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_DOWN, |
1841 | INFINIPATH_IBCC_LINKINITCMD_DISABLE); | |
34b2aafe BS |
1842 | /* don't wait */ |
1843 | ret = 0; | |
1844 | goto bail; | |
1845 | ||
34b2aafe BS |
1846 | case IPATH_IB_LINKARM: |
1847 | if (dd->ipath_flags & IPATH_LINKARMED) { | |
1848 | ret = 0; | |
1849 | goto bail; | |
1850 | } | |
1851 | if (!(dd->ipath_flags & | |
1852 | (IPATH_LINKINIT | IPATH_LINKACTIVE))) { | |
1853 | ret = -EINVAL; | |
1854 | goto bail; | |
1855 | } | |
4330e4da MA |
1856 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED, 0); |
1857 | ||
34b2aafe BS |
1858 | /* |
1859 | * Since the port can transition to ACTIVE by receiving | |
1860 | * a non VL 15 packet, wait for either state. | |
1861 | */ | |
1862 | lstate = IPATH_LINKARMED | IPATH_LINKACTIVE; | |
1863 | break; | |
1864 | ||
1865 | case IPATH_IB_LINKACTIVE: | |
1866 | if (dd->ipath_flags & IPATH_LINKACTIVE) { | |
1867 | ret = 0; | |
1868 | goto bail; | |
1869 | } | |
1870 | if (!(dd->ipath_flags & IPATH_LINKARMED)) { | |
1871 | ret = -EINVAL; | |
1872 | goto bail; | |
1873 | } | |
4330e4da | 1874 | ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE, 0); |
34b2aafe BS |
1875 | lstate = IPATH_LINKACTIVE; |
1876 | break; | |
1877 | ||
946db67f BS |
1878 | case IPATH_IB_LINK_LOOPBACK: |
1879 | dev_info(&dd->pcidev->dev, "Enabling IB local loopback\n"); | |
1880 | dd->ipath_ibcctrl |= INFINIPATH_IBCC_LOOPBACK; | |
1881 | ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, | |
1882 | dd->ipath_ibcctrl); | |
b3e8f541 DO |
1883 | |
1884 | /* turn heartbeat off, as it causes loopback to fail */ | |
1885 | dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, | |
1886 | IPATH_IB_HRTBT_OFF); | |
1887 | /* don't wait */ | |
946db67f | 1888 | ret = 0; |
b3e8f541 | 1889 | goto bail; |
946db67f BS |
1890 | |
1891 | case IPATH_IB_LINK_EXTERNAL: | |
b3e8f541 DO |
1892 | dev_info(&dd->pcidev->dev, |
1893 | "Disabling IB local loopback (normal)\n"); | |
1894 | dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, | |
1895 | IPATH_IB_HRTBT_ON); | |
946db67f BS |
1896 | dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LOOPBACK; |
1897 | ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, | |
1898 | dd->ipath_ibcctrl); | |
b3e8f541 | 1899 | /* don't wait */ |
946db67f | 1900 | ret = 0; |
b3e8f541 DO |
1901 | goto bail; |
1902 | ||
1903 | /* | |
1904 | * Heartbeat can be explicitly enabled by the user via | |
1905 | * "hrtbt_enable" "file", and if disabled, trying to enable here | |
1906 | * will have no effect. Implicit changes (heartbeat off when | |
1907 | * loopback on, and vice versa) are included to ease testing. | |
1908 | */ | |
1909 | case IPATH_IB_LINK_HRTBT: | |
1910 | ret = dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, | |
1911 | IPATH_IB_HRTBT_ON); | |
1912 | goto bail; | |
1913 | ||
1914 | case IPATH_IB_LINK_NO_HRTBT: | |
1915 | ret = dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_HRTBT, | |
1916 | IPATH_IB_HRTBT_OFF); | |
1917 | goto bail; | |
946db67f | 1918 | |
34b2aafe BS |
1919 | default: |
1920 | ipath_dbg("Invalid linkstate 0x%x requested\n", newstate); | |
1921 | ret = -EINVAL; | |
1922 | goto bail; | |
1923 | } | |
1924 | ret = ipath_wait_linkstate(dd, lstate, 2000); | |
1925 | ||
1926 | bail: | |
1927 | return ret; | |
1928 | } | |
1929 | ||
1930 | /** | |
1931 | * ipath_set_mtu - set the MTU | |
1932 | * @dd: the infinipath device | |
1933 | * @arg: the new MTU | |
1934 | * | |
1935 | * we can handle "any" incoming size, the issue here is whether we | |
1936 | * need to restrict our outgoing size. For now, we don't do any | |
1937 | * sanity checking on this, and we don't deal with what happens to | |
1938 | * programs that are already running when the size changes. | |
1939 | * NOTE: changing the MTU will usually cause the IBC to go back to | |
1940 | * link initialize (IPATH_IBSTATE_INIT) state... | |
1941 | */ | |
1942 | int ipath_set_mtu(struct ipath_devdata *dd, u16 arg) | |
1943 | { | |
1944 | u32 piosize; | |
1945 | int changed = 0; | |
1946 | int ret; | |
1947 | ||
1948 | /* | |
1949 | * mtu is IB data payload max. It's the largest power of 2 less | |
1950 | * than piosize (or even larger, since it only really controls the | |
1951 | * largest we can receive; we can send the max of the mtu and | |
1952 | * piosize). We check that it's one of the valid IB sizes. | |
1953 | */ | |
1954 | if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 && | |
826d8010 | 1955 | (arg != 4096 || !ipath_mtu4096)) { |
34b2aafe BS |
1956 | ipath_dbg("Trying to set invalid mtu %u, failing\n", arg); |
1957 | ret = -EINVAL; | |
1958 | goto bail; | |
1959 | } | |
1960 | if (dd->ipath_ibmtu == arg) { | |
1961 | ret = 0; /* same as current */ | |
1962 | goto bail; | |
1963 | } | |
1964 | ||
1965 | piosize = dd->ipath_ibmaxlen; | |
1966 | dd->ipath_ibmtu = arg; | |
1967 | ||
1968 | if (arg >= (piosize - IPATH_PIO_MAXIBHDR)) { | |
1969 | /* Only if it's not the initial value (or reset to it) */ | |
1970 | if (piosize != dd->ipath_init_ibmaxlen) { | |
826d8010 DO |
1971 | if (arg > piosize && arg <= dd->ipath_init_ibmaxlen) |
1972 | piosize = dd->ipath_init_ibmaxlen; | |
34b2aafe BS |
1973 | dd->ipath_ibmaxlen = piosize; |
1974 | changed = 1; | |
1975 | } | |
1976 | } else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) { | |
1977 | piosize = arg + IPATH_PIO_MAXIBHDR; | |
1978 | ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x " | |
1979 | "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize, | |
1980 | arg); | |
1981 | dd->ipath_ibmaxlen = piosize; | |
1982 | changed = 1; | |
1983 | } | |
1984 | ||
1985 | if (changed) { | |
826d8010 | 1986 | u64 ibc = dd->ipath_ibcctrl, ibdw; |
34b2aafe | 1987 | /* |
826d8010 DO |
1988 | * update our housekeeping variables, and set IBC max |
1989 | * size, same as init code; max IBC is max we allow in | |
1990 | * buffer, less the qword pbc, plus 1 for ICRC, in dwords | |
34b2aafe | 1991 | */ |
826d8010 DO |
1992 | dd->ipath_ibmaxlen = piosize - 2 * sizeof(u32); |
1993 | ibdw = (dd->ipath_ibmaxlen >> 2) + 1; | |
34b2aafe | 1994 | ibc &= ~(INFINIPATH_IBCC_MAXPKTLEN_MASK << |
826d8010 DO |
1995 | dd->ibcc_mpl_shift); |
1996 | ibc |= ibdw << dd->ibcc_mpl_shift; | |
34b2aafe BS |
1997 | dd->ipath_ibcctrl = ibc; |
1998 | ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl, | |
1999 | dd->ipath_ibcctrl); | |
2000 | dd->ipath_f_tidtemplate(dd); | |
2001 | } | |
2002 | ||
2003 | ret = 0; | |
2004 | ||
2005 | bail: | |
2006 | return ret; | |
2007 | } | |
2008 | ||
0ab6b2b9 | 2009 | int ipath_set_lid(struct ipath_devdata *dd, u32 lid, u8 lmc) |
34b2aafe | 2010 | { |
0ab6b2b9 | 2011 | dd->ipath_lid = lid; |
34b2aafe BS |
2012 | dd->ipath_lmc = lmc; |
2013 | ||
0ab6b2b9 DO |
2014 | dd->ipath_f_set_ib_cfg(dd, IPATH_IB_CFG_LIDLMC, lid | |
2015 | (~((1U << lmc) - 1)) << 16); | |
2016 | ||
2017 | dev_info(&dd->pcidev->dev, "We got a lid: 0x%x\n", lid); | |
2018 | ||
34b2aafe BS |
2019 | return 0; |
2020 | } | |
2021 | ||
7bb206e3 BS |
2022 | |
2023 | /** | |
2024 | * ipath_write_kreg_port - write a device's per-port 64-bit kernel register | |
2025 | * @dd: the infinipath device | |
2026 | * @regno: the register number to write | |
2027 | * @port: the port containing the register | |
2028 | * @value: the value to write | |
2029 | * | |
2030 | * Registers that vary with the chip implementation constants (port) | |
2031 | * use this routine. | |
2032 | */ | |
2033 | void ipath_write_kreg_port(const struct ipath_devdata *dd, ipath_kreg regno, | |
2034 | unsigned port, u64 value) | |
2035 | { | |
2036 | u16 where; | |
2037 | ||
2038 | if (port < dd->ipath_portcnt && | |
2039 | (regno == dd->ipath_kregs->kr_rcvhdraddr || | |
2040 | regno == dd->ipath_kregs->kr_rcvhdrtailaddr)) | |
2041 | where = regno + port; | |
2042 | else | |
2043 | where = -1; | |
2044 | ||
2045 | ipath_write_kreg(dd, where, value); | |
2046 | } | |
2047 | ||
82466f00 MA |
2048 | /* |
2049 | * Following deal with the "obviously simple" task of overriding the state | |
2050 | * of the LEDS, which normally indicate link physical and logical status. | |
2051 | * The complications arise in dealing with different hardware mappings | |
2052 | * and the board-dependent routine being called from interrupts. | |
2053 | * and then there's the requirement to _flash_ them. | |
2054 | */ | |
2055 | #define LED_OVER_FREQ_SHIFT 8 | |
2056 | #define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT) | |
2057 | /* Below is "non-zero" to force override, but both actual LEDs are off */ | |
2058 | #define LED_OVER_BOTH_OFF (8) | |
2059 | ||
da9aec7b | 2060 | static void ipath_run_led_override(unsigned long opaque) |
82466f00 MA |
2061 | { |
2062 | struct ipath_devdata *dd = (struct ipath_devdata *)opaque; | |
2063 | int timeoff; | |
2064 | int pidx; | |
2065 | u64 lstate, ltstate, val; | |
2066 | ||
2067 | if (!(dd->ipath_flags & IPATH_INITTED)) | |
2068 | return; | |
2069 | ||
2070 | pidx = dd->ipath_led_override_phase++ & 1; | |
2071 | dd->ipath_led_override = dd->ipath_led_override_vals[pidx]; | |
2072 | timeoff = dd->ipath_led_override_timeoff; | |
2073 | ||
2074 | /* | |
2075 | * below potentially restores the LED values per current status, | |
2076 | * should also possibly setup the traffic-blink register, | |
2077 | * but leave that to per-chip functions. | |
2078 | */ | |
2079 | val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus); | |
2080 | ltstate = (val >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) & | |
58411d1c JG |
2081 | dd->ibcs_lts_mask; |
2082 | lstate = (val >> dd->ibcs_ls_shift) & INFINIPATH_IBCS_LINKSTATE_MASK; | |
82466f00 MA |
2083 | |
2084 | dd->ipath_f_setextled(dd, lstate, ltstate); | |
2085 | mod_timer(&dd->ipath_led_override_timer, jiffies + timeoff); | |
2086 | } | |
2087 | ||
2088 | void ipath_set_led_override(struct ipath_devdata *dd, unsigned int val) | |
2089 | { | |
2090 | int timeoff, freq; | |
2091 | ||
2092 | if (!(dd->ipath_flags & IPATH_INITTED)) | |
2093 | return; | |
2094 | ||
2095 | /* First check if we are blinking. If not, use 1HZ polling */ | |
2096 | timeoff = HZ; | |
2097 | freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT; | |
2098 | ||
2099 | if (freq) { | |
2100 | /* For blink, set each phase from one nybble of val */ | |
2101 | dd->ipath_led_override_vals[0] = val & 0xF; | |
2102 | dd->ipath_led_override_vals[1] = (val >> 4) & 0xF; | |
2103 | timeoff = (HZ << 4)/freq; | |
2104 | } else { | |
2105 | /* Non-blink set both phases the same. */ | |
2106 | dd->ipath_led_override_vals[0] = val & 0xF; | |
2107 | dd->ipath_led_override_vals[1] = val & 0xF; | |
2108 | } | |
2109 | dd->ipath_led_override_timeoff = timeoff; | |
2110 | ||
2111 | /* | |
2112 | * If the timer has not already been started, do so. Use a "quick" | |
2113 | * timeout so the function will be called soon, to look at our request. | |
2114 | */ | |
2115 | if (atomic_inc_return(&dd->ipath_led_override_timer_active) == 1) { | |
2116 | /* Need to start timer */ | |
2117 | init_timer(&dd->ipath_led_override_timer); | |
2118 | dd->ipath_led_override_timer.function = | |
2119 | ipath_run_led_override; | |
2120 | dd->ipath_led_override_timer.data = (unsigned long) dd; | |
2121 | dd->ipath_led_override_timer.expires = jiffies + 1; | |
2122 | add_timer(&dd->ipath_led_override_timer); | |
2ba3f56e | 2123 | } else |
82466f00 | 2124 | atomic_dec(&dd->ipath_led_override_timer_active); |
82466f00 MA |
2125 | } |
2126 | ||
7bb206e3 BS |
2127 | /** |
2128 | * ipath_shutdown_device - shut down a device | |
2129 | * @dd: the infinipath device | |
2130 | * | |
2131 | * This is called to make the device quiet when we are about to | |
2132 | * unload the driver, and also when the device is administratively | |
2133 | * disabled. It does not free any data structures. | |
2134 | * Everything it does has to be setup again by ipath_init_chip(dd,1) | |
2135 | */ | |
2136 | void ipath_shutdown_device(struct ipath_devdata *dd) | |
2137 | { | |
e342c119 JG |
2138 | unsigned long flags; |
2139 | ||
7bb206e3 BS |
2140 | ipath_dbg("Shutting down the device\n"); |
2141 | ||
58411d1c JG |
2142 | ipath_hol_up(dd); /* make sure user processes aren't suspended */ |
2143 | ||
7bb206e3 BS |
2144 | dd->ipath_flags |= IPATH_LINKUNK; |
2145 | dd->ipath_flags &= ~(IPATH_INITTED | IPATH_LINKDOWN | | |
2146 | IPATH_LINKINIT | IPATH_LINKARMED | | |
2147 | IPATH_LINKACTIVE); | |
2148 | *dd->ipath_statusp &= ~(IPATH_STATUS_IB_CONF | | |
2149 | IPATH_STATUS_IB_READY); | |
2150 | ||
2151 | /* mask interrupts, but not errors */ | |
2152 | ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, 0ULL); | |
2153 | ||
2154 | dd->ipath_rcvctrl = 0; | |
2155 | ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl, | |
2156 | dd->ipath_rcvctrl); | |
2157 | ||
2158 | /* | |
2159 | * gracefully stop all sends allowing any in progress to trickle out | |
2160 | * first. | |
2161 | */ | |
e342c119 JG |
2162 | spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags); |
2163 | dd->ipath_sendctrl = 0; | |
2164 | ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl); | |
7bb206e3 | 2165 | /* flush it */ |
44f8e3f3 | 2166 | ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch); |
e342c119 JG |
2167 | spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags); |
2168 | ||
7bb206e3 BS |
2169 | /* |
2170 | * enough for anything that's going to trickle out to have actually | |
2171 | * done so. | |
2172 | */ | |
2173 | udelay(5); | |
2174 | ||
4330e4da | 2175 | ipath_set_ib_lstate(dd, 0, INFINIPATH_IBCC_LINKINITCMD_DISABLE); |
3810f2a8 | 2176 | ipath_cancel_sends(dd, 0); |
7bb206e3 | 2177 | |
49739b3e RC |
2178 | signal_ib_event(dd, IB_EVENT_PORT_ERR); |
2179 | ||
7bb206e3 BS |
2180 | /* disable IBC */ |
2181 | dd->ipath_control &= ~INFINIPATH_C_LINKENABLE; | |
2182 | ipath_write_kreg(dd, dd->ipath_kregs->kr_control, | |
a40f55fc | 2183 | dd->ipath_control | INFINIPATH_C_FREEZEMODE); |
7bb206e3 BS |
2184 | |
2185 | /* | |
2186 | * clear SerdesEnable and turn the leds off; do this here because | |
2187 | * we are unloading, so don't count on interrupts to move along | |
2188 | * Turn the LEDs off explictly for the same reason. | |
2189 | */ | |
2190 | dd->ipath_f_quiet_serdes(dd); | |
7bb206e3 | 2191 | |
58411d1c JG |
2192 | /* stop all the timers that might still be running */ |
2193 | del_timer_sync(&dd->ipath_hol_timer); | |
7bb206e3 BS |
2194 | if (dd->ipath_stats_timer_active) { |
2195 | del_timer_sync(&dd->ipath_stats_timer); | |
2196 | dd->ipath_stats_timer_active = 0; | |
2197 | } | |
2198 | ||
2199 | /* | |
2200 | * clear all interrupts and errors, so that the next time the driver | |
2201 | * is loaded or device is enabled, we know that whatever is set | |
2202 | * happened while we were unloaded | |
2203 | */ | |
2204 | ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, | |
2205 | ~0ULL & ~INFINIPATH_HWE_MEMBISTFAILED); | |
2206 | ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, -1LL); | |
2207 | ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, -1LL); | |
aecd3b5a MA |
2208 | |
2209 | ipath_cdbg(VERBOSE, "Flush time and errors to EEPROM\n"); | |
2210 | ipath_update_eeprom_log(dd); | |
7bb206e3 BS |
2211 | } |
2212 | ||
2213 | /** | |
2214 | * ipath_free_pddata - free a port's allocated data | |
2215 | * @dd: the infinipath device | |
f37bda92 | 2216 | * @pd: the portdata structure |
7bb206e3 | 2217 | * |
f37bda92 BS |
2218 | * free up any allocated data for a port |
2219 | * This should not touch anything that would affect a simultaneous | |
2220 | * re-allocation of port data, because it is called after ipath_mutex | |
2221 | * is released (and can be called from reinit as well). | |
2222 | * It should never change any chip state, or global driver state. | |
2223 | * (The only exception to global state is freeing the port0 port0_skbs.) | |
7bb206e3 | 2224 | */ |
f37bda92 | 2225 | void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd) |
7bb206e3 | 2226 | { |
7bb206e3 BS |
2227 | if (!pd) |
2228 | return; | |
f37bda92 BS |
2229 | |
2230 | if (pd->port_rcvhdrq) { | |
7bb206e3 BS |
2231 | ipath_cdbg(VERBOSE, "free closed port %d rcvhdrq @ %p " |
2232 | "(size=%lu)\n", pd->port_port, pd->port_rcvhdrq, | |
2233 | (unsigned long) pd->port_rcvhdrq_size); | |
2234 | dma_free_coherent(&dd->pcidev->dev, pd->port_rcvhdrq_size, | |
2235 | pd->port_rcvhdrq, pd->port_rcvhdrq_phys); | |
2236 | pd->port_rcvhdrq = NULL; | |
f37bda92 BS |
2237 | if (pd->port_rcvhdrtail_kvaddr) { |
2238 | dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE, | |
076fafcd | 2239 | pd->port_rcvhdrtail_kvaddr, |
f37bda92 BS |
2240 | pd->port_rcvhdrqtailaddr_phys); |
2241 | pd->port_rcvhdrtail_kvaddr = NULL; | |
2242 | } | |
7bb206e3 | 2243 | } |
f37bda92 BS |
2244 | if (pd->port_port && pd->port_rcvegrbuf) { |
2245 | unsigned e; | |
2246 | ||
2247 | for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) { | |
2248 | void *base = pd->port_rcvegrbuf[e]; | |
2249 | size_t size = pd->port_rcvegrbuf_size; | |
2250 | ||
2251 | ipath_cdbg(VERBOSE, "egrbuf free(%p, %lu), " | |
2252 | "chunk %u/%u\n", base, | |
2253 | (unsigned long) size, | |
2254 | e, pd->port_rcvegrbuf_chunks); | |
2255 | dma_free_coherent(&dd->pcidev->dev, size, | |
2256 | base, pd->port_rcvegrbuf_phys[e]); | |
7bb206e3 | 2257 | } |
9929b0fb | 2258 | kfree(pd->port_rcvegrbuf); |
f37bda92 | 2259 | pd->port_rcvegrbuf = NULL; |
9929b0fb | 2260 | kfree(pd->port_rcvegrbuf_phys); |
f37bda92 | 2261 | pd->port_rcvegrbuf_phys = NULL; |
7bb206e3 | 2262 | pd->port_rcvegrbuf_chunks = 0; |
1fd3b40f | 2263 | } else if (pd->port_port == 0 && dd->ipath_port0_skbinfo) { |
7bb206e3 | 2264 | unsigned e; |
1fd3b40f | 2265 | struct ipath_skbinfo *skbinfo = dd->ipath_port0_skbinfo; |
7bb206e3 | 2266 | |
1fd3b40f BS |
2267 | dd->ipath_port0_skbinfo = NULL; |
2268 | ipath_cdbg(VERBOSE, "free closed port %d " | |
2269 | "ipath_port0_skbinfo @ %p\n", pd->port_port, | |
2270 | skbinfo); | |
9355fb6a | 2271 | for (e = 0; e < dd->ipath_p0_rcvegrcnt; e++) |
2ba3f56e RC |
2272 | if (skbinfo[e].skb) { |
2273 | pci_unmap_single(dd->pcidev, skbinfo[e].phys, | |
2274 | dd->ipath_ibmaxlen, | |
2275 | PCI_DMA_FROMDEVICE); | |
2276 | dev_kfree_skb(skbinfo[e].skb); | |
2277 | } | |
1fd3b40f | 2278 | vfree(skbinfo); |
7bb206e3 | 2279 | } |
f37bda92 | 2280 | kfree(pd->port_tid_pg_list); |
9929b0fb BS |
2281 | vfree(pd->subport_uregbase); |
2282 | vfree(pd->subport_rcvegrbuf); | |
2283 | vfree(pd->subport_rcvhdr_base); | |
f37bda92 | 2284 | kfree(pd); |
7bb206e3 BS |
2285 | } |
2286 | ||
ac2ae4c9 | 2287 | static int __init infinipath_init(void) |
7bb206e3 BS |
2288 | { |
2289 | int ret; | |
2290 | ||
39c0d0b9 BS |
2291 | if (ipath_debug & __IPATH_DBG) |
2292 | printk(KERN_INFO DRIVER_LOAD_MSG "%s", ib_ipath_version); | |
7bb206e3 BS |
2293 | |
2294 | /* | |
2295 | * These must be called before the driver is registered with | |
2296 | * the PCI subsystem. | |
2297 | */ | |
2298 | idr_init(&unit_table); | |
2299 | if (!idr_pre_get(&unit_table, GFP_KERNEL)) { | |
2300 | ret = -ENOMEM; | |
2301 | goto bail; | |
2302 | } | |
2303 | ||
2304 | ret = pci_register_driver(&ipath_driver); | |
2305 | if (ret < 0) { | |
2306 | printk(KERN_ERR IPATH_DRV_NAME | |
2307 | ": Unable to register driver: error %d\n", -ret); | |
2308 | goto bail_unit; | |
2309 | } | |
2310 | ||
7bb206e3 BS |
2311 | ret = ipath_init_ipathfs(); |
2312 | if (ret < 0) { | |
2313 | printk(KERN_ERR IPATH_DRV_NAME ": Unable to create " | |
2314 | "ipathfs: error %d\n", -ret); | |
23b9c1ab | 2315 | goto bail_pci; |
7bb206e3 BS |
2316 | } |
2317 | ||
2318 | goto bail; | |
2319 | ||
7bb206e3 BS |
2320 | bail_pci: |
2321 | pci_unregister_driver(&ipath_driver); | |
2322 | ||
2323 | bail_unit: | |
2324 | idr_destroy(&unit_table); | |
2325 | ||
2326 | bail: | |
2327 | return ret; | |
2328 | } | |
2329 | ||
7bb206e3 BS |
2330 | static void __exit infinipath_cleanup(void) |
2331 | { | |
7bb206e3 BS |
2332 | ipath_exit_ipathfs(); |
2333 | ||
7bb206e3 BS |
2334 | ipath_cdbg(VERBOSE, "Unregistering pci driver\n"); |
2335 | pci_unregister_driver(&ipath_driver); | |
2336 | ||
2337 | idr_destroy(&unit_table); | |
2338 | } | |
2339 | ||
2340 | /** | |
2341 | * ipath_reset_device - reset the chip if possible | |
2342 | * @unit: the device to reset | |
2343 | * | |
2344 | * Whether or not reset is successful, we attempt to re-initialize the chip | |
2345 | * (that is, much like a driver unload/reload). We clear the INITTED flag | |
2346 | * so that the various entry points will fail until we reinitialize. For | |
2347 | * now, we only allow this if no user ports are open that use chip resources | |
2348 | */ | |
2349 | int ipath_reset_device(int unit) | |
2350 | { | |
2351 | int ret, i; | |
2352 | struct ipath_devdata *dd = ipath_lookup(unit); | |
2353 | ||
2354 | if (!dd) { | |
2355 | ret = -ENODEV; | |
2356 | goto bail; | |
2357 | } | |
2358 | ||
82466f00 MA |
2359 | if (atomic_read(&dd->ipath_led_override_timer_active)) { |
2360 | /* Need to stop LED timer, _then_ shut off LEDs */ | |
2361 | del_timer_sync(&dd->ipath_led_override_timer); | |
2362 | atomic_set(&dd->ipath_led_override_timer_active, 0); | |
2363 | } | |
2364 | ||
2365 | /* Shut off LEDs after we are sure timer is not running */ | |
2366 | dd->ipath_led_override = LED_OVER_BOTH_OFF; | |
2367 | dd->ipath_f_setextled(dd, 0, 0); | |
2368 | ||
7bb206e3 BS |
2369 | dev_info(&dd->pcidev->dev, "Reset on unit %u requested\n", unit); |
2370 | ||
2371 | if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT)) { | |
2372 | dev_info(&dd->pcidev->dev, "Invalid unit number %u or " | |
2373 | "not initialized or not present\n", unit); | |
2374 | ret = -ENXIO; | |
2375 | goto bail; | |
2376 | } | |
2377 | ||
2378 | if (dd->ipath_pd) | |
23e86a45 | 2379 | for (i = 1; i < dd->ipath_cfgports; i++) { |
7bb206e3 BS |
2380 | if (dd->ipath_pd[i] && dd->ipath_pd[i]->port_cnt) { |
2381 | ipath_dbg("unit %u port %d is in use " | |
2382 | "(PID %u cmd %s), can't reset\n", | |
2383 | unit, i, | |
2384 | dd->ipath_pd[i]->port_pid, | |
2385 | dd->ipath_pd[i]->port_comm); | |
2386 | ret = -EBUSY; | |
2387 | goto bail; | |
2388 | } | |
2389 | } | |
2390 | ||
2391 | dd->ipath_flags &= ~IPATH_INITTED; | |
2392 | ret = dd->ipath_f_reset(dd); | |
2393 | if (ret != 1) | |
2394 | ipath_dbg("reset was not successful\n"); | |
2395 | ipath_dbg("Trying to reinitialize unit %u after reset attempt\n", | |
2396 | unit); | |
2397 | ret = ipath_init_chip(dd, 1); | |
2398 | if (ret) | |
2399 | ipath_dev_err(dd, "Reinitialize unit %u after " | |
2400 | "reset failed with %d\n", unit, ret); | |
2401 | else | |
2402 | dev_info(&dd->pcidev->dev, "Reinitialized unit %u after " | |
2403 | "resetting\n", unit); | |
2404 | ||
2405 | bail: | |
2406 | return ret; | |
2407 | } | |
2408 | ||
58411d1c JG |
2409 | /* |
2410 | * send a signal to all the processes that have the driver open | |
2411 | * through the normal interfaces (i.e., everything other than diags | |
2412 | * interface). Returns number of signalled processes. | |
2413 | */ | |
2414 | static int ipath_signal_procs(struct ipath_devdata *dd, int sig) | |
2415 | { | |
2416 | int i, sub, any = 0; | |
2417 | pid_t pid; | |
2418 | ||
2419 | if (!dd->ipath_pd) | |
2420 | return 0; | |
2421 | for (i = 1; i < dd->ipath_cfgports; i++) { | |
2422 | if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt || | |
2423 | !dd->ipath_pd[i]->port_pid) | |
2424 | continue; | |
2425 | pid = dd->ipath_pd[i]->port_pid; | |
2426 | dev_info(&dd->pcidev->dev, "context %d in use " | |
2427 | "(PID %u), sending signal %d\n", | |
2428 | i, pid, sig); | |
2429 | kill_proc(pid, sig, 1); | |
2430 | any++; | |
2431 | for (sub = 0; sub < INFINIPATH_MAX_SUBPORT; sub++) { | |
2432 | pid = dd->ipath_pd[i]->port_subpid[sub]; | |
2433 | if (!pid) | |
2434 | continue; | |
2435 | dev_info(&dd->pcidev->dev, "sub-context " | |
2436 | "%d:%d in use (PID %u), sending " | |
2437 | "signal %d\n", i, sub, pid, sig); | |
2438 | kill_proc(pid, sig, 1); | |
2439 | any++; | |
2440 | } | |
2441 | } | |
2442 | return any; | |
2443 | } | |
2444 | ||
2445 | static void ipath_hol_signal_down(struct ipath_devdata *dd) | |
2446 | { | |
2447 | if (ipath_signal_procs(dd, SIGSTOP)) | |
2448 | ipath_dbg("Stopped some processes\n"); | |
2449 | ipath_cancel_sends(dd, 1); | |
2450 | } | |
2451 | ||
2452 | ||
2453 | static void ipath_hol_signal_up(struct ipath_devdata *dd) | |
2454 | { | |
2455 | if (ipath_signal_procs(dd, SIGCONT)) | |
2456 | ipath_dbg("Continued some processes\n"); | |
2457 | } | |
2458 | ||
2459 | /* | |
2460 | * link is down, stop any users processes, and flush pending sends | |
2461 | * to prevent HoL blocking, then start the HoL timer that | |
2462 | * periodically continues, then stop procs, so they can detect | |
2463 | * link down if they want, and do something about it. | |
2464 | * Timer may already be running, so use __mod_timer, not add_timer. | |
2465 | */ | |
2466 | void ipath_hol_down(struct ipath_devdata *dd) | |
2467 | { | |
2468 | dd->ipath_hol_state = IPATH_HOL_DOWN; | |
2469 | ipath_hol_signal_down(dd); | |
2470 | dd->ipath_hol_next = IPATH_HOL_DOWNCONT; | |
2471 | dd->ipath_hol_timer.expires = jiffies + | |
2472 | msecs_to_jiffies(ipath_hol_timeout_ms); | |
2473 | __mod_timer(&dd->ipath_hol_timer, dd->ipath_hol_timer.expires); | |
2474 | } | |
2475 | ||
2476 | /* | |
2477 | * link is up, continue any user processes, and ensure timer | |
2478 | * is a nop, if running. Let timer keep running, if set; it | |
2479 | * will nop when it sees the link is up | |
2480 | */ | |
2481 | void ipath_hol_up(struct ipath_devdata *dd) | |
2482 | { | |
2483 | ipath_hol_signal_up(dd); | |
2484 | dd->ipath_hol_state = IPATH_HOL_UP; | |
2485 | } | |
2486 | ||
2487 | /* | |
2488 | * toggle the running/not running state of user proceses | |
2489 | * to prevent HoL blocking on chip resources, but still allow | |
2490 | * user processes to do link down special case handling. | |
2491 | * Should only be called via the timer | |
2492 | */ | |
2493 | void ipath_hol_event(unsigned long opaque) | |
2494 | { | |
2495 | struct ipath_devdata *dd = (struct ipath_devdata *)opaque; | |
2496 | ||
2497 | if (dd->ipath_hol_next == IPATH_HOL_DOWNSTOP | |
2498 | && dd->ipath_hol_state != IPATH_HOL_UP) { | |
2499 | dd->ipath_hol_next = IPATH_HOL_DOWNCONT; | |
2500 | ipath_dbg("Stopping processes\n"); | |
2501 | ipath_hol_signal_down(dd); | |
2502 | } else { /* may do "extra" if also in ipath_hol_up() */ | |
2503 | dd->ipath_hol_next = IPATH_HOL_DOWNSTOP; | |
2504 | ipath_dbg("Continuing processes\n"); | |
2505 | ipath_hol_signal_up(dd); | |
2506 | } | |
2507 | if (dd->ipath_hol_state == IPATH_HOL_UP) | |
2508 | ipath_dbg("link's up, don't resched timer\n"); | |
2509 | else { | |
2510 | dd->ipath_hol_timer.expires = jiffies + | |
2511 | msecs_to_jiffies(ipath_hol_timeout_ms); | |
2512 | __mod_timer(&dd->ipath_hol_timer, | |
2513 | dd->ipath_hol_timer.expires); | |
2514 | } | |
2515 | } | |
2516 | ||
30fc5c31 BS |
2517 | int ipath_set_rx_pol_inv(struct ipath_devdata *dd, u8 new_pol_inv) |
2518 | { | |
2519 | u64 val; | |
2ba3f56e RC |
2520 | |
2521 | if (new_pol_inv > INFINIPATH_XGXS_RX_POL_MASK) | |
30fc5c31 | 2522 | return -1; |
2ba3f56e | 2523 | if (dd->ipath_rx_pol_inv != new_pol_inv) { |
30fc5c31 BS |
2524 | dd->ipath_rx_pol_inv = new_pol_inv; |
2525 | val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig); | |
2526 | val &= ~(INFINIPATH_XGXS_RX_POL_MASK << | |
3cd96564 RD |
2527 | INFINIPATH_XGXS_RX_POL_SHIFT); |
2528 | val |= ((u64)dd->ipath_rx_pol_inv) << | |
2529 | INFINIPATH_XGXS_RX_POL_SHIFT; | |
30fc5c31 BS |
2530 | ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val); |
2531 | } | |
2532 | return 0; | |
2533 | } | |
6ac50727 DO |
2534 | |
2535 | /* | |
2536 | * Disable and enable the armlaunch error. Used for PIO bandwidth testing on | |
2537 | * the 7220, which is count-based, rather than trigger-based. Safe for the | |
2538 | * driver check, since it's at init. Not completely safe when used for | |
2539 | * user-mode checking, since some error checking can be lost, but not | |
2540 | * particularly risky, and only has problematic side-effects in the face of | |
2541 | * very buggy user code. There is no reference counting, but that's also | |
2542 | * fine, given the intended use. | |
2543 | */ | |
2544 | void ipath_enable_armlaunch(struct ipath_devdata *dd) | |
2545 | { | |
2546 | dd->ipath_lasterror &= ~INFINIPATH_E_SPIOARMLAUNCH; | |
2547 | ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, | |
2548 | INFINIPATH_E_SPIOARMLAUNCH); | |
2549 | dd->ipath_errormask |= INFINIPATH_E_SPIOARMLAUNCH; | |
2550 | ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, | |
2551 | dd->ipath_errormask); | |
2552 | } | |
2553 | ||
2554 | void ipath_disable_armlaunch(struct ipath_devdata *dd) | |
2555 | { | |
2556 | /* so don't re-enable if already set */ | |
2557 | dd->ipath_maskederrs &= ~INFINIPATH_E_SPIOARMLAUNCH; | |
2558 | dd->ipath_errormask &= ~INFINIPATH_E_SPIOARMLAUNCH; | |
2559 | ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, | |
2560 | dd->ipath_errormask); | |
2561 | } | |
2562 | ||
7bb206e3 BS |
2563 | module_init(infinipath_init); |
2564 | module_exit(infinipath_cleanup); |