]>
Commit | Line | Data |
---|---|---|
77241056 | 1 | /* |
2280740f | 2 | * Copyright(c) 2015-2017 Intel Corporation. |
77241056 MM |
3 | * |
4 | * This file is provided under a dual BSD/GPLv2 license. When using or | |
5 | * redistributing this file, you may do so under either license. | |
6 | * | |
7 | * GPL LICENSE SUMMARY | |
8 | * | |
77241056 MM |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of version 2 of the GNU General Public License as | |
11 | * published by the Free Software Foundation. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, but | |
14 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * General Public License for more details. | |
17 | * | |
18 | * BSD LICENSE | |
19 | * | |
77241056 MM |
20 | * Redistribution and use in source and binary forms, with or without |
21 | * modification, are permitted provided that the following conditions | |
22 | * are met: | |
23 | * | |
24 | * - Redistributions of source code must retain the above copyright | |
25 | * notice, this list of conditions and the following disclaimer. | |
26 | * - Redistributions in binary form must reproduce the above copyright | |
27 | * notice, this list of conditions and the following disclaimer in | |
28 | * the documentation and/or other materials provided with the | |
29 | * distribution. | |
30 | * - Neither the name of Intel Corporation nor the names of its | |
31 | * contributors may be used to endorse or promote products derived | |
32 | * from this software without specific prior written permission. | |
33 | * | |
34 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
35 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
36 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
37 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
38 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
39 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
40 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
41 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
42 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
43 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
44 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
45 | * | |
46 | */ | |
77241056 MM |
47 | #include <linux/poll.h> |
48 | #include <linux/cdev.h> | |
77241056 | 49 | #include <linux/vmalloc.h> |
77241056 | 50 | #include <linux/io.h> |
6e84f315 | 51 | #include <linux/sched/mm.h> |
8737ce95 | 52 | #include <linux/bitmap.h> |
77241056 | 53 | |
e6bd18f5 JG |
54 | #include <rdma/ib.h> |
55 | ||
77241056 MM |
56 | #include "hfi.h" |
57 | #include "pio.h" | |
58 | #include "device.h" | |
59 | #include "common.h" | |
60 | #include "trace.h" | |
61 | #include "user_sdma.h" | |
701e441d | 62 | #include "user_exp_rcv.h" |
affa48de | 63 | #include "aspm.h" |
06e0ffa6 | 64 | #include "mmu_rb.h" |
77241056 MM |
65 | |
66 | #undef pr_fmt | |
67 | #define pr_fmt(fmt) DRIVER_NAME ": " fmt | |
68 | ||
69 | #define SEND_CTXT_HALT_TIMEOUT 1000 /* msecs */ | |
70 | ||
71 | /* | |
72 | * File operation functions | |
73 | */ | |
f4cd8765 MR |
74 | static int hfi1_file_open(struct inode *inode, struct file *fp); |
75 | static int hfi1_file_close(struct inode *inode, struct file *fp); | |
76 | static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from); | |
77 | static unsigned int hfi1_poll(struct file *fp, struct poll_table_struct *pt); | |
78 | static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma); | |
77241056 | 79 | |
f4cd8765 | 80 | static u64 kvirt_to_phys(void *addr); |
5fbded48 | 81 | static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo); |
f4cd8765 MR |
82 | static int init_subctxts(struct hfi1_ctxtdata *uctxt, |
83 | const struct hfi1_user_info *uinfo); | |
9b60d2cb | 84 | static int init_user_ctxt(struct hfi1_filedata *fd); |
62239fc6 | 85 | static void user_init(struct hfi1_ctxtdata *uctxt); |
5042cddf MR |
86 | static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase, |
87 | __u32 len); | |
88 | static int get_base_info(struct hfi1_filedata *fd, void __user *ubase, | |
89 | __u32 len); | |
9b60d2cb | 90 | static int setup_base_ctxt(struct hfi1_filedata *fd); |
f4cd8765 | 91 | static int setup_subctxt(struct hfi1_ctxtdata *uctxt); |
5fbded48 | 92 | |
9b60d2cb MR |
93 | static int find_sub_ctxt(struct hfi1_filedata *fd, |
94 | const struct hfi1_user_info *uinfo); | |
5042cddf | 95 | static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd, |
f4cd8765 | 96 | struct hfi1_user_info *uinfo); |
42492011 | 97 | static void deallocate_ctxt(struct hfi1_ctxtdata *uctxt); |
f4cd8765 MR |
98 | static unsigned int poll_urgent(struct file *fp, struct poll_table_struct *pt); |
99 | static unsigned int poll_next(struct file *fp, struct poll_table_struct *pt); | |
8737ce95 | 100 | static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt, |
f4cd8765 | 101 | unsigned long events); |
8737ce95 MR |
102 | static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey); |
103 | static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt, | |
f4cd8765 MR |
104 | int start_stop); |
105 | static int vma_fault(struct vm_fault *vmf); | |
8d970cf9 DD |
106 | static long hfi1_file_ioctl(struct file *fp, unsigned int cmd, |
107 | unsigned long arg); | |
77241056 MM |
108 | |
109 | static const struct file_operations hfi1_file_ops = { | |
110 | .owner = THIS_MODULE, | |
77241056 MM |
111 | .write_iter = hfi1_write_iter, |
112 | .open = hfi1_file_open, | |
113 | .release = hfi1_file_close, | |
8d970cf9 | 114 | .unlocked_ioctl = hfi1_file_ioctl, |
77241056 MM |
115 | .poll = hfi1_poll, |
116 | .mmap = hfi1_file_mmap, | |
117 | .llseek = noop_llseek, | |
118 | }; | |
119 | ||
120 | static struct vm_operations_struct vm_ops = { | |
121 | .fault = vma_fault, | |
122 | }; | |
123 | ||
124 | /* | |
125 | * Types of memories mapped into user processes' space | |
126 | */ | |
127 | enum mmap_types { | |
128 | PIO_BUFS = 1, | |
129 | PIO_BUFS_SOP, | |
130 | PIO_CRED, | |
131 | RCV_HDRQ, | |
132 | RCV_EGRBUF, | |
133 | UREGS, | |
134 | EVENTS, | |
135 | STATUS, | |
136 | RTAIL, | |
137 | SUBCTXT_UREGS, | |
138 | SUBCTXT_RCV_HDRQ, | |
139 | SUBCTXT_EGRBUF, | |
140 | SDMA_COMP | |
141 | }; | |
142 | ||
143 | /* | |
144 | * Masks and offsets defining the mmap tokens | |
145 | */ | |
146 | #define HFI1_MMAP_OFFSET_MASK 0xfffULL | |
147 | #define HFI1_MMAP_OFFSET_SHIFT 0 | |
148 | #define HFI1_MMAP_SUBCTXT_MASK 0xfULL | |
149 | #define HFI1_MMAP_SUBCTXT_SHIFT 12 | |
150 | #define HFI1_MMAP_CTXT_MASK 0xffULL | |
151 | #define HFI1_MMAP_CTXT_SHIFT 16 | |
152 | #define HFI1_MMAP_TYPE_MASK 0xfULL | |
153 | #define HFI1_MMAP_TYPE_SHIFT 24 | |
154 | #define HFI1_MMAP_MAGIC_MASK 0xffffffffULL | |
155 | #define HFI1_MMAP_MAGIC_SHIFT 32 | |
156 | ||
157 | #define HFI1_MMAP_MAGIC 0xdabbad00 | |
158 | ||
159 | #define HFI1_MMAP_TOKEN_SET(field, val) \ | |
160 | (((val) & HFI1_MMAP_##field##_MASK) << HFI1_MMAP_##field##_SHIFT) | |
161 | #define HFI1_MMAP_TOKEN_GET(field, token) \ | |
162 | (((token) >> HFI1_MMAP_##field##_SHIFT) & HFI1_MMAP_##field##_MASK) | |
163 | #define HFI1_MMAP_TOKEN(type, ctxt, subctxt, addr) \ | |
164 | (HFI1_MMAP_TOKEN_SET(MAGIC, HFI1_MMAP_MAGIC) | \ | |
165 | HFI1_MMAP_TOKEN_SET(TYPE, type) | \ | |
166 | HFI1_MMAP_TOKEN_SET(CTXT, ctxt) | \ | |
167 | HFI1_MMAP_TOKEN_SET(SUBCTXT, subctxt) | \ | |
e260e404 | 168 | HFI1_MMAP_TOKEN_SET(OFFSET, (offset_in_page(addr)))) |
77241056 | 169 | |
77241056 MM |
170 | #define dbg(fmt, ...) \ |
171 | pr_info(fmt, ##__VA_ARGS__) | |
172 | ||
77241056 MM |
173 | static inline int is_valid_mmap(u64 token) |
174 | { | |
175 | return (HFI1_MMAP_TOKEN_GET(MAGIC, token) == HFI1_MMAP_MAGIC); | |
176 | } | |
177 | ||
178 | static int hfi1_file_open(struct inode *inode, struct file *fp) | |
179 | { | |
ea3a0ee5 | 180 | struct hfi1_filedata *fd; |
e11ffbd5 DD |
181 | struct hfi1_devdata *dd = container_of(inode->i_cdev, |
182 | struct hfi1_devdata, | |
183 | user_cdev); | |
184 | ||
cb51c5d2 | 185 | if (!((dd->flags & HFI1_PRESENT) && dd->kregbase1)) |
5fbded48 MR |
186 | return -EINVAL; |
187 | ||
acd7c8fe TS |
188 | if (!atomic_inc_not_zero(&dd->user_refcount)) |
189 | return -ENXIO; | |
190 | ||
e11ffbd5 DD |
191 | /* Just take a ref now. Not all opens result in a context assign */ |
192 | kobject_get(&dd->kobj); | |
193 | ||
77241056 | 194 | /* The real work is performed later in assign_ctxt() */ |
ea3a0ee5 IW |
195 | |
196 | fd = kzalloc(sizeof(*fd), GFP_KERNEL); | |
197 | ||
3faa3d9a IW |
198 | if (fd) { |
199 | fd->rec_cpu_num = -1; /* no cpu affinity by default */ | |
200 | fd->mm = current->mm; | |
f1f10076 | 201 | mmgrab(fd->mm); |
5fbded48 | 202 | fd->dd = dd; |
acd7c8fe TS |
203 | fp->private_data = fd; |
204 | } else { | |
205 | fp->private_data = NULL; | |
206 | ||
207 | if (atomic_dec_and_test(&dd->user_refcount)) | |
208 | complete(&dd->user_comp); | |
ea3a0ee5 | 209 | |
acd7c8fe TS |
210 | return -ENOMEM; |
211 | } | |
ea3a0ee5 | 212 | |
acd7c8fe | 213 | return 0; |
77241056 MM |
214 | } |
215 | ||
8d970cf9 DD |
216 | static long hfi1_file_ioctl(struct file *fp, unsigned int cmd, |
217 | unsigned long arg) | |
218 | { | |
219 | struct hfi1_filedata *fd = fp->private_data; | |
220 | struct hfi1_ctxtdata *uctxt = fd->uctxt; | |
221 | struct hfi1_user_info uinfo; | |
222 | struct hfi1_tid_info tinfo; | |
223 | int ret = 0; | |
224 | unsigned long addr; | |
225 | int uval = 0; | |
226 | unsigned long ul_uval = 0; | |
227 | u16 uval16 = 0; | |
228 | ||
8a1882eb | 229 | hfi1_cdbg(IOCTL, "IOCTL recv: 0x%x", cmd); |
8d970cf9 DD |
230 | if (cmd != HFI1_IOCTL_ASSIGN_CTXT && |
231 | cmd != HFI1_IOCTL_GET_VERS && | |
232 | !uctxt) | |
233 | return -EINVAL; | |
234 | ||
235 | switch (cmd) { | |
236 | case HFI1_IOCTL_ASSIGN_CTXT: | |
ca2f30a0 IW |
237 | if (uctxt) |
238 | return -EINVAL; | |
239 | ||
8d970cf9 DD |
240 | if (copy_from_user(&uinfo, |
241 | (struct hfi1_user_info __user *)arg, | |
242 | sizeof(uinfo))) | |
243 | return -EFAULT; | |
244 | ||
5fbded48 | 245 | ret = assign_ctxt(fd, &uinfo); |
8d970cf9 DD |
246 | break; |
247 | case HFI1_IOCTL_CTXT_INFO: | |
5042cddf | 248 | ret = get_ctxt_info(fd, (void __user *)(unsigned long)arg, |
8d970cf9 DD |
249 | sizeof(struct hfi1_ctxt_info)); |
250 | break; | |
251 | case HFI1_IOCTL_USER_INFO: | |
5042cddf | 252 | ret = get_base_info(fd, (void __user *)(unsigned long)arg, |
8d970cf9 DD |
253 | sizeof(struct hfi1_base_info)); |
254 | break; | |
255 | case HFI1_IOCTL_CREDIT_UPD: | |
f7ca535b | 256 | if (uctxt) |
8d970cf9 DD |
257 | sc_return_credits(uctxt->sc); |
258 | break; | |
259 | ||
260 | case HFI1_IOCTL_TID_UPDATE: | |
261 | if (copy_from_user(&tinfo, | |
262 | (struct hfi11_tid_info __user *)arg, | |
263 | sizeof(tinfo))) | |
264 | return -EFAULT; | |
265 | ||
5042cddf | 266 | ret = hfi1_user_exp_rcv_setup(fd, &tinfo); |
8d970cf9 DD |
267 | if (!ret) { |
268 | /* | |
269 | * Copy the number of tidlist entries we used | |
270 | * and the length of the buffer we registered. | |
271 | * These fields are adjacent in the structure so | |
272 | * we can copy them at the same time. | |
273 | */ | |
274 | addr = arg + offsetof(struct hfi1_tid_info, tidcnt); | |
275 | if (copy_to_user((void __user *)addr, &tinfo.tidcnt, | |
276 | sizeof(tinfo.tidcnt) + | |
277 | sizeof(tinfo.length))) | |
278 | ret = -EFAULT; | |
279 | } | |
280 | break; | |
281 | ||
282 | case HFI1_IOCTL_TID_FREE: | |
283 | if (copy_from_user(&tinfo, | |
284 | (struct hfi11_tid_info __user *)arg, | |
285 | sizeof(tinfo))) | |
286 | return -EFAULT; | |
287 | ||
5042cddf | 288 | ret = hfi1_user_exp_rcv_clear(fd, &tinfo); |
8d970cf9 DD |
289 | if (ret) |
290 | break; | |
291 | addr = arg + offsetof(struct hfi1_tid_info, tidcnt); | |
292 | if (copy_to_user((void __user *)addr, &tinfo.tidcnt, | |
293 | sizeof(tinfo.tidcnt))) | |
294 | ret = -EFAULT; | |
295 | break; | |
296 | ||
297 | case HFI1_IOCTL_TID_INVAL_READ: | |
298 | if (copy_from_user(&tinfo, | |
299 | (struct hfi11_tid_info __user *)arg, | |
300 | sizeof(tinfo))) | |
301 | return -EFAULT; | |
302 | ||
5042cddf | 303 | ret = hfi1_user_exp_rcv_invalid(fd, &tinfo); |
8d970cf9 DD |
304 | if (ret) |
305 | break; | |
306 | addr = arg + offsetof(struct hfi1_tid_info, tidcnt); | |
307 | if (copy_to_user((void __user *)addr, &tinfo.tidcnt, | |
308 | sizeof(tinfo.tidcnt))) | |
309 | ret = -EFAULT; | |
310 | break; | |
311 | ||
312 | case HFI1_IOCTL_RECV_CTRL: | |
313 | ret = get_user(uval, (int __user *)arg); | |
314 | if (ret != 0) | |
315 | return -EFAULT; | |
316 | ret = manage_rcvq(uctxt, fd->subctxt, uval); | |
317 | break; | |
318 | ||
319 | case HFI1_IOCTL_POLL_TYPE: | |
320 | ret = get_user(uval, (int __user *)arg); | |
321 | if (ret != 0) | |
322 | return -EFAULT; | |
323 | uctxt->poll_type = (typeof(uctxt->poll_type))uval; | |
324 | break; | |
325 | ||
326 | case HFI1_IOCTL_ACK_EVENT: | |
327 | ret = get_user(ul_uval, (unsigned long __user *)arg); | |
328 | if (ret != 0) | |
329 | return -EFAULT; | |
330 | ret = user_event_ack(uctxt, fd->subctxt, ul_uval); | |
331 | break; | |
332 | ||
333 | case HFI1_IOCTL_SET_PKEY: | |
334 | ret = get_user(uval16, (u16 __user *)arg); | |
335 | if (ret != 0) | |
336 | return -EFAULT; | |
337 | if (HFI1_CAP_IS_USET(PKEY_CHECK)) | |
338 | ret = set_ctxt_pkey(uctxt, fd->subctxt, uval16); | |
339 | else | |
340 | return -EPERM; | |
341 | break; | |
342 | ||
343 | case HFI1_IOCTL_CTXT_RESET: { | |
344 | struct send_context *sc; | |
345 | struct hfi1_devdata *dd; | |
346 | ||
347 | if (!uctxt || !uctxt->dd || !uctxt->sc) | |
348 | return -EINVAL; | |
349 | ||
350 | /* | |
351 | * There is no protection here. User level has to | |
352 | * guarantee that no one will be writing to the send | |
353 | * context while it is being re-initialized. | |
354 | * If user level breaks that guarantee, it will break | |
355 | * it's own context and no one else's. | |
356 | */ | |
357 | dd = uctxt->dd; | |
358 | sc = uctxt->sc; | |
359 | /* | |
360 | * Wait until the interrupt handler has marked the | |
361 | * context as halted or frozen. Report error if we time | |
362 | * out. | |
363 | */ | |
364 | wait_event_interruptible_timeout( | |
365 | sc->halt_wait, (sc->flags & SCF_HALTED), | |
366 | msecs_to_jiffies(SEND_CTXT_HALT_TIMEOUT)); | |
367 | if (!(sc->flags & SCF_HALTED)) | |
368 | return -ENOLCK; | |
369 | ||
370 | /* | |
371 | * If the send context was halted due to a Freeze, | |
372 | * wait until the device has been "unfrozen" before | |
373 | * resetting the context. | |
374 | */ | |
375 | if (sc->flags & SCF_FROZEN) { | |
376 | wait_event_interruptible_timeout( | |
377 | dd->event_queue, | |
378 | !(ACCESS_ONCE(dd->flags) & HFI1_FROZEN), | |
379 | msecs_to_jiffies(SEND_CTXT_HALT_TIMEOUT)); | |
380 | if (dd->flags & HFI1_FROZEN) | |
381 | return -ENOLCK; | |
382 | ||
383 | if (dd->flags & HFI1_FORCED_FREEZE) | |
384 | /* | |
385 | * Don't allow context reset if we are into | |
386 | * forced freeze | |
387 | */ | |
388 | return -ENODEV; | |
389 | ||
390 | sc_disable(sc); | |
391 | ret = sc_enable(sc); | |
392 | hfi1_rcvctrl(dd, HFI1_RCVCTRL_CTXT_ENB, | |
393 | uctxt->ctxt); | |
394 | } else { | |
395 | ret = sc_restart(sc); | |
396 | } | |
397 | if (!ret) | |
398 | sc_return_credits(sc); | |
399 | break; | |
400 | } | |
401 | ||
402 | case HFI1_IOCTL_GET_VERS: | |
403 | uval = HFI1_USER_SWVERSION; | |
404 | if (put_user(uval, (int __user *)arg)) | |
405 | return -EFAULT; | |
406 | break; | |
407 | ||
408 | default: | |
409 | return -EINVAL; | |
410 | } | |
411 | ||
412 | return ret; | |
413 | } | |
414 | ||
77241056 MM |
415 | static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from) |
416 | { | |
9e10af47 IW |
417 | struct hfi1_filedata *fd = kiocb->ki_filp->private_data; |
418 | struct hfi1_user_sdma_pkt_q *pq = fd->pq; | |
419 | struct hfi1_user_sdma_comp_q *cq = fd->cq; | |
0904f327 | 420 | int done = 0, reqs = 0; |
77241056 MM |
421 | unsigned long dim = from->nr_segs; |
422 | ||
0904f327 IW |
423 | if (!cq || !pq) |
424 | return -EIO; | |
77241056 | 425 | |
0904f327 IW |
426 | if (!iter_is_iovec(from) || !dim) |
427 | return -EINVAL; | |
77241056 MM |
428 | |
429 | hfi1_cdbg(SDMA, "SDMA request from %u:%u (%lu)", | |
9e10af47 | 430 | fd->uctxt->ctxt, fd->subctxt, dim); |
77241056 | 431 | |
0904f327 IW |
432 | if (atomic_read(&pq->n_reqs) == pq->n_max_reqs) |
433 | return -ENOSPC; | |
77241056 MM |
434 | |
435 | while (dim) { | |
0904f327 | 436 | int ret; |
77241056 MM |
437 | unsigned long count = 0; |
438 | ||
439 | ret = hfi1_user_sdma_process_request( | |
5042cddf | 440 | fd, (struct iovec *)(from->iov + done), |
77241056 | 441 | dim, &count); |
0904f327 IW |
442 | if (ret) { |
443 | reqs = ret; | |
444 | break; | |
445 | } | |
77241056 MM |
446 | dim -= count; |
447 | done += count; | |
448 | reqs++; | |
449 | } | |
0904f327 IW |
450 | |
451 | return reqs; | |
77241056 MM |
452 | } |
453 | ||
454 | static int hfi1_file_mmap(struct file *fp, struct vm_area_struct *vma) | |
455 | { | |
9e10af47 IW |
456 | struct hfi1_filedata *fd = fp->private_data; |
457 | struct hfi1_ctxtdata *uctxt = fd->uctxt; | |
77241056 | 458 | struct hfi1_devdata *dd; |
60368186 | 459 | unsigned long flags; |
77241056 MM |
460 | u64 token = vma->vm_pgoff << PAGE_SHIFT, |
461 | memaddr = 0; | |
60368186 | 462 | void *memvirt = NULL; |
77241056 MM |
463 | u8 subctxt, mapio = 0, vmf = 0, type; |
464 | ssize_t memlen = 0; | |
465 | int ret = 0; | |
466 | u16 ctxt; | |
467 | ||
77241056 MM |
468 | if (!is_valid_mmap(token) || !uctxt || |
469 | !(vma->vm_flags & VM_SHARED)) { | |
470 | ret = -EINVAL; | |
471 | goto done; | |
472 | } | |
473 | dd = uctxt->dd; | |
474 | ctxt = HFI1_MMAP_TOKEN_GET(CTXT, token); | |
475 | subctxt = HFI1_MMAP_TOKEN_GET(SUBCTXT, token); | |
476 | type = HFI1_MMAP_TOKEN_GET(TYPE, token); | |
9e10af47 | 477 | if (ctxt != uctxt->ctxt || subctxt != fd->subctxt) { |
77241056 MM |
478 | ret = -EINVAL; |
479 | goto done; | |
480 | } | |
481 | ||
482 | flags = vma->vm_flags; | |
483 | ||
484 | switch (type) { | |
485 | case PIO_BUFS: | |
486 | case PIO_BUFS_SOP: | |
487 | memaddr = ((dd->physaddr + TXE_PIO_SEND) + | |
488 | /* chip pio base */ | |
d32cf44a | 489 | (uctxt->sc->hw_context * BIT(16))) + |
77241056 MM |
490 | /* 64K PIO space / ctxt */ |
491 | (type == PIO_BUFS_SOP ? | |
492 | (TXE_PIO_SIZE / 2) : 0); /* sop? */ | |
493 | /* | |
494 | * Map only the amount allocated to the context, not the | |
495 | * entire available context's PIO space. | |
496 | */ | |
437b29d1 | 497 | memlen = PAGE_ALIGN(uctxt->sc->credits * PIO_BLOCK_SIZE); |
77241056 MM |
498 | flags &= ~VM_MAYREAD; |
499 | flags |= VM_DONTCOPY | VM_DONTEXPAND; | |
500 | vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); | |
501 | mapio = 1; | |
502 | break; | |
503 | case PIO_CRED: | |
504 | if (flags & VM_WRITE) { | |
505 | ret = -EPERM; | |
506 | goto done; | |
507 | } | |
508 | /* | |
509 | * The credit return location for this context could be on the | |
510 | * second or third page allocated for credit returns (if number | |
511 | * of enabled contexts > 64 and 128 respectively). | |
512 | */ | |
60368186 TK |
513 | memvirt = dd->cr_base[uctxt->numa_id].va; |
514 | memaddr = virt_to_phys(memvirt) + | |
77241056 MM |
515 | (((u64)uctxt->sc->hw_free - |
516 | (u64)dd->cr_base[uctxt->numa_id].va) & PAGE_MASK); | |
517 | memlen = PAGE_SIZE; | |
518 | flags &= ~VM_MAYWRITE; | |
519 | flags |= VM_DONTCOPY | VM_DONTEXPAND; | |
520 | /* | |
521 | * The driver has already allocated memory for credit | |
522 | * returns and programmed it into the chip. Has that | |
523 | * memory been flagged as non-cached? | |
524 | */ | |
525 | /* vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); */ | |
526 | mapio = 1; | |
527 | break; | |
528 | case RCV_HDRQ: | |
77241056 | 529 | memlen = uctxt->rcvhdrq_size; |
60368186 | 530 | memvirt = uctxt->rcvhdrq; |
77241056 MM |
531 | break; |
532 | case RCV_EGRBUF: { | |
533 | unsigned long addr; | |
534 | int i; | |
535 | /* | |
536 | * The RcvEgr buffer need to be handled differently | |
537 | * as multiple non-contiguous pages need to be mapped | |
538 | * into the user process. | |
539 | */ | |
540 | memlen = uctxt->egrbufs.size; | |
541 | if ((vma->vm_end - vma->vm_start) != memlen) { | |
542 | dd_dev_err(dd, "Eager buffer map size invalid (%lu != %lu)\n", | |
543 | (vma->vm_end - vma->vm_start), memlen); | |
544 | ret = -EINVAL; | |
545 | goto done; | |
546 | } | |
547 | if (vma->vm_flags & VM_WRITE) { | |
548 | ret = -EPERM; | |
549 | goto done; | |
550 | } | |
551 | vma->vm_flags &= ~VM_MAYWRITE; | |
552 | addr = vma->vm_start; | |
553 | for (i = 0 ; i < uctxt->egrbufs.numbufs; i++) { | |
60368186 TK |
554 | memlen = uctxt->egrbufs.buffers[i].len; |
555 | memvirt = uctxt->egrbufs.buffers[i].addr; | |
77241056 MM |
556 | ret = remap_pfn_range( |
557 | vma, addr, | |
60368186 TK |
558 | /* |
559 | * virt_to_pfn() does the same, but | |
560 | * it's not available on x86_64 | |
561 | * when CONFIG_MMU is enabled. | |
562 | */ | |
563 | PFN_DOWN(__pa(memvirt)), | |
564 | memlen, | |
77241056 MM |
565 | vma->vm_page_prot); |
566 | if (ret < 0) | |
567 | goto done; | |
60368186 | 568 | addr += memlen; |
77241056 MM |
569 | } |
570 | ret = 0; | |
571 | goto done; | |
572 | } | |
573 | case UREGS: | |
574 | /* | |
575 | * Map only the page that contains this context's user | |
576 | * registers. | |
577 | */ | |
578 | memaddr = (unsigned long) | |
579 | (dd->physaddr + RXE_PER_CONTEXT_USER) | |
580 | + (uctxt->ctxt * RXE_PER_CONTEXT_SIZE); | |
581 | /* | |
582 | * TidFlow table is on the same page as the rest of the | |
583 | * user registers. | |
584 | */ | |
585 | memlen = PAGE_SIZE; | |
586 | flags |= VM_DONTCOPY | VM_DONTEXPAND; | |
587 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
588 | mapio = 1; | |
589 | break; | |
590 | case EVENTS: | |
591 | /* | |
592 | * Use the page where this context's flags are. User level | |
593 | * knows where it's own bitmap is within the page. | |
594 | */ | |
3c6c065a | 595 | memaddr = (unsigned long)(dd->events + |
2280740f VN |
596 | ((uctxt->ctxt - dd->first_dyn_alloc_ctxt) * |
597 | HFI1_MAX_SHARED_CTXTS)) & PAGE_MASK; | |
77241056 MM |
598 | memlen = PAGE_SIZE; |
599 | /* | |
600 | * v3.7 removes VM_RESERVED but the effect is kept by | |
601 | * using VM_IO. | |
602 | */ | |
603 | flags |= VM_IO | VM_DONTEXPAND; | |
604 | vmf = 1; | |
605 | break; | |
606 | case STATUS: | |
12220267 IW |
607 | if (flags & (unsigned long)(VM_WRITE | VM_EXEC)) { |
608 | ret = -EPERM; | |
609 | goto done; | |
610 | } | |
77241056 MM |
611 | memaddr = kvirt_to_phys((void *)dd->status); |
612 | memlen = PAGE_SIZE; | |
613 | flags |= VM_IO | VM_DONTEXPAND; | |
614 | break; | |
615 | case RTAIL: | |
616 | if (!HFI1_CAP_IS_USET(DMA_RTAIL)) { | |
617 | /* | |
618 | * If the memory allocation failed, the context alloc | |
619 | * also would have failed, so we would never get here | |
620 | */ | |
621 | ret = -EINVAL; | |
622 | goto done; | |
623 | } | |
624 | if (flags & VM_WRITE) { | |
625 | ret = -EPERM; | |
626 | goto done; | |
627 | } | |
77241056 | 628 | memlen = PAGE_SIZE; |
60368186 | 629 | memvirt = (void *)uctxt->rcvhdrtail_kvaddr; |
77241056 MM |
630 | flags &= ~VM_MAYWRITE; |
631 | break; | |
632 | case SUBCTXT_UREGS: | |
633 | memaddr = (u64)uctxt->subctxt_uregbase; | |
634 | memlen = PAGE_SIZE; | |
635 | flags |= VM_IO | VM_DONTEXPAND; | |
636 | vmf = 1; | |
637 | break; | |
638 | case SUBCTXT_RCV_HDRQ: | |
639 | memaddr = (u64)uctxt->subctxt_rcvhdr_base; | |
640 | memlen = uctxt->rcvhdrq_size * uctxt->subctxt_cnt; | |
641 | flags |= VM_IO | VM_DONTEXPAND; | |
642 | vmf = 1; | |
643 | break; | |
644 | case SUBCTXT_EGRBUF: | |
645 | memaddr = (u64)uctxt->subctxt_rcvegrbuf; | |
646 | memlen = uctxt->egrbufs.size * uctxt->subctxt_cnt; | |
647 | flags |= VM_IO | VM_DONTEXPAND; | |
648 | flags &= ~VM_MAYWRITE; | |
649 | vmf = 1; | |
650 | break; | |
651 | case SDMA_COMP: { | |
9e10af47 | 652 | struct hfi1_user_sdma_comp_q *cq = fd->cq; |
77241056 | 653 | |
9e10af47 | 654 | if (!cq) { |
77241056 MM |
655 | ret = -EFAULT; |
656 | goto done; | |
657 | } | |
77241056 | 658 | memaddr = (u64)cq->comps; |
437b29d1 | 659 | memlen = PAGE_ALIGN(sizeof(*cq->comps) * cq->nentries); |
77241056 MM |
660 | flags |= VM_IO | VM_DONTEXPAND; |
661 | vmf = 1; | |
662 | break; | |
663 | } | |
664 | default: | |
665 | ret = -EINVAL; | |
666 | break; | |
667 | } | |
668 | ||
669 | if ((vma->vm_end - vma->vm_start) != memlen) { | |
670 | hfi1_cdbg(PROC, "%u:%u Memory size mismatch %lu:%lu", | |
9e10af47 | 671 | uctxt->ctxt, fd->subctxt, |
77241056 MM |
672 | (vma->vm_end - vma->vm_start), memlen); |
673 | ret = -EINVAL; | |
674 | goto done; | |
675 | } | |
676 | ||
677 | vma->vm_flags = flags; | |
6c63e423 SS |
678 | hfi1_cdbg(PROC, |
679 | "%u:%u type:%u io/vf:%d/%d, addr:0x%llx, len:%lu(%lu), flags:0x%lx\n", | |
680 | ctxt, subctxt, type, mapio, vmf, memaddr, memlen, | |
77241056 | 681 | vma->vm_end - vma->vm_start, vma->vm_flags); |
77241056 | 682 | if (vmf) { |
60368186 | 683 | vma->vm_pgoff = PFN_DOWN(memaddr); |
77241056 MM |
684 | vma->vm_ops = &vm_ops; |
685 | ret = 0; | |
686 | } else if (mapio) { | |
60368186 TK |
687 | ret = io_remap_pfn_range(vma, vma->vm_start, |
688 | PFN_DOWN(memaddr), | |
689 | memlen, | |
77241056 | 690 | vma->vm_page_prot); |
60368186 TK |
691 | } else if (memvirt) { |
692 | ret = remap_pfn_range(vma, vma->vm_start, | |
693 | PFN_DOWN(__pa(memvirt)), | |
694 | memlen, | |
695 | vma->vm_page_prot); | |
77241056 | 696 | } else { |
60368186 TK |
697 | ret = remap_pfn_range(vma, vma->vm_start, |
698 | PFN_DOWN(memaddr), | |
699 | memlen, | |
77241056 MM |
700 | vma->vm_page_prot); |
701 | } | |
702 | done: | |
703 | return ret; | |
704 | } | |
705 | ||
706 | /* | |
707 | * Local (non-chip) user memory is not mapped right away but as it is | |
708 | * accessed by the user-level code. | |
709 | */ | |
11bac800 | 710 | static int vma_fault(struct vm_fault *vmf) |
77241056 MM |
711 | { |
712 | struct page *page; | |
713 | ||
714 | page = vmalloc_to_page((void *)(vmf->pgoff << PAGE_SHIFT)); | |
715 | if (!page) | |
716 | return VM_FAULT_SIGBUS; | |
717 | ||
718 | get_page(page); | |
719 | vmf->page = page; | |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
724 | static unsigned int hfi1_poll(struct file *fp, struct poll_table_struct *pt) | |
725 | { | |
726 | struct hfi1_ctxtdata *uctxt; | |
727 | unsigned pollflag; | |
728 | ||
9e10af47 | 729 | uctxt = ((struct hfi1_filedata *)fp->private_data)->uctxt; |
77241056 MM |
730 | if (!uctxt) |
731 | pollflag = POLLERR; | |
732 | else if (uctxt->poll_type == HFI1_POLL_TYPE_URGENT) | |
733 | pollflag = poll_urgent(fp, pt); | |
734 | else if (uctxt->poll_type == HFI1_POLL_TYPE_ANYRCV) | |
735 | pollflag = poll_next(fp, pt); | |
736 | else /* invalid */ | |
737 | pollflag = POLLERR; | |
738 | ||
739 | return pollflag; | |
740 | } | |
741 | ||
742 | static int hfi1_file_close(struct inode *inode, struct file *fp) | |
743 | { | |
744 | struct hfi1_filedata *fdata = fp->private_data; | |
745 | struct hfi1_ctxtdata *uctxt = fdata->uctxt; | |
e11ffbd5 DD |
746 | struct hfi1_devdata *dd = container_of(inode->i_cdev, |
747 | struct hfi1_devdata, | |
748 | user_cdev); | |
77241056 MM |
749 | unsigned long flags, *ev; |
750 | ||
751 | fp->private_data = NULL; | |
752 | ||
753 | if (!uctxt) | |
754 | goto done; | |
755 | ||
756 | hfi1_cdbg(PROC, "freeing ctxt %u:%u", uctxt->ctxt, fdata->subctxt); | |
77241056 MM |
757 | mutex_lock(&hfi1_mutex); |
758 | ||
759 | flush_wc(); | |
760 | /* drain user sdma queue */ | |
483119a7 | 761 | hfi1_user_sdma_free_queues(fdata); |
77241056 | 762 | |
957558c9 | 763 | /* release the cpu */ |
b094a36f | 764 | hfi1_put_proc_affinity(fdata->rec_cpu_num); |
957558c9 | 765 | |
224d71f9 MR |
766 | /* clean up rcv side */ |
767 | hfi1_user_exp_rcv_free(fdata); | |
768 | ||
77241056 MM |
769 | /* |
770 | * Clear any left over, unhandled events so the next process that | |
771 | * gets this context doesn't get confused. | |
772 | */ | |
2280740f | 773 | ev = dd->events + ((uctxt->ctxt - dd->first_dyn_alloc_ctxt) * |
77241056 MM |
774 | HFI1_MAX_SHARED_CTXTS) + fdata->subctxt; |
775 | *ev = 0; | |
776 | ||
8737ce95 | 777 | __clear_bit(fdata->subctxt, uctxt->in_use_ctxts); |
f683c80c MR |
778 | fdata->uctxt = NULL; |
779 | hfi1_rcd_put(uctxt); /* fdata reference */ | |
8737ce95 | 780 | if (!bitmap_empty(uctxt->in_use_ctxts, HFI1_MAX_SHARED_CTXTS)) { |
77241056 MM |
781 | mutex_unlock(&hfi1_mutex); |
782 | goto done; | |
783 | } | |
784 | ||
785 | spin_lock_irqsave(&dd->uctxt_lock, flags); | |
786 | /* | |
787 | * Disable receive context and interrupt available, reset all | |
788 | * RcvCtxtCtrl bits to default values. | |
789 | */ | |
790 | hfi1_rcvctrl(dd, HFI1_RCVCTRL_CTXT_DIS | | |
791 | HFI1_RCVCTRL_TIDFLOW_DIS | | |
792 | HFI1_RCVCTRL_INTRAVAIL_DIS | | |
566c157c | 793 | HFI1_RCVCTRL_TAILUPD_DIS | |
77241056 MM |
794 | HFI1_RCVCTRL_ONE_PKT_EGR_DIS | |
795 | HFI1_RCVCTRL_NO_RHQ_DROP_DIS | | |
796 | HFI1_RCVCTRL_NO_EGR_DROP_DIS, uctxt->ctxt); | |
797 | /* Clear the context's J_KEY */ | |
798 | hfi1_clear_ctxt_jkey(dd, uctxt->ctxt); | |
799 | /* | |
f683c80c MR |
800 | * If a send context is allocated, reset context integrity |
801 | * checks to default and disable the send context. | |
77241056 | 802 | */ |
f683c80c MR |
803 | if (uctxt->sc) { |
804 | set_pio_integrity(uctxt->sc); | |
805 | sc_disable(uctxt->sc); | |
806 | } | |
77241056 MM |
807 | spin_unlock_irqrestore(&dd->uctxt_lock, flags); |
808 | ||
9c1a99c3 | 809 | hfi1_free_ctxt_rcv_groups(uctxt); |
637a9a7f | 810 | hfi1_clear_ctxt_pkey(dd, uctxt); |
94158442 | 811 | |
77241056 | 812 | uctxt->event_flags = 0; |
77241056 | 813 | mutex_unlock(&hfi1_mutex); |
42492011 MR |
814 | |
815 | deallocate_ctxt(uctxt); | |
77241056 | 816 | done: |
e0cf75de | 817 | mmdrop(fdata->mm); |
e11ffbd5 | 818 | kobject_put(&dd->kobj); |
acd7c8fe TS |
819 | |
820 | if (atomic_dec_and_test(&dd->user_refcount)) | |
821 | complete(&dd->user_comp); | |
822 | ||
77241056 MM |
823 | kfree(fdata); |
824 | return 0; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Convert kernel *virtual* addresses to physical addresses. | |
829 | * This is used to vmalloc'ed addresses. | |
830 | */ | |
831 | static u64 kvirt_to_phys(void *addr) | |
832 | { | |
833 | struct page *page; | |
834 | u64 paddr = 0; | |
835 | ||
836 | page = vmalloc_to_page(addr); | |
837 | if (page) | |
838 | paddr = page_to_pfn(page) << PAGE_SHIFT; | |
839 | ||
840 | return paddr; | |
841 | } | |
842 | ||
5fbded48 | 843 | static int assign_ctxt(struct hfi1_filedata *fd, struct hfi1_user_info *uinfo) |
77241056 | 844 | { |
62239fc6 | 845 | int ret; |
0eb62659 | 846 | unsigned int swmajor, swminor; |
77241056 MM |
847 | |
848 | swmajor = uinfo->userversion >> 16; | |
9b60d2cb MR |
849 | if (swmajor != HFI1_USER_SWMAJOR) |
850 | return -ENODEV; | |
77241056 MM |
851 | |
852 | swminor = uinfo->userversion & 0xffff; | |
853 | ||
77241056 | 854 | mutex_lock(&hfi1_mutex); |
62239fc6 MR |
855 | /* |
856 | * Get a sub context if necessary. | |
857 | * ret < 0 error, 0 no context, 1 sub-context found | |
858 | */ | |
859 | ret = 0; | |
957558c9 | 860 | if (uinfo->subctxt_cnt) { |
9b60d2cb | 861 | ret = find_sub_ctxt(fd, uinfo); |
62239fc6 | 862 | if (ret > 0) |
b094a36f SS |
863 | fd->rec_cpu_num = |
864 | hfi1_get_proc_affinity(fd->uctxt->numa_id); | |
957558c9 | 865 | } |
77241056 MM |
866 | |
867 | /* | |
8737ce95 | 868 | * Allocate a base context if context sharing is not required or we |
9b60d2cb | 869 | * couldn't find a sub context. |
77241056 | 870 | */ |
5fbded48 MR |
871 | if (!ret) |
872 | ret = allocate_ctxt(fd, fd->dd, uinfo); | |
873 | ||
77241056 | 874 | mutex_unlock(&hfi1_mutex); |
9b60d2cb MR |
875 | |
876 | /* Depending on the context type, do the appropriate init */ | |
877 | if (ret > 0) { | |
878 | /* | |
879 | * sub-context info can only be set up after the base | |
880 | * context has been completed. | |
881 | */ | |
882 | ret = wait_event_interruptible(fd->uctxt->wait, !test_bit( | |
883 | HFI1_CTXT_BASE_UNINIT, | |
884 | &fd->uctxt->event_flags)); | |
f683c80c MR |
885 | if (test_bit(HFI1_CTXT_BASE_FAILED, &fd->uctxt->event_flags)) |
886 | ret = -ENOMEM; | |
887 | ||
9b60d2cb MR |
888 | /* The only thing a sub context needs is the user_xxx stuff */ |
889 | if (!ret) | |
62239fc6 MR |
890 | ret = init_user_ctxt(fd); |
891 | ||
42492011 | 892 | if (ret) |
62239fc6 | 893 | clear_bit(fd->subctxt, fd->uctxt->in_use_ctxts); |
42492011 | 894 | |
9b60d2cb MR |
895 | } else if (!ret) { |
896 | ret = setup_base_ctxt(fd); | |
62239fc6 MR |
897 | if (fd->uctxt->subctxt_cnt) { |
898 | /* If there is an error, set the failed bit. */ | |
899 | if (ret) | |
900 | set_bit(HFI1_CTXT_BASE_FAILED, | |
901 | &fd->uctxt->event_flags); | |
902 | /* | |
903 | * Base context is done, notify anybody using a | |
904 | * sub-context that is waiting for this completion | |
905 | */ | |
9b60d2cb MR |
906 | clear_bit(HFI1_CTXT_BASE_UNINIT, |
907 | &fd->uctxt->event_flags); | |
908 | wake_up(&fd->uctxt->wait); | |
909 | } | |
42492011 MR |
910 | if (ret) |
911 | deallocate_ctxt(fd->uctxt); | |
912 | } | |
913 | ||
914 | /* If an error occurred, clear the reference */ | |
915 | if (ret && fd->uctxt) { | |
916 | hfi1_rcd_put(fd->uctxt); | |
917 | fd->uctxt = NULL; | |
9b60d2cb MR |
918 | } |
919 | ||
77241056 MM |
920 | return ret; |
921 | } | |
922 | ||
8737ce95 MR |
923 | /* |
924 | * The hfi1_mutex must be held when this function is called. It is | |
925 | * necessary to ensure serialized access to the bitmask in_use_ctxts. | |
926 | */ | |
9b60d2cb MR |
927 | static int find_sub_ctxt(struct hfi1_filedata *fd, |
928 | const struct hfi1_user_info *uinfo) | |
77241056 | 929 | { |
5fbded48 MR |
930 | int i; |
931 | struct hfi1_devdata *dd = fd->dd; | |
8737ce95 | 932 | u16 subctxt; |
77241056 | 933 | |
5fbded48 MR |
934 | for (i = dd->first_dyn_alloc_ctxt; i < dd->num_rcv_contexts; i++) { |
935 | struct hfi1_ctxtdata *uctxt = dd->rcd[i]; | |
77241056 | 936 | |
5fbded48 | 937 | /* Skip ctxts which are not yet open */ |
8737ce95 MR |
938 | if (!uctxt || |
939 | bitmap_empty(uctxt->in_use_ctxts, | |
940 | HFI1_MAX_SHARED_CTXTS)) | |
5fbded48 | 941 | continue; |
77241056 | 942 | |
5fbded48 MR |
943 | /* Skip dynamically allocted kernel contexts */ |
944 | if (uctxt->sc && (uctxt->sc->type == SC_KERNEL)) | |
77241056 | 945 | continue; |
5fbded48 MR |
946 | |
947 | /* Skip ctxt if it doesn't match the requested one */ | |
948 | if (memcmp(uctxt->uuid, uinfo->uuid, | |
949 | sizeof(uctxt->uuid)) || | |
950 | uctxt->jkey != generate_jkey(current_uid()) || | |
951 | uctxt->subctxt_id != uinfo->subctxt_id || | |
952 | uctxt->subctxt_cnt != uinfo->subctxt_cnt) | |
953 | continue; | |
954 | ||
955 | /* Verify the sharing process matches the master */ | |
8737ce95 | 956 | if (uctxt->userversion != uinfo->userversion) |
5fbded48 | 957 | return -EINVAL; |
8737ce95 MR |
958 | |
959 | /* Find an unused context */ | |
960 | subctxt = find_first_zero_bit(uctxt->in_use_ctxts, | |
961 | HFI1_MAX_SHARED_CTXTS); | |
962 | if (subctxt >= uctxt->subctxt_cnt) | |
62239fc6 | 963 | return -EBUSY; |
8737ce95 | 964 | |
5fbded48 | 965 | fd->uctxt = uctxt; |
8737ce95 | 966 | fd->subctxt = subctxt; |
f683c80c MR |
967 | |
968 | hfi1_rcd_get(uctxt); | |
8737ce95 MR |
969 | __set_bit(fd->subctxt, uctxt->in_use_ctxts); |
970 | ||
5fbded48 | 971 | return 1; |
77241056 MM |
972 | } |
973 | ||
5fbded48 | 974 | return 0; |
77241056 MM |
975 | } |
976 | ||
5042cddf | 977 | static int allocate_ctxt(struct hfi1_filedata *fd, struct hfi1_devdata *dd, |
77241056 MM |
978 | struct hfi1_user_info *uinfo) |
979 | { | |
980 | struct hfi1_ctxtdata *uctxt; | |
5fbded48 | 981 | unsigned int ctxt; |
957558c9 | 982 | int ret, numa; |
77241056 MM |
983 | |
984 | if (dd->flags & HFI1_FROZEN) { | |
985 | /* | |
986 | * Pick an error that is unique from all other errors | |
987 | * that are returned so the user process knows that | |
988 | * it tried to allocate while the SPC was frozen. It | |
989 | * it should be able to retry with success in a short | |
990 | * while. | |
991 | */ | |
992 | return -EIO; | |
993 | } | |
994 | ||
5fbded48 MR |
995 | /* |
996 | * This check is sort of redundant to the next EBUSY error. It would | |
997 | * also indicate an inconsistancy in the driver if this value was | |
998 | * zero, but there were still contexts available. | |
999 | */ | |
1000 | if (!dd->freectxts) | |
1001 | return -EBUSY; | |
1002 | ||
2280740f VN |
1003 | for (ctxt = dd->first_dyn_alloc_ctxt; |
1004 | ctxt < dd->num_rcv_contexts; ctxt++) | |
77241056 MM |
1005 | if (!dd->rcd[ctxt]) |
1006 | break; | |
1007 | ||
1008 | if (ctxt == dd->num_rcv_contexts) | |
1009 | return -EBUSY; | |
1010 | ||
b094a36f SS |
1011 | /* |
1012 | * If we don't have a NUMA node requested, preference is towards | |
1013 | * device NUMA node. | |
1014 | */ | |
1015 | fd->rec_cpu_num = hfi1_get_proc_affinity(dd->node); | |
957558c9 MH |
1016 | if (fd->rec_cpu_num != -1) |
1017 | numa = cpu_to_node(fd->rec_cpu_num); | |
1018 | else | |
1019 | numa = numa_node_id(); | |
1020 | uctxt = hfi1_create_ctxtdata(dd->pport, ctxt, numa); | |
77241056 MM |
1021 | if (!uctxt) { |
1022 | dd_dev_err(dd, | |
1023 | "Unable to allocate ctxtdata memory, failing open\n"); | |
1024 | return -ENOMEM; | |
1025 | } | |
957558c9 MH |
1026 | hfi1_cdbg(PROC, "[%u:%u] pid %u assigned to CPU %d (NUMA %u)", |
1027 | uctxt->ctxt, fd->subctxt, current->pid, fd->rec_cpu_num, | |
1028 | uctxt->numa_id); | |
1029 | ||
77241056 MM |
1030 | /* |
1031 | * Allocate and enable a PIO send context. | |
1032 | */ | |
1033 | uctxt->sc = sc_alloc(dd, SC_USER, uctxt->rcvhdrqentsize, | |
cc57236f | 1034 | uctxt->dd->node); |
3a6982df JP |
1035 | if (!uctxt->sc) { |
1036 | ret = -ENOMEM; | |
1037 | goto ctxdata_free; | |
1038 | } | |
6c63e423 SS |
1039 | hfi1_cdbg(PROC, "allocated send context %u(%u)\n", uctxt->sc->sw_index, |
1040 | uctxt->sc->hw_context); | |
77241056 MM |
1041 | ret = sc_enable(uctxt->sc); |
1042 | if (ret) | |
3a6982df JP |
1043 | goto ctxdata_free; |
1044 | ||
77241056 | 1045 | /* |
9b60d2cb MR |
1046 | * Setup sub context resources if the user-level has requested |
1047 | * sub contexts. | |
77241056 MM |
1048 | * This has to be done here so the rest of the sub-contexts find the |
1049 | * proper master. | |
1050 | */ | |
9b60d2cb | 1051 | if (uinfo->subctxt_cnt) { |
77241056 MM |
1052 | ret = init_subctxts(uctxt, uinfo); |
1053 | /* | |
1054 | * On error, we don't need to disable and de-allocate the | |
1055 | * send context because it will be done during file close | |
1056 | */ | |
1057 | if (ret) | |
3a6982df | 1058 | goto ctxdata_free; |
77241056 MM |
1059 | } |
1060 | uctxt->userversion = uinfo->userversion; | |
bdf7752e | 1061 | uctxt->flags = hfi1_cap_mask; /* save current flag state */ |
77241056 MM |
1062 | init_waitqueue_head(&uctxt->wait); |
1063 | strlcpy(uctxt->comm, current->comm, sizeof(uctxt->comm)); | |
1064 | memcpy(uctxt->uuid, uinfo->uuid, sizeof(uctxt->uuid)); | |
1065 | uctxt->jkey = generate_jkey(current_uid()); | |
77241056 | 1066 | hfi1_stats.sps_ctxts++; |
affa48de AD |
1067 | /* |
1068 | * Disable ASPM when there are open user/PSM contexts to avoid | |
1069 | * issues with ASPM L1 exit latency | |
1070 | */ | |
1071 | if (dd->freectxts-- == dd->num_user_contexts) | |
1072 | aspm_disable_all(dd); | |
9e10af47 | 1073 | fd->uctxt = uctxt; |
77241056 | 1074 | |
f683c80c MR |
1075 | /* Count the reference for the fd */ |
1076 | hfi1_rcd_get(uctxt); | |
1077 | ||
77241056 | 1078 | return 0; |
3a6982df JP |
1079 | |
1080 | ctxdata_free: | |
1081 | dd->rcd[ctxt] = NULL; | |
f683c80c | 1082 | hfi1_rcd_put(uctxt); |
3a6982df | 1083 | return ret; |
77241056 MM |
1084 | } |
1085 | ||
42492011 MR |
1086 | static void deallocate_ctxt(struct hfi1_ctxtdata *uctxt) |
1087 | { | |
1088 | mutex_lock(&hfi1_mutex); | |
1089 | hfi1_stats.sps_ctxts--; | |
1090 | if (++uctxt->dd->freectxts == uctxt->dd->num_user_contexts) | |
1091 | aspm_enable_all(uctxt->dd); | |
1092 | ||
1093 | /* _rcd_put() should be done after releasing mutex */ | |
1094 | uctxt->dd->rcd[uctxt->ctxt] = NULL; | |
1095 | mutex_unlock(&hfi1_mutex); | |
1096 | hfi1_rcd_put(uctxt); /* dd reference */ | |
1097 | } | |
1098 | ||
77241056 MM |
1099 | static int init_subctxts(struct hfi1_ctxtdata *uctxt, |
1100 | const struct hfi1_user_info *uinfo) | |
1101 | { | |
8737ce95 | 1102 | u16 num_subctxts; |
77241056 MM |
1103 | |
1104 | num_subctxts = uinfo->subctxt_cnt; | |
acac10fd MH |
1105 | if (num_subctxts > HFI1_MAX_SHARED_CTXTS) |
1106 | return -EINVAL; | |
77241056 MM |
1107 | |
1108 | uctxt->subctxt_cnt = uinfo->subctxt_cnt; | |
1109 | uctxt->subctxt_id = uinfo->subctxt_id; | |
77241056 | 1110 | uctxt->redirect_seq_cnt = 1; |
9b60d2cb | 1111 | set_bit(HFI1_CTXT_BASE_UNINIT, &uctxt->event_flags); |
acac10fd MH |
1112 | |
1113 | return 0; | |
77241056 MM |
1114 | } |
1115 | ||
1116 | static int setup_subctxt(struct hfi1_ctxtdata *uctxt) | |
1117 | { | |
1118 | int ret = 0; | |
8737ce95 | 1119 | u16 num_subctxts = uctxt->subctxt_cnt; |
77241056 MM |
1120 | |
1121 | uctxt->subctxt_uregbase = vmalloc_user(PAGE_SIZE); | |
9b60d2cb MR |
1122 | if (!uctxt->subctxt_uregbase) |
1123 | return -ENOMEM; | |
1124 | ||
77241056 MM |
1125 | /* We can take the size of the RcvHdr Queue from the master */ |
1126 | uctxt->subctxt_rcvhdr_base = vmalloc_user(uctxt->rcvhdrq_size * | |
1127 | num_subctxts); | |
1128 | if (!uctxt->subctxt_rcvhdr_base) { | |
1129 | ret = -ENOMEM; | |
1130 | goto bail_ureg; | |
1131 | } | |
1132 | ||
1133 | uctxt->subctxt_rcvegrbuf = vmalloc_user(uctxt->egrbufs.size * | |
1134 | num_subctxts); | |
1135 | if (!uctxt->subctxt_rcvegrbuf) { | |
1136 | ret = -ENOMEM; | |
1137 | goto bail_rhdr; | |
1138 | } | |
9b60d2cb MR |
1139 | |
1140 | return 0; | |
1141 | ||
77241056 MM |
1142 | bail_rhdr: |
1143 | vfree(uctxt->subctxt_rcvhdr_base); | |
9b60d2cb | 1144 | uctxt->subctxt_rcvhdr_base = NULL; |
77241056 MM |
1145 | bail_ureg: |
1146 | vfree(uctxt->subctxt_uregbase); | |
1147 | uctxt->subctxt_uregbase = NULL; | |
9b60d2cb | 1148 | |
77241056 MM |
1149 | return ret; |
1150 | } | |
1151 | ||
62239fc6 | 1152 | static void user_init(struct hfi1_ctxtdata *uctxt) |
77241056 | 1153 | { |
77241056 | 1154 | unsigned int rcvctrl_ops = 0; |
77241056 MM |
1155 | |
1156 | /* initialize poll variables... */ | |
1157 | uctxt->urgent = 0; | |
1158 | uctxt->urgent_poll = 0; | |
1159 | ||
1160 | /* | |
1161 | * Now enable the ctxt for receive. | |
1162 | * For chips that are set to DMA the tail register to memory | |
1163 | * when they change (and when the update bit transitions from | |
1164 | * 0 to 1. So for those chips, we turn it off and then back on. | |
1165 | * This will (very briefly) affect any other open ctxts, but the | |
1166 | * duration is very short, and therefore isn't an issue. We | |
1167 | * explicitly set the in-memory tail copy to 0 beforehand, so we | |
1168 | * don't have to wait to be sure the DMA update has happened | |
1169 | * (chip resets head/tail to 0 on transition to enable). | |
1170 | */ | |
1171 | if (uctxt->rcvhdrtail_kvaddr) | |
1172 | clear_rcvhdrtail(uctxt); | |
1173 | ||
1174 | /* Setup J_KEY before enabling the context */ | |
1175 | hfi1_set_ctxt_jkey(uctxt->dd, uctxt->ctxt, uctxt->jkey); | |
1176 | ||
1177 | rcvctrl_ops = HFI1_RCVCTRL_CTXT_ENB; | |
bdf7752e | 1178 | if (HFI1_CAP_UGET_MASK(uctxt->flags, HDRSUPP)) |
77241056 MM |
1179 | rcvctrl_ops |= HFI1_RCVCTRL_TIDFLOW_ENB; |
1180 | /* | |
1181 | * Ignore the bit in the flags for now until proper | |
1182 | * support for multiple packet per rcv array entry is | |
1183 | * added. | |
1184 | */ | |
bdf7752e | 1185 | if (!HFI1_CAP_UGET_MASK(uctxt->flags, MULTI_PKT_EGR)) |
77241056 | 1186 | rcvctrl_ops |= HFI1_RCVCTRL_ONE_PKT_EGR_ENB; |
bdf7752e | 1187 | if (HFI1_CAP_UGET_MASK(uctxt->flags, NODROP_EGR_FULL)) |
77241056 | 1188 | rcvctrl_ops |= HFI1_RCVCTRL_NO_EGR_DROP_ENB; |
bdf7752e | 1189 | if (HFI1_CAP_UGET_MASK(uctxt->flags, NODROP_RHQ_FULL)) |
77241056 | 1190 | rcvctrl_ops |= HFI1_RCVCTRL_NO_RHQ_DROP_ENB; |
566c157c MH |
1191 | /* |
1192 | * The RcvCtxtCtrl.TailUpd bit has to be explicitly written. | |
1193 | * We can't rely on the correct value to be set from prior | |
1194 | * uses of the chip or ctxt. Therefore, add the rcvctrl op | |
1195 | * for both cases. | |
1196 | */ | |
bdf7752e | 1197 | if (HFI1_CAP_UGET_MASK(uctxt->flags, DMA_RTAIL)) |
77241056 | 1198 | rcvctrl_ops |= HFI1_RCVCTRL_TAILUPD_ENB; |
566c157c MH |
1199 | else |
1200 | rcvctrl_ops |= HFI1_RCVCTRL_TAILUPD_DIS; | |
77241056 | 1201 | hfi1_rcvctrl(uctxt->dd, rcvctrl_ops, uctxt->ctxt); |
77241056 MM |
1202 | } |
1203 | ||
5042cddf MR |
1204 | static int get_ctxt_info(struct hfi1_filedata *fd, void __user *ubase, |
1205 | __u32 len) | |
77241056 MM |
1206 | { |
1207 | struct hfi1_ctxt_info cinfo; | |
9e10af47 | 1208 | struct hfi1_ctxtdata *uctxt = fd->uctxt; |
77241056 MM |
1209 | int ret = 0; |
1210 | ||
ebe6b2e8 | 1211 | memset(&cinfo, 0, sizeof(cinfo)); |
bdf7752e DL |
1212 | cinfo.runtime_flags = (((uctxt->flags >> HFI1_CAP_MISC_SHIFT) & |
1213 | HFI1_CAP_MISC_MASK) << HFI1_CAP_USER_SHIFT) | | |
1214 | HFI1_CAP_UGET_MASK(uctxt->flags, MASK) | | |
1215 | HFI1_CAP_KGET_MASK(uctxt->flags, K2U); | |
622c202c DL |
1216 | /* adjust flag if this fd is not able to cache */ |
1217 | if (!fd->handler) | |
1218 | cinfo.runtime_flags |= HFI1_CAP_TID_UNMAP; /* no caching */ | |
1219 | ||
77241056 MM |
1220 | cinfo.num_active = hfi1_count_active_units(); |
1221 | cinfo.unit = uctxt->dd->unit; | |
1222 | cinfo.ctxt = uctxt->ctxt; | |
9e10af47 | 1223 | cinfo.subctxt = fd->subctxt; |
77241056 MM |
1224 | cinfo.rcvtids = roundup(uctxt->egrbufs.alloced, |
1225 | uctxt->dd->rcv_entries.group_size) + | |
1226 | uctxt->expected_count; | |
1227 | cinfo.credits = uctxt->sc->credits; | |
1228 | cinfo.numa_node = uctxt->numa_id; | |
1229 | cinfo.rec_cpu = fd->rec_cpu_num; | |
1230 | cinfo.send_ctxt = uctxt->sc->hw_context; | |
1231 | ||
1232 | cinfo.egrtids = uctxt->egrbufs.alloced; | |
1233 | cinfo.rcvhdrq_cnt = uctxt->rcvhdrq_cnt; | |
1234 | cinfo.rcvhdrq_entsize = uctxt->rcvhdrqentsize << 2; | |
9e10af47 | 1235 | cinfo.sdma_ring_size = fd->cq->nentries; |
77241056 MM |
1236 | cinfo.rcvegr_size = uctxt->egrbufs.rcvtid_size; |
1237 | ||
9e10af47 | 1238 | trace_hfi1_ctxt_info(uctxt->dd, uctxt->ctxt, fd->subctxt, cinfo); |
77241056 MM |
1239 | if (copy_to_user(ubase, &cinfo, sizeof(cinfo))) |
1240 | ret = -EFAULT; | |
bdf7752e | 1241 | |
77241056 MM |
1242 | return ret; |
1243 | } | |
1244 | ||
9b60d2cb MR |
1245 | static int init_user_ctxt(struct hfi1_filedata *fd) |
1246 | { | |
1247 | struct hfi1_ctxtdata *uctxt = fd->uctxt; | |
1248 | int ret; | |
1249 | ||
1250 | ret = hfi1_user_sdma_alloc_queues(uctxt, fd); | |
1251 | if (ret) | |
1252 | return ret; | |
1253 | ||
1254 | ret = hfi1_user_exp_rcv_init(fd); | |
1255 | ||
1256 | return ret; | |
1257 | } | |
1258 | ||
1259 | static int setup_base_ctxt(struct hfi1_filedata *fd) | |
77241056 | 1260 | { |
9e10af47 | 1261 | struct hfi1_ctxtdata *uctxt = fd->uctxt; |
77241056 MM |
1262 | struct hfi1_devdata *dd = uctxt->dd; |
1263 | int ret = 0; | |
1264 | ||
9b60d2cb | 1265 | hfi1_init_ctxt(uctxt->sc); |
77241056 | 1266 | |
9b60d2cb MR |
1267 | /* Now allocate the RcvHdr queue and eager buffers. */ |
1268 | ret = hfi1_create_rcvhdrq(dd, uctxt); | |
1269 | if (ret) | |
62239fc6 | 1270 | return ret; |
94158442 | 1271 | |
9b60d2cb | 1272 | ret = hfi1_setup_eagerbufs(uctxt); |
94158442 | 1273 | if (ret) |
62239fc6 | 1274 | goto setup_failed; |
9b60d2cb MR |
1275 | |
1276 | /* If sub-contexts are enabled, do the appropriate setup */ | |
1277 | if (uctxt->subctxt_cnt) | |
1278 | ret = setup_subctxt(uctxt); | |
1279 | if (ret) | |
62239fc6 | 1280 | goto setup_failed; |
9b60d2cb | 1281 | |
9c1a99c3 | 1282 | ret = hfi1_alloc_ctxt_rcv_groups(uctxt); |
9b60d2cb | 1283 | if (ret) |
62239fc6 | 1284 | goto setup_failed; |
9b60d2cb MR |
1285 | |
1286 | ret = init_user_ctxt(fd); | |
77241056 | 1287 | if (ret) |
62239fc6 | 1288 | goto setup_failed; |
77241056 | 1289 | |
62239fc6 MR |
1290 | user_init(uctxt); |
1291 | ||
1292 | return 0; | |
1293 | ||
1294 | setup_failed: | |
f683c80c | 1295 | /* Call _free_ctxtdata, not _rcd_put(). We still need the context. */ |
62239fc6 | 1296 | hfi1_free_ctxtdata(dd, uctxt); |
77241056 MM |
1297 | return ret; |
1298 | } | |
1299 | ||
5042cddf MR |
1300 | static int get_base_info(struct hfi1_filedata *fd, void __user *ubase, |
1301 | __u32 len) | |
77241056 MM |
1302 | { |
1303 | struct hfi1_base_info binfo; | |
9e10af47 | 1304 | struct hfi1_ctxtdata *uctxt = fd->uctxt; |
77241056 MM |
1305 | struct hfi1_devdata *dd = uctxt->dd; |
1306 | ssize_t sz; | |
1307 | unsigned offset; | |
1308 | int ret = 0; | |
1309 | ||
9b60d2cb | 1310 | trace_hfi1_uctxtdata(uctxt->dd, uctxt, fd->subctxt); |
77241056 MM |
1311 | |
1312 | memset(&binfo, 0, sizeof(binfo)); | |
1313 | binfo.hw_version = dd->revision; | |
1314 | binfo.sw_version = HFI1_KERN_SWVERSION; | |
1315 | binfo.bthqp = kdeth_qp; | |
1316 | binfo.jkey = uctxt->jkey; | |
1317 | /* | |
1318 | * If more than 64 contexts are enabled the allocated credit | |
1319 | * return will span two or three contiguous pages. Since we only | |
1320 | * map the page containing the context's credit return address, | |
1321 | * we need to calculate the offset in the proper page. | |
1322 | */ | |
1323 | offset = ((u64)uctxt->sc->hw_free - | |
1324 | (u64)dd->cr_base[uctxt->numa_id].va) % PAGE_SIZE; | |
1325 | binfo.sc_credits_addr = HFI1_MMAP_TOKEN(PIO_CRED, uctxt->ctxt, | |
9e10af47 | 1326 | fd->subctxt, offset); |
77241056 | 1327 | binfo.pio_bufbase = HFI1_MMAP_TOKEN(PIO_BUFS, uctxt->ctxt, |
9e10af47 | 1328 | fd->subctxt, |
77241056 MM |
1329 | uctxt->sc->base_addr); |
1330 | binfo.pio_bufbase_sop = HFI1_MMAP_TOKEN(PIO_BUFS_SOP, | |
1331 | uctxt->ctxt, | |
9e10af47 | 1332 | fd->subctxt, |
77241056 MM |
1333 | uctxt->sc->base_addr); |
1334 | binfo.rcvhdr_bufbase = HFI1_MMAP_TOKEN(RCV_HDRQ, uctxt->ctxt, | |
9e10af47 | 1335 | fd->subctxt, |
77241056 MM |
1336 | uctxt->rcvhdrq); |
1337 | binfo.rcvegr_bufbase = HFI1_MMAP_TOKEN(RCV_EGRBUF, uctxt->ctxt, | |
9e10af47 | 1338 | fd->subctxt, |
60368186 | 1339 | uctxt->egrbufs.rcvtids[0].dma); |
77241056 | 1340 | binfo.sdma_comp_bufbase = HFI1_MMAP_TOKEN(SDMA_COMP, uctxt->ctxt, |
9e10af47 | 1341 | fd->subctxt, 0); |
77241056 MM |
1342 | /* |
1343 | * user regs are at | |
1344 | * (RXE_PER_CONTEXT_USER + (ctxt * RXE_PER_CONTEXT_SIZE)) | |
1345 | */ | |
1346 | binfo.user_regbase = HFI1_MMAP_TOKEN(UREGS, uctxt->ctxt, | |
9e10af47 | 1347 | fd->subctxt, 0); |
2280740f | 1348 | offset = offset_in_page((((uctxt->ctxt - dd->first_dyn_alloc_ctxt) * |
9e10af47 | 1349 | HFI1_MAX_SHARED_CTXTS) + fd->subctxt) * |
e260e404 | 1350 | sizeof(*dd->events)); |
77241056 | 1351 | binfo.events_bufbase = HFI1_MMAP_TOKEN(EVENTS, uctxt->ctxt, |
9e10af47 | 1352 | fd->subctxt, |
77241056 MM |
1353 | offset); |
1354 | binfo.status_bufbase = HFI1_MMAP_TOKEN(STATUS, uctxt->ctxt, | |
9e10af47 | 1355 | fd->subctxt, |
77241056 MM |
1356 | dd->status); |
1357 | if (HFI1_CAP_IS_USET(DMA_RTAIL)) | |
1358 | binfo.rcvhdrtail_base = HFI1_MMAP_TOKEN(RTAIL, uctxt->ctxt, | |
9e10af47 | 1359 | fd->subctxt, 0); |
77241056 MM |
1360 | if (uctxt->subctxt_cnt) { |
1361 | binfo.subctxt_uregbase = HFI1_MMAP_TOKEN(SUBCTXT_UREGS, | |
1362 | uctxt->ctxt, | |
9e10af47 | 1363 | fd->subctxt, 0); |
77241056 MM |
1364 | binfo.subctxt_rcvhdrbuf = HFI1_MMAP_TOKEN(SUBCTXT_RCV_HDRQ, |
1365 | uctxt->ctxt, | |
9e10af47 | 1366 | fd->subctxt, 0); |
77241056 MM |
1367 | binfo.subctxt_rcvegrbuf = HFI1_MMAP_TOKEN(SUBCTXT_EGRBUF, |
1368 | uctxt->ctxt, | |
9e10af47 | 1369 | fd->subctxt, 0); |
77241056 MM |
1370 | } |
1371 | sz = (len < sizeof(binfo)) ? len : sizeof(binfo); | |
1372 | if (copy_to_user(ubase, &binfo, sz)) | |
1373 | ret = -EFAULT; | |
1374 | return ret; | |
1375 | } | |
1376 | ||
1377 | static unsigned int poll_urgent(struct file *fp, | |
1378 | struct poll_table_struct *pt) | |
1379 | { | |
9e10af47 IW |
1380 | struct hfi1_filedata *fd = fp->private_data; |
1381 | struct hfi1_ctxtdata *uctxt = fd->uctxt; | |
77241056 MM |
1382 | struct hfi1_devdata *dd = uctxt->dd; |
1383 | unsigned pollflag; | |
1384 | ||
1385 | poll_wait(fp, &uctxt->wait, pt); | |
1386 | ||
1387 | spin_lock_irq(&dd->uctxt_lock); | |
1388 | if (uctxt->urgent != uctxt->urgent_poll) { | |
1389 | pollflag = POLLIN | POLLRDNORM; | |
1390 | uctxt->urgent_poll = uctxt->urgent; | |
1391 | } else { | |
1392 | pollflag = 0; | |
1393 | set_bit(HFI1_CTXT_WAITING_URG, &uctxt->event_flags); | |
1394 | } | |
1395 | spin_unlock_irq(&dd->uctxt_lock); | |
1396 | ||
1397 | return pollflag; | |
1398 | } | |
1399 | ||
1400 | static unsigned int poll_next(struct file *fp, | |
1401 | struct poll_table_struct *pt) | |
1402 | { | |
9e10af47 IW |
1403 | struct hfi1_filedata *fd = fp->private_data; |
1404 | struct hfi1_ctxtdata *uctxt = fd->uctxt; | |
77241056 MM |
1405 | struct hfi1_devdata *dd = uctxt->dd; |
1406 | unsigned pollflag; | |
1407 | ||
1408 | poll_wait(fp, &uctxt->wait, pt); | |
1409 | ||
1410 | spin_lock_irq(&dd->uctxt_lock); | |
1411 | if (hdrqempty(uctxt)) { | |
1412 | set_bit(HFI1_CTXT_WAITING_RCV, &uctxt->event_flags); | |
1413 | hfi1_rcvctrl(dd, HFI1_RCVCTRL_INTRAVAIL_ENB, uctxt->ctxt); | |
1414 | pollflag = 0; | |
e490974e | 1415 | } else { |
77241056 | 1416 | pollflag = POLLIN | POLLRDNORM; |
e490974e | 1417 | } |
77241056 MM |
1418 | spin_unlock_irq(&dd->uctxt_lock); |
1419 | ||
1420 | return pollflag; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Find all user contexts in use, and set the specified bit in their | |
1425 | * event mask. | |
1426 | * See also find_ctxt() for a similar use, that is specific to send buffers. | |
1427 | */ | |
1428 | int hfi1_set_uevent_bits(struct hfi1_pportdata *ppd, const int evtbit) | |
1429 | { | |
1430 | struct hfi1_ctxtdata *uctxt; | |
1431 | struct hfi1_devdata *dd = ppd->dd; | |
1432 | unsigned ctxt; | |
1433 | int ret = 0; | |
1434 | unsigned long flags; | |
1435 | ||
1436 | if (!dd->events) { | |
1437 | ret = -EINVAL; | |
1438 | goto done; | |
1439 | } | |
1440 | ||
1441 | spin_lock_irqsave(&dd->uctxt_lock, flags); | |
2280740f | 1442 | for (ctxt = dd->first_dyn_alloc_ctxt; ctxt < dd->num_rcv_contexts; |
77241056 MM |
1443 | ctxt++) { |
1444 | uctxt = dd->rcd[ctxt]; | |
1445 | if (uctxt) { | |
1446 | unsigned long *evs = dd->events + | |
2280740f | 1447 | (uctxt->ctxt - dd->first_dyn_alloc_ctxt) * |
77241056 MM |
1448 | HFI1_MAX_SHARED_CTXTS; |
1449 | int i; | |
1450 | /* | |
1451 | * subctxt_cnt is 0 if not shared, so do base | |
1452 | * separately, first, then remaining subctxt, if any | |
1453 | */ | |
1454 | set_bit(evtbit, evs); | |
1455 | for (i = 1; i < uctxt->subctxt_cnt; i++) | |
1456 | set_bit(evtbit, evs + i); | |
1457 | } | |
1458 | } | |
1459 | spin_unlock_irqrestore(&dd->uctxt_lock, flags); | |
1460 | done: | |
1461 | return ret; | |
1462 | } | |
1463 | ||
1464 | /** | |
1465 | * manage_rcvq - manage a context's receive queue | |
1466 | * @uctxt: the context | |
1467 | * @subctxt: the sub-context | |
1468 | * @start_stop: action to carry out | |
1469 | * | |
1470 | * start_stop == 0 disables receive on the context, for use in queue | |
1471 | * overflow conditions. start_stop==1 re-enables, to be used to | |
1472 | * re-init the software copy of the head register | |
1473 | */ | |
8737ce95 | 1474 | static int manage_rcvq(struct hfi1_ctxtdata *uctxt, u16 subctxt, |
77241056 MM |
1475 | int start_stop) |
1476 | { | |
1477 | struct hfi1_devdata *dd = uctxt->dd; | |
1478 | unsigned int rcvctrl_op; | |
1479 | ||
1480 | if (subctxt) | |
1481 | goto bail; | |
1482 | /* atomically clear receive enable ctxt. */ | |
1483 | if (start_stop) { | |
1484 | /* | |
1485 | * On enable, force in-memory copy of the tail register to | |
1486 | * 0, so that protocol code doesn't have to worry about | |
1487 | * whether or not the chip has yet updated the in-memory | |
1488 | * copy or not on return from the system call. The chip | |
1489 | * always resets it's tail register back to 0 on a | |
1490 | * transition from disabled to enabled. | |
1491 | */ | |
1492 | if (uctxt->rcvhdrtail_kvaddr) | |
1493 | clear_rcvhdrtail(uctxt); | |
1494 | rcvctrl_op = HFI1_RCVCTRL_CTXT_ENB; | |
e490974e | 1495 | } else { |
77241056 | 1496 | rcvctrl_op = HFI1_RCVCTRL_CTXT_DIS; |
e490974e | 1497 | } |
77241056 MM |
1498 | hfi1_rcvctrl(dd, rcvctrl_op, uctxt->ctxt); |
1499 | /* always; new head should be equal to new tail; see above */ | |
1500 | bail: | |
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * clear the event notifier events for this context. | |
1506 | * User process then performs actions appropriate to bit having been | |
1507 | * set, if desired, and checks again in future. | |
1508 | */ | |
8737ce95 | 1509 | static int user_event_ack(struct hfi1_ctxtdata *uctxt, u16 subctxt, |
77241056 MM |
1510 | unsigned long events) |
1511 | { | |
1512 | int i; | |
1513 | struct hfi1_devdata *dd = uctxt->dd; | |
1514 | unsigned long *evs; | |
1515 | ||
1516 | if (!dd->events) | |
1517 | return 0; | |
1518 | ||
2280740f | 1519 | evs = dd->events + ((uctxt->ctxt - dd->first_dyn_alloc_ctxt) * |
77241056 MM |
1520 | HFI1_MAX_SHARED_CTXTS) + subctxt; |
1521 | ||
1522 | for (i = 0; i <= _HFI1_MAX_EVENT_BIT; i++) { | |
1523 | if (!test_bit(i, &events)) | |
1524 | continue; | |
1525 | clear_bit(i, evs); | |
1526 | } | |
1527 | return 0; | |
1528 | } | |
1529 | ||
8737ce95 | 1530 | static int set_ctxt_pkey(struct hfi1_ctxtdata *uctxt, u16 subctxt, u16 pkey) |
77241056 MM |
1531 | { |
1532 | int ret = -ENOENT, i, intable = 0; | |
1533 | struct hfi1_pportdata *ppd = uctxt->ppd; | |
1534 | struct hfi1_devdata *dd = uctxt->dd; | |
1535 | ||
1536 | if (pkey == LIM_MGMT_P_KEY || pkey == FULL_MGMT_P_KEY) { | |
1537 | ret = -EINVAL; | |
1538 | goto done; | |
1539 | } | |
1540 | ||
1541 | for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) | |
1542 | if (pkey == ppd->pkeys[i]) { | |
1543 | intable = 1; | |
1544 | break; | |
1545 | } | |
1546 | ||
1547 | if (intable) | |
1548 | ret = hfi1_set_ctxt_pkey(dd, uctxt->ctxt, pkey); | |
1549 | done: | |
1550 | return ret; | |
1551 | } | |
1552 | ||
77241056 MM |
1553 | static void user_remove(struct hfi1_devdata *dd) |
1554 | { | |
77241056 MM |
1555 | |
1556 | hfi1_cdev_cleanup(&dd->user_cdev, &dd->user_device); | |
77241056 MM |
1557 | } |
1558 | ||
1559 | static int user_add(struct hfi1_devdata *dd) | |
1560 | { | |
1561 | char name[10]; | |
1562 | int ret; | |
1563 | ||
77241056 | 1564 | snprintf(name, sizeof(name), "%s_%d", class_name(), dd->unit); |
0eb62659 | 1565 | ret = hfi1_cdev_init(dd->unit, name, &hfi1_file_ops, |
e116a64f | 1566 | &dd->user_cdev, &dd->user_device, |
e11ffbd5 | 1567 | true, &dd->kobj); |
77241056 | 1568 | if (ret) |
7312f29d | 1569 | user_remove(dd); |
77241056 | 1570 | |
77241056 MM |
1571 | return ret; |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * Create per-unit files in /dev | |
1576 | */ | |
1577 | int hfi1_device_create(struct hfi1_devdata *dd) | |
1578 | { | |
0f7b1f91 | 1579 | return user_add(dd); |
77241056 MM |
1580 | } |
1581 | ||
1582 | /* | |
1583 | * Remove per-unit files in /dev | |
1584 | * void, core kernel returns no errors for this stuff | |
1585 | */ | |
1586 | void hfi1_device_remove(struct hfi1_devdata *dd) | |
1587 | { | |
1588 | user_remove(dd); | |
77241056 | 1589 | } |