0d1c8b8c1c12d297d5153fb354fc46ebf2010f01
[linux-2.6.git] / drivers / misc / sgi-gru / grufile.c
1 /*
2  * SN Platform GRU Driver
3  *
4  *              FILE OPERATIONS & DRIVER INITIALIZATION
5  *
6  * This file supports the user system call for file open, close, mmap, etc.
7  * This also incudes the driver initialization code.
8  *
9  *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
10  *
11  *  This program is free software; you can redistribute it and/or modify
12  *  it under the terms of the GNU General Public License as published by
13  *  the Free Software Foundation; either version 2 of the License, or
14  *  (at your option) any later version.
15  *
16  *  This program is distributed in the hope that it will be useful,
17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  *  GNU General Public License for more details.
20  *
21  *  You should have received a copy of the GNU General Public License
22  *  along with this program; if not, write to the Free Software
23  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
24  */
25
26 #include <linux/module.h>
27 #include <linux/kernel.h>
28 #include <linux/errno.h>
29 #include <linux/slab.h>
30 #include <linux/mm.h>
31 #include <linux/io.h>
32 #include <linux/smp_lock.h>
33 #include <linux/spinlock.h>
34 #include <linux/device.h>
35 #include <linux/miscdevice.h>
36 #include <linux/interrupt.h>
37 #include <linux/proc_fs.h>
38 #include <linux/uaccess.h>
39 #include <asm/uv/uv.h>
40 #include "gru.h"
41 #include "grulib.h"
42 #include "grutables.h"
43
44 #include <asm/uv/uv_hub.h>
45 #include <asm/uv/uv_mmrs.h>
46
47 struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
48 unsigned long gru_start_paddr __read_mostly;
49 void *gru_start_vaddr __read_mostly;
50 unsigned long gru_end_paddr __read_mostly;
51 unsigned int gru_max_gids __read_mostly;
52 struct gru_stats_s gru_stats;
53
54 /* Guaranteed user available resources on each node */
55 static int max_user_cbrs, max_user_dsr_bytes;
56
57 static struct file_operations gru_fops;
58 static struct miscdevice gru_miscdev;
59
60
61 /*
62  * gru_vma_close
63  *
64  * Called when unmapping a device mapping. Frees all gru resources
65  * and tables belonging to the vma.
66  */
67 static void gru_vma_close(struct vm_area_struct *vma)
68 {
69         struct gru_vma_data *vdata;
70         struct gru_thread_state *gts;
71         struct list_head *entry, *next;
72
73         if (!vma->vm_private_data)
74                 return;
75
76         vdata = vma->vm_private_data;
77         vma->vm_private_data = NULL;
78         gru_dbg(grudev, "vma %p, file %p, vdata %p\n", vma, vma->vm_file,
79                                 vdata);
80         list_for_each_safe(entry, next, &vdata->vd_head) {
81                 gts =
82                     list_entry(entry, struct gru_thread_state, ts_next);
83                 list_del(&gts->ts_next);
84                 mutex_lock(&gts->ts_ctxlock);
85                 if (gts->ts_gru)
86                         gru_unload_context(gts, 0);
87                 mutex_unlock(&gts->ts_ctxlock);
88                 gts_drop(gts);
89         }
90         kfree(vdata);
91         STAT(vdata_free);
92 }
93
94 /*
95  * gru_file_mmap
96  *
97  * Called when mmaping the device.  Initializes the vma with a fault handler
98  * and private data structure necessary to allocate, track, and free the
99  * underlying pages.
100  */
101 static int gru_file_mmap(struct file *file, struct vm_area_struct *vma)
102 {
103         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) != (VM_SHARED | VM_WRITE))
104                 return -EPERM;
105
106         if (vma->vm_start & (GRU_GSEG_PAGESIZE - 1) ||
107                                 vma->vm_end & (GRU_GSEG_PAGESIZE - 1))
108                 return -EINVAL;
109
110         vma->vm_flags |=
111             (VM_IO | VM_DONTCOPY | VM_LOCKED | VM_DONTEXPAND | VM_PFNMAP |
112                         VM_RESERVED);
113         vma->vm_page_prot = PAGE_SHARED;
114         vma->vm_ops = &gru_vm_ops;
115
116         vma->vm_private_data = gru_alloc_vma_data(vma, 0);
117         if (!vma->vm_private_data)
118                 return -ENOMEM;
119
120         gru_dbg(grudev, "file %p, vaddr 0x%lx, vma %p, vdata %p\n",
121                 file, vma->vm_start, vma, vma->vm_private_data);
122         return 0;
123 }
124
125 /*
126  * Create a new GRU context
127  */
128 static int gru_create_new_context(unsigned long arg)
129 {
130         struct gru_create_context_req req;
131         struct vm_area_struct *vma;
132         struct gru_vma_data *vdata;
133         int ret = -EINVAL;
134
135
136         if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
137                 return -EFAULT;
138
139         if (req.data_segment_bytes > max_user_dsr_bytes)
140                 return -EINVAL;
141         if (req.control_blocks > max_user_cbrs || !req.maximum_thread_count)
142                 return -EINVAL;
143
144         if (!(req.options & GRU_OPT_MISS_MASK))
145                 req.options |= GRU_OPT_MISS_FMM_INTR;
146
147         down_write(&current->mm->mmap_sem);
148         vma = gru_find_vma(req.gseg);
149         if (vma) {
150                 vdata = vma->vm_private_data;
151                 vdata->vd_user_options = req.options;
152                 vdata->vd_dsr_au_count =
153                     GRU_DS_BYTES_TO_AU(req.data_segment_bytes);
154                 vdata->vd_cbr_au_count = GRU_CB_COUNT_TO_AU(req.control_blocks);
155                 ret = 0;
156         }
157         up_write(&current->mm->mmap_sem);
158
159         return ret;
160 }
161
162 /*
163  * Get GRU configuration info (temp - for emulator testing)
164  */
165 static long gru_get_config_info(unsigned long arg)
166 {
167         struct gru_config_info info;
168         int nodesperblade;
169
170         if (num_online_nodes() > 1 &&
171                         (uv_node_to_blade_id(1) == uv_node_to_blade_id(0)))
172                 nodesperblade = 2;
173         else
174                 nodesperblade = 1;
175         info.cpus = num_online_cpus();
176         info.nodes = num_online_nodes();
177         info.blades = info.nodes / nodesperblade;
178         info.chiplets = GRU_CHIPLETS_PER_BLADE * info.blades;
179
180         if (copy_to_user((void __user *)arg, &info, sizeof(info)))
181                 return -EFAULT;
182         return 0;
183 }
184
185 /*
186  * Get GRU chiplet status
187  */
188 static long gru_get_chiplet_status(unsigned long arg)
189 {
190         struct gru_state *gru;
191         struct gru_chiplet_info info;
192
193         if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
194                 return -EFAULT;
195
196         if (info.node == -1)
197                 info.node = numa_node_id();
198         if (info.node >= num_possible_nodes() ||
199                         info.chiplet >= GRU_CHIPLETS_PER_HUB ||
200                         info.node < 0 || info.chiplet < 0)
201                 return -EINVAL;
202
203         info.blade = uv_node_to_blade_id(info.node);
204         gru = get_gru(info.blade, info.chiplet);
205
206         info.total_dsr_bytes = GRU_NUM_DSR_BYTES;
207         info.total_cbr = GRU_NUM_CB;
208         info.total_user_dsr_bytes = GRU_NUM_DSR_BYTES -
209                 gru->gs_reserved_dsr_bytes;
210         info.total_user_cbr = GRU_NUM_CB - gru->gs_reserved_cbrs;
211         info.free_user_dsr_bytes = hweight64(gru->gs_dsr_map) *
212                         GRU_DSR_AU_BYTES;
213         info.free_user_cbr = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
214
215         if (copy_to_user((void __user *)arg, &info, sizeof(info)))
216                 return -EFAULT;
217         return 0;
218 }
219
220 /*
221  * gru_file_unlocked_ioctl
222  *
223  * Called to update file attributes via IOCTL calls.
224  */
225 static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
226                                     unsigned long arg)
227 {
228         int err = -EBADRQC;
229
230         gru_dbg(grudev, "file %p\n", file);
231
232         switch (req) {
233         case GRU_CREATE_CONTEXT:
234                 err = gru_create_new_context(arg);
235                 break;
236         case GRU_SET_TASK_SLICE:
237                 err = gru_set_task_slice(arg);
238                 break;
239         case GRU_USER_GET_EXCEPTION_DETAIL:
240                 err = gru_get_exception_detail(arg);
241                 break;
242         case GRU_USER_UNLOAD_CONTEXT:
243                 err = gru_user_unload_context(arg);
244                 break;
245         case GRU_GET_CHIPLET_STATUS:
246                 err = gru_get_chiplet_status(arg);
247                 break;
248         case GRU_USER_FLUSH_TLB:
249                 err = gru_user_flush_tlb(arg);
250                 break;
251         case GRU_USER_CALL_OS:
252                 err = gru_handle_user_call_os(arg);
253                 break;
254         case GRU_GET_GSEG_STATISTICS:
255                 err = gru_get_gseg_statistics(arg);
256                 break;
257         case GRU_KTEST:
258                 err = gru_ktest(arg);
259                 break;
260         case GRU_GET_CONFIG_INFO:
261                 err = gru_get_config_info(arg);
262                 break;
263         case GRU_DUMP_CHIPLET_STATE:
264                 err = gru_dump_chiplet_request(arg);
265                 break;
266         }
267         return err;
268 }
269
270 /*
271  * Called at init time to build tables for all GRUs that are present in the
272  * system.
273  */
274 static void gru_init_chiplet(struct gru_state *gru, unsigned long paddr,
275                              void *vaddr, int nid, int bid, int grunum)
276 {
277         spin_lock_init(&gru->gs_lock);
278         spin_lock_init(&gru->gs_asid_lock);
279         gru->gs_gru_base_paddr = paddr;
280         gru->gs_gru_base_vaddr = vaddr;
281         gru->gs_gid = bid * GRU_CHIPLETS_PER_BLADE + grunum;
282         gru->gs_blade = gru_base[bid];
283         gru->gs_blade_id = bid;
284         gru->gs_cbr_map = (GRU_CBR_AU == 64) ? ~0 : (1UL << GRU_CBR_AU) - 1;
285         gru->gs_dsr_map = (1UL << GRU_DSR_AU) - 1;
286         gru->gs_asid_limit = MAX_ASID;
287         gru_tgh_flush_init(gru);
288         if (gru->gs_gid >= gru_max_gids)
289                 gru_max_gids = gru->gs_gid + 1;
290         gru_dbg(grudev, "bid %d, nid %d, gid %d, vaddr %p (0x%lx)\n",
291                 bid, nid, gru->gs_gid, gru->gs_gru_base_vaddr,
292                 gru->gs_gru_base_paddr);
293 }
294
295 static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
296 {
297         int pnode, nid, bid, chip;
298         int cbrs, dsrbytes, n;
299         int order = get_order(sizeof(struct gru_blade_state));
300         struct page *page;
301         struct gru_state *gru;
302         unsigned long paddr;
303         void *vaddr;
304
305         max_user_cbrs = GRU_NUM_CB;
306         max_user_dsr_bytes = GRU_NUM_DSR_BYTES;
307         for_each_online_node(nid) {
308                 bid = uv_node_to_blade_id(nid);
309                 pnode = uv_node_to_pnode(nid);
310                 if (bid < 0 || gru_base[bid])
311                         continue;
312                 page = alloc_pages_exact_node(nid, GFP_KERNEL, order);
313                 if (!page)
314                         goto fail;
315                 gru_base[bid] = page_address(page);
316                 memset(gru_base[bid], 0, sizeof(struct gru_blade_state));
317                 gru_base[bid]->bs_lru_gru = &gru_base[bid]->bs_grus[0];
318                 spin_lock_init(&gru_base[bid]->bs_lock);
319                 init_rwsem(&gru_base[bid]->bs_kgts_sema);
320
321                 dsrbytes = 0;
322                 cbrs = 0;
323                 for (gru = gru_base[bid]->bs_grus, chip = 0;
324                                 chip < GRU_CHIPLETS_PER_BLADE;
325                                 chip++, gru++) {
326                         paddr = gru_chiplet_paddr(gru_base_paddr, pnode, chip);
327                         vaddr = gru_chiplet_vaddr(gru_base_vaddr, pnode, chip);
328                         gru_init_chiplet(gru, paddr, vaddr, nid, bid, chip);
329                         n = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
330                         cbrs = max(cbrs, n);
331                         n = hweight64(gru->gs_dsr_map) * GRU_DSR_AU_BYTES;
332                         dsrbytes = max(dsrbytes, n);
333                 }
334                 max_user_cbrs = min(max_user_cbrs, cbrs);
335                 max_user_dsr_bytes = min(max_user_dsr_bytes, dsrbytes);
336         }
337
338         return 0;
339
340 fail:
341         for (nid--; nid >= 0; nid--)
342                 free_pages((unsigned long)gru_base[nid], order);
343         return -ENOMEM;
344 }
345
346 #ifdef CONFIG_IA64
347
348 static int get_base_irq(void)
349 {
350         return IRQ_GRU;
351 }
352
353 #elif defined CONFIG_X86_64
354
355 static void noop(unsigned int irq)
356 {
357 }
358
359 static struct irq_chip gru_chip = {
360         .name           = "gru",
361         .mask           = noop,
362         .unmask         = noop,
363         .ack            = noop,
364 };
365
366 static int get_base_irq(void)
367 {
368         set_irq_chip(IRQ_GRU, &gru_chip);
369         set_irq_chip(IRQ_GRU + 1, &gru_chip);
370         return IRQ_GRU;
371 }
372 #endif
373
374 /*
375  * gru_init
376  *
377  * Called at boot or module load time to initialize the GRUs.
378  */
379 static int __init gru_init(void)
380 {
381         int ret, irq, chip;
382         char id[10];
383
384         if (!is_uv_system())
385                 return 0;
386
387 #if defined CONFIG_IA64
388         gru_start_paddr = 0xd000000000UL; /* ZZZZZZZZZZZZZZZZZZZ fixme */
389 #else
390         gru_start_paddr = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR) &
391                                 0x7fffffffffffUL;
392 #endif
393         gru_start_vaddr = __va(gru_start_paddr);
394         gru_end_paddr = gru_start_paddr + GRU_MAX_BLADES * GRU_SIZE;
395         printk(KERN_INFO "GRU space: 0x%lx - 0x%lx\n",
396                gru_start_paddr, gru_end_paddr);
397         irq = get_base_irq();
398         for (chip = 0; chip < GRU_CHIPLETS_PER_BLADE; chip++) {
399                 ret = request_irq(irq + chip, gru_intr, 0, id, NULL);
400                 /* TODO: fix irq handling on x86. For now ignore failure because
401                  * interrupts are not required & not yet fully supported */
402                 if (ret) {
403                         printk(KERN_WARNING
404                                "!!!WARNING: GRU ignoring request failure!!!\n");
405                         ret = 0;
406                 }
407                 if (ret) {
408                         printk(KERN_ERR "%s: request_irq failed\n",
409                                GRU_DRIVER_ID_STR);
410                         goto exit1;
411                 }
412         }
413
414         ret = misc_register(&gru_miscdev);
415         if (ret) {
416                 printk(KERN_ERR "%s: misc_register failed\n",
417                        GRU_DRIVER_ID_STR);
418                 goto exit1;
419         }
420
421         ret = gru_proc_init();
422         if (ret) {
423                 printk(KERN_ERR "%s: proc init failed\n", GRU_DRIVER_ID_STR);
424                 goto exit2;
425         }
426
427         ret = gru_init_tables(gru_start_paddr, gru_start_vaddr);
428         if (ret) {
429                 printk(KERN_ERR "%s: init tables failed\n", GRU_DRIVER_ID_STR);
430                 goto exit3;
431         }
432         gru_kservices_init();
433
434         printk(KERN_INFO "%s: v%s\n", GRU_DRIVER_ID_STR,
435                GRU_DRIVER_VERSION_STR);
436         return 0;
437
438 exit3:
439         gru_proc_exit();
440 exit2:
441         misc_deregister(&gru_miscdev);
442 exit1:
443         for (--chip; chip >= 0; chip--)
444                 free_irq(irq + chip, NULL);
445         return ret;
446
447 }
448
449 static void __exit gru_exit(void)
450 {
451         int i, bid;
452         int order = get_order(sizeof(struct gru_state) *
453                               GRU_CHIPLETS_PER_BLADE);
454
455         if (!is_uv_system())
456                 return;
457
458         for (i = 0; i < GRU_CHIPLETS_PER_BLADE; i++)
459                 free_irq(IRQ_GRU + i, NULL);
460         gru_kservices_exit();
461         for (bid = 0; bid < GRU_MAX_BLADES; bid++)
462                 free_pages((unsigned long)gru_base[bid], order);
463
464         misc_deregister(&gru_miscdev);
465         gru_proc_exit();
466 }
467
468 static struct file_operations gru_fops = {
469         .owner          = THIS_MODULE,
470         .unlocked_ioctl = gru_file_unlocked_ioctl,
471         .mmap           = gru_file_mmap,
472 };
473
474 static struct miscdevice gru_miscdev = {
475         .minor          = MISC_DYNAMIC_MINOR,
476         .name           = "gru",
477         .fops           = &gru_fops,
478 };
479
480 struct vm_operations_struct gru_vm_ops = {
481         .close          = gru_vma_close,
482         .fault          = gru_fault,
483 };
484
485 #ifndef MODULE
486 fs_initcall(gru_init);
487 #else
488 module_init(gru_init);
489 #endif
490 module_exit(gru_exit);
491
492 module_param(gru_options, ulong, 0644);
493 MODULE_PARM_DESC(gru_options, "Various debug options");
494
495 MODULE_AUTHOR("Silicon Graphics, Inc.");
496 MODULE_LICENSE("GPL");
497 MODULE_DESCRIPTION(GRU_DRIVER_ID_STR GRU_DRIVER_VERSION_STR);
498 MODULE_VERSION(GRU_DRIVER_VERSION_STR);
499