Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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) 2008-2014 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/spinlock.h>
33 #include <linux/device.h>
34 #include <linux/miscdevice.h>
35 #include <linux/interrupt.h>
36 #include <linux/proc_fs.h>
37 #include <linux/uaccess.h>
38 #ifdef CONFIG_X86_64
39 #include <asm/uv/uv_irq.h>
40 #endif
41 #include <asm/uv/uv.h>
42 #include "gru.h"
43 #include "grulib.h"
44 #include "grutables.h"
45
46 #include <asm/uv/uv_hub.h>
47 #include <asm/uv/uv_mmrs.h>
48
49 struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
50 unsigned long gru_start_paddr __read_mostly;
51 void *gru_start_vaddr __read_mostly;
52 unsigned long gru_end_paddr __read_mostly;
53 unsigned int gru_max_gids __read_mostly;
54 struct gru_stats_s gru_stats;
55
56 /* Guaranteed user available resources on each node */
57 static int max_user_cbrs, max_user_dsr_bytes;
58
59 static struct miscdevice gru_miscdev;
60
61 static int gru_supported(void)
62 {
63         return is_uv_system() &&
64                 (uv_hub_info->hub_revision < UV3_HUB_REVISION_BASE);
65 }
66
67 /*
68  * gru_vma_close
69  *
70  * Called when unmapping a device mapping. Frees all gru resources
71  * and tables belonging to the vma.
72  */
73 static void gru_vma_close(struct vm_area_struct *vma)
74 {
75         struct gru_vma_data *vdata;
76         struct gru_thread_state *gts;
77         struct list_head *entry, *next;
78
79         if (!vma->vm_private_data)
80                 return;
81
82         vdata = vma->vm_private_data;
83         vma->vm_private_data = NULL;
84         gru_dbg(grudev, "vma %p, file %p, vdata %p\n", vma, vma->vm_file,
85                                 vdata);
86         list_for_each_safe(entry, next, &vdata->vd_head) {
87                 gts =
88                     list_entry(entry, struct gru_thread_state, ts_next);
89                 list_del(&gts->ts_next);
90                 mutex_lock(&gts->ts_ctxlock);
91                 if (gts->ts_gru)
92                         gru_unload_context(gts, 0);
93                 mutex_unlock(&gts->ts_ctxlock);
94                 gts_drop(gts);
95         }
96         kfree(vdata);
97         STAT(vdata_free);
98 }
99
100 /*
101  * gru_file_mmap
102  *
103  * Called when mmapping the device.  Initializes the vma with a fault handler
104  * and private data structure necessary to allocate, track, and free the
105  * underlying pages.
106  */
107 static int gru_file_mmap(struct file *file, struct vm_area_struct *vma)
108 {
109         if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) != (VM_SHARED | VM_WRITE))
110                 return -EPERM;
111
112         if (vma->vm_start & (GRU_GSEG_PAGESIZE - 1) ||
113                                 vma->vm_end & (GRU_GSEG_PAGESIZE - 1))
114                 return -EINVAL;
115
116         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_LOCKED |
117                          VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP;
118         vma->vm_page_prot = PAGE_SHARED;
119         vma->vm_ops = &gru_vm_ops;
120
121         vma->vm_private_data = gru_alloc_vma_data(vma, 0);
122         if (!vma->vm_private_data)
123                 return -ENOMEM;
124
125         gru_dbg(grudev, "file %p, vaddr 0x%lx, vma %p, vdata %p\n",
126                 file, vma->vm_start, vma, vma->vm_private_data);
127         return 0;
128 }
129
130 /*
131  * Create a new GRU context
132  */
133 static int gru_create_new_context(unsigned long arg)
134 {
135         struct gru_create_context_req req;
136         struct vm_area_struct *vma;
137         struct gru_vma_data *vdata;
138         int ret = -EINVAL;
139
140         if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
141                 return -EFAULT;
142
143         if (req.data_segment_bytes > max_user_dsr_bytes)
144                 return -EINVAL;
145         if (req.control_blocks > max_user_cbrs || !req.maximum_thread_count)
146                 return -EINVAL;
147
148         if (!(req.options & GRU_OPT_MISS_MASK))
149                 req.options |= GRU_OPT_MISS_FMM_INTR;
150
151         down_write(&current->mm->mmap_sem);
152         vma = gru_find_vma(req.gseg);
153         if (vma) {
154                 vdata = vma->vm_private_data;
155                 vdata->vd_user_options = req.options;
156                 vdata->vd_dsr_au_count =
157                     GRU_DS_BYTES_TO_AU(req.data_segment_bytes);
158                 vdata->vd_cbr_au_count = GRU_CB_COUNT_TO_AU(req.control_blocks);
159                 vdata->vd_tlb_preload_count = req.tlb_preload_count;
160                 ret = 0;
161         }
162         up_write(&current->mm->mmap_sem);
163
164         return ret;
165 }
166
167 /*
168  * Get GRU configuration info (temp - for emulator testing)
169  */
170 static long gru_get_config_info(unsigned long arg)
171 {
172         struct gru_config_info info;
173         int nodesperblade;
174
175         if (num_online_nodes() > 1 &&
176                         (uv_node_to_blade_id(1) == uv_node_to_blade_id(0)))
177                 nodesperblade = 2;
178         else
179                 nodesperblade = 1;
180         memset(&info, 0, sizeof(info));
181         info.cpus = num_online_cpus();
182         info.nodes = num_online_nodes();
183         info.blades = info.nodes / nodesperblade;
184         info.chiplets = GRU_CHIPLETS_PER_BLADE * info.blades;
185
186         if (copy_to_user((void __user *)arg, &info, sizeof(info)))
187                 return -EFAULT;
188         return 0;
189 }
190
191 /*
192  * gru_file_unlocked_ioctl
193  *
194  * Called to update file attributes via IOCTL calls.
195  */
196 static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
197                                     unsigned long arg)
198 {
199         int err = -EBADRQC;
200
201         gru_dbg(grudev, "file %p, req 0x%x, 0x%lx\n", file, req, arg);
202
203         switch (req) {
204         case GRU_CREATE_CONTEXT:
205                 err = gru_create_new_context(arg);
206                 break;
207         case GRU_SET_CONTEXT_OPTION:
208                 err = gru_set_context_option(arg);
209                 break;
210         case GRU_USER_GET_EXCEPTION_DETAIL:
211                 err = gru_get_exception_detail(arg);
212                 break;
213         case GRU_USER_UNLOAD_CONTEXT:
214                 err = gru_user_unload_context(arg);
215                 break;
216         case GRU_USER_FLUSH_TLB:
217                 err = gru_user_flush_tlb(arg);
218                 break;
219         case GRU_USER_CALL_OS:
220                 err = gru_handle_user_call_os(arg);
221                 break;
222         case GRU_GET_GSEG_STATISTICS:
223                 err = gru_get_gseg_statistics(arg);
224                 break;
225         case GRU_KTEST:
226                 err = gru_ktest(arg);
227                 break;
228         case GRU_GET_CONFIG_INFO:
229                 err = gru_get_config_info(arg);
230                 break;
231         case GRU_DUMP_CHIPLET_STATE:
232                 err = gru_dump_chiplet_request(arg);
233                 break;
234         }
235         return err;
236 }
237
238 /*
239  * Called at init time to build tables for all GRUs that are present in the
240  * system.
241  */
242 static void gru_init_chiplet(struct gru_state *gru, unsigned long paddr,
243                              void *vaddr, int blade_id, int chiplet_id)
244 {
245         spin_lock_init(&gru->gs_lock);
246         spin_lock_init(&gru->gs_asid_lock);
247         gru->gs_gru_base_paddr = paddr;
248         gru->gs_gru_base_vaddr = vaddr;
249         gru->gs_gid = blade_id * GRU_CHIPLETS_PER_BLADE + chiplet_id;
250         gru->gs_blade = gru_base[blade_id];
251         gru->gs_blade_id = blade_id;
252         gru->gs_chiplet_id = chiplet_id;
253         gru->gs_cbr_map = (GRU_CBR_AU == 64) ? ~0 : (1UL << GRU_CBR_AU) - 1;
254         gru->gs_dsr_map = (1UL << GRU_DSR_AU) - 1;
255         gru->gs_asid_limit = MAX_ASID;
256         gru_tgh_flush_init(gru);
257         if (gru->gs_gid >= gru_max_gids)
258                 gru_max_gids = gru->gs_gid + 1;
259         gru_dbg(grudev, "bid %d, gid %d, vaddr %p (0x%lx)\n",
260                 blade_id, gru->gs_gid, gru->gs_gru_base_vaddr,
261                 gru->gs_gru_base_paddr);
262 }
263
264 static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
265 {
266         int pnode, nid, bid, chip;
267         int cbrs, dsrbytes, n;
268         int order = get_order(sizeof(struct gru_blade_state));
269         struct page *page;
270         struct gru_state *gru;
271         unsigned long paddr;
272         void *vaddr;
273
274         max_user_cbrs = GRU_NUM_CB;
275         max_user_dsr_bytes = GRU_NUM_DSR_BYTES;
276         for_each_possible_blade(bid) {
277                 pnode = uv_blade_to_pnode(bid);
278                 nid = uv_blade_to_memory_nid(bid);/* -1 if no memory on blade */
279                 page = alloc_pages_node(nid, GFP_KERNEL, order);
280                 if (!page)
281                         goto fail;
282                 gru_base[bid] = page_address(page);
283                 memset(gru_base[bid], 0, sizeof(struct gru_blade_state));
284                 gru_base[bid]->bs_lru_gru = &gru_base[bid]->bs_grus[0];
285                 spin_lock_init(&gru_base[bid]->bs_lock);
286                 init_rwsem(&gru_base[bid]->bs_kgts_sema);
287
288                 dsrbytes = 0;
289                 cbrs = 0;
290                 for (gru = gru_base[bid]->bs_grus, chip = 0;
291                                 chip < GRU_CHIPLETS_PER_BLADE;
292                                 chip++, gru++) {
293                         paddr = gru_chiplet_paddr(gru_base_paddr, pnode, chip);
294                         vaddr = gru_chiplet_vaddr(gru_base_vaddr, pnode, chip);
295                         gru_init_chiplet(gru, paddr, vaddr, bid, chip);
296                         n = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
297                         cbrs = max(cbrs, n);
298                         n = hweight64(gru->gs_dsr_map) * GRU_DSR_AU_BYTES;
299                         dsrbytes = max(dsrbytes, n);
300                 }
301                 max_user_cbrs = min(max_user_cbrs, cbrs);
302                 max_user_dsr_bytes = min(max_user_dsr_bytes, dsrbytes);
303         }
304
305         return 0;
306
307 fail:
308         for (bid--; bid >= 0; bid--)
309                 free_pages((unsigned long)gru_base[bid], order);
310         return -ENOMEM;
311 }
312
313 static void gru_free_tables(void)
314 {
315         int bid;
316         int order = get_order(sizeof(struct gru_state) *
317                               GRU_CHIPLETS_PER_BLADE);
318
319         for (bid = 0; bid < GRU_MAX_BLADES; bid++)
320                 free_pages((unsigned long)gru_base[bid], order);
321 }
322
323 static unsigned long gru_chiplet_cpu_to_mmr(int chiplet, int cpu, int *corep)
324 {
325         unsigned long mmr = 0;
326         int core;
327
328         /*
329          * We target the cores of a blade and not the hyperthreads themselves.
330          * There is a max of 8 cores per socket and 2 sockets per blade,
331          * making for a max total of 16 cores (i.e., 16 CPUs without
332          * hyperthreading and 32 CPUs with hyperthreading).
333          */
334         core = uv_cpu_core_number(cpu) + UV_MAX_INT_CORES * uv_cpu_socket_number(cpu);
335         if (core >= GRU_NUM_TFM || uv_cpu_ht_number(cpu))
336                 return 0;
337
338         if (chiplet == 0) {
339                 mmr = UVH_GR0_TLB_INT0_CONFIG +
340                     core * (UVH_GR0_TLB_INT1_CONFIG - UVH_GR0_TLB_INT0_CONFIG);
341         } else if (chiplet == 1) {
342                 mmr = UVH_GR1_TLB_INT0_CONFIG +
343                     core * (UVH_GR1_TLB_INT1_CONFIG - UVH_GR1_TLB_INT0_CONFIG);
344         } else {
345                 BUG();
346         }
347
348         *corep = core;
349         return mmr;
350 }
351
352 #ifdef CONFIG_IA64
353
354 static int gru_irq_count[GRU_CHIPLETS_PER_BLADE];
355
356 static void gru_noop(struct irq_data *d)
357 {
358 }
359
360 static struct irq_chip gru_chip[GRU_CHIPLETS_PER_BLADE] = {
361         [0 ... GRU_CHIPLETS_PER_BLADE - 1] {
362                 .irq_mask       = gru_noop,
363                 .irq_unmask     = gru_noop,
364                 .irq_ack        = gru_noop
365         }
366 };
367
368 static int gru_chiplet_setup_tlb_irq(int chiplet, char *irq_name,
369                         irq_handler_t irq_handler, int cpu, int blade)
370 {
371         unsigned long mmr;
372         int irq = IRQ_GRU + chiplet;
373         int ret, core;
374
375         mmr = gru_chiplet_cpu_to_mmr(chiplet, cpu, &core);
376         if (mmr == 0)
377                 return 0;
378
379         if (gru_irq_count[chiplet] == 0) {
380                 gru_chip[chiplet].name = irq_name;
381                 ret = irq_set_chip(irq, &gru_chip[chiplet]);
382                 if (ret) {
383                         printk(KERN_ERR "%s: set_irq_chip failed, errno=%d\n",
384                                GRU_DRIVER_ID_STR, -ret);
385                         return ret;
386                 }
387
388                 ret = request_irq(irq, irq_handler, 0, irq_name, NULL);
389                 if (ret) {
390                         printk(KERN_ERR "%s: request_irq failed, errno=%d\n",
391                                GRU_DRIVER_ID_STR, -ret);
392                         return ret;
393                 }
394         }
395         gru_irq_count[chiplet]++;
396
397         return 0;
398 }
399
400 static void gru_chiplet_teardown_tlb_irq(int chiplet, int cpu, int blade)
401 {
402         unsigned long mmr;
403         int core, irq = IRQ_GRU + chiplet;
404
405         if (gru_irq_count[chiplet] == 0)
406                 return;
407
408         mmr = gru_chiplet_cpu_to_mmr(chiplet, cpu, &core);
409         if (mmr == 0)
410                 return;
411
412         if (--gru_irq_count[chiplet] == 0)
413                 free_irq(irq, NULL);
414 }
415
416 #elif defined CONFIG_X86_64
417
418 static int gru_chiplet_setup_tlb_irq(int chiplet, char *irq_name,
419                         irq_handler_t irq_handler, int cpu, int blade)
420 {
421         unsigned long mmr;
422         int irq, core;
423         int ret;
424
425         mmr = gru_chiplet_cpu_to_mmr(chiplet, cpu, &core);
426         if (mmr == 0)
427                 return 0;
428
429         irq = uv_setup_irq(irq_name, cpu, blade, mmr, UV_AFFINITY_CPU);
430         if (irq < 0) {
431                 printk(KERN_ERR "%s: uv_setup_irq failed, errno=%d\n",
432                        GRU_DRIVER_ID_STR, -irq);
433                 return irq;
434         }
435
436         ret = request_irq(irq, irq_handler, 0, irq_name, NULL);
437         if (ret) {
438                 uv_teardown_irq(irq);
439                 printk(KERN_ERR "%s: request_irq failed, errno=%d\n",
440                        GRU_DRIVER_ID_STR, -ret);
441                 return ret;
442         }
443         gru_base[blade]->bs_grus[chiplet].gs_irq[core] = irq;
444         return 0;
445 }
446
447 static void gru_chiplet_teardown_tlb_irq(int chiplet, int cpu, int blade)
448 {
449         int irq, core;
450         unsigned long mmr;
451
452         mmr = gru_chiplet_cpu_to_mmr(chiplet, cpu, &core);
453         if (mmr) {
454                 irq = gru_base[blade]->bs_grus[chiplet].gs_irq[core];
455                 if (irq) {
456                         free_irq(irq, NULL);
457                         uv_teardown_irq(irq);
458                 }
459         }
460 }
461
462 #endif
463
464 static void gru_teardown_tlb_irqs(void)
465 {
466         int blade;
467         int cpu;
468
469         for_each_online_cpu(cpu) {
470                 blade = uv_cpu_to_blade_id(cpu);
471                 gru_chiplet_teardown_tlb_irq(0, cpu, blade);
472                 gru_chiplet_teardown_tlb_irq(1, cpu, blade);
473         }
474         for_each_possible_blade(blade) {
475                 if (uv_blade_nr_possible_cpus(blade))
476                         continue;
477                 gru_chiplet_teardown_tlb_irq(0, 0, blade);
478                 gru_chiplet_teardown_tlb_irq(1, 0, blade);
479         }
480 }
481
482 static int gru_setup_tlb_irqs(void)
483 {
484         int blade;
485         int cpu;
486         int ret;
487
488         for_each_online_cpu(cpu) {
489                 blade = uv_cpu_to_blade_id(cpu);
490                 ret = gru_chiplet_setup_tlb_irq(0, "GRU0_TLB", gru0_intr, cpu, blade);
491                 if (ret != 0)
492                         goto exit1;
493
494                 ret = gru_chiplet_setup_tlb_irq(1, "GRU1_TLB", gru1_intr, cpu, blade);
495                 if (ret != 0)
496                         goto exit1;
497         }
498         for_each_possible_blade(blade) {
499                 if (uv_blade_nr_possible_cpus(blade))
500                         continue;
501                 ret = gru_chiplet_setup_tlb_irq(0, "GRU0_TLB", gru_intr_mblade, 0, blade);
502                 if (ret != 0)
503                         goto exit1;
504
505                 ret = gru_chiplet_setup_tlb_irq(1, "GRU1_TLB", gru_intr_mblade, 0, blade);
506                 if (ret != 0)
507                         goto exit1;
508         }
509
510         return 0;
511
512 exit1:
513         gru_teardown_tlb_irqs();
514         return ret;
515 }
516
517 /*
518  * gru_init
519  *
520  * Called at boot or module load time to initialize the GRUs.
521  */
522 static int __init gru_init(void)
523 {
524         int ret;
525
526         if (!gru_supported())
527                 return 0;
528
529 #if defined CONFIG_IA64
530         gru_start_paddr = 0xd000000000UL; /* ZZZZZZZZZZZZZZZZZZZ fixme */
531 #else
532         gru_start_paddr = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR) &
533                                 0x7fffffffffffUL;
534 #endif
535         gru_start_vaddr = __va(gru_start_paddr);
536         gru_end_paddr = gru_start_paddr + GRU_MAX_BLADES * GRU_SIZE;
537         printk(KERN_INFO "GRU space: 0x%lx - 0x%lx\n",
538                gru_start_paddr, gru_end_paddr);
539         ret = misc_register(&gru_miscdev);
540         if (ret) {
541                 printk(KERN_ERR "%s: misc_register failed\n",
542                        GRU_DRIVER_ID_STR);
543                 goto exit0;
544         }
545
546         ret = gru_proc_init();
547         if (ret) {
548                 printk(KERN_ERR "%s: proc init failed\n", GRU_DRIVER_ID_STR);
549                 goto exit1;
550         }
551
552         ret = gru_init_tables(gru_start_paddr, gru_start_vaddr);
553         if (ret) {
554                 printk(KERN_ERR "%s: init tables failed\n", GRU_DRIVER_ID_STR);
555                 goto exit2;
556         }
557
558         ret = gru_setup_tlb_irqs();
559         if (ret != 0)
560                 goto exit3;
561
562         gru_kservices_init();
563
564         printk(KERN_INFO "%s: v%s\n", GRU_DRIVER_ID_STR,
565                GRU_DRIVER_VERSION_STR);
566         return 0;
567
568 exit3:
569         gru_free_tables();
570 exit2:
571         gru_proc_exit();
572 exit1:
573         misc_deregister(&gru_miscdev);
574 exit0:
575         return ret;
576
577 }
578
579 static void __exit gru_exit(void)
580 {
581         if (!gru_supported())
582                 return;
583
584         gru_teardown_tlb_irqs();
585         gru_kservices_exit();
586         gru_free_tables();
587         misc_deregister(&gru_miscdev);
588         gru_proc_exit();
589 }
590
591 static const struct file_operations gru_fops = {
592         .owner          = THIS_MODULE,
593         .unlocked_ioctl = gru_file_unlocked_ioctl,
594         .mmap           = gru_file_mmap,
595         .llseek         = noop_llseek,
596 };
597
598 static struct miscdevice gru_miscdev = {
599         .minor          = MISC_DYNAMIC_MINOR,
600         .name           = "gru",
601         .fops           = &gru_fops,
602 };
603
604 const struct vm_operations_struct gru_vm_ops = {
605         .close          = gru_vma_close,
606         .fault          = gru_fault,
607 };
608
609 #ifndef MODULE
610 fs_initcall(gru_init);
611 #else
612 module_init(gru_init);
613 #endif
614 module_exit(gru_exit);
615
616 module_param(gru_options, ulong, 0644);
617 MODULE_PARM_DESC(gru_options, "Various debug options");
618
619 MODULE_AUTHOR("Silicon Graphics, Inc.");
620 MODULE_LICENSE("GPL");
621 MODULE_DESCRIPTION(GRU_DRIVER_ID_STR GRU_DRIVER_VERSION_STR);
622 MODULE_VERSION(GRU_DRIVER_VERSION_STR);
623