Commit | Line | Data |
---|---|---|
13d19498 JS |
1 | /* |
2 | * SN Platform GRU Driver | |
3 | * | |
4 | * GRU DRIVER TABLES, MACROS, externs, etc | |
5 | * | |
6 | * Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved. | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or modify | |
9 | * it under the terms of the GNU General Public License as published by | |
10 | * the Free Software Foundation; either version 2 of the License, or | |
11 | * (at your option) any later version. | |
12 | * | |
13 | * This program is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | * GNU General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU General Public License | |
19 | * along with this program; if not, write to the Free Software | |
20 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
21 | */ | |
22 | ||
23 | #ifndef __GRUTABLES_H__ | |
24 | #define __GRUTABLES_H__ | |
25 | ||
26 | /* | |
9ca8e40c JS |
27 | * GRU Chiplet: |
28 | * The GRU is a user addressible memory accelerator. It provides | |
29 | * several forms of load, store, memset, bcopy instructions. In addition, it | |
30 | * contains special instructions for AMOs, sending messages to message | |
31 | * queues, etc. | |
32 | * | |
33 | * The GRU is an integral part of the node controller. It connects | |
34 | * directly to the cpu socket. In its current implementation, there are 2 | |
35 | * GRU chiplets in the node controller on each blade (~node). | |
36 | * | |
37 | * The entire GRU memory space is fully coherent and cacheable by the cpus. | |
38 | * | |
39 | * Each GRU chiplet has a physical memory map that looks like the following: | |
40 | * | |
41 | * +-----------------+ | |
42 | * |/////////////////| | |
43 | * |/////////////////| | |
44 | * |/////////////////| | |
45 | * |/////////////////| | |
46 | * |/////////////////| | |
47 | * |/////////////////| | |
48 | * |/////////////////| | |
49 | * |/////////////////| | |
50 | * +-----------------+ | |
51 | * | system control | | |
52 | * +-----------------+ _______ +-------------+ | |
53 | * |/////////////////| / | | | |
54 | * |/////////////////| / | | | |
55 | * |/////////////////| / | instructions| | |
56 | * |/////////////////| / | | | |
57 | * |/////////////////| / | | | |
58 | * |/////////////////| / |-------------| | |
59 | * |/////////////////| / | | | |
60 | * +-----------------+ | | | |
61 | * | context 15 | | data | | |
62 | * +-----------------+ | | | |
63 | * | ...... | \ | | | |
64 | * +-----------------+ \____________ +-------------+ | |
65 | * | context 1 | | |
66 | * +-----------------+ | |
67 | * | context 0 | | |
68 | * +-----------------+ | |
69 | * | |
70 | * Each of the "contexts" is a chunk of memory that can be mmaped into user | |
71 | * space. The context consists of 2 parts: | |
72 | * | |
73 | * - an instruction space that can be directly accessed by the user | |
74 | * to issue GRU instructions and to check instruction status. | |
75 | * | |
76 | * - a data area that acts as normal RAM. | |
77 | * | |
78 | * User instructions contain virtual addresses of data to be accessed by the | |
79 | * GRU. The GRU contains a TLB that is used to convert these user virtual | |
80 | * addresses to physical addresses. | |
81 | * | |
82 | * The "system control" area of the GRU chiplet is used by the kernel driver | |
83 | * to manage user contexts and to perform functions such as TLB dropin and | |
84 | * purging. | |
85 | * | |
86 | * One context may be reserved for the kernel and used for cross-partition | |
87 | * communication. The GRU will also be used to asynchronously zero out | |
88 | * large blocks of memory (not currently implemented). | |
89 | * | |
90 | * | |
13d19498 JS |
91 | * Tables: |
92 | * | |
93 | * VDATA-VMA Data - Holds a few parameters. Head of linked list of | |
94 | * GTS tables for threads using the GSEG | |
95 | * GTS - Gru Thread State - contains info for managing a GSEG context. A | |
96 | * GTS is allocated for each thread accessing a | |
97 | * GSEG. | |
98 | * GTD - GRU Thread Data - contains shadow copy of GRU data when GSEG is | |
99 | * not loaded into a GRU | |
100 | * GMS - GRU Memory Struct - Used to manage TLB shootdowns. Tracks GRUs | |
101 | * where a GSEG has been loaded. Similar to | |
102 | * an mm_struct but for GRU. | |
103 | * | |
104 | * GS - GRU State - Used to manage the state of a GRU chiplet | |
105 | * BS - Blade State - Used to manage state of all GRU chiplets | |
106 | * on a blade | |
107 | * | |
108 | * | |
109 | * Normal task tables for task using GRU. | |
110 | * - 2 threads in process | |
111 | * - 2 GSEGs open in process | |
112 | * - GSEG1 is being used by both threads | |
113 | * - GSEG2 is used only by thread 2 | |
114 | * | |
115 | * task -->| | |
116 | * task ---+---> mm ->------ (notifier) -------+-> gms | |
117 | * | | | |
118 | * |--> vma -> vdata ---> gts--->| GSEG1 (thread1) | |
119 | * | | | | |
120 | * | +-> gts--->| GSEG1 (thread2) | |
121 | * | | | |
122 | * |--> vma -> vdata ---> gts--->| GSEG2 (thread2) | |
123 | * . | |
124 | * . | |
125 | * | |
126 | * GSEGs are marked DONTCOPY on fork | |
127 | * | |
128 | * At open | |
129 | * file.private_data -> NULL | |
130 | * | |
131 | * At mmap, | |
132 | * vma -> vdata | |
133 | * | |
134 | * After gseg reference | |
135 | * vma -> vdata ->gts | |
136 | * | |
137 | * After fork | |
138 | * parent | |
139 | * vma -> vdata -> gts | |
140 | * child | |
141 | * (vma is not copied) | |
142 | * | |
143 | */ | |
144 | ||
145 | #include <linux/rmap.h> | |
146 | #include <linux/interrupt.h> | |
147 | #include <linux/mutex.h> | |
148 | #include <linux/wait.h> | |
149 | #include <linux/mmu_notifier.h> | |
150 | #include "gru.h" | |
151 | #include "gruhandles.h" | |
152 | ||
153 | extern struct gru_stats_s gru_stats; | |
154 | extern struct gru_blade_state *gru_base[]; | |
155 | extern unsigned long gru_start_paddr, gru_end_paddr; | |
156 | ||
157 | #define GRU_MAX_BLADES MAX_NUMNODES | |
158 | #define GRU_MAX_GRUS (GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE) | |
159 | ||
160 | #define GRU_DRIVER_ID_STR "SGI GRU Device Driver" | |
161 | #define GRU_DRIVER_VERSION_STR "0.80" | |
162 | ||
163 | /* | |
164 | * GRU statistics. | |
165 | */ | |
166 | struct gru_stats_s { | |
167 | atomic_long_t vdata_alloc; | |
168 | atomic_long_t vdata_free; | |
169 | atomic_long_t gts_alloc; | |
170 | atomic_long_t gts_free; | |
171 | atomic_long_t vdata_double_alloc; | |
172 | atomic_long_t gts_double_allocate; | |
173 | atomic_long_t assign_context; | |
174 | atomic_long_t assign_context_failed; | |
175 | atomic_long_t free_context; | |
176 | atomic_long_t load_context; | |
177 | atomic_long_t unload_context; | |
178 | atomic_long_t steal_context; | |
179 | atomic_long_t steal_context_failed; | |
180 | atomic_long_t nopfn; | |
181 | atomic_long_t break_cow; | |
182 | atomic_long_t asid_new; | |
183 | atomic_long_t asid_next; | |
184 | atomic_long_t asid_wrap; | |
185 | atomic_long_t asid_reuse; | |
186 | atomic_long_t intr; | |
43884604 | 187 | atomic_long_t intr_mm_lock_failed; |
13d19498 | 188 | atomic_long_t call_os; |
43884604 | 189 | atomic_long_t call_os_offnode_reference; |
13d19498 JS |
190 | atomic_long_t call_os_check_for_bug; |
191 | atomic_long_t call_os_wait_queue; | |
192 | atomic_long_t user_flush_tlb; | |
193 | atomic_long_t user_unload_context; | |
194 | atomic_long_t user_exception; | |
195 | atomic_long_t set_task_slice; | |
196 | atomic_long_t migrate_check; | |
197 | atomic_long_t migrated_retarget; | |
198 | atomic_long_t migrated_unload; | |
199 | atomic_long_t migrated_unload_delay; | |
200 | atomic_long_t migrated_nopfn_retarget; | |
201 | atomic_long_t migrated_nopfn_unload; | |
202 | atomic_long_t tlb_dropin; | |
203 | atomic_long_t tlb_dropin_fail_no_asid; | |
204 | atomic_long_t tlb_dropin_fail_upm; | |
205 | atomic_long_t tlb_dropin_fail_invalid; | |
206 | atomic_long_t tlb_dropin_fail_range_active; | |
207 | atomic_long_t tlb_dropin_fail_idle; | |
208 | atomic_long_t tlb_dropin_fail_fmm; | |
209 | atomic_long_t mmu_invalidate_range; | |
210 | atomic_long_t mmu_invalidate_page; | |
211 | atomic_long_t mmu_clear_flush_young; | |
212 | atomic_long_t flush_tlb; | |
213 | atomic_long_t flush_tlb_gru; | |
214 | atomic_long_t flush_tlb_gru_tgh; | |
215 | atomic_long_t flush_tlb_gru_zero_asid; | |
216 | ||
217 | atomic_long_t copy_gpa; | |
218 | ||
219 | atomic_long_t mesq_receive; | |
220 | atomic_long_t mesq_receive_none; | |
221 | atomic_long_t mesq_send; | |
222 | atomic_long_t mesq_send_failed; | |
223 | atomic_long_t mesq_noop; | |
224 | atomic_long_t mesq_send_unexpected_error; | |
225 | atomic_long_t mesq_send_lb_overflow; | |
226 | atomic_long_t mesq_send_qlimit_reached; | |
227 | atomic_long_t mesq_send_amo_nacked; | |
228 | atomic_long_t mesq_send_put_nacked; | |
229 | atomic_long_t mesq_qf_not_full; | |
230 | atomic_long_t mesq_qf_locked; | |
231 | atomic_long_t mesq_qf_noop_not_full; | |
232 | atomic_long_t mesq_qf_switch_head_failed; | |
233 | atomic_long_t mesq_qf_unexpected_error; | |
234 | atomic_long_t mesq_noop_unexpected_error; | |
235 | atomic_long_t mesq_noop_lb_overflow; | |
236 | atomic_long_t mesq_noop_qlimit_reached; | |
237 | atomic_long_t mesq_noop_amo_nacked; | |
238 | atomic_long_t mesq_noop_put_nacked; | |
239 | ||
240 | }; | |
241 | ||
a24e5e1c JS |
242 | enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync, |
243 | cchop_deallocate, tghop_invalidate, mcsop_last}; | |
244 | ||
e56484da JS |
245 | struct mcs_op_statistic { |
246 | atomic_long_t count; | |
247 | atomic_long_t total; | |
248 | unsigned long max; | |
249 | }; | |
250 | ||
251 | extern struct mcs_op_statistic mcs_op_statistics[mcsop_last]; | |
252 | ||
13d19498 JS |
253 | #define OPT_DPRINT 1 |
254 | #define OPT_STATS 2 | |
255 | #define GRU_QUICKLOOK 4 | |
256 | ||
257 | ||
258 | #define IRQ_GRU 110 /* Starting IRQ number for interrupts */ | |
259 | ||
260 | /* Delay in jiffies between attempts to assign a GRU context */ | |
261 | #define GRU_ASSIGN_DELAY ((HZ * 20) / 1000) | |
262 | ||
263 | /* | |
264 | * If a process has it's context stolen, min delay in jiffies before trying to | |
265 | * steal a context from another process. | |
266 | */ | |
267 | #define GRU_STEAL_DELAY ((HZ * 200) / 1000) | |
268 | ||
269 | #define STAT(id) do { \ | |
9ca8e40c | 270 | if (gru_options & OPT_STATS) \ |
13d19498 JS |
271 | atomic_long_inc(&gru_stats.id); \ |
272 | } while (0) | |
273 | ||
274 | #ifdef CONFIG_SGI_GRU_DEBUG | |
275 | #define gru_dbg(dev, fmt, x...) \ | |
276 | do { \ | |
9ca8e40c | 277 | if (gru_options & OPT_DPRINT) \ |
13d19498 JS |
278 | dev_dbg(dev, "%s: " fmt, __func__, x); \ |
279 | } while (0) | |
280 | #else | |
281 | #define gru_dbg(x...) | |
282 | #endif | |
283 | ||
284 | /*----------------------------------------------------------------------------- | |
285 | * ASID management | |
286 | */ | |
287 | #define MAX_ASID 0xfffff0 | |
288 | #define MIN_ASID 8 | |
289 | #define ASID_INC 8 /* number of regions */ | |
290 | ||
291 | /* Generate a GRU asid value from a GRU base asid & a virtual address. */ | |
292 | #if defined CONFIG_IA64 | |
293 | #define VADDR_HI_BIT 64 | |
6a4ad39b | 294 | #elif defined CONFIG_X86_64 |
13d19498 | 295 | #define VADDR_HI_BIT 48 |
13d19498 JS |
296 | #else |
297 | #error "Unsupported architecture" | |
298 | #endif | |
fe5bb6b0 | 299 | #define GRUREGION(addr) ((addr) >> (VADDR_HI_BIT - 3) & 3) |
13d19498 JS |
300 | #define GRUASID(asid, addr) ((asid) + GRUREGION(addr)) |
301 | ||
302 | /*------------------------------------------------------------------------------ | |
303 | * File & VMS Tables | |
304 | */ | |
305 | ||
306 | struct gru_state; | |
307 | ||
308 | /* | |
309 | * This structure is pointed to from the mmstruct via the notifier pointer. | |
310 | * There is one of these per address space. | |
311 | */ | |
fe5bb6b0 JS |
312 | struct gru_mm_tracker { /* pack to reduce size */ |
313 | unsigned int mt_asid_gen:24; /* ASID wrap count */ | |
314 | unsigned int mt_asid:24; /* current base ASID for gru */ | |
315 | unsigned short mt_ctxbitmap:16;/* bitmap of contexts using | |
13d19498 | 316 | asid */ |
fe5bb6b0 | 317 | } __attribute__ ((packed)); |
13d19498 JS |
318 | |
319 | struct gru_mm_struct { | |
320 | struct mmu_notifier ms_notifier; | |
321 | atomic_t ms_refcnt; | |
322 | spinlock_t ms_asid_lock; /* protects ASID assignment */ | |
323 | atomic_t ms_range_active;/* num range_invals active */ | |
324 | char ms_released; | |
325 | wait_queue_head_t ms_wait_queue; | |
326 | DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS); | |
327 | struct gru_mm_tracker ms_asids[GRU_MAX_GRUS]; | |
328 | }; | |
329 | ||
330 | /* | |
331 | * One of these structures is allocated when a GSEG is mmaped. The | |
332 | * structure is pointed to by the vma->vm_private_data field in the vma struct. | |
333 | */ | |
334 | struct gru_vma_data { | |
335 | spinlock_t vd_lock; /* Serialize access to vma */ | |
336 | struct list_head vd_head; /* head of linked list of gts */ | |
337 | long vd_user_options;/* misc user option flags */ | |
338 | int vd_cbr_au_count; | |
339 | int vd_dsr_au_count; | |
340 | }; | |
341 | ||
342 | /* | |
343 | * One of these is allocated for each thread accessing a mmaped GRU. A linked | |
344 | * list of these structure is hung off the struct gru_vma_data in the mm_struct. | |
345 | */ | |
346 | struct gru_thread_state { | |
347 | struct list_head ts_next; /* list - head at vma-private */ | |
348 | struct mutex ts_ctxlock; /* load/unload CTX lock */ | |
349 | struct mm_struct *ts_mm; /* mm currently mapped to | |
350 | context */ | |
351 | struct vm_area_struct *ts_vma; /* vma of GRU context */ | |
352 | struct gru_state *ts_gru; /* GRU where the context is | |
353 | loaded */ | |
354 | struct gru_mm_struct *ts_gms; /* asid & ioproc struct */ | |
355 | unsigned long ts_cbr_map; /* map of allocated CBRs */ | |
356 | unsigned long ts_dsr_map; /* map of allocated DATA | |
357 | resources */ | |
358 | unsigned long ts_steal_jiffies;/* jiffies when context last | |
359 | stolen */ | |
360 | long ts_user_options;/* misc user option flags */ | |
361 | pid_t ts_tgid_owner; /* task that is using the | |
362 | context - for migration */ | |
363 | int ts_tsid; /* thread that owns the | |
364 | structure */ | |
365 | int ts_tlb_int_select;/* target cpu if interrupts | |
366 | enabled */ | |
367 | int ts_ctxnum; /* context number where the | |
368 | context is loaded */ | |
369 | atomic_t ts_refcnt; /* reference count GTS */ | |
370 | unsigned char ts_dsr_au_count;/* Number of DSR resources | |
371 | required for contest */ | |
372 | unsigned char ts_cbr_au_count;/* Number of CBR resources | |
373 | required for contest */ | |
fe5bb6b0 JS |
374 | char ts_blade; /* If >= 0, migrate context if |
375 | ref from diferent blade */ | |
13d19498 JS |
376 | char ts_force_unload;/* force context to be unloaded |
377 | after migration */ | |
378 | char ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each | |
379 | allocated CB */ | |
380 | unsigned long ts_gdata[0]; /* save area for GRU data (CB, | |
381 | DS, CBE) */ | |
382 | }; | |
383 | ||
384 | /* | |
385 | * Threaded programs actually allocate an array of GSEGs when a context is | |
386 | * created. Each thread uses a separate GSEG. TSID is the index into the GSEG | |
387 | * array. | |
388 | */ | |
389 | #define TSID(a, v) (((a) - (v)->vm_start) / GRU_GSEG_PAGESIZE) | |
390 | #define UGRUADDR(gts) ((gts)->ts_vma->vm_start + \ | |
391 | (gts)->ts_tsid * GRU_GSEG_PAGESIZE) | |
392 | ||
393 | #define NULLCTX (-1) /* if context not loaded into GRU */ | |
394 | ||
395 | /*----------------------------------------------------------------------------- | |
396 | * GRU State Tables | |
397 | */ | |
398 | ||
399 | /* | |
400 | * One of these exists for each GRU chiplet. | |
401 | */ | |
402 | struct gru_state { | |
403 | struct gru_blade_state *gs_blade; /* GRU state for entire | |
404 | blade */ | |
405 | unsigned long gs_gru_base_paddr; /* Physical address of | |
406 | gru segments (64) */ | |
407 | void *gs_gru_base_vaddr; /* Virtual address of | |
408 | gru segments (64) */ | |
409 | unsigned char gs_gid; /* unique GRU number */ | |
410 | unsigned char gs_tgh_local_shift; /* used to pick TGH for | |
411 | local flush */ | |
412 | unsigned char gs_tgh_first_remote; /* starting TGH# for | |
413 | remote flush */ | |
414 | unsigned short gs_blade_id; /* blade of GRU */ | |
415 | spinlock_t gs_asid_lock; /* lock used for | |
416 | assigning asids */ | |
417 | spinlock_t gs_lock; /* lock used for | |
418 | assigning contexts */ | |
419 | ||
420 | /* -- the following are protected by the gs_asid_lock spinlock ---- */ | |
421 | unsigned int gs_asid; /* Next availe ASID */ | |
422 | unsigned int gs_asid_limit; /* Limit of available | |
423 | ASIDs */ | |
424 | unsigned int gs_asid_gen; /* asid generation. | |
425 | Inc on wrap */ | |
426 | ||
427 | /* --- the following fields are protected by the gs_lock spinlock --- */ | |
428 | unsigned long gs_context_map; /* bitmap to manage | |
429 | contexts in use */ | |
430 | unsigned long gs_cbr_map; /* bitmap to manage CB | |
431 | resources */ | |
432 | unsigned long gs_dsr_map; /* bitmap used to manage | |
433 | DATA resources */ | |
434 | unsigned int gs_reserved_cbrs; /* Number of kernel- | |
435 | reserved cbrs */ | |
436 | unsigned int gs_reserved_dsr_bytes; /* Bytes of kernel- | |
437 | reserved dsrs */ | |
438 | unsigned short gs_active_contexts; /* number of contexts | |
439 | in use */ | |
440 | struct gru_thread_state *gs_gts[GRU_NUM_CCH]; /* GTS currently using | |
441 | the context */ | |
442 | }; | |
443 | ||
444 | /* | |
445 | * This structure contains the GRU state for all the GRUs on a blade. | |
446 | */ | |
447 | struct gru_blade_state { | |
448 | void *kernel_cb; /* First kernel | |
449 | reserved cb */ | |
450 | void *kernel_dsr; /* First kernel | |
451 | reserved DSR */ | |
452 | /* ---- the following are protected by the bs_lock spinlock ---- */ | |
453 | spinlock_t bs_lock; /* lock used for | |
454 | stealing contexts */ | |
455 | int bs_lru_ctxnum; /* STEAL - last context | |
456 | stolen */ | |
457 | struct gru_state *bs_lru_gru; /* STEAL - last gru | |
458 | stolen */ | |
459 | ||
460 | struct gru_state bs_grus[GRU_CHIPLETS_PER_BLADE]; | |
461 | }; | |
462 | ||
463 | /*----------------------------------------------------------------------------- | |
464 | * Address Primitives | |
465 | */ | |
466 | #define get_tfm_for_cpu(g, c) \ | |
467 | ((struct gru_tlb_fault_map *)get_tfm((g)->gs_gru_base_vaddr, (c))) | |
468 | #define get_tfh_by_index(g, i) \ | |
469 | ((struct gru_tlb_fault_handle *)get_tfh((g)->gs_gru_base_vaddr, (i))) | |
470 | #define get_tgh_by_index(g, i) \ | |
471 | ((struct gru_tlb_global_handle *)get_tgh((g)->gs_gru_base_vaddr, (i))) | |
472 | #define get_cbe_by_index(g, i) \ | |
473 | ((struct gru_control_block_extended *)get_cbe((g)->gs_gru_base_vaddr,\ | |
474 | (i))) | |
475 | ||
476 | /*----------------------------------------------------------------------------- | |
477 | * Useful Macros | |
478 | */ | |
479 | ||
480 | /* Given a blade# & chiplet#, get a pointer to the GRU */ | |
481 | #define get_gru(b, c) (&gru_base[b]->bs_grus[c]) | |
482 | ||
483 | /* Number of bytes to save/restore when unloading/loading GRU contexts */ | |
484 | #define DSR_BYTES(dsr) ((dsr) * GRU_DSR_AU_BYTES) | |
485 | #define CBR_BYTES(cbr) ((cbr) * GRU_HANDLE_BYTES * GRU_CBR_AU_SIZE * 2) | |
486 | ||
487 | /* Convert a user CB number to the actual CBRNUM */ | |
488 | #define thread_cbr_number(gts, n) ((gts)->ts_cbr_idx[(n) / GRU_CBR_AU_SIZE] \ | |
489 | * GRU_CBR_AU_SIZE + (n) % GRU_CBR_AU_SIZE) | |
490 | ||
491 | /* Convert a gid to a pointer to the GRU */ | |
492 | #define GID_TO_GRU(gid) \ | |
493 | (gru_base[(gid) / GRU_CHIPLETS_PER_BLADE] ? \ | |
494 | (&gru_base[(gid) / GRU_CHIPLETS_PER_BLADE]-> \ | |
495 | bs_grus[(gid) % GRU_CHIPLETS_PER_BLADE]) : \ | |
496 | NULL) | |
497 | ||
498 | /* Scan all active GRUs in a GRU bitmap */ | |
499 | #define for_each_gru_in_bitmap(gid, map) \ | |
500 | for ((gid) = find_first_bit((map), GRU_MAX_GRUS); (gid) < GRU_MAX_GRUS;\ | |
501 | (gid)++, (gid) = find_next_bit((map), GRU_MAX_GRUS, (gid))) | |
502 | ||
503 | /* Scan all active GRUs on a specific blade */ | |
504 | #define for_each_gru_on_blade(gru, nid, i) \ | |
505 | for ((gru) = gru_base[nid]->bs_grus, (i) = 0; \ | |
506 | (i) < GRU_CHIPLETS_PER_BLADE; \ | |
507 | (i)++, (gru)++) | |
508 | ||
509 | /* Scan all active GTSs on a gru. Note: must hold ss_lock to use this macro. */ | |
510 | #define for_each_gts_on_gru(gts, gru, ctxnum) \ | |
511 | for ((ctxnum) = 0; (ctxnum) < GRU_NUM_CCH; (ctxnum)++) \ | |
512 | if (((gts) = (gru)->gs_gts[ctxnum])) | |
513 | ||
514 | /* Scan each CBR whose bit is set in a TFM (or copy of) */ | |
515 | #define for_each_cbr_in_tfm(i, map) \ | |
516 | for ((i) = find_first_bit(map, GRU_NUM_CBE); \ | |
517 | (i) < GRU_NUM_CBE; \ | |
518 | (i)++, (i) = find_next_bit(map, GRU_NUM_CBE, i)) | |
519 | ||
520 | /* Scan each CBR in a CBR bitmap. Note: multiple CBRs in an allocation unit */ | |
521 | #define for_each_cbr_in_allocation_map(i, map, k) \ | |
522 | for ((k) = find_first_bit(map, GRU_CBR_AU); (k) < GRU_CBR_AU; \ | |
523 | (k) = find_next_bit(map, GRU_CBR_AU, (k) + 1)) \ | |
524 | for ((i) = (k)*GRU_CBR_AU_SIZE; \ | |
525 | (i) < ((k) + 1) * GRU_CBR_AU_SIZE; (i)++) | |
526 | ||
527 | /* Scan each DSR in a DSR bitmap. Note: multiple DSRs in an allocation unit */ | |
528 | #define for_each_dsr_in_allocation_map(i, map, k) \ | |
529 | for ((k) = find_first_bit((const unsigned long *)map, GRU_DSR_AU);\ | |
530 | (k) < GRU_DSR_AU; \ | |
531 | (k) = find_next_bit((const unsigned long *)map, \ | |
532 | GRU_DSR_AU, (k) + 1)) \ | |
533 | for ((i) = (k) * GRU_DSR_AU_CL; \ | |
534 | (i) < ((k) + 1) * GRU_DSR_AU_CL; (i)++) | |
535 | ||
536 | #define gseg_physical_address(gru, ctxnum) \ | |
537 | ((gru)->gs_gru_base_paddr + ctxnum * GRU_GSEG_STRIDE) | |
538 | #define gseg_virtual_address(gru, ctxnum) \ | |
539 | ((gru)->gs_gru_base_vaddr + ctxnum * GRU_GSEG_STRIDE) | |
540 | ||
541 | /*----------------------------------------------------------------------------- | |
542 | * Lock / Unlock GRU handles | |
543 | * Use the "delresp" bit in the handle as a "lock" bit. | |
544 | */ | |
545 | ||
546 | /* Lock hierarchy checking enabled only in emulator */ | |
547 | ||
548 | static inline void __lock_handle(void *h) | |
549 | { | |
550 | while (test_and_set_bit(1, h)) | |
551 | cpu_relax(); | |
552 | } | |
553 | ||
554 | static inline void __unlock_handle(void *h) | |
555 | { | |
556 | clear_bit(1, h); | |
557 | } | |
558 | ||
559 | static inline void lock_cch_handle(struct gru_context_configuration_handle *cch) | |
560 | { | |
561 | __lock_handle(cch); | |
562 | } | |
563 | ||
564 | static inline void unlock_cch_handle(struct gru_context_configuration_handle | |
565 | *cch) | |
566 | { | |
567 | __unlock_handle(cch); | |
568 | } | |
569 | ||
570 | static inline void lock_tgh_handle(struct gru_tlb_global_handle *tgh) | |
571 | { | |
572 | __lock_handle(tgh); | |
573 | } | |
574 | ||
575 | static inline void unlock_tgh_handle(struct gru_tlb_global_handle *tgh) | |
576 | { | |
577 | __unlock_handle(tgh); | |
578 | } | |
579 | ||
580 | /*----------------------------------------------------------------------------- | |
581 | * Function prototypes & externs | |
582 | */ | |
583 | struct gru_unload_context_req; | |
584 | ||
585 | extern struct vm_operations_struct gru_vm_ops; | |
586 | extern struct device *grudev; | |
587 | ||
588 | extern struct gru_vma_data *gru_alloc_vma_data(struct vm_area_struct *vma, | |
589 | int tsid); | |
590 | extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct | |
591 | *vma, int tsid); | |
592 | extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct | |
593 | *vma, int tsid); | |
594 | extern void gru_unload_context(struct gru_thread_state *gts, int savestate); | |
595 | extern void gts_drop(struct gru_thread_state *gts); | |
596 | extern void gru_tgh_flush_init(struct gru_state *gru); | |
597 | extern int gru_kservices_init(struct gru_state *gru); | |
598 | extern irqreturn_t gru_intr(int irq, void *dev_id); | |
599 | extern int gru_handle_user_call_os(unsigned long address); | |
600 | extern int gru_user_flush_tlb(unsigned long arg); | |
601 | extern int gru_user_unload_context(unsigned long arg); | |
602 | extern int gru_get_exception_detail(unsigned long arg); | |
603 | extern int gru_set_task_slice(long address); | |
604 | extern int gru_cpu_fault_map_id(void); | |
605 | extern struct vm_area_struct *gru_find_vma(unsigned long vaddr); | |
606 | extern void gru_flush_all_tlb(struct gru_state *gru); | |
607 | extern int gru_proc_init(void); | |
608 | extern void gru_proc_exit(void); | |
609 | ||
9ca8e40c | 610 | extern unsigned long gru_reserve_cb_resources(struct gru_state *gru, |
13d19498 | 611 | int cbr_au_count, char *cbmap); |
9ca8e40c | 612 | extern unsigned long gru_reserve_ds_resources(struct gru_state *gru, |
13d19498 JS |
613 | int dsr_au_count, char *dsmap); |
614 | extern int gru_fault(struct vm_area_struct *, struct vm_fault *vmf); | |
615 | extern struct gru_mm_struct *gru_register_mmu_notifier(void); | |
616 | extern void gru_drop_mmu_notifier(struct gru_mm_struct *gms); | |
617 | ||
618 | extern void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start, | |
619 | unsigned long len); | |
620 | ||
9ca8e40c | 621 | extern unsigned long gru_options; |
13d19498 JS |
622 | |
623 | #endif /* __GRUTABLES_H__ */ |