xref: /openbmc/linux/drivers/misc/sgi-gru/grutables.h (revision e8e0929d)
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 /*
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  *
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 "grulib.h"
152 #include "gruhandles.h"
153 
154 extern struct gru_stats_s gru_stats;
155 extern struct gru_blade_state *gru_base[];
156 extern unsigned long gru_start_paddr, gru_end_paddr;
157 extern void *gru_start_vaddr;
158 extern unsigned int gru_max_gids;
159 
160 #define GRU_MAX_BLADES		MAX_NUMNODES
161 #define GRU_MAX_GRUS		(GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE)
162 
163 #define GRU_DRIVER_ID_STR	"SGI GRU Device Driver"
164 #define GRU_DRIVER_VERSION_STR	"0.80"
165 
166 /*
167  * GRU statistics.
168  */
169 struct gru_stats_s {
170 	atomic_long_t vdata_alloc;
171 	atomic_long_t vdata_free;
172 	atomic_long_t gts_alloc;
173 	atomic_long_t gts_free;
174 	atomic_long_t vdata_double_alloc;
175 	atomic_long_t gts_double_allocate;
176 	atomic_long_t assign_context;
177 	atomic_long_t assign_context_failed;
178 	atomic_long_t free_context;
179 	atomic_long_t load_user_context;
180 	atomic_long_t load_kernel_context;
181 	atomic_long_t lock_kernel_context;
182 	atomic_long_t unlock_kernel_context;
183 	atomic_long_t steal_user_context;
184 	atomic_long_t steal_kernel_context;
185 	atomic_long_t steal_context_failed;
186 	atomic_long_t nopfn;
187 	atomic_long_t break_cow;
188 	atomic_long_t asid_new;
189 	atomic_long_t asid_next;
190 	atomic_long_t asid_wrap;
191 	atomic_long_t asid_reuse;
192 	atomic_long_t intr;
193 	atomic_long_t intr_mm_lock_failed;
194 	atomic_long_t call_os;
195 	atomic_long_t call_os_offnode_reference;
196 	atomic_long_t call_os_check_for_bug;
197 	atomic_long_t call_os_wait_queue;
198 	atomic_long_t user_flush_tlb;
199 	atomic_long_t user_unload_context;
200 	atomic_long_t user_exception;
201 	atomic_long_t set_context_option;
202 	atomic_long_t migrate_check;
203 	atomic_long_t migrated_retarget;
204 	atomic_long_t migrated_unload;
205 	atomic_long_t migrated_unload_delay;
206 	atomic_long_t migrated_nopfn_retarget;
207 	atomic_long_t migrated_nopfn_unload;
208 	atomic_long_t tlb_dropin;
209 	atomic_long_t tlb_dropin_fail_no_asid;
210 	atomic_long_t tlb_dropin_fail_upm;
211 	atomic_long_t tlb_dropin_fail_invalid;
212 	atomic_long_t tlb_dropin_fail_range_active;
213 	atomic_long_t tlb_dropin_fail_idle;
214 	atomic_long_t tlb_dropin_fail_fmm;
215 	atomic_long_t tlb_dropin_fail_no_exception;
216 	atomic_long_t tlb_dropin_fail_no_exception_war;
217 	atomic_long_t tfh_stale_on_fault;
218 	atomic_long_t mmu_invalidate_range;
219 	atomic_long_t mmu_invalidate_page;
220 	atomic_long_t mmu_clear_flush_young;
221 	atomic_long_t flush_tlb;
222 	atomic_long_t flush_tlb_gru;
223 	atomic_long_t flush_tlb_gru_tgh;
224 	atomic_long_t flush_tlb_gru_zero_asid;
225 
226 	atomic_long_t copy_gpa;
227 
228 	atomic_long_t mesq_receive;
229 	atomic_long_t mesq_receive_none;
230 	atomic_long_t mesq_send;
231 	atomic_long_t mesq_send_failed;
232 	atomic_long_t mesq_noop;
233 	atomic_long_t mesq_send_unexpected_error;
234 	atomic_long_t mesq_send_lb_overflow;
235 	atomic_long_t mesq_send_qlimit_reached;
236 	atomic_long_t mesq_send_amo_nacked;
237 	atomic_long_t mesq_send_put_nacked;
238 	atomic_long_t mesq_qf_not_full;
239 	atomic_long_t mesq_qf_locked;
240 	atomic_long_t mesq_qf_noop_not_full;
241 	atomic_long_t mesq_qf_switch_head_failed;
242 	atomic_long_t mesq_qf_unexpected_error;
243 	atomic_long_t mesq_noop_unexpected_error;
244 	atomic_long_t mesq_noop_lb_overflow;
245 	atomic_long_t mesq_noop_qlimit_reached;
246 	atomic_long_t mesq_noop_amo_nacked;
247 	atomic_long_t mesq_noop_put_nacked;
248 
249 };
250 
251 enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
252 	cchop_deallocate, tghop_invalidate, mcsop_last};
253 
254 struct mcs_op_statistic {
255 	atomic_long_t	count;
256 	atomic_long_t	total;
257 	unsigned long	max;
258 };
259 
260 extern struct mcs_op_statistic mcs_op_statistics[mcsop_last];
261 
262 #define OPT_DPRINT	1
263 #define OPT_STATS	2
264 
265 
266 #define IRQ_GRU			110	/* Starting IRQ number for interrupts */
267 
268 /* Delay in jiffies between attempts to assign a GRU context */
269 #define GRU_ASSIGN_DELAY	((HZ * 20) / 1000)
270 
271 /*
272  * If a process has it's context stolen, min delay in jiffies before trying to
273  * steal a context from another process.
274  */
275 #define GRU_STEAL_DELAY		((HZ * 200) / 1000)
276 
277 #define STAT(id)	do {						\
278 				if (gru_options & OPT_STATS)		\
279 					atomic_long_inc(&gru_stats.id);	\
280 			} while (0)
281 
282 #ifdef CONFIG_SGI_GRU_DEBUG
283 #define gru_dbg(dev, fmt, x...)						\
284 	do {								\
285 		if (gru_options & OPT_DPRINT)				\
286 			dev_dbg(dev, "%s: " fmt, __func__, x);		\
287 	} while (0)
288 #else
289 #define gru_dbg(x...)
290 #endif
291 
292 /*-----------------------------------------------------------------------------
293  * ASID management
294  */
295 #define MAX_ASID	0xfffff0
296 #define MIN_ASID	8
297 #define ASID_INC	8	/* number of regions */
298 
299 /* Generate a GRU asid value from a GRU base asid & a virtual address. */
300 #if defined CONFIG_IA64
301 #define VADDR_HI_BIT		64
302 #elif defined CONFIG_X86_64
303 #define VADDR_HI_BIT		48
304 #else
305 #error "Unsupported architecture"
306 #endif
307 #define GRUREGION(addr)		((addr) >> (VADDR_HI_BIT - 3) & 3)
308 #define GRUASID(asid, addr)	((asid) + GRUREGION(addr))
309 
310 /*------------------------------------------------------------------------------
311  *  File & VMS Tables
312  */
313 
314 struct gru_state;
315 
316 /*
317  * This structure is pointed to from the mmstruct via the notifier pointer.
318  * There is one of these per address space.
319  */
320 struct gru_mm_tracker {				/* pack to reduce size */
321 	unsigned int		mt_asid_gen:24;	/* ASID wrap count */
322 	unsigned int		mt_asid:24;	/* current base ASID for gru */
323 	unsigned short		mt_ctxbitmap:16;/* bitmap of contexts using
324 						   asid */
325 } __attribute__ ((packed));
326 
327 struct gru_mm_struct {
328 	struct mmu_notifier	ms_notifier;
329 	atomic_t		ms_refcnt;
330 	spinlock_t		ms_asid_lock;	/* protects ASID assignment */
331 	atomic_t		ms_range_active;/* num range_invals active */
332 	char			ms_released;
333 	wait_queue_head_t	ms_wait_queue;
334 	DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS);
335 	struct gru_mm_tracker	ms_asids[GRU_MAX_GRUS];
336 };
337 
338 /*
339  * One of these structures is allocated when a GSEG is mmaped. The
340  * structure is pointed to by the vma->vm_private_data field in the vma struct.
341  */
342 struct gru_vma_data {
343 	spinlock_t		vd_lock;	/* Serialize access to vma */
344 	struct list_head	vd_head;	/* head of linked list of gts */
345 	long			vd_user_options;/* misc user option flags */
346 	int			vd_cbr_au_count;
347 	int			vd_dsr_au_count;
348 };
349 
350 /*
351  * One of these is allocated for each thread accessing a mmaped GRU. A linked
352  * list of these structure is hung off the struct gru_vma_data in the mm_struct.
353  */
354 struct gru_thread_state {
355 	struct list_head	ts_next;	/* list - head at vma-private */
356 	struct mutex		ts_ctxlock;	/* load/unload CTX lock */
357 	struct mm_struct	*ts_mm;		/* mm currently mapped to
358 						   context */
359 	struct vm_area_struct	*ts_vma;	/* vma of GRU context */
360 	struct gru_state	*ts_gru;	/* GRU where the context is
361 						   loaded */
362 	struct gru_mm_struct	*ts_gms;	/* asid & ioproc struct */
363 	unsigned long		ts_cbr_map;	/* map of allocated CBRs */
364 	unsigned long		ts_dsr_map;	/* map of allocated DATA
365 						   resources */
366 	unsigned long		ts_steal_jiffies;/* jiffies when context last
367 						    stolen */
368 	long			ts_user_options;/* misc user option flags */
369 	pid_t			ts_tgid_owner;	/* task that is using the
370 						   context - for migration */
371 	unsigned short		ts_sizeavail;	/* Pagesizes in use */
372 	int			ts_tsid;	/* thread that owns the
373 						   structure */
374 	int			ts_tlb_int_select;/* target cpu if interrupts
375 						     enabled */
376 	int			ts_ctxnum;	/* context number where the
377 						   context is loaded */
378 	atomic_t		ts_refcnt;	/* reference count GTS */
379 	unsigned char		ts_dsr_au_count;/* Number of DSR resources
380 						   required for contest */
381 	unsigned char		ts_cbr_au_count;/* Number of CBR resources
382 						   required for contest */
383 	char			ts_cch_req_slice;/* CCH packet slice */
384 	char			ts_blade;	/* If >= 0, migrate context if
385 						   ref from diferent blade */
386 	char			ts_force_cch_reload;
387 	char			ts_force_unload;/* force context to be unloaded
388 						   after migration */
389 	char			ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each
390 							  allocated CB */
391 	int			ts_data_valid;	/* Indicates if ts_gdata has
392 						   valid data */
393 	struct gts_statistics	ustats;		/* User statistics */
394 	unsigned long		ts_gdata[0];	/* save area for GRU data (CB,
395 						   DS, CBE) */
396 };
397 
398 /*
399  * Threaded programs actually allocate an array of GSEGs when a context is
400  * created. Each thread uses a separate GSEG. TSID is the index into the GSEG
401  * array.
402  */
403 #define TSID(a, v)		(((a) - (v)->vm_start) / GRU_GSEG_PAGESIZE)
404 #define UGRUADDR(gts)		((gts)->ts_vma->vm_start +		\
405 					(gts)->ts_tsid * GRU_GSEG_PAGESIZE)
406 
407 #define NULLCTX			(-1)	/* if context not loaded into GRU */
408 
409 /*-----------------------------------------------------------------------------
410  *  GRU State Tables
411  */
412 
413 /*
414  * One of these exists for each GRU chiplet.
415  */
416 struct gru_state {
417 	struct gru_blade_state	*gs_blade;		/* GRU state for entire
418 							   blade */
419 	unsigned long		gs_gru_base_paddr;	/* Physical address of
420 							   gru segments (64) */
421 	void			*gs_gru_base_vaddr;	/* Virtual address of
422 							   gru segments (64) */
423 	unsigned short		gs_gid;			/* unique GRU number */
424 	unsigned short		gs_blade_id;		/* blade of GRU */
425 	unsigned char		gs_tgh_local_shift;	/* used to pick TGH for
426 							   local flush */
427 	unsigned char		gs_tgh_first_remote;	/* starting TGH# for
428 							   remote flush */
429 	spinlock_t		gs_asid_lock;		/* lock used for
430 							   assigning asids */
431 	spinlock_t		gs_lock;		/* lock used for
432 							   assigning contexts */
433 
434 	/* -- the following are protected by the gs_asid_lock spinlock ---- */
435 	unsigned int		gs_asid;		/* Next availe ASID */
436 	unsigned int		gs_asid_limit;		/* Limit of available
437 							   ASIDs */
438 	unsigned int		gs_asid_gen;		/* asid generation.
439 							   Inc on wrap */
440 
441 	/* --- the following fields are protected by the gs_lock spinlock --- */
442 	unsigned long		gs_context_map;		/* bitmap to manage
443 							   contexts in use */
444 	unsigned long		gs_cbr_map;		/* bitmap to manage CB
445 							   resources */
446 	unsigned long		gs_dsr_map;		/* bitmap used to manage
447 							   DATA resources */
448 	unsigned int		gs_reserved_cbrs;	/* Number of kernel-
449 							   reserved cbrs */
450 	unsigned int		gs_reserved_dsr_bytes;	/* Bytes of kernel-
451 							   reserved dsrs */
452 	unsigned short		gs_active_contexts;	/* number of contexts
453 							   in use */
454 	struct gru_thread_state	*gs_gts[GRU_NUM_CCH];	/* GTS currently using
455 							   the context */
456 };
457 
458 /*
459  * This structure contains the GRU state for all the GRUs on a blade.
460  */
461 struct gru_blade_state {
462 	void			*kernel_cb;		/* First kernel
463 							   reserved cb */
464 	void			*kernel_dsr;		/* First kernel
465 							   reserved DSR */
466 	struct rw_semaphore	bs_kgts_sema;		/* lock for kgts */
467 	struct gru_thread_state *bs_kgts;		/* GTS for kernel use */
468 
469 	/* ---- the following are used for managing kernel async GRU CBRs --- */
470 	int			bs_async_dsr_bytes;	/* DSRs for async */
471 	int			bs_async_cbrs;		/* CBRs AU for async */
472 	struct completion	*bs_async_wq;
473 
474 	/* ---- the following are protected by the bs_lock spinlock ---- */
475 	spinlock_t		bs_lock;		/* lock used for
476 							   stealing contexts */
477 	int			bs_lru_ctxnum;		/* STEAL - last context
478 							   stolen */
479 	struct gru_state	*bs_lru_gru;		/* STEAL - last gru
480 							   stolen */
481 
482 	struct gru_state	bs_grus[GRU_CHIPLETS_PER_BLADE];
483 };
484 
485 /*-----------------------------------------------------------------------------
486  * Address Primitives
487  */
488 #define get_tfm_for_cpu(g, c)						\
489 	((struct gru_tlb_fault_map *)get_tfm((g)->gs_gru_base_vaddr, (c)))
490 #define get_tfh_by_index(g, i)						\
491 	((struct gru_tlb_fault_handle *)get_tfh((g)->gs_gru_base_vaddr, (i)))
492 #define get_tgh_by_index(g, i)						\
493 	((struct gru_tlb_global_handle *)get_tgh((g)->gs_gru_base_vaddr, (i)))
494 #define get_cbe_by_index(g, i)						\
495 	((struct gru_control_block_extended *)get_cbe((g)->gs_gru_base_vaddr,\
496 			(i)))
497 
498 /*-----------------------------------------------------------------------------
499  * Useful Macros
500  */
501 
502 /* Given a blade# & chiplet#, get a pointer to the GRU */
503 #define get_gru(b, c)		(&gru_base[b]->bs_grus[c])
504 
505 /* Number of bytes to save/restore when unloading/loading GRU contexts */
506 #define DSR_BYTES(dsr)		((dsr) * GRU_DSR_AU_BYTES)
507 #define CBR_BYTES(cbr)		((cbr) * GRU_HANDLE_BYTES * GRU_CBR_AU_SIZE * 2)
508 
509 /* Convert a user CB number to the actual CBRNUM */
510 #define thread_cbr_number(gts, n) ((gts)->ts_cbr_idx[(n) / GRU_CBR_AU_SIZE] \
511 				  * GRU_CBR_AU_SIZE + (n) % GRU_CBR_AU_SIZE)
512 
513 /* Convert a gid to a pointer to the GRU */
514 #define GID_TO_GRU(gid)							\
515 	(gru_base[(gid) / GRU_CHIPLETS_PER_BLADE] ?			\
516 		(&gru_base[(gid) / GRU_CHIPLETS_PER_BLADE]->		\
517 			bs_grus[(gid) % GRU_CHIPLETS_PER_BLADE]) :	\
518 	 NULL)
519 
520 /* Scan all active GRUs in a GRU bitmap */
521 #define for_each_gru_in_bitmap(gid, map)				\
522 	for ((gid) = find_first_bit((map), GRU_MAX_GRUS); (gid) < GRU_MAX_GRUS;\
523 		(gid)++, (gid) = find_next_bit((map), GRU_MAX_GRUS, (gid)))
524 
525 /* Scan all active GRUs on a specific blade */
526 #define for_each_gru_on_blade(gru, nid, i)				\
527 	for ((gru) = gru_base[nid]->bs_grus, (i) = 0;			\
528 			(i) < GRU_CHIPLETS_PER_BLADE;			\
529 			(i)++, (gru)++)
530 
531 /* Scan all GRUs */
532 #define foreach_gid(gid)						\
533 	for ((gid) = 0; (gid) < gru_max_gids; (gid)++)
534 
535 /* Scan all active GTSs on a gru. Note: must hold ss_lock to use this macro. */
536 #define for_each_gts_on_gru(gts, gru, ctxnum)				\
537 	for ((ctxnum) = 0; (ctxnum) < GRU_NUM_CCH; (ctxnum)++)		\
538 		if (((gts) = (gru)->gs_gts[ctxnum]))
539 
540 /* Scan each CBR whose bit is set in a TFM (or copy of) */
541 #define for_each_cbr_in_tfm(i, map)					\
542 	for ((i) = find_first_bit(map, GRU_NUM_CBE);			\
543 			(i) < GRU_NUM_CBE;				\
544 			(i)++, (i) = find_next_bit(map, GRU_NUM_CBE, i))
545 
546 /* Scan each CBR in a CBR bitmap. Note: multiple CBRs in an allocation unit */
547 #define for_each_cbr_in_allocation_map(i, map, k)			\
548 	for ((k) = find_first_bit(map, GRU_CBR_AU); (k) < GRU_CBR_AU;	\
549 			(k) = find_next_bit(map, GRU_CBR_AU, (k) + 1)) 	\
550 		for ((i) = (k)*GRU_CBR_AU_SIZE;				\
551 				(i) < ((k) + 1) * GRU_CBR_AU_SIZE; (i)++)
552 
553 /* Scan each DSR in a DSR bitmap. Note: multiple DSRs in an allocation unit */
554 #define for_each_dsr_in_allocation_map(i, map, k)			\
555 	for ((k) = find_first_bit((const unsigned long *)map, GRU_DSR_AU);\
556 			(k) < GRU_DSR_AU;				\
557 			(k) = find_next_bit((const unsigned long *)map,	\
558 					  GRU_DSR_AU, (k) + 1))		\
559 		for ((i) = (k) * GRU_DSR_AU_CL;				\
560 				(i) < ((k) + 1) * GRU_DSR_AU_CL; (i)++)
561 
562 #define gseg_physical_address(gru, ctxnum)				\
563 		((gru)->gs_gru_base_paddr + ctxnum * GRU_GSEG_STRIDE)
564 #define gseg_virtual_address(gru, ctxnum)				\
565 		((gru)->gs_gru_base_vaddr + ctxnum * GRU_GSEG_STRIDE)
566 
567 /*-----------------------------------------------------------------------------
568  * Lock / Unlock GRU handles
569  * 	Use the "delresp" bit in the handle as a "lock" bit.
570  */
571 
572 /* Lock hierarchy checking enabled only in emulator */
573 
574 /* 0 = lock failed, 1 = locked */
575 static inline int __trylock_handle(void *h)
576 {
577 	return !test_and_set_bit(1, h);
578 }
579 
580 static inline void __lock_handle(void *h)
581 {
582 	while (test_and_set_bit(1, h))
583 		cpu_relax();
584 }
585 
586 static inline void __unlock_handle(void *h)
587 {
588 	clear_bit(1, h);
589 }
590 
591 static inline int trylock_cch_handle(struct gru_context_configuration_handle *cch)
592 {
593 	return __trylock_handle(cch);
594 }
595 
596 static inline void lock_cch_handle(struct gru_context_configuration_handle *cch)
597 {
598 	__lock_handle(cch);
599 }
600 
601 static inline void unlock_cch_handle(struct gru_context_configuration_handle
602 				     *cch)
603 {
604 	__unlock_handle(cch);
605 }
606 
607 static inline void lock_tgh_handle(struct gru_tlb_global_handle *tgh)
608 {
609 	__lock_handle(tgh);
610 }
611 
612 static inline void unlock_tgh_handle(struct gru_tlb_global_handle *tgh)
613 {
614 	__unlock_handle(tgh);
615 }
616 
617 static inline int is_kernel_context(struct gru_thread_state *gts)
618 {
619 	return !gts->ts_mm;
620 }
621 
622 /*-----------------------------------------------------------------------------
623  * Function prototypes & externs
624  */
625 struct gru_unload_context_req;
626 
627 extern const struct vm_operations_struct gru_vm_ops;
628 extern struct device *grudev;
629 
630 extern struct gru_vma_data *gru_alloc_vma_data(struct vm_area_struct *vma,
631 				int tsid);
632 extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct
633 				*vma, int tsid);
634 extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct
635 				*vma, int tsid);
636 extern struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts,
637 		int blade);
638 extern void gru_load_context(struct gru_thread_state *gts);
639 extern void gru_steal_context(struct gru_thread_state *gts, int blade_id);
640 extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
641 extern int gru_update_cch(struct gru_thread_state *gts, int force_unload);
642 extern void gts_drop(struct gru_thread_state *gts);
643 extern void gru_tgh_flush_init(struct gru_state *gru);
644 extern int gru_kservices_init(void);
645 extern void gru_kservices_exit(void);
646 extern int gru_dump_chiplet_request(unsigned long arg);
647 extern long gru_get_gseg_statistics(unsigned long arg);
648 extern irqreturn_t gru_intr(int irq, void *dev_id);
649 extern int gru_handle_user_call_os(unsigned long address);
650 extern int gru_user_flush_tlb(unsigned long arg);
651 extern int gru_user_unload_context(unsigned long arg);
652 extern int gru_get_exception_detail(unsigned long arg);
653 extern int gru_set_context_option(unsigned long address);
654 extern int gru_cpu_fault_map_id(void);
655 extern struct vm_area_struct *gru_find_vma(unsigned long vaddr);
656 extern void gru_flush_all_tlb(struct gru_state *gru);
657 extern int gru_proc_init(void);
658 extern void gru_proc_exit(void);
659 
660 extern struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
661 		int cbr_au_count, int dsr_au_count, int options, int tsid);
662 extern unsigned long gru_reserve_cb_resources(struct gru_state *gru,
663 		int cbr_au_count, char *cbmap);
664 extern unsigned long gru_reserve_ds_resources(struct gru_state *gru,
665 		int dsr_au_count, char *dsmap);
666 extern int gru_fault(struct vm_area_struct *, struct vm_fault *vmf);
667 extern struct gru_mm_struct *gru_register_mmu_notifier(void);
668 extern void gru_drop_mmu_notifier(struct gru_mm_struct *gms);
669 
670 extern int gru_ktest(unsigned long arg);
671 extern void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start,
672 					unsigned long len);
673 
674 extern unsigned long gru_options;
675 
676 #endif /* __GRUTABLES_H__ */
677