xref: /openbmc/linux/drivers/misc/sgi-gru/grutables.h (revision a09d2831)
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	"0.85"
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 gms_alloc;
175 	atomic_long_t gms_free;
176 	atomic_long_t gts_double_allocate;
177 	atomic_long_t assign_context;
178 	atomic_long_t assign_context_failed;
179 	atomic_long_t free_context;
180 	atomic_long_t load_user_context;
181 	atomic_long_t load_kernel_context;
182 	atomic_long_t lock_kernel_context;
183 	atomic_long_t unlock_kernel_context;
184 	atomic_long_t steal_user_context;
185 	atomic_long_t steal_kernel_context;
186 	atomic_long_t steal_context_failed;
187 	atomic_long_t nopfn;
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_cbr;
194 	atomic_long_t intr_tfh;
195 	atomic_long_t intr_spurious;
196 	atomic_long_t intr_mm_lock_failed;
197 	atomic_long_t call_os;
198 	atomic_long_t call_os_wait_queue;
199 	atomic_long_t user_flush_tlb;
200 	atomic_long_t user_unload_context;
201 	atomic_long_t user_exception;
202 	atomic_long_t set_context_option;
203 	atomic_long_t check_context_retarget_intr;
204 	atomic_long_t check_context_unload;
205 	atomic_long_t tlb_dropin;
206 	atomic_long_t tlb_preload_page;
207 	atomic_long_t tlb_dropin_fail_no_asid;
208 	atomic_long_t tlb_dropin_fail_upm;
209 	atomic_long_t tlb_dropin_fail_invalid;
210 	atomic_long_t tlb_dropin_fail_range_active;
211 	atomic_long_t tlb_dropin_fail_idle;
212 	atomic_long_t tlb_dropin_fail_fmm;
213 	atomic_long_t tlb_dropin_fail_no_exception;
214 	atomic_long_t tfh_stale_on_fault;
215 	atomic_long_t mmu_invalidate_range;
216 	atomic_long_t mmu_invalidate_page;
217 	atomic_long_t flush_tlb;
218 	atomic_long_t flush_tlb_gru;
219 	atomic_long_t flush_tlb_gru_tgh;
220 	atomic_long_t flush_tlb_gru_zero_asid;
221 
222 	atomic_long_t copy_gpa;
223 	atomic_long_t read_gpa;
224 
225 	atomic_long_t mesq_receive;
226 	atomic_long_t mesq_receive_none;
227 	atomic_long_t mesq_send;
228 	atomic_long_t mesq_send_failed;
229 	atomic_long_t mesq_noop;
230 	atomic_long_t mesq_send_unexpected_error;
231 	atomic_long_t mesq_send_lb_overflow;
232 	atomic_long_t mesq_send_qlimit_reached;
233 	atomic_long_t mesq_send_amo_nacked;
234 	atomic_long_t mesq_send_put_nacked;
235 	atomic_long_t mesq_page_overflow;
236 	atomic_long_t mesq_qf_locked;
237 	atomic_long_t mesq_qf_noop_not_full;
238 	atomic_long_t mesq_qf_switch_head_failed;
239 	atomic_long_t mesq_qf_unexpected_error;
240 	atomic_long_t mesq_noop_unexpected_error;
241 	atomic_long_t mesq_noop_lb_overflow;
242 	atomic_long_t mesq_noop_qlimit_reached;
243 	atomic_long_t mesq_noop_amo_nacked;
244 	atomic_long_t mesq_noop_put_nacked;
245 	atomic_long_t mesq_noop_page_overflow;
246 
247 };
248 
249 enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
250 	cchop_deallocate, tfhop_write_only, tfhop_write_restart,
251 	tghop_invalidate, mcsop_last};
252 
253 struct mcs_op_statistic {
254 	atomic_long_t	count;
255 	atomic_long_t	total;
256 	unsigned long	max;
257 };
258 
259 extern struct mcs_op_statistic mcs_op_statistics[mcsop_last];
260 
261 #define OPT_DPRINT		1
262 #define OPT_STATS		2
263 
264 
265 #define IRQ_GRU			110	/* Starting IRQ number for interrupts */
266 
267 /* Delay in jiffies between attempts to assign a GRU context */
268 #define GRU_ASSIGN_DELAY	((HZ * 20) / 1000)
269 
270 /*
271  * If a process has it's context stolen, min delay in jiffies before trying to
272  * steal a context from another process.
273  */
274 #define GRU_STEAL_DELAY		((HZ * 200) / 1000)
275 
276 #define STAT(id)	do {						\
277 				if (gru_options & OPT_STATS)		\
278 					atomic_long_inc(&gru_stats.id);	\
279 			} while (0)
280 
281 #ifdef CONFIG_SGI_GRU_DEBUG
282 #define gru_dbg(dev, fmt, x...)						\
283 	do {								\
284 		if (gru_options & OPT_DPRINT)				\
285 			printk(KERN_DEBUG "GRU:%d %s: " fmt, smp_processor_id(), __func__, x);\
286 	} while (0)
287 #else
288 #define gru_dbg(x...)
289 #endif
290 
291 /*-----------------------------------------------------------------------------
292  * ASID management
293  */
294 #define MAX_ASID	0xfffff0
295 #define MIN_ASID	8
296 #define ASID_INC	8	/* number of regions */
297 
298 /* Generate a GRU asid value from a GRU base asid & a virtual address. */
299 #define VADDR_HI_BIT		64
300 #define GRUREGION(addr)		((addr) >> (VADDR_HI_BIT - 3) & 3)
301 #define GRUASID(asid, addr)	((asid) + GRUREGION(addr))
302 
303 /*------------------------------------------------------------------------------
304  *  File & VMS Tables
305  */
306 
307 struct gru_state;
308 
309 /*
310  * This structure is pointed to from the mmstruct via the notifier pointer.
311  * There is one of these per address space.
312  */
313 struct gru_mm_tracker {				/* pack to reduce size */
314 	unsigned int		mt_asid_gen:24;	/* ASID wrap count */
315 	unsigned int		mt_asid:24;	/* current base ASID for gru */
316 	unsigned short		mt_ctxbitmap:16;/* bitmap of contexts using
317 						   asid */
318 } __attribute__ ((packed));
319 
320 struct gru_mm_struct {
321 	struct mmu_notifier	ms_notifier;
322 	atomic_t		ms_refcnt;
323 	spinlock_t		ms_asid_lock;	/* protects ASID assignment */
324 	atomic_t		ms_range_active;/* num range_invals active */
325 	char			ms_released;
326 	wait_queue_head_t	ms_wait_queue;
327 	DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS);
328 	struct gru_mm_tracker	ms_asids[GRU_MAX_GRUS];
329 };
330 
331 /*
332  * One of these structures is allocated when a GSEG is mmaped. The
333  * structure is pointed to by the vma->vm_private_data field in the vma struct.
334  */
335 struct gru_vma_data {
336 	spinlock_t		vd_lock;	/* Serialize access to vma */
337 	struct list_head	vd_head;	/* head of linked list of gts */
338 	long			vd_user_options;/* misc user option flags */
339 	int			vd_cbr_au_count;
340 	int			vd_dsr_au_count;
341 	unsigned char		vd_tlb_preload_count;
342 };
343 
344 /*
345  * One of these is allocated for each thread accessing a mmaped GRU. A linked
346  * list of these structure is hung off the struct gru_vma_data in the mm_struct.
347  */
348 struct gru_thread_state {
349 	struct list_head	ts_next;	/* list - head at vma-private */
350 	struct mutex		ts_ctxlock;	/* load/unload CTX lock */
351 	struct mm_struct	*ts_mm;		/* mm currently mapped to
352 						   context */
353 	struct vm_area_struct	*ts_vma;	/* vma of GRU context */
354 	struct gru_state	*ts_gru;	/* GRU where the context is
355 						   loaded */
356 	struct gru_mm_struct	*ts_gms;	/* asid & ioproc struct */
357 	unsigned char		ts_tlb_preload_count; /* TLB preload pages */
358 	unsigned long		ts_cbr_map;	/* map of allocated CBRs */
359 	unsigned long		ts_dsr_map;	/* map of allocated DATA
360 						   resources */
361 	unsigned long		ts_steal_jiffies;/* jiffies when context last
362 						    stolen */
363 	long			ts_user_options;/* misc user option flags */
364 	pid_t			ts_tgid_owner;	/* task that is using the
365 						   context - for migration */
366 	short			ts_user_blade_id;/* user selected blade */
367 	char			ts_user_chiplet_id;/* user selected chiplet */
368 	unsigned short		ts_sizeavail;	/* Pagesizes in use */
369 	int			ts_tsid;	/* thread that owns the
370 						   structure */
371 	int			ts_tlb_int_select;/* target cpu if interrupts
372 						     enabled */
373 	int			ts_ctxnum;	/* context number where the
374 						   context is loaded */
375 	atomic_t		ts_refcnt;	/* reference count GTS */
376 	unsigned char		ts_dsr_au_count;/* Number of DSR resources
377 						   required for contest */
378 	unsigned char		ts_cbr_au_count;/* Number of CBR resources
379 						   required for contest */
380 	char			ts_cch_req_slice;/* CCH packet slice */
381 	char			ts_blade;	/* If >= 0, migrate context if
382 						   ref from diferent blade */
383 	char			ts_force_cch_reload;
384 	char			ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each
385 							  allocated CB */
386 	int			ts_data_valid;	/* Indicates if ts_gdata has
387 						   valid data */
388 	struct gru_gseg_statistics ustats;	/* User statistics */
389 	unsigned long		ts_gdata[0];	/* save area for GRU data (CB,
390 						   DS, CBE) */
391 };
392 
393 /*
394  * Threaded programs actually allocate an array of GSEGs when a context is
395  * created. Each thread uses a separate GSEG. TSID is the index into the GSEG
396  * array.
397  */
398 #define TSID(a, v)		(((a) - (v)->vm_start) / GRU_GSEG_PAGESIZE)
399 #define UGRUADDR(gts)		((gts)->ts_vma->vm_start +		\
400 					(gts)->ts_tsid * GRU_GSEG_PAGESIZE)
401 
402 #define NULLCTX			(-1)	/* if context not loaded into GRU */
403 
404 /*-----------------------------------------------------------------------------
405  *  GRU State Tables
406  */
407 
408 /*
409  * One of these exists for each GRU chiplet.
410  */
411 struct gru_state {
412 	struct gru_blade_state	*gs_blade;		/* GRU state for entire
413 							   blade */
414 	unsigned long		gs_gru_base_paddr;	/* Physical address of
415 							   gru segments (64) */
416 	void			*gs_gru_base_vaddr;	/* Virtual address of
417 							   gru segments (64) */
418 	unsigned short		gs_gid;			/* unique GRU number */
419 	unsigned short		gs_blade_id;		/* blade of GRU */
420 	unsigned char		gs_chiplet_id;		/* blade chiplet of GRU */
421 	unsigned char		gs_tgh_local_shift;	/* used to pick TGH for
422 							   local flush */
423 	unsigned char		gs_tgh_first_remote;	/* starting TGH# for
424 							   remote flush */
425 	spinlock_t		gs_asid_lock;		/* lock used for
426 							   assigning asids */
427 	spinlock_t		gs_lock;		/* lock used for
428 							   assigning contexts */
429 
430 	/* -- the following are protected by the gs_asid_lock spinlock ---- */
431 	unsigned int		gs_asid;		/* Next availe ASID */
432 	unsigned int		gs_asid_limit;		/* Limit of available
433 							   ASIDs */
434 	unsigned int		gs_asid_gen;		/* asid generation.
435 							   Inc on wrap */
436 
437 	/* --- the following fields are protected by the gs_lock spinlock --- */
438 	unsigned long		gs_context_map;		/* bitmap to manage
439 							   contexts in use */
440 	unsigned long		gs_cbr_map;		/* bitmap to manage CB
441 							   resources */
442 	unsigned long		gs_dsr_map;		/* bitmap used to manage
443 							   DATA resources */
444 	unsigned int		gs_reserved_cbrs;	/* Number of kernel-
445 							   reserved cbrs */
446 	unsigned int		gs_reserved_dsr_bytes;	/* Bytes of kernel-
447 							   reserved dsrs */
448 	unsigned short		gs_active_contexts;	/* number of contexts
449 							   in use */
450 	struct gru_thread_state	*gs_gts[GRU_NUM_CCH];	/* GTS currently using
451 							   the context */
452 	int			gs_irq[GRU_NUM_TFM];	/* Interrupt irqs */
453 };
454 
455 /*
456  * This structure contains the GRU state for all the GRUs on a blade.
457  */
458 struct gru_blade_state {
459 	void			*kernel_cb;		/* First kernel
460 							   reserved cb */
461 	void			*kernel_dsr;		/* First kernel
462 							   reserved DSR */
463 	struct rw_semaphore	bs_kgts_sema;		/* lock for kgts */
464 	struct gru_thread_state *bs_kgts;		/* GTS for kernel use */
465 
466 	/* ---- the following are used for managing kernel async GRU CBRs --- */
467 	int			bs_async_dsr_bytes;	/* DSRs for async */
468 	int			bs_async_cbrs;		/* CBRs AU for async */
469 	struct completion	*bs_async_wq;
470 
471 	/* ---- the following are protected by the bs_lock spinlock ---- */
472 	spinlock_t		bs_lock;		/* lock used for
473 							   stealing contexts */
474 	int			bs_lru_ctxnum;		/* STEAL - last context
475 							   stolen */
476 	struct gru_state	*bs_lru_gru;		/* STEAL - last gru
477 							   stolen */
478 
479 	struct gru_state	bs_grus[GRU_CHIPLETS_PER_BLADE];
480 };
481 
482 /*-----------------------------------------------------------------------------
483  * Address Primitives
484  */
485 #define get_tfm_for_cpu(g, c)						\
486 	((struct gru_tlb_fault_map *)get_tfm((g)->gs_gru_base_vaddr, (c)))
487 #define get_tfh_by_index(g, i)						\
488 	((struct gru_tlb_fault_handle *)get_tfh((g)->gs_gru_base_vaddr, (i)))
489 #define get_tgh_by_index(g, i)						\
490 	((struct gru_tlb_global_handle *)get_tgh((g)->gs_gru_base_vaddr, (i)))
491 #define get_cbe_by_index(g, i)						\
492 	((struct gru_control_block_extended *)get_cbe((g)->gs_gru_base_vaddr,\
493 			(i)))
494 
495 /*-----------------------------------------------------------------------------
496  * Useful Macros
497  */
498 
499 /* Given a blade# & chiplet#, get a pointer to the GRU */
500 #define get_gru(b, c)		(&gru_base[b]->bs_grus[c])
501 
502 /* Number of bytes to save/restore when unloading/loading GRU contexts */
503 #define DSR_BYTES(dsr)		((dsr) * GRU_DSR_AU_BYTES)
504 #define CBR_BYTES(cbr)		((cbr) * GRU_HANDLE_BYTES * GRU_CBR_AU_SIZE * 2)
505 
506 /* Convert a user CB number to the actual CBRNUM */
507 #define thread_cbr_number(gts, n) ((gts)->ts_cbr_idx[(n) / GRU_CBR_AU_SIZE] \
508 				  * GRU_CBR_AU_SIZE + (n) % GRU_CBR_AU_SIZE)
509 
510 /* Convert a gid to a pointer to the GRU */
511 #define GID_TO_GRU(gid)							\
512 	(gru_base[(gid) / GRU_CHIPLETS_PER_BLADE] ?			\
513 		(&gru_base[(gid) / GRU_CHIPLETS_PER_BLADE]->		\
514 			bs_grus[(gid) % GRU_CHIPLETS_PER_BLADE]) :	\
515 	 NULL)
516 
517 /* Scan all active GRUs in a GRU bitmap */
518 #define for_each_gru_in_bitmap(gid, map)				\
519 	for ((gid) = find_first_bit((map), GRU_MAX_GRUS); (gid) < GRU_MAX_GRUS;\
520 		(gid)++, (gid) = find_next_bit((map), GRU_MAX_GRUS, (gid)))
521 
522 /* Scan all active GRUs on a specific blade */
523 #define for_each_gru_on_blade(gru, nid, i)				\
524 	for ((gru) = gru_base[nid]->bs_grus, (i) = 0;			\
525 			(i) < GRU_CHIPLETS_PER_BLADE;			\
526 			(i)++, (gru)++)
527 
528 /* Scan all GRUs */
529 #define foreach_gid(gid)						\
530 	for ((gid) = 0; (gid) < gru_max_gids; (gid)++)
531 
532 /* Scan all active GTSs on a gru. Note: must hold ss_lock to use this macro. */
533 #define for_each_gts_on_gru(gts, gru, ctxnum)				\
534 	for ((ctxnum) = 0; (ctxnum) < GRU_NUM_CCH; (ctxnum)++)		\
535 		if (((gts) = (gru)->gs_gts[ctxnum]))
536 
537 /* Scan each CBR whose bit is set in a TFM (or copy of) */
538 #define for_each_cbr_in_tfm(i, map)					\
539 	for ((i) = find_first_bit(map, GRU_NUM_CBE);			\
540 			(i) < GRU_NUM_CBE;				\
541 			(i)++, (i) = find_next_bit(map, GRU_NUM_CBE, i))
542 
543 /* Scan each CBR in a CBR bitmap. Note: multiple CBRs in an allocation unit */
544 #define for_each_cbr_in_allocation_map(i, map, k)			\
545 	for ((k) = find_first_bit(map, GRU_CBR_AU); (k) < GRU_CBR_AU;	\
546 			(k) = find_next_bit(map, GRU_CBR_AU, (k) + 1)) 	\
547 		for ((i) = (k)*GRU_CBR_AU_SIZE;				\
548 				(i) < ((k) + 1) * GRU_CBR_AU_SIZE; (i)++)
549 
550 /* Scan each DSR in a DSR bitmap. Note: multiple DSRs in an allocation unit */
551 #define for_each_dsr_in_allocation_map(i, map, k)			\
552 	for ((k) = find_first_bit((const unsigned long *)map, GRU_DSR_AU);\
553 			(k) < GRU_DSR_AU;				\
554 			(k) = find_next_bit((const unsigned long *)map,	\
555 					  GRU_DSR_AU, (k) + 1))		\
556 		for ((i) = (k) * GRU_DSR_AU_CL;				\
557 				(i) < ((k) + 1) * GRU_DSR_AU_CL; (i)++)
558 
559 #define gseg_physical_address(gru, ctxnum)				\
560 		((gru)->gs_gru_base_paddr + ctxnum * GRU_GSEG_STRIDE)
561 #define gseg_virtual_address(gru, ctxnum)				\
562 		((gru)->gs_gru_base_vaddr + ctxnum * GRU_GSEG_STRIDE)
563 
564 /*-----------------------------------------------------------------------------
565  * Lock / Unlock GRU handles
566  * 	Use the "delresp" bit in the handle as a "lock" bit.
567  */
568 
569 /* Lock hierarchy checking enabled only in emulator */
570 
571 /* 0 = lock failed, 1 = locked */
572 static inline int __trylock_handle(void *h)
573 {
574 	return !test_and_set_bit(1, h);
575 }
576 
577 static inline void __lock_handle(void *h)
578 {
579 	while (test_and_set_bit(1, h))
580 		cpu_relax();
581 }
582 
583 static inline void __unlock_handle(void *h)
584 {
585 	clear_bit(1, h);
586 }
587 
588 static inline int trylock_cch_handle(struct gru_context_configuration_handle *cch)
589 {
590 	return __trylock_handle(cch);
591 }
592 
593 static inline void lock_cch_handle(struct gru_context_configuration_handle *cch)
594 {
595 	__lock_handle(cch);
596 }
597 
598 static inline void unlock_cch_handle(struct gru_context_configuration_handle
599 				     *cch)
600 {
601 	__unlock_handle(cch);
602 }
603 
604 static inline void lock_tgh_handle(struct gru_tlb_global_handle *tgh)
605 {
606 	__lock_handle(tgh);
607 }
608 
609 static inline void unlock_tgh_handle(struct gru_tlb_global_handle *tgh)
610 {
611 	__unlock_handle(tgh);
612 }
613 
614 static inline int is_kernel_context(struct gru_thread_state *gts)
615 {
616 	return !gts->ts_mm;
617 }
618 
619 /*
620  * The following are for Nehelem-EX. A more general scheme is needed for
621  * future processors.
622  */
623 #define UV_MAX_INT_CORES		8
624 #define uv_cpu_socket_number(p)		((cpu_physical_id(p) >> 5) & 1)
625 #define uv_cpu_ht_number(p)		(cpu_physical_id(p) & 1)
626 #define uv_cpu_core_number(p)		(((cpu_physical_id(p) >> 2) & 4) |	\
627 					((cpu_physical_id(p) >> 1) & 3))
628 /*-----------------------------------------------------------------------------
629  * Function prototypes & externs
630  */
631 struct gru_unload_context_req;
632 
633 extern const struct vm_operations_struct gru_vm_ops;
634 extern struct device *grudev;
635 
636 extern struct gru_vma_data *gru_alloc_vma_data(struct vm_area_struct *vma,
637 				int tsid);
638 extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct
639 				*vma, int tsid);
640 extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct
641 				*vma, int tsid);
642 extern struct gru_state *gru_assign_gru_context(struct gru_thread_state *gts);
643 extern void gru_load_context(struct gru_thread_state *gts);
644 extern void gru_steal_context(struct gru_thread_state *gts);
645 extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
646 extern int gru_update_cch(struct gru_thread_state *gts);
647 extern void gts_drop(struct gru_thread_state *gts);
648 extern void gru_tgh_flush_init(struct gru_state *gru);
649 extern int gru_kservices_init(void);
650 extern void gru_kservices_exit(void);
651 extern irqreturn_t gru0_intr(int irq, void *dev_id);
652 extern irqreturn_t gru1_intr(int irq, void *dev_id);
653 extern irqreturn_t gru_intr_mblade(int irq, void *dev_id);
654 extern int gru_dump_chiplet_request(unsigned long arg);
655 extern long gru_get_gseg_statistics(unsigned long arg);
656 extern int gru_handle_user_call_os(unsigned long address);
657 extern int gru_user_flush_tlb(unsigned long arg);
658 extern int gru_user_unload_context(unsigned long arg);
659 extern int gru_get_exception_detail(unsigned long arg);
660 extern int gru_set_context_option(unsigned long address);
661 extern void gru_check_context_placement(struct gru_thread_state *gts);
662 extern int gru_cpu_fault_map_id(void);
663 extern struct vm_area_struct *gru_find_vma(unsigned long vaddr);
664 extern void gru_flush_all_tlb(struct gru_state *gru);
665 extern int gru_proc_init(void);
666 extern void gru_proc_exit(void);
667 
668 extern struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
669 		int cbr_au_count, int dsr_au_count,
670 		unsigned char tlb_preload_count, int options, int tsid);
671 extern unsigned long gru_reserve_cb_resources(struct gru_state *gru,
672 		int cbr_au_count, char *cbmap);
673 extern unsigned long gru_reserve_ds_resources(struct gru_state *gru,
674 		int dsr_au_count, char *dsmap);
675 extern int gru_fault(struct vm_area_struct *, struct vm_fault *vmf);
676 extern struct gru_mm_struct *gru_register_mmu_notifier(void);
677 extern void gru_drop_mmu_notifier(struct gru_mm_struct *gms);
678 
679 extern int gru_ktest(unsigned long arg);
680 extern void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start,
681 					unsigned long len);
682 
683 extern unsigned long gru_options;
684 
685 #endif /* __GRUTABLES_H__ */
686