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 <linux/mm_types.h>
151 #include "gru.h"
152 #include "grulib.h"
153 #include "gruhandles.h"
154 
155 extern struct gru_stats_s gru_stats;
156 extern struct gru_blade_state *gru_base[];
157 extern unsigned long gru_start_paddr, gru_end_paddr;
158 extern void *gru_start_vaddr;
159 extern unsigned int gru_max_gids;
160 
161 #define GRU_MAX_BLADES		MAX_NUMNODES
162 #define GRU_MAX_GRUS		(GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE)
163 
164 #define GRU_DRIVER_ID_STR	"SGI GRU Device Driver"
165 #define GRU_DRIVER_VERSION_STR	"0.85"
166 
167 /*
168  * GRU statistics.
169  */
170 struct gru_stats_s {
171 	atomic_long_t vdata_alloc;
172 	atomic_long_t vdata_free;
173 	atomic_long_t gts_alloc;
174 	atomic_long_t gts_free;
175 	atomic_long_t gms_alloc;
176 	atomic_long_t gms_free;
177 	atomic_long_t gts_double_allocate;
178 	atomic_long_t assign_context;
179 	atomic_long_t assign_context_failed;
180 	atomic_long_t free_context;
181 	atomic_long_t load_user_context;
182 	atomic_long_t load_kernel_context;
183 	atomic_long_t lock_kernel_context;
184 	atomic_long_t unlock_kernel_context;
185 	atomic_long_t steal_user_context;
186 	atomic_long_t steal_kernel_context;
187 	atomic_long_t steal_context_failed;
188 	atomic_long_t nopfn;
189 	atomic_long_t asid_new;
190 	atomic_long_t asid_next;
191 	atomic_long_t asid_wrap;
192 	atomic_long_t asid_reuse;
193 	atomic_long_t intr;
194 	atomic_long_t intr_cbr;
195 	atomic_long_t intr_tfh;
196 	atomic_long_t intr_spurious;
197 	atomic_long_t intr_mm_lock_failed;
198 	atomic_long_t call_os;
199 	atomic_long_t call_os_wait_queue;
200 	atomic_long_t user_flush_tlb;
201 	atomic_long_t user_unload_context;
202 	atomic_long_t user_exception;
203 	atomic_long_t set_context_option;
204 	atomic_long_t check_context_retarget_intr;
205 	atomic_long_t check_context_unload;
206 	atomic_long_t tlb_dropin;
207 	atomic_long_t tlb_preload_page;
208 	atomic_long_t tlb_dropin_fail_no_asid;
209 	atomic_long_t tlb_dropin_fail_upm;
210 	atomic_long_t tlb_dropin_fail_invalid;
211 	atomic_long_t tlb_dropin_fail_range_active;
212 	atomic_long_t tlb_dropin_fail_idle;
213 	atomic_long_t tlb_dropin_fail_fmm;
214 	atomic_long_t tlb_dropin_fail_no_exception;
215 	atomic_long_t tfh_stale_on_fault;
216 	atomic_long_t mmu_invalidate_range;
217 	atomic_long_t mmu_invalidate_page;
218 	atomic_long_t flush_tlb;
219 	atomic_long_t flush_tlb_gru;
220 	atomic_long_t flush_tlb_gru_tgh;
221 	atomic_long_t flush_tlb_gru_zero_asid;
222 
223 	atomic_long_t copy_gpa;
224 	atomic_long_t read_gpa;
225 
226 	atomic_long_t mesq_receive;
227 	atomic_long_t mesq_receive_none;
228 	atomic_long_t mesq_send;
229 	atomic_long_t mesq_send_failed;
230 	atomic_long_t mesq_noop;
231 	atomic_long_t mesq_send_unexpected_error;
232 	atomic_long_t mesq_send_lb_overflow;
233 	atomic_long_t mesq_send_qlimit_reached;
234 	atomic_long_t mesq_send_amo_nacked;
235 	atomic_long_t mesq_send_put_nacked;
236 	atomic_long_t mesq_page_overflow;
237 	atomic_long_t mesq_qf_locked;
238 	atomic_long_t mesq_qf_noop_not_full;
239 	atomic_long_t mesq_qf_switch_head_failed;
240 	atomic_long_t mesq_qf_unexpected_error;
241 	atomic_long_t mesq_noop_unexpected_error;
242 	atomic_long_t mesq_noop_lb_overflow;
243 	atomic_long_t mesq_noop_qlimit_reached;
244 	atomic_long_t mesq_noop_amo_nacked;
245 	atomic_long_t mesq_noop_put_nacked;
246 	atomic_long_t mesq_noop_page_overflow;
247 
248 };
249 
250 enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
251 	cchop_deallocate, tfhop_write_only, tfhop_write_restart,
252 	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 			printk(KERN_DEBUG "GRU:%d %s: " fmt, smp_processor_id(), __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 #define VADDR_HI_BIT		64
301 #define GRUREGION(addr)		((addr) >> (VADDR_HI_BIT - 3) & 3)
302 #define GRUASID(asid, addr)	((asid) + GRUREGION(addr))
303 
304 /*------------------------------------------------------------------------------
305  *  File & VMS Tables
306  */
307 
308 struct gru_state;
309 
310 /*
311  * This structure is pointed to from the mmstruct via the notifier pointer.
312  * There is one of these per address space.
313  */
314 struct gru_mm_tracker {				/* pack to reduce size */
315 	unsigned int		mt_asid_gen:24;	/* ASID wrap count */
316 	unsigned int		mt_asid:24;	/* current base ASID for gru */
317 	unsigned short		mt_ctxbitmap:16;/* bitmap of contexts using
318 						   asid */
319 } __attribute__ ((packed));
320 
321 struct gru_mm_struct {
322 	struct mmu_notifier	ms_notifier;
323 	atomic_t		ms_refcnt;
324 	spinlock_t		ms_asid_lock;	/* protects ASID assignment */
325 	atomic_t		ms_range_active;/* num range_invals active */
326 	char			ms_released;
327 	wait_queue_head_t	ms_wait_queue;
328 	DECLARE_BITMAP(ms_asidmap, GRU_MAX_GRUS);
329 	struct gru_mm_tracker	ms_asids[GRU_MAX_GRUS];
330 };
331 
332 /*
333  * One of these structures is allocated when a GSEG is mmaped. The
334  * structure is pointed to by the vma->vm_private_data field in the vma struct.
335  */
336 struct gru_vma_data {
337 	spinlock_t		vd_lock;	/* Serialize access to vma */
338 	struct list_head	vd_head;	/* head of linked list of gts */
339 	long			vd_user_options;/* misc user option flags */
340 	int			vd_cbr_au_count;
341 	int			vd_dsr_au_count;
342 	unsigned char		vd_tlb_preload_count;
343 };
344 
345 /*
346  * One of these is allocated for each thread accessing a mmaped GRU. A linked
347  * list of these structure is hung off the struct gru_vma_data in the mm_struct.
348  */
349 struct gru_thread_state {
350 	struct list_head	ts_next;	/* list - head at vma-private */
351 	struct mutex		ts_ctxlock;	/* load/unload CTX lock */
352 	struct mm_struct	*ts_mm;		/* mm currently mapped to
353 						   context */
354 	struct vm_area_struct	*ts_vma;	/* vma of GRU context */
355 	struct gru_state	*ts_gru;	/* GRU where the context is
356 						   loaded */
357 	struct gru_mm_struct	*ts_gms;	/* asid & ioproc struct */
358 	unsigned char		ts_tlb_preload_count; /* TLB preload pages */
359 	unsigned long		ts_cbr_map;	/* map of allocated CBRs */
360 	unsigned long		ts_dsr_map;	/* map of allocated DATA
361 						   resources */
362 	unsigned long		ts_steal_jiffies;/* jiffies when context last
363 						    stolen */
364 	long			ts_user_options;/* misc user option flags */
365 	pid_t			ts_tgid_owner;	/* task that is using the
366 						   context - for migration */
367 	short			ts_user_blade_id;/* user selected blade */
368 	char			ts_user_chiplet_id;/* user selected chiplet */
369 	unsigned short		ts_sizeavail;	/* Pagesizes in use */
370 	int			ts_tsid;	/* thread that owns the
371 						   structure */
372 	int			ts_tlb_int_select;/* target cpu if interrupts
373 						     enabled */
374 	int			ts_ctxnum;	/* context number where the
375 						   context is loaded */
376 	atomic_t		ts_refcnt;	/* reference count GTS */
377 	unsigned char		ts_dsr_au_count;/* Number of DSR resources
378 						   required for contest */
379 	unsigned char		ts_cbr_au_count;/* Number of CBR resources
380 						   required for contest */
381 	char			ts_cch_req_slice;/* CCH packet slice */
382 	char			ts_blade;	/* If >= 0, migrate context if
383 						   ref from different blade */
384 	char			ts_force_cch_reload;
385 	char			ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each
386 							  allocated CB */
387 	int			ts_data_valid;	/* Indicates if ts_gdata has
388 						   valid data */
389 	struct gru_gseg_statistics ustats;	/* User statistics */
390 	unsigned long		ts_gdata[0];	/* save area for GRU data (CB,
391 						   DS, CBE) */
392 };
393 
394 /*
395  * Threaded programs actually allocate an array of GSEGs when a context is
396  * created. Each thread uses a separate GSEG. TSID is the index into the GSEG
397  * array.
398  */
399 #define TSID(a, v)		(((a) - (v)->vm_start) / GRU_GSEG_PAGESIZE)
400 #define UGRUADDR(gts)		((gts)->ts_vma->vm_start +		\
401 					(gts)->ts_tsid * GRU_GSEG_PAGESIZE)
402 
403 #define NULLCTX			(-1)	/* if context not loaded into GRU */
404 
405 /*-----------------------------------------------------------------------------
406  *  GRU State Tables
407  */
408 
409 /*
410  * One of these exists for each GRU chiplet.
411  */
412 struct gru_state {
413 	struct gru_blade_state	*gs_blade;		/* GRU state for entire
414 							   blade */
415 	unsigned long		gs_gru_base_paddr;	/* Physical address of
416 							   gru segments (64) */
417 	void			*gs_gru_base_vaddr;	/* Virtual address of
418 							   gru segments (64) */
419 	unsigned short		gs_gid;			/* unique GRU number */
420 	unsigned short		gs_blade_id;		/* blade of GRU */
421 	unsigned char		gs_chiplet_id;		/* blade chiplet of GRU */
422 	unsigned char		gs_tgh_local_shift;	/* used to pick TGH for
423 							   local flush */
424 	unsigned char		gs_tgh_first_remote;	/* starting TGH# for
425 							   remote flush */
426 	spinlock_t		gs_asid_lock;		/* lock used for
427 							   assigning asids */
428 	spinlock_t		gs_lock;		/* lock used for
429 							   assigning contexts */
430 
431 	/* -- the following are protected by the gs_asid_lock spinlock ---- */
432 	unsigned int		gs_asid;		/* Next availe ASID */
433 	unsigned int		gs_asid_limit;		/* Limit of available
434 							   ASIDs */
435 	unsigned int		gs_asid_gen;		/* asid generation.
436 							   Inc on wrap */
437 
438 	/* --- the following fields are protected by the gs_lock spinlock --- */
439 	unsigned long		gs_context_map;		/* bitmap to manage
440 							   contexts in use */
441 	unsigned long		gs_cbr_map;		/* bitmap to manage CB
442 							   resources */
443 	unsigned long		gs_dsr_map;		/* bitmap used to manage
444 							   DATA resources */
445 	unsigned int		gs_reserved_cbrs;	/* Number of kernel-
446 							   reserved cbrs */
447 	unsigned int		gs_reserved_dsr_bytes;	/* Bytes of kernel-
448 							   reserved dsrs */
449 	unsigned short		gs_active_contexts;	/* number of contexts
450 							   in use */
451 	struct gru_thread_state	*gs_gts[GRU_NUM_CCH];	/* GTS currently using
452 							   the context */
453 	int			gs_irq[GRU_NUM_TFM];	/* Interrupt irqs */
454 };
455 
456 /*
457  * This structure contains the GRU state for all the GRUs on a blade.
458  */
459 struct gru_blade_state {
460 	void			*kernel_cb;		/* First kernel
461 							   reserved cb */
462 	void			*kernel_dsr;		/* First kernel
463 							   reserved DSR */
464 	struct rw_semaphore	bs_kgts_sema;		/* lock for kgts */
465 	struct gru_thread_state *bs_kgts;		/* GTS for kernel use */
466 
467 	/* ---- the following are used for managing kernel async GRU CBRs --- */
468 	int			bs_async_dsr_bytes;	/* DSRs for async */
469 	int			bs_async_cbrs;		/* CBRs AU for async */
470 	struct completion	*bs_async_wq;
471 
472 	/* ---- the following are protected by the bs_lock spinlock ---- */
473 	spinlock_t		bs_lock;		/* lock used for
474 							   stealing contexts */
475 	int			bs_lru_ctxnum;		/* STEAL - last context
476 							   stolen */
477 	struct gru_state	*bs_lru_gru;		/* STEAL - last gru
478 							   stolen */
479 
480 	struct gru_state	bs_grus[GRU_CHIPLETS_PER_BLADE];
481 };
482 
483 /*-----------------------------------------------------------------------------
484  * Address Primitives
485  */
486 #define get_tfm_for_cpu(g, c)						\
487 	((struct gru_tlb_fault_map *)get_tfm((g)->gs_gru_base_vaddr, (c)))
488 #define get_tfh_by_index(g, i)						\
489 	((struct gru_tlb_fault_handle *)get_tfh((g)->gs_gru_base_vaddr, (i)))
490 #define get_tgh_by_index(g, i)						\
491 	((struct gru_tlb_global_handle *)get_tgh((g)->gs_gru_base_vaddr, (i)))
492 #define get_cbe_by_index(g, i)						\
493 	((struct gru_control_block_extended *)get_cbe((g)->gs_gru_base_vaddr,\
494 			(i)))
495 
496 /*-----------------------------------------------------------------------------
497  * Useful Macros
498  */
499 
500 /* Given a blade# & chiplet#, get a pointer to the GRU */
501 #define get_gru(b, c)		(&gru_base[b]->bs_grus[c])
502 
503 /* Number of bytes to save/restore when unloading/loading GRU contexts */
504 #define DSR_BYTES(dsr)		((dsr) * GRU_DSR_AU_BYTES)
505 #define CBR_BYTES(cbr)		((cbr) * GRU_HANDLE_BYTES * GRU_CBR_AU_SIZE * 2)
506 
507 /* Convert a user CB number to the actual CBRNUM */
508 #define thread_cbr_number(gts, n) ((gts)->ts_cbr_idx[(n) / GRU_CBR_AU_SIZE] \
509 				  * GRU_CBR_AU_SIZE + (n) % GRU_CBR_AU_SIZE)
510 
511 /* Convert a gid to a pointer to the GRU */
512 #define GID_TO_GRU(gid)							\
513 	(gru_base[(gid) / GRU_CHIPLETS_PER_BLADE] ?			\
514 		(&gru_base[(gid) / GRU_CHIPLETS_PER_BLADE]->		\
515 			bs_grus[(gid) % GRU_CHIPLETS_PER_BLADE]) :	\
516 	 NULL)
517 
518 /* Scan all active GRUs in a GRU bitmap */
519 #define for_each_gru_in_bitmap(gid, map)				\
520 	for_each_set_bit((gid), (map), GRU_MAX_GRUS)
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_each_set_bit((i), (map), GRU_NUM_CBE)
540 
541 /* Scan each CBR in a CBR bitmap. Note: multiple CBRs in an allocation unit */
542 #define for_each_cbr_in_allocation_map(i, map, k)			\
543 	for_each_set_bit((k), (map), GRU_CBR_AU)			\
544 		for ((i) = (k)*GRU_CBR_AU_SIZE;				\
545 				(i) < ((k) + 1) * GRU_CBR_AU_SIZE; (i)++)
546 
547 /* Scan each DSR in a DSR bitmap. Note: multiple DSRs in an allocation unit */
548 #define for_each_dsr_in_allocation_map(i, map, k)			\
549 	for_each_set_bit((k), (const unsigned long *)(map), GRU_DSR_AU)	\
550 		for ((i) = (k) * GRU_DSR_AU_CL;				\
551 				(i) < ((k) + 1) * GRU_DSR_AU_CL; (i)++)
552 
553 #define gseg_physical_address(gru, ctxnum)				\
554 		((gru)->gs_gru_base_paddr + ctxnum * GRU_GSEG_STRIDE)
555 #define gseg_virtual_address(gru, ctxnum)				\
556 		((gru)->gs_gru_base_vaddr + ctxnum * GRU_GSEG_STRIDE)
557 
558 /*-----------------------------------------------------------------------------
559  * Lock / Unlock GRU handles
560  * 	Use the "delresp" bit in the handle as a "lock" bit.
561  */
562 
563 /* Lock hierarchy checking enabled only in emulator */
564 
565 /* 0 = lock failed, 1 = locked */
__trylock_handle(void * h)566 static inline int __trylock_handle(void *h)
567 {
568 	return !test_and_set_bit(1, h);
569 }
570 
__lock_handle(void * h)571 static inline void __lock_handle(void *h)
572 {
573 	while (test_and_set_bit(1, h))
574 		cpu_relax();
575 }
576 
__unlock_handle(void * h)577 static inline void __unlock_handle(void *h)
578 {
579 	clear_bit(1, h);
580 }
581 
trylock_cch_handle(struct gru_context_configuration_handle * cch)582 static inline int trylock_cch_handle(struct gru_context_configuration_handle *cch)
583 {
584 	return __trylock_handle(cch);
585 }
586 
lock_cch_handle(struct gru_context_configuration_handle * cch)587 static inline void lock_cch_handle(struct gru_context_configuration_handle *cch)
588 {
589 	__lock_handle(cch);
590 }
591 
unlock_cch_handle(struct gru_context_configuration_handle * cch)592 static inline void unlock_cch_handle(struct gru_context_configuration_handle
593 				     *cch)
594 {
595 	__unlock_handle(cch);
596 }
597 
lock_tgh_handle(struct gru_tlb_global_handle * tgh)598 static inline void lock_tgh_handle(struct gru_tlb_global_handle *tgh)
599 {
600 	__lock_handle(tgh);
601 }
602 
unlock_tgh_handle(struct gru_tlb_global_handle * tgh)603 static inline void unlock_tgh_handle(struct gru_tlb_global_handle *tgh)
604 {
605 	__unlock_handle(tgh);
606 }
607 
is_kernel_context(struct gru_thread_state * gts)608 static inline int is_kernel_context(struct gru_thread_state *gts)
609 {
610 	return !gts->ts_mm;
611 }
612 
613 /*
614  * The following are for Nehelem-EX. A more general scheme is needed for
615  * future processors.
616  */
617 #define UV_MAX_INT_CORES		8
618 #define uv_cpu_socket_number(p)		((cpu_physical_id(p) >> 5) & 1)
619 #define uv_cpu_ht_number(p)		(cpu_physical_id(p) & 1)
620 #define uv_cpu_core_number(p)		(((cpu_physical_id(p) >> 2) & 4) |	\
621 					((cpu_physical_id(p) >> 1) & 3))
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 extern void gru_load_context(struct gru_thread_state *gts);
638 extern void gru_steal_context(struct gru_thread_state *gts);
639 extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
640 extern int gru_update_cch(struct gru_thread_state *gts);
641 extern void gts_drop(struct gru_thread_state *gts);
642 extern void gru_tgh_flush_init(struct gru_state *gru);
643 extern int gru_kservices_init(void);
644 extern void gru_kservices_exit(void);
645 extern irqreturn_t gru0_intr(int irq, void *dev_id);
646 extern irqreturn_t gru1_intr(int irq, void *dev_id);
647 extern irqreturn_t gru_intr_mblade(int irq, void *dev_id);
648 extern int gru_dump_chiplet_request(unsigned long arg);
649 extern long gru_get_gseg_statistics(unsigned long arg);
650 extern int gru_handle_user_call_os(unsigned long address);
651 extern int gru_user_flush_tlb(unsigned long arg);
652 extern int gru_user_unload_context(unsigned long arg);
653 extern int gru_get_exception_detail(unsigned long arg);
654 extern int gru_set_context_option(unsigned long address);
655 extern void gru_check_context_placement(struct gru_thread_state *gts);
656 extern int gru_cpu_fault_map_id(void);
657 extern struct vm_area_struct *gru_find_vma(unsigned long vaddr);
658 extern void gru_flush_all_tlb(struct gru_state *gru);
659 extern int gru_proc_init(void);
660 extern void gru_proc_exit(void);
661 
662 extern struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
663 		int cbr_au_count, int dsr_au_count,
664 		unsigned char tlb_preload_count, int options, int tsid);
665 extern unsigned long gru_reserve_cb_resources(struct gru_state *gru,
666 		int cbr_au_count, char *cbmap);
667 extern unsigned long gru_reserve_ds_resources(struct gru_state *gru,
668 		int dsr_au_count, char *dsmap);
669 extern vm_fault_t gru_fault(struct vm_fault *vmf);
670 extern struct gru_mm_struct *gru_register_mmu_notifier(void);
671 extern void gru_drop_mmu_notifier(struct gru_mm_struct *gms);
672 
673 extern int gru_ktest(unsigned long arg);
674 extern void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start,
675 					unsigned long len);
676 
677 extern unsigned long gru_options;
678 
679 #endif /* __GRUTABLES_H__ */
680