2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * SGI UV Broadcast Assist Unit definitions
8 * Copyright (C) 2008 Silicon Graphics, Inc. All rights reserved.
11 #ifndef _ASM_X86_UV_UV_BAU_H
12 #define _ASM_X86_UV_UV_BAU_H
14 #include <linux/bitmap.h>
18 * Broadcast Assist Unit messaging structures
20 * Selective Broadcast activations are induced by software action
21 * specifying a particular 8-descriptor "set" via a 6-bit index written
23 * Thus there are 64 unique 512-byte sets of SB descriptors - one set for
24 * each 6-bit index value. These descriptor sets are mapped in sequence
25 * starting with set 0 located at the address specified in the
26 * BAU_SB_DESCRIPTOR_BASE register, set 1 is located at BASE + 512,
27 * set 2 is at BASE + 2*512, set 3 at BASE + 3*512, and so on.
29 * We will use 31 sets, one for sending BAU messages from each of the 32
32 * TLB shootdown will use the first of the 8 descriptors of each set.
33 * Each of the descriptors is 64 bytes in size (8*64 = 512 bytes in a set).
36 #define UV_ITEMS_PER_DESCRIPTOR 8
37 /* the 'throttle' to prevent the hardware stay-busy bug */
38 #define MAX_BAU_CONCURRENT 3
39 #define UV_CPUS_PER_ACT_STATUS 32
40 #define UV_ACT_STATUS_MASK 0x3
41 #define UV_ACT_STATUS_SIZE 2
42 #define UV_ADP_SIZE 32
43 #define UV_DISTRIBUTION_SIZE 256
44 #define UV_SW_ACK_NPENDING 8
45 #define UV_NET_ENDPOINT_INTD 0x38
46 #define UV_DESC_BASE_PNODE_SHIFT 49
47 #define UV_PAYLOADQ_PNODE_SHIFT 49
48 #define UV_PTC_BASENAME "sgi_uv/ptc_statistics"
49 #define UV_BAU_BASENAME "sgi_uv/bau_tunables"
50 #define UV_BAU_TUNABLES_DIR "sgi_uv"
51 #define UV_BAU_TUNABLES_FILE "bau_tunables"
52 #define WHITESPACE " \t\n"
53 #define uv_physnodeaddr(x) ((__pa((unsigned long)(x)) & uv_mmask))
54 #define UV_ENABLE_INTD_SOFT_ACK_MODE_SHIFT 15
55 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD_SHIFT 16
56 #define UV_INTD_SOFT_ACK_TIMEOUT_PERIOD 0x0000000009UL
57 /* [19:16] SOFT_ACK timeout period 19: 1 is urgency 7 17:16 1 is multiplier */
58 #define BAU_MISC_CONTROL_MULT_MASK 3
60 #define UVH_AGING_PRESCALE_SEL 0x000000b000UL
61 /* [30:28] URGENCY_7 an index into a table of times */
62 #define BAU_URGENCY_7_SHIFT 28
63 #define BAU_URGENCY_7_MASK 7
65 #define UVH_TRANSACTION_TIMEOUT 0x000000b200UL
66 /* [45:40] BAU - BAU transaction timeout select - a multiplier */
67 #define BAU_TRANS_SHIFT 40
68 #define BAU_TRANS_MASK 0x3f
71 * bits in UVH_LB_BAU_SB_ACTIVATION_STATUS_0/1
73 #define DESC_STATUS_IDLE 0
74 #define DESC_STATUS_ACTIVE 1
75 #define DESC_STATUS_DESTINATION_TIMEOUT 2
76 #define DESC_STATUS_SOURCE_TIMEOUT 3
79 * delay for 'plugged' timeout retries, in microseconds
81 #define PLUGGED_DELAY 10
84 * threshholds at which to use IPI to free resources
86 /* after this # consecutive 'plugged' timeouts, use IPI to release resources */
87 #define PLUGSB4RESET 100
88 /* after this many consecutive timeouts, use IPI to release resources */
89 #define TIMEOUTSB4RESET 1
90 /* at this number uses of IPI to release resources, giveup the request */
91 #define IPI_RESET_LIMIT 1
92 /* after this # consecutive successes, bump up the throttle if it was lowered */
93 #define COMPLETE_THRESHOLD 5
96 * number of entries in the destination side payload queue
98 #define DEST_Q_SIZE 20
100 * number of destination side software ack resources
102 #define DEST_NUM_RESOURCES 8
103 #define MAX_CPUS_PER_NODE 32
105 * completion statuses for sending a TLB flush message
107 #define FLUSH_RETRY_PLUGGED 1
108 #define FLUSH_RETRY_TIMEOUT 2
109 #define FLUSH_GIVEUP 3
110 #define FLUSH_COMPLETE 4
113 * tuning the action when the numalink network is extremely delayed
115 #define CONGESTED_RESPONSE_US 1000 /* 'long' response time, in microseconds */
116 #define CONGESTED_REPS 10 /* long delays averaged over this many broadcasts */
117 #define CONGESTED_PERIOD 30 /* time for the bau to be disabled, in seconds */
120 * Distribution: 32 bytes (256 bits) (bytes 0-0x1f of descriptor)
121 * If the 'multilevel' flag in the header portion of the descriptor
122 * has been set to 0, then endpoint multi-unicast mode is selected.
123 * The distribution specification (32 bytes) is interpreted as a 256-bit
124 * distribution vector. Adjacent bits correspond to consecutive even numbered
125 * nodeIDs. The result of adding the index of a given bit to the 15-bit
126 * 'base_dest_nodeid' field of the header corresponds to the
127 * destination nodeID associated with that specified bit.
129 struct bau_target_uvhubmask
{
130 unsigned long bits
[BITS_TO_LONGS(UV_DISTRIBUTION_SIZE
)];
134 * mask of cpu's on a uvhub
135 * (during initialization we need to check that unsigned long has
136 * enough bits for max. cpu's per uvhub)
138 struct bau_local_cpumask
{
143 * Payload: 16 bytes (128 bits) (bytes 0x20-0x2f of descriptor)
144 * only 12 bytes (96 bits) of the payload area are usable.
145 * An additional 3 bytes (bits 27:4) of the header address are carried
146 * to the next bytes of the destination payload queue.
147 * And an additional 2 bytes of the header Suppl_A field are also
148 * carried to the destination payload queue.
149 * But the first byte of the Suppl_A becomes bits 127:120 (the 16th byte)
150 * of the destination payload queue, which is written by the hardware
151 * with the s/w ack resource bit vector.
152 * [ effective message contents (16 bytes (128 bits) maximum), not counting
153 * the s/w ack bit vector ]
157 * The payload is software-defined for INTD transactions
159 struct bau_msg_payload
{
160 unsigned long address
; /* signifies a page or all TLB's
163 unsigned short sending_cpu
; /* filled in by sender */
165 unsigned short acknowledge_count
;/* filled in by destination */
167 unsigned int reserved1
:32; /* not usable */
172 * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
173 * see table 4.2.3.0.1 in broacast_assist spec.
175 struct bau_msg_header
{
176 unsigned int dest_subnodeid
:6; /* must be 0x10, for the LB */
178 unsigned int base_dest_nodeid
:15; /* nasid (pnode<<1) of */
179 /* bits 20:6 */ /* first bit in uvhub map */
180 unsigned int command
:8; /* message type */
182 /* 0x38: SN3net EndPoint Message */
183 unsigned int rsvd_1
:3; /* must be zero */
185 /* int will align on 32 bits */
186 unsigned int rsvd_2
:9; /* must be zero */
188 /* Suppl_A is 56-41 */
189 unsigned int sequence
:16;/* message sequence number */
190 /* bits 56:41 */ /* becomes bytes 16-17 of msg */
191 /* Address field (96:57) is never used as an
192 address (these are address bits 42:3) */
194 unsigned int rsvd_3
:1; /* must be zero */
196 /* address bits 27:4 are payload */
197 /* these next 24 (58-81) bits become bytes 12-14 of msg */
199 /* bits 65:58 land in byte 12 */
200 unsigned int replied_to
:1;/* sent as 0 by the source to byte 12 */
202 unsigned int msg_type
:3; /* software type of the message*/
204 unsigned int canceled
:1; /* message canceled, resource to be freed*/
206 unsigned int payload_1a
:1;/* not currently used */
208 unsigned int payload_1b
:2;/* not currently used */
211 /* bits 73:66 land in byte 13 */
212 unsigned int payload_1ca
:6;/* not currently used */
214 unsigned int payload_1c
:2;/* not currently used */
217 /* bits 81:74 land in byte 14 */
218 unsigned int payload_1d
:6;/* not currently used */
220 unsigned int payload_1e
:2;/* not currently used */
223 unsigned int rsvd_4
:7; /* must be zero */
225 unsigned int sw_ack_flag
:1;/* software acknowledge flag */
227 /* INTD trasactions at destination are to
228 wait for software acknowledge */
229 unsigned int rsvd_5
:6; /* must be zero */
231 unsigned int rsvd_6
:5; /* must be zero */
233 unsigned int int_both
:1;/* if 1, interrupt both sockets on the uvhub */
235 unsigned int fairness
:3;/* usually zero */
237 unsigned int multilevel
:1; /* multi-level multicast format */
239 /* 0 for TLB: endpoint multi-unicast messages */
240 unsigned int chaining
:1;/* next descriptor is part of this activation*/
242 unsigned int rsvd_7
:21; /* must be zero */
248 #define MSG_REGULAR 1
252 * The activation descriptor:
253 * The format of the message to send, plus all accompanying control
257 struct bau_target_uvhubmask distribution
;
259 * message template, consisting of header and payload:
261 struct bau_msg_header header
;
262 struct bau_msg_payload payload
;
265 * -payload-- ---------header------
266 * bytes 0-11 bits 41-56 bits 58-81
269 * A/B/C are moved to:
271 * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
272 * ------------payload queue-----------
276 * The payload queue on the destination side is an array of these.
277 * With BAU_MISC_CONTROL set for software acknowledge mode, the messages
278 * are 32 bytes (2 micropackets) (256 bits) in length, but contain only 17
279 * bytes of usable data, including the sw ack vector in byte 15 (bits 127:120)
280 * (12 bytes come from bau_msg_payload, 3 from payload_1, 2 from
281 * sw_ack_vector and payload_2)
282 * "Enabling Software Acknowledgment mode (see Section 4.3.3 Software
283 * Acknowledge Processing) also selects 32 byte (17 bytes usable) payload
286 struct bau_payload_queue_entry
{
287 unsigned long address
; /* signifies a page or all TLB's
289 /* 64 bits, bytes 0-7 */
291 unsigned short sending_cpu
; /* cpu that sent the message */
292 /* 16 bits, bytes 8-9 */
294 unsigned short acknowledge_count
; /* filled in by destination */
295 /* 16 bits, bytes 10-11 */
297 /* these next 3 bytes come from bits 58-81 of the message header */
298 unsigned short replied_to
:1; /* sent as 0 by the source */
299 unsigned short msg_type
:3; /* software message type */
300 unsigned short canceled
:1; /* sent as 0 by the source */
301 unsigned short unused1
:3; /* not currently using */
304 unsigned char unused2a
; /* not currently using */
306 unsigned char unused2
; /* not currently using */
309 unsigned char sw_ack_vector
; /* filled in by the hardware */
310 /* byte 15 (bits 127:120) */
312 unsigned short sequence
; /* message sequence number */
314 unsigned char unused4
[2]; /* not currently using bytes 18-19 */
317 int number_of_cpus
; /* filled in at destination */
318 /* 32 bits, bytes 20-23 (aligned) */
320 unsigned char unused5
[8]; /* not using */
325 struct bau_payload_queue_entry
*msg
;
328 struct bau_payload_queue_entry
*va_queue_first
;
329 struct bau_payload_queue_entry
*va_queue_last
;
337 * This structure is allocated per_cpu for UV TLB shootdown statistics.
340 /* sender statistics */
341 unsigned long s_giveup
; /* number of fall backs to IPI-style flushes */
342 unsigned long s_requestor
; /* number of shootdown requests */
343 unsigned long s_stimeout
; /* source side timeouts */
344 unsigned long s_dtimeout
; /* destination side timeouts */
345 unsigned long s_time
; /* time spent in sending side */
346 unsigned long s_retriesok
; /* successful retries */
347 unsigned long s_ntargcpu
; /* total number of cpu's targeted */
348 unsigned long s_ntargself
; /* times the sending cpu was targeted */
349 unsigned long s_ntarglocals
; /* targets of cpus on the local blade */
350 unsigned long s_ntargremotes
; /* targets of cpus on remote blades */
351 unsigned long s_ntarglocaluvhub
; /* targets of the local hub */
352 unsigned long s_ntargremoteuvhub
; /* remotes hubs targeted */
353 unsigned long s_ntarguvhub
; /* total number of uvhubs targeted */
354 unsigned long s_ntarguvhub16
; /* number of times target hubs >= 16*/
355 unsigned long s_ntarguvhub8
; /* number of times target hubs >= 8 */
356 unsigned long s_ntarguvhub4
; /* number of times target hubs >= 4 */
357 unsigned long s_ntarguvhub2
; /* number of times target hubs >= 2 */
358 unsigned long s_ntarguvhub1
; /* number of times target hubs == 1 */
359 unsigned long s_resets_plug
; /* ipi-style resets from plug state */
360 unsigned long s_resets_timeout
; /* ipi-style resets from timeouts */
361 unsigned long s_busy
; /* status stayed busy past s/w timer */
362 unsigned long s_throttles
; /* waits in throttle */
363 unsigned long s_retry_messages
; /* retry broadcasts */
364 unsigned long s_bau_reenabled
; /* for bau enable/disable */
365 unsigned long s_bau_disabled
; /* for bau enable/disable */
366 /* destination statistics */
367 unsigned long d_alltlb
; /* times all tlb's on this cpu were flushed */
368 unsigned long d_onetlb
; /* times just one tlb on this cpu was flushed */
369 unsigned long d_multmsg
; /* interrupts with multiple messages */
370 unsigned long d_nomsg
; /* interrupts with no message */
371 unsigned long d_time
; /* time spent on destination side */
372 unsigned long d_requestee
; /* number of messages processed */
373 unsigned long d_retries
; /* number of retry messages processed */
374 unsigned long d_canceled
; /* number of messages canceled by retries */
375 unsigned long d_nocanceled
; /* retries that found nothing to cancel */
376 unsigned long d_resets
; /* number of ipi-style requests processed */
377 unsigned long d_rcanceled
; /* number of messages canceled by resets */
381 * one per-cpu; to locate the software tables
384 struct bau_desc
*descriptor_base
;
385 struct bau_payload_queue_entry
*va_queue_first
;
386 struct bau_payload_queue_entry
*va_queue_last
;
387 struct bau_payload_queue_entry
*bau_msg_head
;
388 struct bau_control
*uvhub_master
;
389 struct bau_control
*socket_master
;
390 struct ptc_stats
*statp
;
391 unsigned long timeout_interval
;
392 unsigned long set_bau_on_time
;
393 atomic_t active_descriptor_count
;
403 short cpus_in_socket
;
405 unsigned short message_number
;
406 unsigned short uvhub_quiesce
;
407 short socket_acknowledge_count
[DEST_Q_SIZE
];
408 cycles_t send_message
;
409 spinlock_t uvhub_lock
;
410 spinlock_t queue_lock
;
412 int max_bau_concurrent
;
413 int max_bau_concurrent_constant
;
418 int complete_threshold
;
419 int congested_response_us
;
421 int congested_period
;
422 cycles_t period_time
;
423 long period_requests
;
426 static inline int bau_uvhub_isset(int uvhub
, struct bau_target_uvhubmask
*dstp
)
428 return constant_test_bit(uvhub
, &dstp
->bits
[0]);
430 static inline void bau_uvhub_set(int uvhub
, struct bau_target_uvhubmask
*dstp
)
432 __set_bit(uvhub
, &dstp
->bits
[0]);
434 static inline void bau_uvhubs_clear(struct bau_target_uvhubmask
*dstp
,
437 bitmap_zero(&dstp
->bits
[0], nbits
);
439 static inline int bau_uvhub_weight(struct bau_target_uvhubmask
*dstp
)
441 return bitmap_weight((unsigned long *)&dstp
->bits
[0],
442 UV_DISTRIBUTION_SIZE
);
445 static inline void bau_cpubits_clear(struct bau_local_cpumask
*dstp
, int nbits
)
447 bitmap_zero(&dstp
->bits
, nbits
);
450 #define cpubit_isset(cpu, bau_local_cpumask) \
451 test_bit((cpu), (bau_local_cpumask).bits)
453 extern void uv_bau_message_intr1(void);
454 extern void uv_bau_timeout_intr1(void);
456 struct atomic_short
{
461 * atomic_read_short - read a short atomic variable
462 * @v: pointer of type atomic_short
464 * Atomically reads the value of @v.
466 static inline int atomic_read_short(const struct atomic_short
*v
)
472 * atomic_add_short_return - add and return a short int
473 * @i: short value to add
474 * @v: pointer of type atomic_short
476 * Atomically adds @i to @v and returns @i + @v
478 static inline int atomic_add_short_return(short i
, struct atomic_short
*v
)
481 asm volatile(LOCK_PREFIX
"xaddw %0, %1"
482 : "+r" (i
), "+m" (v
->counter
)
487 #endif /* _ASM_X86_UV_UV_BAU_H */