blob: 66553bdb0a4791af3e31dd6207703d445cef6dc6 [file] [log] [blame]
Zou Nan hai8187a2b2010-05-21 09:08:55 +08001#ifndef _INTEL_RINGBUFFER_H_
2#define _INTEL_RINGBUFFER_H_
3
Brad Volkin44e895a2014-05-10 14:10:43 -07004#include <linux/hashtable.h>
Chris Wilson06fbca72015-04-07 16:20:36 +01005#include "i915_gem_batch_pool.h"
Chris Wilsondcff85c2016-08-05 10:14:11 +01006#include "i915_gem_request.h"
Brad Volkin44e895a2014-05-10 14:10:43 -07007
8#define I915_CMD_HASH_ORDER 9
9
Oscar Mateo47122742014-07-24 17:04:28 +010010/* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
11 * but keeps the logic simple. Indeed, the whole purpose of this macro is just
12 * to give some inclination as to some of the magic values used in the various
13 * workarounds!
14 */
15#define CACHELINE_BYTES 64
Arun Siluvery17ee9502015-06-19 19:07:01 +010016#define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(uint32_t))
Oscar Mateo47122742014-07-24 17:04:28 +010017
Ville Syrjälä633cf8f2012-12-03 18:43:32 +020018/*
19 * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
20 * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
21 * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
22 *
23 * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
24 * cacheline, the Head Pointer must not be greater than the Tail
25 * Pointer."
26 */
27#define I915_RING_FREE_SPACE 64
28
Chris Wilson57e88532016-08-15 10:48:57 +010029struct intel_hw_status_page {
30 struct i915_vma *vma;
31 u32 *page_addr;
32 u32 ggtt_offset;
Zou Nan hai8187a2b2010-05-21 09:08:55 +080033};
34
Dave Gordonbbdc070a2016-07-20 18:16:05 +010035#define I915_READ_TAIL(engine) I915_READ(RING_TAIL((engine)->mmio_base))
36#define I915_WRITE_TAIL(engine, val) I915_WRITE(RING_TAIL((engine)->mmio_base), val)
Zou Nan haicae58522010-11-09 17:17:32 +080037
Dave Gordonbbdc070a2016-07-20 18:16:05 +010038#define I915_READ_START(engine) I915_READ(RING_START((engine)->mmio_base))
39#define I915_WRITE_START(engine, val) I915_WRITE(RING_START((engine)->mmio_base), val)
Zou Nan haicae58522010-11-09 17:17:32 +080040
Dave Gordonbbdc070a2016-07-20 18:16:05 +010041#define I915_READ_HEAD(engine) I915_READ(RING_HEAD((engine)->mmio_base))
42#define I915_WRITE_HEAD(engine, val) I915_WRITE(RING_HEAD((engine)->mmio_base), val)
Zou Nan haicae58522010-11-09 17:17:32 +080043
Dave Gordonbbdc070a2016-07-20 18:16:05 +010044#define I915_READ_CTL(engine) I915_READ(RING_CTL((engine)->mmio_base))
45#define I915_WRITE_CTL(engine, val) I915_WRITE(RING_CTL((engine)->mmio_base), val)
Zou Nan haicae58522010-11-09 17:17:32 +080046
Dave Gordonbbdc070a2016-07-20 18:16:05 +010047#define I915_READ_IMR(engine) I915_READ(RING_IMR((engine)->mmio_base))
48#define I915_WRITE_IMR(engine, val) I915_WRITE(RING_IMR((engine)->mmio_base), val)
Daniel Vetter870e86d2010-08-02 16:29:44 +020049
Dave Gordonbbdc070a2016-07-20 18:16:05 +010050#define I915_READ_MODE(engine) I915_READ(RING_MI_MODE((engine)->mmio_base))
51#define I915_WRITE_MODE(engine, val) I915_WRITE(RING_MI_MODE((engine)->mmio_base), val)
Naresh Kumar Kachhie9fea572014-03-12 16:39:41 +053052
Ben Widawsky3e789982014-06-30 09:53:37 -070053/* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
54 * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
55 */
Chris Wilson8c12672e2016-04-07 07:29:14 +010056#define gen8_semaphore_seqno_size sizeof(uint64_t)
57#define GEN8_SEMAPHORE_OFFSET(__from, __to) \
58 (((__from) * I915_NUM_ENGINES + (__to)) * gen8_semaphore_seqno_size)
Ben Widawsky3e789982014-06-30 09:53:37 -070059#define GEN8_SIGNAL_OFFSET(__ring, to) \
Chris Wilson51d545d2016-08-15 10:49:02 +010060 (dev_priv->semaphore->node.start + \
Chris Wilson8c12672e2016-04-07 07:29:14 +010061 GEN8_SEMAPHORE_OFFSET((__ring)->id, (to)))
Ben Widawsky3e789982014-06-30 09:53:37 -070062#define GEN8_WAIT_OFFSET(__ring, from) \
Chris Wilson51d545d2016-08-15 10:49:02 +010063 (dev_priv->semaphore->node.start + \
Chris Wilson8c12672e2016-04-07 07:29:14 +010064 GEN8_SEMAPHORE_OFFSET(from, (__ring)->id))
Ben Widawsky3e789982014-06-30 09:53:37 -070065
Chris Wilson7e37f882016-08-02 22:50:21 +010066enum intel_engine_hangcheck_action {
Mika Kuoppalada661462013-09-06 16:03:28 +030067 HANGCHECK_IDLE = 0,
Jani Nikulaf2f4d822013-08-11 12:44:01 +030068 HANGCHECK_WAIT,
69 HANGCHECK_ACTIVE,
70 HANGCHECK_KICK,
71 HANGCHECK_HUNG,
72};
Mika Kuoppalaad8beae2013-06-12 12:35:32 +030073
Mika Kuoppalab6b0fac2014-01-30 19:04:43 +020074#define HANGCHECK_SCORE_RING_HUNG 31
75
Ben Widawskyf9e61372016-09-20 16:54:33 +030076#define I915_MAX_SLICES 3
77#define I915_MAX_SUBSLICES 3
78
79#define instdone_slice_mask(dev_priv__) \
80 (INTEL_GEN(dev_priv__) == 7 ? \
81 1 : INTEL_INFO(dev_priv__)->sseu.slice_mask)
82
83#define instdone_subslice_mask(dev_priv__) \
84 (INTEL_GEN(dev_priv__) == 7 ? \
85 1 : INTEL_INFO(dev_priv__)->sseu.subslice_mask)
86
87#define for_each_instdone_slice_subslice(dev_priv__, slice__, subslice__) \
88 for ((slice__) = 0, (subslice__) = 0; \
89 (slice__) < I915_MAX_SLICES; \
90 (subslice__) = ((subslice__) + 1) < I915_MAX_SUBSLICES ? (subslice__) + 1 : 0, \
91 (slice__) += ((subslice__) == 0)) \
92 for_each_if((BIT(slice__) & instdone_slice_mask(dev_priv__)) && \
93 (BIT(subslice__) & instdone_subslice_mask(dev_priv__)))
94
Ben Widawskyd6369512016-09-20 16:54:32 +030095struct intel_instdone {
96 u32 instdone;
97 /* The following exist only in the RCS engine */
98 u32 slice_common;
Ben Widawskyf9e61372016-09-20 16:54:33 +030099 u32 sampler[I915_MAX_SLICES][I915_MAX_SUBSLICES];
100 u32 row[I915_MAX_SLICES][I915_MAX_SUBSLICES];
Ben Widawskyd6369512016-09-20 16:54:32 +0300101};
102
Chris Wilson7e37f882016-08-02 22:50:21 +0100103struct intel_engine_hangcheck {
Chris Wilson50877442014-03-21 12:41:53 +0000104 u64 acthd;
Mika Kuoppala92cab732013-05-24 17:16:07 +0300105 u32 seqno;
Mika Kuoppala05407ff2013-05-30 09:04:29 +0300106 int score;
Chris Wilson7e37f882016-08-02 22:50:21 +0100107 enum intel_engine_hangcheck_action action;
Chris Wilson4be17382014-06-06 10:22:29 +0100108 int deadlock;
Ben Widawskyd6369512016-09-20 16:54:32 +0300109 struct intel_instdone instdone;
Mika Kuoppala92cab732013-05-24 17:16:07 +0300110};
111
Chris Wilson7e37f882016-08-02 22:50:21 +0100112struct intel_ring {
Tvrtko Ursulin0eb973d2016-01-15 15:10:28 +0000113 struct i915_vma *vma;
Chris Wilson57e88532016-08-15 10:48:57 +0100114 void *vaddr;
Oscar Mateo8ee14972014-05-22 14:13:34 +0100115
Tvrtko Ursulin4a570db2016-03-16 11:00:38 +0000116 struct intel_engine_cs *engine;
Daniel Vetter0c7dd532014-08-11 16:17:44 +0200117
Chris Wilson675d9ad2016-08-04 07:52:36 +0100118 struct list_head request_list;
119
Oscar Mateo8ee14972014-05-22 14:13:34 +0100120 u32 head;
121 u32 tail;
122 int space;
123 int size;
124 int effective_size;
125
126 /** We track the position of the requests in the ring buffer, and
127 * when each is retired we increment last_retired_head as the GPU
128 * must have finished processing the request and so we know we
129 * can advance the ringbuffer up to that position.
130 *
131 * last_retired_head is set to -1 after the value is consumed so
132 * we can detect new retirements.
133 */
134 u32 last_retired_head;
135};
136
Chris Wilsone2efd132016-05-24 14:53:34 +0100137struct i915_gem_context;
Jordan Justen361b0272016-03-06 23:30:27 -0800138struct drm_i915_reg_table;
Nick Hoath21076372015-01-15 13:10:38 +0000139
Arun Siluvery17ee9502015-06-19 19:07:01 +0100140/*
141 * we use a single page to load ctx workarounds so all of these
142 * values are referred in terms of dwords
143 *
144 * struct i915_wa_ctx_bb:
145 * offset: specifies batch starting position, also helpful in case
146 * if we want to have multiple batches at different offsets based on
147 * some criteria. It is not a requirement at the moment but provides
148 * an option for future use.
149 * size: size of the batch in DWORDS
150 */
Chris Wilson48bb74e2016-08-15 10:49:04 +0100151struct i915_ctx_workarounds {
Arun Siluvery17ee9502015-06-19 19:07:01 +0100152 struct i915_wa_ctx_bb {
153 u32 offset;
154 u32 size;
155 } indirect_ctx, per_ctx;
Chris Wilson48bb74e2016-08-15 10:49:04 +0100156 struct i915_vma *vma;
Arun Siluvery17ee9502015-06-19 19:07:01 +0100157};
158
Chris Wilsonc81d4612016-07-01 17:23:25 +0100159struct drm_i915_gem_request;
160
Chris Wilsonc0336662016-05-06 15:40:21 +0100161struct intel_engine_cs {
162 struct drm_i915_private *i915;
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800163 const char *name;
Tvrtko Ursulin117897f2016-03-16 11:00:40 +0000164 enum intel_engine_id {
Tvrtko Ursulinde1add32016-01-15 15:12:50 +0000165 RCS = 0,
Daniel Vetter96154f22011-12-14 13:57:00 +0100166 BCS,
Tvrtko Ursulinde1add32016-01-15 15:12:50 +0000167 VCS,
168 VCS2, /* Keep instances of the same type engine together. */
169 VECS
Chris Wilson92204342010-09-18 11:02:01 +0100170 } id;
Tvrtko Ursulin666796d2016-03-16 11:00:39 +0000171#define I915_NUM_ENGINES 5
Tvrtko Ursulinde1add32016-01-15 15:12:50 +0000172#define _VCS(n) (VCS + (n))
Chris Wilson426960b2016-01-15 16:51:46 +0000173 unsigned int exec_id;
Tvrtko Ursulin5ec2cf72016-08-16 17:04:20 +0100174 enum intel_engine_hw_id {
175 RCS_HW = 0,
176 VCS_HW,
177 BCS_HW,
178 VECS_HW,
179 VCS2_HW
180 } hw_id;
181 enum intel_engine_hw_id guc_id; /* XXX same as hw_id? */
Chris Wilson04769652016-07-20 09:21:11 +0100182 u64 fence_context;
Daniel Vetter333e9fe2010-08-02 16:24:01 +0200183 u32 mmio_base;
Dave Gordonc2c7f242016-07-13 16:03:35 +0100184 unsigned int irq_shift;
Chris Wilson7e37f882016-08-02 22:50:21 +0100185 struct intel_ring *buffer;
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800186
Chris Wilson688e6c72016-07-01 17:23:15 +0100187 /* Rather than have every client wait upon all user interrupts,
188 * with the herd waking after every interrupt and each doing the
189 * heavyweight seqno dance, we delegate the task (of being the
190 * bottom-half of the user interrupt) to the first client. After
191 * every interrupt, we wake up one client, who does the heavyweight
192 * coherent seqno read and either goes back to sleep (if incomplete),
193 * or wakes up all the completed clients in parallel, before then
194 * transferring the bottom-half status to the next client in the queue.
195 *
196 * Compared to walking the entire list of waiters in a single dedicated
197 * bottom-half, we reduce the latency of the first waiter by avoiding
198 * a context switch, but incur additional coherent seqno reads when
199 * following the chain of request breadcrumbs. Since it is most likely
200 * that we have a single client waiting on each seqno, then reducing
201 * the overhead of waking that client is much preferred.
202 */
203 struct intel_breadcrumbs {
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100204 struct task_struct __rcu *irq_seqno_bh; /* bh for interrupts */
Chris Wilsonaca34b62016-07-06 12:39:02 +0100205 bool irq_posted;
206
Chris Wilson688e6c72016-07-01 17:23:15 +0100207 spinlock_t lock; /* protects the lists of requests */
208 struct rb_root waiters; /* sorted by retirement, priority */
Chris Wilsonc81d4612016-07-01 17:23:25 +0100209 struct rb_root signals; /* sorted by retirement */
Chris Wilson688e6c72016-07-01 17:23:15 +0100210 struct intel_wait *first_wait; /* oldest waiter by retirement */
Chris Wilsonc81d4612016-07-01 17:23:25 +0100211 struct task_struct *signaler; /* used for fence signalling */
Chris Wilsonb3850852016-07-01 17:23:26 +0100212 struct drm_i915_gem_request *first_signal;
Chris Wilson688e6c72016-07-01 17:23:15 +0100213 struct timer_list fake_irq; /* used after a missed interrupt */
Chris Wilson83348ba2016-08-09 17:47:51 +0100214 struct timer_list hangcheck; /* detect missed interrupts */
215
216 unsigned long timeout;
Chris Wilsonaca34b62016-07-06 12:39:02 +0100217
218 bool irq_enabled : 1;
219 bool rpm_wakelock : 1;
Chris Wilson688e6c72016-07-01 17:23:15 +0100220 } breadcrumbs;
221
Chris Wilson06fbca72015-04-07 16:20:36 +0100222 /*
223 * A pool of objects to use as shadow copies of client batch buffers
224 * when the command parser is enabled. Prevents the client from
225 * modifying the batch contents after software parsing.
226 */
227 struct i915_gem_batch_pool batch_pool;
228
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800229 struct intel_hw_status_page status_page;
Arun Siluvery17ee9502015-06-19 19:07:01 +0100230 struct i915_ctx_workarounds wa_ctx;
Chris Wilson56c0f1a2016-08-15 10:48:58 +0100231 struct i915_vma *scratch;
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800232
Chris Wilson61ff75a2016-07-01 17:23:28 +0100233 u32 irq_keep_mask; /* always keep these interrupts */
234 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
Dave Gordon38a0f2d2016-07-20 18:16:06 +0100235 void (*irq_enable)(struct intel_engine_cs *engine);
236 void (*irq_disable)(struct intel_engine_cs *engine);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800237
Dave Gordon38a0f2d2016-07-20 18:16:06 +0100238 int (*init_hw)(struct intel_engine_cs *engine);
Chris Wilson821ed7d2016-09-09 14:11:53 +0100239 void (*reset_hw)(struct intel_engine_cs *engine,
240 struct drm_i915_gem_request *req);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800241
John Harrison87531812015-05-29 17:43:44 +0100242 int (*init_context)(struct drm_i915_gem_request *req);
Arun Siluvery86d7f232014-08-26 14:44:50 +0100243
Chris Wilsonddd66c52016-08-02 22:50:31 +0100244 int (*emit_flush)(struct drm_i915_gem_request *request,
245 u32 mode);
246#define EMIT_INVALIDATE BIT(0)
247#define EMIT_FLUSH BIT(1)
248#define EMIT_BARRIER (EMIT_INVALIDATE | EMIT_FLUSH)
249 int (*emit_bb_start)(struct drm_i915_gem_request *req,
250 u64 offset, u32 length,
251 unsigned int dispatch_flags);
252#define I915_DISPATCH_SECURE BIT(0)
253#define I915_DISPATCH_PINNED BIT(1)
254#define I915_DISPATCH_RS BIT(2)
255 int (*emit_request)(struct drm_i915_gem_request *req);
Chris Wilson5590af32016-09-09 14:11:54 +0100256
257 /* Pass the request to the hardware queue (e.g. directly into
258 * the legacy ringbuffer or to the end of an execlist).
259 *
260 * This is called from an atomic context with irqs disabled; must
261 * be irq safe.
262 */
Chris Wilsonddd66c52016-08-02 22:50:31 +0100263 void (*submit_request)(struct drm_i915_gem_request *req);
Chris Wilson5590af32016-09-09 14:11:54 +0100264
Chris Wilsonb2eadbc2012-08-09 10:58:30 +0100265 /* Some chipsets are not quite as coherent as advertised and need
266 * an expensive kick to force a true read of the up-to-date seqno.
267 * However, the up-to-date seqno is not always required and the last
268 * seen value is good enough. Note that the seqno will always be
269 * monotonic, even if not coherent.
270 */
Dave Gordon38a0f2d2016-07-20 18:16:06 +0100271 void (*irq_seqno_barrier)(struct intel_engine_cs *engine);
Dave Gordon38a0f2d2016-07-20 18:16:06 +0100272 void (*cleanup)(struct intel_engine_cs *engine);
Ben Widawskyebc348b2014-04-29 14:52:28 -0700273
Ben Widawsky3e789982014-06-30 09:53:37 -0700274 /* GEN8 signal/wait table - never trust comments!
275 * signal to signal to signal to signal to signal to
276 * RCS VCS BCS VECS VCS2
277 * --------------------------------------------------------------------
278 * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
279 * |-------------------------------------------------------------------
280 * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
281 * |-------------------------------------------------------------------
282 * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
283 * |-------------------------------------------------------------------
284 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) |
285 * |-------------------------------------------------------------------
286 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) |
287 * |-------------------------------------------------------------------
288 *
289 * Generalization:
290 * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
291 * ie. transpose of g(x, y)
292 *
293 * sync from sync from sync from sync from sync from
294 * RCS VCS BCS VECS VCS2
295 * --------------------------------------------------------------------
296 * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
297 * |-------------------------------------------------------------------
298 * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
299 * |-------------------------------------------------------------------
300 * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
301 * |-------------------------------------------------------------------
302 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) |
303 * |-------------------------------------------------------------------
304 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) |
305 * |-------------------------------------------------------------------
306 *
307 * Generalization:
308 * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
309 * ie. transpose of f(x, y)
310 */
Ben Widawskyebc348b2014-04-29 14:52:28 -0700311 struct {
Tvrtko Ursulin666796d2016-03-16 11:00:39 +0000312 u32 sync_seqno[I915_NUM_ENGINES-1];
Ben Widawsky78325f22014-04-29 14:52:29 -0700313
Ben Widawsky3e789982014-06-30 09:53:37 -0700314 union {
Tvrtko Ursulin318f89c2016-08-16 17:04:21 +0100315#define GEN6_SEMAPHORE_LAST VECS_HW
316#define GEN6_NUM_SEMAPHORES (GEN6_SEMAPHORE_LAST + 1)
317#define GEN6_SEMAPHORES_MASK GENMASK(GEN6_SEMAPHORE_LAST, 0)
Ben Widawsky3e789982014-06-30 09:53:37 -0700318 struct {
319 /* our mbox written by others */
Tvrtko Ursulin318f89c2016-08-16 17:04:21 +0100320 u32 wait[GEN6_NUM_SEMAPHORES];
Ben Widawsky3e789982014-06-30 09:53:37 -0700321 /* mboxes this ring signals to */
Tvrtko Ursulin318f89c2016-08-16 17:04:21 +0100322 i915_reg_t signal[GEN6_NUM_SEMAPHORES];
Ben Widawsky3e789982014-06-30 09:53:37 -0700323 } mbox;
Tvrtko Ursulin666796d2016-03-16 11:00:39 +0000324 u64 signal_ggtt[I915_NUM_ENGINES];
Ben Widawsky3e789982014-06-30 09:53:37 -0700325 };
Ben Widawsky78325f22014-04-29 14:52:29 -0700326
327 /* AKA wait() */
Chris Wilsonad7bdb22016-08-02 22:50:40 +0100328 int (*sync_to)(struct drm_i915_gem_request *req,
329 struct drm_i915_gem_request *signal);
330 int (*signal)(struct drm_i915_gem_request *req);
Ben Widawskyebc348b2014-04-29 14:52:28 -0700331 } semaphore;
Ben Widawskyad776f82013-05-28 19:22:18 -0700332
Oscar Mateo4da46e12014-07-24 17:04:27 +0100333 /* Execlists */
Tvrtko Ursulin27af5ee2016-04-04 12:11:56 +0100334 struct tasklet_struct irq_tasklet;
335 spinlock_t execlist_lock; /* used inside tasklet, use spin_lock_bh */
Chris Wilson70c2a242016-09-09 14:11:46 +0100336 struct execlist_port {
337 struct drm_i915_gem_request *request;
338 unsigned int count;
339 } execlist_port[2];
Michel Thierryacdd8842014-07-24 17:04:38 +0100340 struct list_head execlist_queue;
Tvrtko Ursulin37566852016-04-12 14:37:31 +0100341 unsigned int fw_domains;
Tvrtko Ursulinca825802016-01-15 15:10:27 +0000342 bool disable_lite_restore_wa;
Chris Wilson70c2a242016-09-09 14:11:46 +0100343 bool preempt_wa;
Tvrtko Ursulinca825802016-01-15 15:10:27 +0000344 u32 ctx_desc_template;
Oscar Mateo4da46e12014-07-24 17:04:27 +0100345
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800346 /**
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800347 * List of breadcrumbs associated with GPU requests currently
348 * outstanding.
349 */
350 struct list_head request_list;
351
Chris Wilsona56ba562010-09-28 10:07:56 +0100352 /**
Tomas Elf94f7bbe2015-07-09 15:30:57 +0100353 * Seqno of request most recently submitted to request_list.
354 * Used exclusively by hang checker to avoid grabbing lock while
355 * inspecting request list.
356 */
357 u32 last_submitted_seqno;
358
Chris Wilsondcff85c2016-08-05 10:14:11 +0100359 /* An RCU guarded pointer to the last request. No reference is
360 * held to the request, users must carefully acquire a reference to
Chris Wilson1426f712016-08-09 17:03:22 +0100361 * the request using i915_gem_active_get_rcu(), or hold the
Chris Wilsondcff85c2016-08-05 10:14:11 +0100362 * struct_mutex.
363 */
364 struct i915_gem_active last_request;
365
Chris Wilsone2efd132016-05-24 14:53:34 +0100366 struct i915_gem_context *last_context;
Ben Widawsky40521052012-06-04 14:42:43 -0700367
Chris Wilson7e37f882016-08-02 22:50:21 +0100368 struct intel_engine_hangcheck hangcheck;
Mika Kuoppala92cab732013-05-24 17:16:07 +0300369
Brad Volkin44e895a2014-05-10 14:10:43 -0700370 bool needs_cmd_parser;
371
Brad Volkin351e3db2014-02-18 10:15:46 -0800372 /*
Brad Volkin44e895a2014-05-10 14:10:43 -0700373 * Table of commands the command parser needs to know about
Chris Wilson33a051a2016-07-27 09:07:26 +0100374 * for this engine.
Brad Volkin351e3db2014-02-18 10:15:46 -0800375 */
Brad Volkin44e895a2014-05-10 14:10:43 -0700376 DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
Brad Volkin351e3db2014-02-18 10:15:46 -0800377
378 /*
379 * Table of registers allowed in commands that read/write registers.
380 */
Jordan Justen361b0272016-03-06 23:30:27 -0800381 const struct drm_i915_reg_table *reg_tables;
382 int reg_table_count;
Brad Volkin351e3db2014-02-18 10:15:46 -0800383
384 /*
385 * Returns the bitmask for the length field of the specified command.
386 * Return 0 for an unrecognized/invalid command.
387 *
Chris Wilson33a051a2016-07-27 09:07:26 +0100388 * If the command parser finds an entry for a command in the engine's
Brad Volkin351e3db2014-02-18 10:15:46 -0800389 * cmd_tables, it gets the command's length based on the table entry.
Chris Wilson33a051a2016-07-27 09:07:26 +0100390 * If not, it calls this function to determine the per-engine length
391 * field encoding for the command (i.e. different opcode ranges use
392 * certain bits to encode the command length in the header).
Brad Volkin351e3db2014-02-18 10:15:46 -0800393 */
394 u32 (*get_cmd_length_mask)(u32 cmd_header);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800395};
396
Dave Gordonb0366a52015-12-08 15:02:36 +0000397static inline bool
Chris Wilson67d97da2016-07-04 08:08:31 +0100398intel_engine_initialized(const struct intel_engine_cs *engine)
Dave Gordonb0366a52015-12-08 15:02:36 +0000399{
Chris Wilsonc0336662016-05-06 15:40:21 +0100400 return engine->i915 != NULL;
Dave Gordonb0366a52015-12-08 15:02:36 +0000401}
Chris Wilsonb4519512012-05-11 14:29:30 +0100402
Daniel Vetter96154f22011-12-14 13:57:00 +0100403static inline unsigned
Chris Wilson67d97da2016-07-04 08:08:31 +0100404intel_engine_flag(const struct intel_engine_cs *engine)
Daniel Vetter96154f22011-12-14 13:57:00 +0100405{
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000406 return 1 << engine->id;
Daniel Vetter96154f22011-12-14 13:57:00 +0100407}
408
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800409static inline u32
Chris Wilson7e37f882016-08-02 22:50:21 +0100410intel_engine_sync_index(struct intel_engine_cs *engine,
411 struct intel_engine_cs *other)
Chris Wilson1ec14ad2010-12-04 11:30:53 +0000412{
413 int idx;
414
415 /*
Rodrigo Vividdd4dbc2014-06-30 09:51:11 -0700416 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
417 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
418 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
419 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
420 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
Chris Wilson1ec14ad2010-12-04 11:30:53 +0000421 */
422
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000423 idx = (other - engine) - 1;
Chris Wilson1ec14ad2010-12-04 11:30:53 +0000424 if (idx < 0)
Tvrtko Ursulin666796d2016-03-16 11:00:39 +0000425 idx += I915_NUM_ENGINES;
Chris Wilson1ec14ad2010-12-04 11:30:53 +0000426
427 return idx;
428}
429
Imre Deak319404d2015-08-14 18:35:27 +0300430static inline void
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000431intel_flush_status_page(struct intel_engine_cs *engine, int reg)
Imre Deak319404d2015-08-14 18:35:27 +0300432{
Chris Wilson0d317ce2016-04-09 10:57:56 +0100433 mb();
434 clflush(&engine->status_page.page_addr[reg]);
435 mb();
Imre Deak319404d2015-08-14 18:35:27 +0300436}
437
Chris Wilson1ec14ad2010-12-04 11:30:53 +0000438static inline u32
Chris Wilson5dd8e502016-04-09 10:57:57 +0100439intel_read_status_page(struct intel_engine_cs *engine, int reg)
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800440{
Daniel Vetter4225d0f2012-04-26 23:28:16 +0200441 /* Ensure that the compiler doesn't optimize away the load. */
Chris Wilson5dd8e502016-04-09 10:57:57 +0100442 return READ_ONCE(engine->status_page.page_addr[reg]);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800443}
444
Mika Kuoppalab70ec5b2012-12-19 11:13:05 +0200445static inline void
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000446intel_write_status_page(struct intel_engine_cs *engine,
Mika Kuoppalab70ec5b2012-12-19 11:13:05 +0200447 int reg, u32 value)
448{
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000449 engine->status_page.page_addr[reg] = value;
Mika Kuoppalab70ec5b2012-12-19 11:13:05 +0200450}
451
Jani Nikulae2828912016-01-18 09:19:47 +0200452/*
Chris Wilson311bd682011-01-13 19:06:50 +0000453 * Reads a dword out of the status page, which is written to from the command
454 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
455 * MI_STORE_DATA_IMM.
456 *
457 * The following dwords have a reserved meaning:
458 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
459 * 0x04: ring 0 head pointer
460 * 0x05: ring 1 head pointer (915-class)
461 * 0x06: ring 2 head pointer (915-class)
462 * 0x10-0x1b: Context status DWords (GM45)
463 * 0x1f: Last written status offset. (GM45)
Thomas Danielb07da532015-02-18 11:48:21 +0000464 * 0x20-0x2f: Reserved (Gen6+)
Chris Wilson311bd682011-01-13 19:06:50 +0000465 *
Thomas Danielb07da532015-02-18 11:48:21 +0000466 * The area from dword 0x30 to 0x3ff is available for driver usage.
Chris Wilson311bd682011-01-13 19:06:50 +0000467 */
Thomas Danielb07da532015-02-18 11:48:21 +0000468#define I915_GEM_HWS_INDEX 0x30
Chris Wilson7c17d372016-01-20 15:43:35 +0200469#define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
Thomas Danielb07da532015-02-18 11:48:21 +0000470#define I915_GEM_HWS_SCRATCH_INDEX 0x40
Jesse Barnes9a289772012-10-26 09:42:42 -0700471#define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
Chris Wilson311bd682011-01-13 19:06:50 +0000472
Chris Wilson7e37f882016-08-02 22:50:21 +0100473struct intel_ring *
474intel_engine_create_ring(struct intel_engine_cs *engine, int size);
Chris Wilsonaad29fb2016-08-02 22:50:23 +0100475int intel_ring_pin(struct intel_ring *ring);
476void intel_ring_unpin(struct intel_ring *ring);
Chris Wilson7e37f882016-08-02 22:50:21 +0100477void intel_ring_free(struct intel_ring *ring);
Oscar Mateo84c23772014-07-24 17:04:15 +0100478
Chris Wilson7e37f882016-08-02 22:50:21 +0100479void intel_engine_stop(struct intel_engine_cs *engine);
480void intel_engine_cleanup(struct intel_engine_cs *engine);
Ben Widawsky96f298a2011-03-19 18:14:27 -0700481
Chris Wilson821ed7d2016-09-09 14:11:53 +0100482void intel_legacy_submission_resume(struct drm_i915_private *dev_priv);
483
John Harrison6689cb22015-03-19 12:30:08 +0000484int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
485
John Harrison5fb9de12015-05-29 17:44:07 +0100486int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
John Harrisonbba09b12015-05-29 17:44:06 +0100487int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
Chris Wilson406ea8d2016-07-20 13:31:55 +0100488
Chris Wilson7e37f882016-08-02 22:50:21 +0100489static inline void intel_ring_emit(struct intel_ring *ring, u32 data)
Chris Wilsone898cd22010-08-04 15:18:14 +0100490{
Chris Wilsonb5321f32016-08-02 22:50:18 +0100491 *(uint32_t *)(ring->vaddr + ring->tail) = data;
492 ring->tail += 4;
Chris Wilsone898cd22010-08-04 15:18:14 +0100493}
Chris Wilson406ea8d2016-07-20 13:31:55 +0100494
Chris Wilson7e37f882016-08-02 22:50:21 +0100495static inline void intel_ring_emit_reg(struct intel_ring *ring, i915_reg_t reg)
Ville Syrjäläf92a9162015-11-04 23:20:07 +0200496{
Chris Wilsonb5321f32016-08-02 22:50:18 +0100497 intel_ring_emit(ring, i915_mmio_reg_offset(reg));
Ville Syrjäläf92a9162015-11-04 23:20:07 +0200498}
Chris Wilson406ea8d2016-07-20 13:31:55 +0100499
Chris Wilson7e37f882016-08-02 22:50:21 +0100500static inline void intel_ring_advance(struct intel_ring *ring)
Chris Wilson09246732013-08-10 22:16:32 +0100501{
Chris Wilson8f942012016-08-02 22:50:30 +0100502 /* Dummy function.
503 *
504 * This serves as a placeholder in the code so that the reader
505 * can compare against the preceding intel_ring_begin() and
506 * check that the number of dwords emitted matches the space
507 * reserved for the command packet (i.e. the value passed to
508 * intel_ring_begin()).
Chris Wilsonc5efa1a2016-08-02 22:50:29 +0100509 */
Chris Wilson8f942012016-08-02 22:50:30 +0100510}
511
512static inline u32 intel_ring_offset(struct intel_ring *ring, u32 value)
513{
514 /* Don't write ring->size (equivalent to 0) as that hangs some GPUs. */
515 return value & (ring->size - 1);
Chris Wilson09246732013-08-10 22:16:32 +0100516}
Chris Wilson406ea8d2016-07-20 13:31:55 +0100517
Oscar Mateo82e104c2014-07-24 17:04:26 +0100518int __intel_ring_space(int head, int tail, int size);
Chris Wilson32c04f12016-08-02 22:50:22 +0100519void intel_ring_update_space(struct intel_ring *ring);
Chris Wilson09246732013-08-10 22:16:32 +0100520
Chris Wilson7e37f882016-08-02 22:50:21 +0100521void intel_engine_init_seqno(struct intel_engine_cs *engine, u32 seqno);
Chris Wilson821ed7d2016-09-09 14:11:53 +0100522void intel_engine_reset_irq(struct intel_engine_cs *engine);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800523
Tvrtko Ursulin019bf272016-07-13 16:03:41 +0100524void intel_engine_setup_common(struct intel_engine_cs *engine);
525int intel_engine_init_common(struct intel_engine_cs *engine);
Chris Wilsonadc320c2016-08-15 10:48:59 +0100526int intel_engine_create_scratch(struct intel_engine_cs *engine, int size);
Chris Wilson96a945a2016-08-03 13:19:16 +0100527void intel_engine_cleanup_common(struct intel_engine_cs *engine);
Tvrtko Ursulin019bf272016-07-13 16:03:41 +0100528
Chris Wilsondcff85c2016-08-05 10:14:11 +0100529static inline int intel_engine_idle(struct intel_engine_cs *engine,
Chris Wilsonea746f32016-09-09 14:11:49 +0100530 unsigned int flags)
Chris Wilsondcff85c2016-08-05 10:14:11 +0100531{
532 /* Wait upon the last request to be completed */
533 return i915_gem_active_wait_unlocked(&engine->last_request,
Chris Wilsonea746f32016-09-09 14:11:49 +0100534 flags, NULL, NULL);
Chris Wilsondcff85c2016-08-05 10:14:11 +0100535}
536
Tvrtko Ursulin8b3e2d32016-07-13 16:03:37 +0100537int intel_init_render_ring_buffer(struct intel_engine_cs *engine);
538int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine);
539int intel_init_bsd2_ring_buffer(struct intel_engine_cs *engine);
540int intel_init_blt_ring_buffer(struct intel_engine_cs *engine);
541int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine);
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800542
Chris Wilson7e37f882016-08-02 22:50:21 +0100543u64 intel_engine_get_active_head(struct intel_engine_cs *engine);
Chris Wilson1b7744e2016-07-01 17:23:17 +0100544static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
545{
546 return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
547}
Daniel Vetter79f321b2010-09-24 21:20:10 +0200548
Tvrtko Ursulin0bc40be2016-03-16 11:00:37 +0000549int init_workarounds_ring(struct intel_engine_cs *engine);
Michel Thierry771b9a52014-11-11 16:47:33 +0000550
John Harrison29b1b412015-06-18 13:10:09 +0100551/*
552 * Arbitrary size for largest possible 'add request' sequence. The code paths
553 * are complex and variable. Empirical measurement shows that the worst case
Chris Wilson596e5ef2016-04-29 09:07:04 +0100554 * is BDW at 192 bytes (6 + 6 + 36 dwords), then ILK at 136 bytes. However,
555 * we need to allocate double the largest single packet within that emission
556 * to account for tail wraparound (so 6 + 6 + 72 dwords for BDW).
John Harrison29b1b412015-06-18 13:10:09 +0100557 */
Chris Wilson596e5ef2016-04-29 09:07:04 +0100558#define MIN_SPACE_FOR_ADD_REQUEST 336
John Harrison29b1b412015-06-18 13:10:09 +0100559
Chris Wilsona58c01a2016-04-29 13:18:21 +0100560static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
561{
Chris Wilson57e88532016-08-15 10:48:57 +0100562 return engine->status_page.ggtt_offset + I915_GEM_HWS_INDEX_ADDR;
Chris Wilsona58c01a2016-04-29 13:18:21 +0100563}
564
Chris Wilson688e6c72016-07-01 17:23:15 +0100565/* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
Chris Wilson688e6c72016-07-01 17:23:15 +0100566int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
567
568static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
569{
570 wait->tsk = current;
571 wait->seqno = seqno;
572}
573
574static inline bool intel_wait_complete(const struct intel_wait *wait)
575{
576 return RB_EMPTY_NODE(&wait->node);
577}
578
579bool intel_engine_add_wait(struct intel_engine_cs *engine,
580 struct intel_wait *wait);
581void intel_engine_remove_wait(struct intel_engine_cs *engine,
582 struct intel_wait *wait);
Chris Wilsonb3850852016-07-01 17:23:26 +0100583void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
Chris Wilson688e6c72016-07-01 17:23:15 +0100584
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100585static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
Chris Wilson688e6c72016-07-01 17:23:15 +0100586{
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100587 return rcu_access_pointer(engine->breadcrumbs.irq_seqno_bh);
Chris Wilson688e6c72016-07-01 17:23:15 +0100588}
589
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100590static inline bool intel_engine_wakeup(const struct intel_engine_cs *engine)
Chris Wilson688e6c72016-07-01 17:23:15 +0100591{
592 bool wakeup = false;
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100593
Chris Wilson688e6c72016-07-01 17:23:15 +0100594 /* Note that for this not to dangerously chase a dangling pointer,
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100595 * we must hold the rcu_read_lock here.
Chris Wilson688e6c72016-07-01 17:23:15 +0100596 *
597 * Also note that tsk is likely to be in !TASK_RUNNING state so an
598 * early test for tsk->state != TASK_RUNNING before wake_up_process()
599 * is unlikely to be beneficial.
600 */
Chris Wilsondbd6ef22016-08-09 17:47:52 +0100601 if (intel_engine_has_waiter(engine)) {
602 struct task_struct *tsk;
603
604 rcu_read_lock();
605 tsk = rcu_dereference(engine->breadcrumbs.irq_seqno_bh);
606 if (tsk)
607 wakeup = wake_up_process(tsk);
608 rcu_read_unlock();
609 }
610
Chris Wilson688e6c72016-07-01 17:23:15 +0100611 return wakeup;
612}
613
Chris Wilson688e6c72016-07-01 17:23:15 +0100614void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
615unsigned int intel_kick_waiters(struct drm_i915_private *i915);
Chris Wilsonc81d4612016-07-01 17:23:25 +0100616unsigned int intel_kick_signalers(struct drm_i915_private *i915);
Chris Wilson688e6c72016-07-01 17:23:15 +0100617
Chris Wilsondcff85c2016-08-05 10:14:11 +0100618static inline bool intel_engine_is_active(struct intel_engine_cs *engine)
619{
620 return i915_gem_active_isset(&engine->last_request);
621}
622
Zou Nan hai8187a2b2010-05-21 09:08:55 +0800623#endif /* _INTEL_RINGBUFFER_H_ */