blob: 940e7f7df69a69ae6652269759880579c1b3447d [file] [log] [blame]
/*
* Copyright © 2014 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*
* Authors:
* Ben Widawsky <ben@bwidawsk.net>
* Michel Thierry <michel.thierry@intel.com>
* Thomas Daniel <thomas.daniel@intel.com>
* Oscar Mateo <oscar.mateo@intel.com>
*
*/
/**
* DOC: Logical Rings, Logical Ring Contexts and Execlists
*
* Motivation:
* GEN8 brings an expansion of the HW contexts: "Logical Ring Contexts".
* These expanded contexts enable a number of new abilities, especially
* "Execlists" (also implemented in this file).
*
* One of the main differences with the legacy HW contexts is that logical
* ring contexts incorporate many more things to the context's state, like
* PDPs or ringbuffer control registers:
*
* The reason why PDPs are included in the context is straightforward: as
* PPGTTs (per-process GTTs) are actually per-context, having the PDPs
* contained there mean you don't need to do a ppgtt->switch_mm yourself,
* instead, the GPU will do it for you on the context switch.
*
* But, what about the ringbuffer control registers (head, tail, etc..)?
* shouldn't we just need a set of those per engine command streamer? This is
* where the name "Logical Rings" starts to make sense: by virtualizing the
* rings, the engine cs shifts to a new "ring buffer" with every context
* switch. When you want to submit a workload to the GPU you: A) choose your
* context, B) find its appropriate virtualized ring, C) write commands to it
* and then, finally, D) tell the GPU to switch to that context.
*
* Instead of the legacy MI_SET_CONTEXT, the way you tell the GPU to switch
* to a contexts is via a context execution list, ergo "Execlists".
*
* LRC implementation:
* Regarding the creation of contexts, we have:
*
* - One global default context.
* - One local default context for each opened fd.
* - One local extra context for each context create ioctl call.
*
* Now that ringbuffers belong per-context (and not per-engine, like before)
* and that contexts are uniquely tied to a given engine (and not reusable,
* like before) we need:
*
* - One ringbuffer per-engine inside each context.
* - One backing object per-engine inside each context.
*
* The global default context starts its life with these new objects fully
* allocated and populated. The local default context for each opened fd is
* more complex, because we don't know at creation time which engine is going
* to use them. To handle this, we have implemented a deferred creation of LR
* contexts:
*
* The local context starts its life as a hollow or blank holder, that only
* gets populated for a given engine once we receive an execbuffer. If later
* on we receive another execbuffer ioctl for the same context but a different
* engine, we allocate/populate a new ringbuffer and context backing object and
* so on.
*
* Finally, regarding local contexts created using the ioctl call: as they are
* only allowed with the render ring, we can allocate & populate them right
* away (no need to defer anything, at least for now).
*
* Execlists implementation:
* Execlists are the new method by which, on gen8+ hardware, workloads are
* submitted for execution (as opposed to the legacy, ringbuffer-based, method).
* This method works as follows:
*
* When a request is committed, its commands (the BB start and any leading or
* trailing commands, like the seqno breadcrumbs) are placed in the ringbuffer
* for the appropriate context. The tail pointer in the hardware context is not
* updated at this time, but instead, kept by the driver in the ringbuffer
* structure. A structure representing this request is added to a request queue
* for the appropriate engine: this structure contains a copy of the context's
* tail after the request was written to the ring buffer and a pointer to the
* context itself.
*
* If the engine's request queue was empty before the request was added, the
* queue is processed immediately. Otherwise the queue will be processed during
* a context switch interrupt. In any case, elements on the queue will get sent
* (in pairs) to the GPU's ExecLists Submit Port (ELSP, for short) with a
* globally unique 20-bits submission ID.
*
* When execution of a request completes, the GPU updates the context status
* buffer with a context complete event and generates a context switch interrupt.
* During the interrupt handling, the driver examines the events in the buffer:
* for each context complete event, if the announced ID matches that on the head
* of the request queue, then that request is retired and removed from the queue.
*
* After processing, if any requests were retired and the queue is not empty
* then a new execution list can be submitted. The two requests at the front of
* the queue are next to be submitted but since a context may not occur twice in
* an execution list, if subsequent requests have the same ID as the first then
* the two requests must be combined. This is done simply by discarding requests
* at the head of the queue until either only one requests is left (in which case
* we use a NULL second context) or the first two requests have unique IDs.
*
* By always executing the first two requests in the queue the driver ensures
* that the GPU is kept as busy as possible. In the case where a single context
* completes but a second context is still executing, the request for this second
* context will be at the head of the queue when we remove the first one. This
* request will then be resubmitted along with a new request for a different context,
* which will cause the hardware to continue executing the second request and queue
* the new request (the GPU detects the condition of a context getting preempted
* with the same context and optimizes the context switch flow by not doing
* preemption, but just sampling the new tail pointer).
*
*/
#include <linux/interrupt.h>
#include "i915_drv.h"
#include "i915_perf.h"
#include "i915_trace.h"
#include "i915_vgpu.h"
#include "intel_context.h"
#include "intel_engine_pm.h"
#include "intel_gt.h"
#include "intel_gt_pm.h"
#include "intel_gt_requests.h"
#include "intel_lrc_reg.h"
#include "intel_mocs.h"
#include "intel_reset.h"
#include "intel_ring.h"
#include "intel_workarounds.h"
#define RING_EXECLIST_QFULL (1 << 0x2)
#define RING_EXECLIST1_VALID (1 << 0x3)
#define RING_EXECLIST0_VALID (1 << 0x4)
#define RING_EXECLIST_ACTIVE_STATUS (3 << 0xE)
#define RING_EXECLIST1_ACTIVE (1 << 0x11)
#define RING_EXECLIST0_ACTIVE (1 << 0x12)
#define GEN8_CTX_STATUS_IDLE_ACTIVE (1 << 0)
#define GEN8_CTX_STATUS_PREEMPTED (1 << 1)
#define GEN8_CTX_STATUS_ELEMENT_SWITCH (1 << 2)
#define GEN8_CTX_STATUS_ACTIVE_IDLE (1 << 3)
#define GEN8_CTX_STATUS_COMPLETE (1 << 4)
#define GEN8_CTX_STATUS_LITE_RESTORE (1 << 15)
#define GEN8_CTX_STATUS_COMPLETED_MASK \
(GEN8_CTX_STATUS_COMPLETE | GEN8_CTX_STATUS_PREEMPTED)
#define CTX_DESC_FORCE_RESTORE BIT_ULL(2)
#define GEN12_CTX_STATUS_SWITCHED_TO_NEW_QUEUE (0x1) /* lower csb dword */
#define GEN12_CTX_SWITCH_DETAIL(csb_dw) ((csb_dw) & 0xF) /* upper csb dword */
#define GEN12_CSB_SW_CTX_ID_MASK GENMASK(25, 15)
#define GEN12_IDLE_CTX_ID 0x7FF
#define GEN12_CSB_CTX_VALID(csb_dw) \
(FIELD_GET(GEN12_CSB_SW_CTX_ID_MASK, csb_dw) != GEN12_IDLE_CTX_ID)
/* Typical size of the average request (2 pipecontrols and a MI_BB) */
#define EXECLISTS_REQUEST_SIZE 64 /* bytes */
#define WA_TAIL_DWORDS 2
#define WA_TAIL_BYTES (sizeof(u32) * WA_TAIL_DWORDS)
struct virtual_engine {
struct intel_engine_cs base;
struct intel_context context;
/*
* We allow only a single request through the virtual engine at a time
* (each request in the timeline waits for the completion fence of
* the previous before being submitted). By restricting ourselves to
* only submitting a single request, each request is placed on to a
* physical to maximise load spreading (by virtue of the late greedy
* scheduling -- each real engine takes the next available request
* upon idling).
*/
struct i915_request *request;
/*
* We keep a rbtree of available virtual engines inside each physical
* engine, sorted by priority. Here we preallocate the nodes we need
* for the virtual engine, indexed by physical_engine->id.
*/
struct ve_node {
struct rb_node rb;
int prio;
} nodes[I915_NUM_ENGINES];
/*
* Keep track of bonded pairs -- restrictions upon on our selection
* of physical engines any particular request may be submitted to.
* If we receive a submit-fence from a master engine, we will only
* use one of sibling_mask physical engines.
*/
struct ve_bond {
const struct intel_engine_cs *master;
intel_engine_mask_t sibling_mask;
} *bonds;
unsigned int num_bonds;
/* And finally, which physical engines this virtual engine maps onto. */
unsigned int num_siblings;
struct intel_engine_cs *siblings[0];
};
static struct virtual_engine *to_virtual_engine(struct intel_engine_cs *engine)
{
GEM_BUG_ON(!intel_engine_is_virtual(engine));
return container_of(engine, struct virtual_engine, base);
}
static int __execlists_context_alloc(struct intel_context *ce,
struct intel_engine_cs *engine);
static void execlists_init_reg_state(u32 *reg_state,
const struct intel_context *ce,
const struct intel_engine_cs *engine,
const struct intel_ring *ring,
bool close);
static void
__execlists_update_reg_state(const struct intel_context *ce,
const struct intel_engine_cs *engine,
u32 head);
static void mark_eio(struct i915_request *rq)
{
if (i915_request_completed(rq))
return;
GEM_BUG_ON(i915_request_signaled(rq));
dma_fence_set_error(&rq->fence, -EIO);
i915_request_mark_complete(rq);
}
static struct i915_request *
active_request(const struct intel_timeline * const tl, struct i915_request *rq)
{
struct i915_request *active = rq;
rcu_read_lock();
list_for_each_entry_continue_reverse(rq, &tl->requests, link) {
if (i915_request_completed(rq))
break;
active = rq;
}
rcu_read_unlock();
return active;
}
static inline u32 intel_hws_preempt_address(struct intel_engine_cs *engine)
{
return (i915_ggtt_offset(engine->status_page.vma) +
I915_GEM_HWS_PREEMPT_ADDR);
}
static inline void
ring_set_paused(const struct intel_engine_cs *engine, int state)
{
/*
* We inspect HWS_PREEMPT with a semaphore inside
* engine->emit_fini_breadcrumb. If the dword is true,
* the ring is paused as the semaphore will busywait
* until the dword is false.
*/
engine->status_page.addr[I915_GEM_HWS_PREEMPT] = state;
if (state)
wmb();
}
static inline struct i915_priolist *to_priolist(struct rb_node *rb)
{
return rb_entry(rb, struct i915_priolist, node);
}
static inline int rq_prio(const struct i915_request *rq)
{
return rq->sched.attr.priority;
}
static int effective_prio(const struct i915_request *rq)
{
int prio = rq_prio(rq);
/*
* If this request is special and must not be interrupted at any
* cost, so be it. Note we are only checking the most recent request
* in the context and so may be masking an earlier vip request. It
* is hoped that under the conditions where nopreempt is used, this
* will not matter (i.e. all requests to that context will be
* nopreempt for as long as desired).
*/
if (i915_request_has_nopreempt(rq))
prio = I915_PRIORITY_UNPREEMPTABLE;
/*
* On unwinding the active request, we give it a priority bump
* if it has completed waiting on any semaphore. If we know that
* the request has already started, we can prevent an unwanted
* preempt-to-idle cycle by taking that into account now.
*/
if (__i915_request_has_started(rq))
prio |= I915_PRIORITY_NOSEMAPHORE;
/* Restrict mere WAIT boosts from triggering preemption */
BUILD_BUG_ON(__NO_PREEMPTION & ~I915_PRIORITY_MASK); /* only internal */
return prio | __NO_PREEMPTION;
}
static int queue_prio(const struct intel_engine_execlists *execlists)
{
struct i915_priolist *p;
struct rb_node *rb;
rb = rb_first_cached(&execlists->queue);
if (!rb)
return INT_MIN;
/*
* As the priolist[] are inverted, with the highest priority in [0],
* we have to flip the index value to become priority.
*/
p = to_priolist(rb);
return ((p->priority + 1) << I915_USER_PRIORITY_SHIFT) - ffs(p->used);
}
static inline bool need_preempt(const struct intel_engine_cs *engine,
const struct i915_request *rq,
struct rb_node *rb)
{
int last_prio;
if (!intel_engine_has_semaphores(engine))
return false;
/*
* Check if the current priority hint merits a preemption attempt.
*
* We record the highest value priority we saw during rescheduling
* prior to this dequeue, therefore we know that if it is strictly
* less than the current tail of ESLP[0], we do not need to force
* a preempt-to-idle cycle.
*
* However, the priority hint is a mere hint that we may need to
* preempt. If that hint is stale or we may be trying to preempt
* ourselves, ignore the request.
*
* More naturally we would write
* prio >= max(0, last);
* except that we wish to prevent triggering preemption at the same
* priority level: the task that is running should remain running
* to preserve FIFO ordering of dependencies.
*/
last_prio = max(effective_prio(rq), I915_PRIORITY_NORMAL - 1);
if (engine->execlists.queue_priority_hint <= last_prio)
return false;
/*
* Check against the first request in ELSP[1], it will, thanks to the
* power of PI, be the highest priority of that context.
*/
if (!list_is_last(&rq->sched.link, &engine->active.requests) &&
rq_prio(list_next_entry(rq, sched.link)) > last_prio)
return true;
if (rb) {
struct virtual_engine *ve =
rb_entry(rb, typeof(*ve), nodes[engine->id].rb);
bool preempt = false;
if (engine == ve->siblings[0]) { /* only preempt one sibling */
struct i915_request *next;
rcu_read_lock();
next = READ_ONCE(ve->request);
if (next)
preempt = rq_prio(next) > last_prio;
rcu_read_unlock();
}
if (preempt)
return preempt;
}
/*
* If the inflight context did not trigger the preemption, then maybe
* it was the set of queued requests? Pick the highest priority in
* the queue (the first active priolist) and see if it deserves to be
* running instead of ELSP[0].
*
* The highest priority request in the queue can not be either
* ELSP[0] or ELSP[1] as, thanks again to PI, if it was the same
* context, it's priority would not exceed ELSP[0] aka last_prio.
*/
return queue_prio(&engine->execlists) > last_prio;
}
__maybe_unused static inline bool
assert_priority_queue(const struct i915_request *prev,
const struct i915_request *next)
{
/*
* Without preemption, the prev may refer to the still active element
* which we refuse to let go.
*
* Even with preemption, there are times when we think it is better not
* to preempt and leave an ostensibly lower priority request in flight.
*/
if (i915_request_is_active(prev))
return true;
return rq_prio(prev) >= rq_prio(next);
}
/*
* The context descriptor encodes various attributes of a context,
* including its GTT address and some flags. Because it's fairly
* expensive to calculate, we'll just do it once and cache the result,
* which remains valid until the context is unpinned.
*
* This is what a descriptor looks like, from LSB to MSB::
*
* bits 0-11: flags, GEN8_CTX_* (cached in ctx->desc_template)
* bits 12-31: LRCA, GTT address of (the HWSP of) this context
* bits 32-52: ctx ID, a globally unique tag (highest bit used by GuC)
* bits 53-54: mbz, reserved for use by hardware
* bits 55-63: group ID, currently unused and set to 0
*
* Starting from Gen11, the upper dword of the descriptor has a new format:
*
* bits 32-36: reserved
* bits 37-47: SW context ID
* bits 48:53: engine instance
* bit 54: mbz, reserved for use by hardware
* bits 55-60: SW counter
* bits 61-63: engine class
*
* engine info, SW context ID and SW counter need to form a unique number
* (Context ID) per lrc.
*/
static u64
lrc_descriptor(struct intel_context *ce, struct intel_engine_cs *engine)
{
u64 desc;
desc = INTEL_LEGACY_32B_CONTEXT;
if (i915_vm_is_4lvl(ce->vm))
desc = INTEL_LEGACY_64B_CONTEXT;
desc <<= GEN8_CTX_ADDRESSING_MODE_SHIFT;
desc |= GEN8_CTX_VALID | GEN8_CTX_PRIVILEGE;
if (IS_GEN(engine->i915, 8))
desc |= GEN8_CTX_L3LLC_COHERENT;
desc |= i915_ggtt_offset(ce->state); /* bits 12-31 */
/*
* The following 32bits are copied into the OA reports (dword 2).
* Consider updating oa_get_render_ctx_id in i915_perf.c when changing
* anything below.
*/
if (INTEL_GEN(engine->i915) >= 11) {
desc |= (u64)engine->instance << GEN11_ENGINE_INSTANCE_SHIFT;
/* bits 48-53 */
desc |= (u64)engine->class << GEN11_ENGINE_CLASS_SHIFT;
/* bits 61-63 */
}
return desc;
}
static inline unsigned int dword_in_page(void *addr)
{
return offset_in_page(addr) / sizeof(u32);
}
static void set_offsets(u32 *regs,
const u8 *data,
const struct intel_engine_cs *engine,
bool clear)
#define NOP(x) (BIT(7) | (x))
#define LRI(count, flags) ((flags) << 6 | (count) | BUILD_BUG_ON_ZERO(count >= BIT(6)))
#define POSTED BIT(0)
#define REG(x) (((x) >> 2) | BUILD_BUG_ON_ZERO(x >= 0x200))
#define REG16(x) \
(((x) >> 9) | BIT(7) | BUILD_BUG_ON_ZERO(x >= 0x10000)), \
(((x) >> 2) & 0x7f)
#define END(x) 0, (x)
{
const u32 base = engine->mmio_base;
while (*data) {
u8 count, flags;
if (*data & BIT(7)) { /* skip */
count = *data++ & ~BIT(7);
if (clear)
memset32(regs, MI_NOOP, count);
regs += count;
continue;
}
count = *data & 0x3f;
flags = *data >> 6;
data++;
*regs = MI_LOAD_REGISTER_IMM(count);
if (flags & POSTED)
*regs |= MI_LRI_FORCE_POSTED;
if (INTEL_GEN(engine->i915) >= 11)
*regs |= MI_LRI_CS_MMIO;
regs++;
GEM_BUG_ON(!count);
do {
u32 offset = 0;
u8 v;
do {
v = *data++;
offset <<= 7;
offset |= v & ~BIT(7);
} while (v & BIT(7));
regs[0] = base + (offset << 2);
if (clear)
regs[1] = 0;
regs += 2;
} while (--count);
}
if (clear) {
u8 count = *++data;
/* Clear past the tail for HW access */
GEM_BUG_ON(dword_in_page(regs) > count);
memset32(regs, MI_NOOP, count - dword_in_page(regs));
/* Close the batch; used mainly by live_lrc_layout() */
*regs = MI_BATCH_BUFFER_END;
if (INTEL_GEN(engine->i915) >= 10)
*regs |= BIT(0);
}
}
static const u8 gen8_xcs_offsets[] = {
NOP(1),
LRI(11, 0),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x11c),
REG(0x114),
REG(0x118),
NOP(9),
LRI(9, 0),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
NOP(13),
LRI(2, 0),
REG16(0x200),
REG(0x028),
END(80)
};
static const u8 gen9_xcs_offsets[] = {
NOP(1),
LRI(14, POSTED),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x11c),
REG(0x114),
REG(0x118),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
NOP(3),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
NOP(13),
LRI(1, POSTED),
REG16(0x200),
NOP(13),
LRI(44, POSTED),
REG(0x028),
REG(0x09c),
REG(0x0c0),
REG(0x178),
REG(0x17c),
REG16(0x358),
REG(0x170),
REG(0x150),
REG(0x154),
REG(0x158),
REG16(0x41c),
REG16(0x600),
REG16(0x604),
REG16(0x608),
REG16(0x60c),
REG16(0x610),
REG16(0x614),
REG16(0x618),
REG16(0x61c),
REG16(0x620),
REG16(0x624),
REG16(0x628),
REG16(0x62c),
REG16(0x630),
REG16(0x634),
REG16(0x638),
REG16(0x63c),
REG16(0x640),
REG16(0x644),
REG16(0x648),
REG16(0x64c),
REG16(0x650),
REG16(0x654),
REG16(0x658),
REG16(0x65c),
REG16(0x660),
REG16(0x664),
REG16(0x668),
REG16(0x66c),
REG16(0x670),
REG16(0x674),
REG16(0x678),
REG16(0x67c),
REG(0x068),
END(176)
};
static const u8 gen12_xcs_offsets[] = {
NOP(1),
LRI(13, POSTED),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
REG(0x180),
REG16(0x2b4),
NOP(5),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
END(80)
};
static const u8 gen8_rcs_offsets[] = {
NOP(1),
LRI(14, POSTED),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x11c),
REG(0x114),
REG(0x118),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
NOP(3),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
NOP(13),
LRI(1, 0),
REG(0x0c8),
END(80)
};
static const u8 gen9_rcs_offsets[] = {
NOP(1),
LRI(14, POSTED),
REG16(0x244),
REG(0x34),
REG(0x30),
REG(0x38),
REG(0x3c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x11c),
REG(0x114),
REG(0x118),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
NOP(3),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
NOP(13),
LRI(1, 0),
REG(0xc8),
NOP(13),
LRI(44, POSTED),
REG(0x28),
REG(0x9c),
REG(0xc0),
REG(0x178),
REG(0x17c),
REG16(0x358),
REG(0x170),
REG(0x150),
REG(0x154),
REG(0x158),
REG16(0x41c),
REG16(0x600),
REG16(0x604),
REG16(0x608),
REG16(0x60c),
REG16(0x610),
REG16(0x614),
REG16(0x618),
REG16(0x61c),
REG16(0x620),
REG16(0x624),
REG16(0x628),
REG16(0x62c),
REG16(0x630),
REG16(0x634),
REG16(0x638),
REG16(0x63c),
REG16(0x640),
REG16(0x644),
REG16(0x648),
REG16(0x64c),
REG16(0x650),
REG16(0x654),
REG16(0x658),
REG16(0x65c),
REG16(0x660),
REG16(0x664),
REG16(0x668),
REG16(0x66c),
REG16(0x670),
REG16(0x674),
REG16(0x678),
REG16(0x67c),
REG(0x68),
END(176)
};
static const u8 gen11_rcs_offsets[] = {
NOP(1),
LRI(15, POSTED),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x11c),
REG(0x114),
REG(0x118),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
REG(0x180),
NOP(1),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
LRI(1, POSTED),
REG(0x1b0),
NOP(10),
LRI(1, 0),
REG(0x0c8),
END(80)
};
static const u8 gen12_rcs_offsets[] = {
NOP(1),
LRI(13, POSTED),
REG16(0x244),
REG(0x034),
REG(0x030),
REG(0x038),
REG(0x03c),
REG(0x168),
REG(0x140),
REG(0x110),
REG(0x1c0),
REG(0x1c4),
REG(0x1c8),
REG(0x180),
REG16(0x2b4),
NOP(5),
LRI(9, POSTED),
REG16(0x3a8),
REG16(0x28c),
REG16(0x288),
REG16(0x284),
REG16(0x280),
REG16(0x27c),
REG16(0x278),
REG16(0x274),
REG16(0x270),
LRI(3, POSTED),
REG(0x1b0),
REG16(0x5a8),
REG16(0x5ac),
NOP(6),
LRI(1, 0),
REG(0x0c8),
END(80)
};
#undef END
#undef REG16
#undef REG
#undef LRI
#undef NOP
static const u8 *reg_offsets(const struct intel_engine_cs *engine)
{
/*
* The gen12+ lists only have the registers we program in the basic
* default state. We rely on the context image using relative
* addressing to automatic fixup the register state between the
* physical engines for virtual engine.
*/
GEM_BUG_ON(INTEL_GEN(engine->i915) >= 12 &&
!intel_engine_has_relative_mmio(engine));
if (engine->class == RENDER_CLASS) {
if (INTEL_GEN(engine->i915) >= 12)
return gen12_rcs_offsets;
else if (INTEL_GEN(engine->i915) >= 11)
return gen11_rcs_offsets;
else if (INTEL_GEN(engine->i915) >= 9)
return gen9_rcs_offsets;
else
return gen8_rcs_offsets;
} else {
if (INTEL_GEN(engine->i915) >= 12)
return gen12_xcs_offsets;
else if (INTEL_GEN(engine->i915) >= 9)
return gen9_xcs_offsets;
else
return gen8_xcs_offsets;
}
}
static struct i915_request *
__unwind_incomplete_requests(struct intel_engine_cs *engine)
{
struct i915_request *rq, *rn, *active = NULL;
struct list_head *uninitialized_var(pl);
int prio = I915_PRIORITY_INVALID;
lockdep_assert_held(&engine->active.lock);
list_for_each_entry_safe_reverse(rq, rn,
&engine->active.requests,
sched.link) {
if (i915_request_completed(rq))
continue; /* XXX */
__i915_request_unsubmit(rq);
/*
* Push the request back into the queue for later resubmission.
* If this request is not native to this physical engine (i.e.
* it came from a virtual source), push it back onto the virtual
* engine so that it can be moved across onto another physical
* engine as load dictates.
*/
if (likely(rq->execution_mask == engine->mask)) {
GEM_BUG_ON(rq_prio(rq) == I915_PRIORITY_INVALID);
if (rq_prio(rq) != prio) {
prio = rq_prio(rq);
pl = i915_sched_lookup_priolist(engine, prio);
}
GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root));
list_move(&rq->sched.link, pl);
set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags);
active = rq;
} else {
struct intel_engine_cs *owner = rq->context->engine;
/*
* Decouple the virtual breadcrumb before moving it
* back to the virtual engine -- we don't want the
* request to complete in the background and try
* and cancel the breadcrumb on the virtual engine
* (instead of the old engine where it is linked)!
*/
if (test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
&rq->fence.flags)) {
spin_lock_nested(&rq->lock,
SINGLE_DEPTH_NESTING);
i915_request_cancel_breadcrumb(rq);
spin_unlock(&rq->lock);
}
rq->engine = owner;
owner->submit_request(rq);
active = NULL;
}
}
return active;
}
struct i915_request *
execlists_unwind_incomplete_requests(struct intel_engine_execlists *execlists)
{
struct intel_engine_cs *engine =
container_of(execlists, typeof(*engine), execlists);
return __unwind_incomplete_requests(engine);
}
static inline void
execlists_context_status_change(struct i915_request *rq, unsigned long status)
{
/*
* Only used when GVT-g is enabled now. When GVT-g is disabled,
* The compiler should eliminate this function as dead-code.
*/
if (!IS_ENABLED(CONFIG_DRM_I915_GVT))
return;
atomic_notifier_call_chain(&rq->engine->context_status_notifier,
status, rq);
}
static void intel_engine_context_in(struct intel_engine_cs *engine)
{
unsigned long flags;
if (READ_ONCE(engine->stats.enabled) == 0)
return;
write_seqlock_irqsave(&engine->stats.lock, flags);
if (engine->stats.enabled > 0) {
if (engine->stats.active++ == 0)
engine->stats.start = ktime_get();
GEM_BUG_ON(engine->stats.active == 0);
}
write_sequnlock_irqrestore(&engine->stats.lock, flags);
}
static void intel_engine_context_out(struct intel_engine_cs *engine)
{
unsigned long flags;
if (READ_ONCE(engine->stats.enabled) == 0)
return;
write_seqlock_irqsave(&engine->stats.lock, flags);
if (engine->stats.enabled > 0) {
ktime_t last;
if (engine->stats.active && --engine->stats.active == 0) {
/*
* Decrement the active context count and in case GPU
* is now idle add up to the running total.
*/
last = ktime_sub(ktime_get(), engine->stats.start);
engine->stats.total = ktime_add(engine->stats.total,
last);
} else if (engine->stats.active == 0) {
/*
* After turning on engine stats, context out might be
* the first event in which case we account from the
* time stats gathering was turned on.
*/
last = ktime_sub(ktime_get(), engine->stats.enabled_at);
engine->stats.total = ktime_add(engine->stats.total,
last);
}
}
write_sequnlock_irqrestore(&engine->stats.lock, flags);
}
static int lrc_ring_mi_mode(const struct intel_engine_cs *engine)
{
if (INTEL_GEN(engine->i915) >= 12)
return 0x60;
else if (INTEL_GEN(engine->i915) >= 9)
return 0x54;
else if (engine->class == RENDER_CLASS)
return 0x58;
else
return -1;
}
static void
execlists_check_context(const struct intel_context *ce,
const struct intel_engine_cs *engine)
{
const struct intel_ring *ring = ce->ring;
u32 *regs = ce->lrc_reg_state;
bool valid = true;
int x;
if (regs[CTX_RING_START] != i915_ggtt_offset(ring->vma)) {
pr_err("%s: context submitted with incorrect RING_START [%08x], expected %08x\n",
engine->name,
regs[CTX_RING_START],
i915_ggtt_offset(ring->vma));
regs[CTX_RING_START] = i915_ggtt_offset(ring->vma);
valid = false;
}
if ((regs[CTX_RING_CTL] & ~(RING_WAIT | RING_WAIT_SEMAPHORE)) !=
(RING_CTL_SIZE(ring->size) | RING_VALID)) {
pr_err("%s: context submitted with incorrect RING_CTL [%08x], expected %08x\n",
engine->name,
regs[CTX_RING_CTL],
(u32)(RING_CTL_SIZE(ring->size) | RING_VALID));
regs[CTX_RING_CTL] = RING_CTL_SIZE(ring->size) | RING_VALID;
valid = false;
}
x = lrc_ring_mi_mode(engine);
if (x != -1 && regs[x + 1] & (regs[x + 1] >> 16) & STOP_RING) {
pr_err("%s: context submitted with STOP_RING [%08x] in RING_MI_MODE\n",
engine->name, regs[x + 1]);
regs[x + 1] &= ~STOP_RING;
regs[x + 1] |= STOP_RING << 16;
valid = false;
}
WARN_ONCE(!valid, "Invalid lrc state found before submission\n");
}
static void restore_default_state(struct intel_context *ce,
struct intel_engine_cs *engine)
{
u32 *regs = ce->lrc_reg_state;
if (engine->pinned_default_state)
memcpy(regs, /* skip restoring the vanilla PPHWSP */
engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE,
engine->context_size - PAGE_SIZE);
execlists_init_reg_state(regs, ce, engine, ce->ring, false);
}
static void reset_active(struct i915_request *rq,
struct intel_engine_cs *engine)
{
struct intel_context * const ce = rq->context;
u32 head;
/*
* The executing context has been cancelled. We want to prevent
* further execution along this context and propagate the error on
* to anything depending on its results.
*
* In __i915_request_submit(), we apply the -EIO and remove the
* requests' payloads for any banned requests. But first, we must
* rewind the context back to the start of the incomplete request so
* that we do not jump back into the middle of the batch.
*
* We preserve the breadcrumbs and semaphores of the incomplete
* requests so that inter-timeline dependencies (i.e other timelines)
* remain correctly ordered. And we defer to __i915_request_submit()
* so that all asynchronous waits are correctly handled.
*/
ENGINE_TRACE(engine, "{ rq=%llx:%lld }\n",
rq->fence.context, rq->fence.seqno);
/* On resubmission of the active request, payload will be scrubbed */
if (i915_request_completed(rq))
head = rq->tail;
else
head = active_request(ce->timeline, rq)->head;
head = intel_ring_wrap(ce->ring, head);
/* Scrub the context image to prevent replaying the previous batch */
restore_default_state(ce, engine);
__execlists_update_reg_state(ce, engine, head);
/* We've switched away, so this should be a no-op, but intent matters */
ce->lrc_desc |= CTX_DESC_FORCE_RESTORE;
}
static inline struct intel_engine_cs *
__execlists_schedule_in(struct i915_request *rq)
{
struct intel_engine_cs * const engine = rq->engine;
struct intel_context * const ce = rq->context;
intel_context_get(ce);
if (unlikely(intel_context_is_banned(ce)))
reset_active(rq, engine);
if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
execlists_check_context(ce, engine);
if (ce->tag) {
/* Use a fixed tag for OA and friends */
ce->lrc_desc |= (u64)ce->tag << 32;
} else {
/* We don't need a strict matching tag, just different values */
ce->lrc_desc &= ~GENMASK_ULL(47, 37);
ce->lrc_desc |=
(u64)(++engine->context_tag % NUM_CONTEXT_TAG) <<
GEN11_SW_CTX_ID_SHIFT;
BUILD_BUG_ON(NUM_CONTEXT_TAG > GEN12_MAX_CONTEXT_HW_ID);
}
__intel_gt_pm_get(engine->gt);
execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_IN);
intel_engine_context_in(engine);
return engine;
}
static inline struct i915_request *
execlists_schedule_in(struct i915_request *rq, int idx)
{
struct intel_context * const ce = rq->context;
struct intel_engine_cs *old;
GEM_BUG_ON(!intel_engine_pm_is_awake(rq->engine));
trace_i915_request_in(rq, idx);
old = READ_ONCE(ce->inflight);
do {
if (!old) {
WRITE_ONCE(ce->inflight, __execlists_schedule_in(rq));
break;
}
} while (!try_cmpxchg(&ce->inflight, &old, ptr_inc(old)));
GEM_BUG_ON(intel_context_inflight(ce) != rq->engine);
return i915_request_get(rq);
}
static void kick_siblings(struct i915_request *rq, struct intel_context *ce)
{
struct virtual_engine *ve = container_of(ce, typeof(*ve), context);
struct i915_request *next = READ_ONCE(ve->request);
if (next && next->execution_mask & ~rq->execution_mask)
tasklet_schedule(&ve->base.execlists.tasklet);
}
static inline void
__execlists_schedule_out(struct i915_request *rq,
struct intel_engine_cs * const engine)
{
struct intel_context * const ce = rq->context;
/*
* NB process_csb() is not under the engine->active.lock and hence
* schedule_out can race with schedule_in meaning that we should
* refrain from doing non-trivial work here.
*/
/*
* If we have just completed this context, the engine may now be
* idle and we want to re-enter powersaving.
*/
if (list_is_last(&rq->link, &ce->timeline->requests) &&
i915_request_completed(rq))
intel_engine_add_retire(engine, ce->timeline);
intel_engine_context_out(engine);
execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT);
intel_gt_pm_put_async(engine->gt);
/*
* If this is part of a virtual engine, its next request may
* have been blocked waiting for access to the active context.
* We have to kick all the siblings again in case we need to
* switch (e.g. the next request is not runnable on this
* engine). Hopefully, we will already have submitted the next
* request before the tasklet runs and do not need to rebuild
* each virtual tree and kick everyone again.
*/
if (ce->engine != engine)
kick_siblings(rq, ce);
intel_context_put(ce);
}
static inline void
execlists_schedule_out(struct i915_request *rq)
{
struct intel_context * const ce = rq->context;
struct intel_engine_cs *cur, *old;
trace_i915_request_out(rq);
old = READ_ONCE(ce->inflight);
do
cur = ptr_unmask_bits(old, 2) ? ptr_dec(old) : NULL;
while (!try_cmpxchg(&ce->inflight, &old, cur));
if (!cur)
__execlists_schedule_out(rq, old);
i915_request_put(rq);
}
static u64 execlists_update_context(struct i915_request *rq)
{
struct intel_context *ce = rq->context;
u64 desc = ce->lrc_desc;
u32 tail, prev;
/*
* WaIdleLiteRestore:bdw,skl
*
* We should never submit the context with the same RING_TAIL twice
* just in case we submit an empty ring, which confuses the HW.
*
* We append a couple of NOOPs (gen8_emit_wa_tail) after the end of
* the normal request to be able to always advance the RING_TAIL on
* subsequent resubmissions (for lite restore). Should that fail us,
* and we try and submit the same tail again, force the context
* reload.
*
* If we need to return to a preempted context, we need to skip the
* lite-restore and force it to reload the RING_TAIL. Otherwise, the
* HW has a tendency to ignore us rewinding the TAIL to the end of
* an earlier request.
*/
tail = intel_ring_set_tail(rq->ring, rq->tail);
prev = ce->lrc_reg_state[CTX_RING_TAIL];
if (unlikely(intel_ring_direction(rq->ring, tail, prev) <= 0))
desc |= CTX_DESC_FORCE_RESTORE;
ce->lrc_reg_state[CTX_RING_TAIL] = tail;
rq->tail = rq->wa_tail;
/*
* Make sure the context image is complete before we submit it to HW.
*
* Ostensibly, writes (including the WCB) should be flushed prior to
* an uncached write such as our mmio register access, the empirical
* evidence (esp. on Braswell) suggests that the WC write into memory
* may not be visible to the HW prior to the completion of the UC
* register write and that we may begin execution from the context
* before its image is complete leading to invalid PD chasing.
*/
wmb();
ce->lrc_desc &= ~CTX_DESC_FORCE_RESTORE;
return desc;
}
static inline void write_desc(struct intel_engine_execlists *execlists, u64 desc, u32 port)
{
if (execlists->ctrl_reg) {
writel(lower_32_bits(desc), execlists->submit_reg + port * 2);
writel(upper_32_bits(desc), execlists->submit_reg + port * 2 + 1);
} else {
writel(upper_32_bits(desc), execlists->submit_reg);
writel(lower_32_bits(desc), execlists->submit_reg);
}
}
static __maybe_unused void
trace_ports(const struct intel_engine_execlists *execlists,
const char *msg,
struct i915_request * const *ports)
{
const struct intel_engine_cs *engine =
container_of(execlists, typeof(*engine), execlists);
if (!ports[0])
return;
ENGINE_TRACE(engine, "%s { %llx:%lld%s, %llx:%lld }\n", msg,
ports[0]->fence.context,
ports[0]->fence.seqno,
i915_request_completed(ports[0]) ? "!" :
i915_request_started(ports[0]) ? "*" :
"",
ports[1] ? ports[1]->fence.context : 0,
ports[1] ? ports[1]->fence.seqno : 0);
}
static __maybe_unused bool
assert_pending_valid(const struct intel_engine_execlists *execlists,
const char *msg)
{
struct i915_request * const *port, *rq;
struct intel_context *ce = NULL;
trace_ports(execlists, msg, execlists->pending);
if (!execlists->pending[0]) {
GEM_TRACE_ERR("Nothing pending for promotion!\n");
return false;
}
if (execlists->pending[execlists_num_ports(execlists)]) {
GEM_TRACE_ERR("Excess pending[%d] for promotion!\n",
execlists_num_ports(execlists));
return false;
}
for (port = execlists->pending; (rq = *port); port++) {
unsigned long flags;
bool ok = true;
GEM_BUG_ON(!kref_read(&rq->fence.refcount));
GEM_BUG_ON(!i915_request_is_active(rq));
if (ce == rq->context) {
GEM_TRACE_ERR("Dup context:%llx in pending[%zd]\n",
ce->timeline->fence_context,
port - execlists->pending);
return false;
}
ce = rq->context;
/* Hold tightly onto the lock to prevent concurrent retires! */
if (!spin_trylock_irqsave(&rq->lock, flags))
continue;
if (i915_request_completed(rq))
goto unlock;
if (i915_active_is_idle(&ce->active) &&
!intel_context_is_barrier(ce)) {
GEM_TRACE_ERR("Inactive context:%llx in pending[%zd]\n",
ce->timeline->fence_context,
port - execlists->pending);
ok = false;
goto unlock;
}
if (!i915_vma_is_pinned(ce->state)) {
GEM_TRACE_ERR("Unpinned context:%llx in pending[%zd]\n",
ce->timeline->fence_context,
port - execlists->pending);
ok = false;
goto unlock;
}
if (!i915_vma_is_pinned(ce->ring->vma)) {
GEM_TRACE_ERR("Unpinned ring:%llx in pending[%zd]\n",
ce->timeline->fence_context,
port - execlists->pending);
ok = false;
goto unlock;
}
unlock:
spin_unlock_irqrestore(&rq->lock, flags);
if (!ok)
return false;
}
return ce;
}
static void execlists_submit_ports(struct intel_engine_cs *engine)
{
struct intel_engine_execlists *execlists = &engine->execlists;
unsigned int n;
GEM_BUG_ON(!assert_pending_valid(execlists, "submit"));
/*
* We can skip acquiring intel_runtime_pm_get() here as it was taken
* on our behalf by the request (see i915_gem_mark_busy()) and it will
* not be relinquished until the device is idle (see
* i915_gem_idle_work_handler()). As a precaution, we make sure
* that all ELSP are drained i.e. we have processed the CSB,
* before allowing ourselves to idle and calling intel_runtime_pm_put().
*/
GEM_BUG_ON(!intel_engine_pm_is_awake(engine));
/*
* ELSQ note: the submit queue is not cleared after being submitted
* to the HW so we need to make sure we always clean it up. This is
* currently ensured by the fact that we always write the same number
* of elsq entries, keep this in mind before changing the loop below.
*/
for (n = execlists_num_ports(execlists); n--; ) {
struct i915_request *rq = execlists->pending[n];
write_desc(execlists,
rq ? execlists_update_context(rq) : 0,
n);
}
/* we need to manually load the submit queue */
if (execlists->ctrl_reg)
writel(EL_CTRL_LOAD, execlists->ctrl_reg);
}
static bool ctx_single_port_submission(const struct intel_context *ce)
{
return (IS_ENABLED(CONFIG_DRM_I915_GVT) &&
intel_context_force_single_submission(ce));
}
static bool can_merge_ctx(const struct intel_context *prev,
const struct intel_context *next)
{
if (prev != next)
return false;
if (ctx_single_port_submission(prev))
return false;
return true;
}
static bool can_merge_rq(const struct i915_request *prev,
const struct i915_request *next)
{
GEM_BUG_ON(prev == next);
GEM_BUG_ON(!assert_priority_queue(prev, next));
/*
* We do not submit known completed requests. Therefore if the next
* request is already completed, we can pretend to merge it in
* with the previous context (and we will skip updating the ELSP
* and tracking). Thus hopefully keeping the ELSP full with active
* contexts, despite the best efforts of preempt-to-busy to confuse
* us.
*/
if (i915_request_completed(next))
return true;
if (unlikely((prev->fence.flags ^ next->fence.flags) &
(BIT(I915_FENCE_FLAG_NOPREEMPT) |
BIT(I915_FENCE_FLAG_SENTINEL))))
return false;
if (!can_merge_ctx(prev->context, next->context))
return false;
return true;
}
static void virtual_update_register_offsets(u32 *regs,
struct intel_engine_cs *engine)
{
set_offsets(regs, reg_offsets(engine), engine, false);
}
static bool virtual_matches(const struct virtual_engine *ve,
const struct i915_request *rq,
const struct intel_engine_cs *engine)
{
const struct intel_engine_cs *inflight;
if (!(rq->execution_mask & engine->mask)) /* We peeked too soon! */
return false;
/*
* We track when the HW has completed saving the context image
* (i.e. when we have seen the final CS event switching out of
* the context) and must not overwrite the context image before
* then. This restricts us to only using the active engine
* while the previous virtualized request is inflight (so
* we reuse the register offsets). This is a very small
* hystersis on the greedy seelction algorithm.
*/
inflight = intel_context_inflight(&ve->context);
if (inflight && inflight != engine)
return false;
return true;
}
static void virtual_xfer_breadcrumbs(struct virtual_engine *ve,
struct intel_engine_cs *engine)
{
struct intel_engine_cs *old = ve->siblings[0];
/* All unattached (rq->engine == old) must already be completed */
spin_lock(&old->breadcrumbs.irq_lock);
if (!list_empty(&ve->context.signal_link)) {
list_move_tail(&ve->context.signal_link,
&engine->breadcrumbs.signalers);
intel_engine_signal_breadcrumbs(engine);
}
spin_unlock(&old->breadcrumbs.irq_lock);
}
static struct i915_request *
last_active(const struct intel_engine_execlists *execlists)
{
struct i915_request * const *last = READ_ONCE(execlists->active);
while (*last && i915_request_completed(*last))
last++;
return *last;
}
#define for_each_waiter(p__, rq__) \
list_for_each_entry_lockless(p__, \
&(rq__)->sched.waiters_list, \
wait_link)
static void defer_request(struct i915_request *rq, struct list_head * const pl)
{
LIST_HEAD(list);
/*
* We want to move the interrupted request to the back of
* the round-robin list (i.e. its priority level), but
* in doing so, we must then move all requests that were in
* flight and were waiting for the interrupted request to
* be run after it again.
*/
do {
struct i915_dependency *p;
GEM_BUG_ON(i915_request_is_active(rq));
list_move_tail(&rq->sched.link, pl);
for_each_waiter(p, rq) {
struct i915_request *w =
container_of(p->waiter, typeof(*w), sched);
/* Leave semaphores spinning on the other engines */
if (w->engine != rq->engine)
continue;
/* No waiter should start before its signaler */
GEM_BUG_ON(i915_request_started(w) &&
!i915_request_completed(rq));
GEM_BUG_ON(i915_request_is_active(w));
if (!i915_request_is_ready(w))
continue;
if (rq_prio(w) < rq_prio(rq))
continue;
GEM_BUG_ON(rq_prio(w) > rq_prio(rq));
list_move_tail(&w->sched.link, &list);
}
rq = list_first_entry_or_null(&list, typeof(*rq), sched.link);
} while (rq);
}
static void defer_active(struct intel_engine_cs *engine)
{
struct i915_request *rq;
rq = __unwind_incomplete_requests(engine);
if (!rq)
return;
defer_request(rq, i915_sched_lookup_priolist(engine, rq_prio(rq)));
}
static bool
need_timeslice(struct intel_engine_cs *engine, const struct i915_request *rq)
{
int hint;
if (!intel_engine_has_timeslices(engine))
return false;
hint = engine->execlists.queue_priority_hint;
if (!list_is_last(&rq->sched.link, &engine->active.requests))
hint = max(hint, rq_prio(list_next_entry(rq, sched.link)));
return hint >= effective_prio(rq);
}
static int
switch_prio(struct intel_engine_cs *engine, const struct i915_request *rq)
{
if (list_is_last(&rq->sched.link, &engine->active.requests))
return INT_MIN;
return rq_prio(list_next_entry(rq, sched.link));
}
static inline unsigned long
timeslice(const struct intel_engine_cs *engine)
{
return READ_ONCE(engine->props.timeslice_duration_ms);
}
static unsigned long
active_timeslice(const struct intel_engine_cs *engine)
{
const struct i915_request *rq = *engine->execlists.active;
if (!rq || i915_request_completed(rq))
return 0;
if (engine->execlists.switch_priority_hint < effective_prio(rq))
return 0;
return timeslice(engine);
}
static void set_timeslice(struct intel_engine_cs *engine)
{
if (!intel_engine_has_timeslices(engine))
return;
set_timer_ms(&engine->execlists.timer, active_timeslice(engine));
}
static void start_timeslice(struct intel_engine_cs *engine)
{
struct intel_engine_execlists *execlists = &engine->execlists;
execlists->switch_priority_hint = execlists->queue_priority_hint;
if (timer_pending(&execlists->timer))
return;
set_timer_ms(&execlists->timer, timeslice(engine));
}
static void record_preemption(struct intel_engine_execlists *execlists)
{
(void)I915_SELFTEST_ONLY(execlists->preempt_hang.count++);
}
static unsigned long active_preempt_timeout(struct intel_engine_cs *engine)
{
struct i915_request *rq;
rq = last_active(&engine->execlists);
if (!rq)
return 0;
/* Force a fast reset for terminated contexts (ignoring sysfs!) */
if (unlikely(intel_context_is_banned(rq->context)))
return 1;
return READ_ONCE(engine->props.preempt_timeout_ms);
}
static void set_preempt_timeout(struct intel_engine_cs *engine)
{
if (!intel_engine_has_preempt_reset(engine))
return;
set_timer_ms(&engine->execlists.preempt,
active_preempt_timeout(engine));
}
static inline void clear_ports(struct i915_request **ports, int count)
{
memset_p((void **)ports, NULL, count);
}
static void execlists_dequeue(struct intel_engine_cs *engine)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
struct i915_request **port = execlists->pending;
struct i915_request ** const last_port = port + execlists->port_mask;
struct i915_request *last;
struct rb_node *rb;
bool submit = false;
/*
* Hardware submission is through 2 ports. Conceptually each port
* has a (RING_START, RING_HEAD, RING_TAIL) tuple. RING_START is
* static for a context, and unique to each, so we only execute
* requests belonging to a single context from each ring. RING_HEAD
* is maintained by the CS in the context image, it marks the place
* where it got up to last time, and through RING_TAIL we tell the CS
* where we want to execute up to this time.
*
* In this list the requests are in order of execution. Consecutive
* requests from the same context are adjacent in the ringbuffer. We
* can combine these requests into a single RING_TAIL update:
*
* RING_HEAD...req1...req2
* ^- RING_TAIL
* since to execute req2 the CS must first execute req1.
*
* Our goal then is to point each port to the end of a consecutive
* sequence of requests as being the most optimal (fewest wake ups
* and context switches) submission.
*/
for (rb = rb_first_cached(&execlists->virtual); rb; ) {
struct virtual_engine *ve =
rb_entry(rb, typeof(*ve), nodes[engine->id].rb);
struct i915_request *rq = READ_ONCE(ve->request);
if (!rq) { /* lazily cleanup after another engine handled rq */
rb_erase_cached(rb, &execlists->virtual);
RB_CLEAR_NODE(rb);
rb = rb_first_cached(&execlists->virtual);
continue;
}
if (!virtual_matches(ve, rq, engine)) {
rb = rb_next(rb);
continue;
}
break;
}
/*
* If the queue is higher priority than the last
* request in the currently active context, submit afresh.
* We will resubmit again afterwards in case we need to split
* the active context to interject the preemption request,
* i.e. we will retrigger preemption following the ack in case
* of trouble.
*/
last = last_active(execlists);
if (last) {
if (need_preempt(engine, last, rb)) {
ENGINE_TRACE(engine,
"preempting last=%llx:%lld, prio=%d, hint=%d\n",
last->fence.context,
last->fence.seqno,
last->sched.attr.priority,
execlists->queue_priority_hint);
record_preemption(execlists);
/*
* Don't let the RING_HEAD advance past the breadcrumb
* as we unwind (and until we resubmit) so that we do
* not accidentally tell it to go backwards.
*/
ring_set_paused(engine, 1);
/*
* Note that we have not stopped the GPU at this point,
* so we are unwinding the incomplete requests as they
* remain inflight and so by the time we do complete
* the preemption, some of the unwound requests may
* complete!
*/
__unwind_incomplete_requests(engine);
last = NULL;
} else if (need_timeslice(engine, last) &&
timer_expired(&engine->execlists.timer)) {
ENGINE_TRACE(engine,
"expired last=%llx:%lld, prio=%d, hint=%d\n",
last->fence.context,
last->fence.seqno,
last->sched.attr.priority,
execlists->queue_priority_hint);
ring_set_paused(engine, 1);
defer_active(engine);
/*
* Unlike for preemption, if we rewind and continue
* executing the same context as previously active,
* the order of execution will remain the same and
* the tail will only advance. We do not need to
* force a full context restore, as a lite-restore
* is sufficient to resample the monotonic TAIL.
*
* If we switch to any other context, similarly we
* will not rewind TAIL of current context, and
* normal save/restore will preserve state and allow
* us to later continue executing the same request.
*/
last = NULL;
} else {
/*
* Otherwise if we already have a request pending
* for execution after the current one, we can
* just wait until the next CS event before
* queuing more. In either case we will force a
* lite-restore preemption event, but if we wait
* we hopefully coalesce several updates into a single
* submission.
*/
if (!list_is_last(&last->sched.link,
&engine->active.requests)) {
/*
* Even if ELSP[1] is occupied and not worthy
* of timeslices, our queue might be.
*/
start_timeslice(engine);
return;
}
}
}
while (rb) { /* XXX virtual is always taking precedence */
struct virtual_engine *ve =
rb_entry(rb, typeof(*ve), nodes[engine->id].rb);
struct i915_request *rq;
spin_lock(&ve->base.active.lock);
rq = ve->request;
if (unlikely(!rq)) { /* lost the race to a sibling */
spin_unlock(&ve->base.active.lock);
rb_erase_cached(rb, &execlists->virtual);
RB_CLEAR_NODE(rb);
rb = rb_first_cached(&execlists->virtual);
continue;
}
GEM_BUG_ON(rq != ve->request);
GEM_BUG_ON(rq->engine != &ve->base);
GEM_BUG_ON(rq->context != &ve->context);
if (rq_prio(rq) >= queue_prio(execlists)) {
if (!virtual_matches(ve, rq, engine)) {
spin_unlock(&ve->base.active.lock);
rb = rb_next(rb);
continue;
}
if (last && !can_merge_rq(last, rq)) {
spin_unlock(&ve->base.active.lock);
start_timeslice(engine);
return; /* leave this for another sibling */
}
ENGINE_TRACE(engine,
"virtual rq=%llx:%lld%s, new engine? %s\n",
rq->fence.context,
rq->fence.seqno,
i915_request_completed(rq) ? "!" :
i915_request_started(rq) ? "*" :
"",
yesno(engine != ve->siblings[0]));
ve->request = NULL;
ve->base.execlists.queue_priority_hint = INT_MIN;
rb_erase_cached(rb, &execlists->virtual);
RB_CLEAR_NODE(rb);
GEM_BUG_ON(!(rq->execution_mask & engine->mask));
rq->engine = engine;
if (engine != ve->siblings[0]) {
u32 *regs = ve->context.lrc_reg_state;
unsigned int n;
GEM_BUG_ON(READ_ONCE(ve->context.inflight));
if (!intel_engine_has_relative_mmio(engine))
virtual_update_register_offsets(regs,
engine);
if (!list_empty(&ve->context.signals))
virtual_xfer_breadcrumbs(ve, engine);
/*
* Move the bound engine to the top of the list
* for future execution. We then kick this
* tasklet first before checking others, so that
* we preferentially reuse this set of bound
* registers.
*/
for (n = 1; n < ve->num_siblings; n++) {
if (ve->siblings[n] == engine) {
swap(ve->siblings[n],
ve->siblings[0]);
break;
}
}
GEM_BUG_ON(ve->siblings[0] != engine);
}
if (__i915_request_submit(rq)) {
submit = true;
last = rq;
}
i915_request_put(rq);
/*
* Hmm, we have a bunch of virtual engine requests,
* but the first one was already completed (thanks
* preempt-to-busy!). Keep looking at the veng queue
* until we have no more relevant requests (i.e.
* the normal submit queue has higher priority).
*/
if (!submit) {
spin_unlock(&ve->base.active.lock);
rb = rb_first_cached(&execlists->virtual);
continue;
}
}
spin_unlock(&ve->base.active.lock);
break;
}
while ((rb = rb_first_cached(&execlists->queue))) {
struct i915_priolist *p = to_priolist(rb);
struct i915_request *rq, *rn;
int i;
priolist_for_each_request_consume(rq, rn, p, i) {
bool merge = true;
/*
* Can we combine this request with the current port?
* It has to be the same context/ringbuffer and not
* have any exceptions (e.g. GVT saying never to
* combine contexts).
*
* If we can combine the requests, we can execute both
* by updating the RING_TAIL to point to the end of the
* second request, and so we never need to tell the
* hardware about the first.
*/
if (last && !can_merge_rq(last, rq)) {
/*
* If we are on the second port and cannot
* combine this request with the last, then we
* are done.
*/
if (port == last_port)
goto done;
/*
* We must not populate both ELSP[] with the
* same LRCA, i.e. we must submit 2 different
* contexts if we submit 2 ELSP.
*/
if (last->context == rq->context)
goto done;
if (i915_request_has_sentinel(last))
goto done;
/*
* If GVT overrides us we only ever submit
* port[0], leaving port[1] empty. Note that we
* also have to be careful that we don't queue
* the same context (even though a different
* request) to the second port.
*/
if (ctx_single_port_submission(last->context) ||
ctx_single_port_submission(rq->context))
goto done;
merge = false;
}
if (__i915_request_submit(rq)) {
if (!merge) {
*port = execlists_schedule_in(last, port - execlists->pending);
port++;
last = NULL;
}
GEM_BUG_ON(last &&
!can_merge_ctx(last->context,
rq->context));
submit = true;
last = rq;
}
}
rb_erase_cached(&p->node, &execlists->queue);
i915_priolist_free(p);
}
done:
/*
* Here be a bit of magic! Or sleight-of-hand, whichever you prefer.
*
* We choose the priority hint such that if we add a request of greater
* priority than this, we kick the submission tasklet to decide on
* the right order of submitting the requests to hardware. We must
* also be prepared to reorder requests as they are in-flight on the
* HW. We derive the priority hint then as the first "hole" in
* the HW submission ports and if there are no available slots,
* the priority of the lowest executing request, i.e. last.
*
* When we do receive a higher priority request ready to run from the
* user, see queue_request(), the priority hint is bumped to that
* request triggering preemption on the next dequeue (or subsequent
* interrupt for secondary ports).
*/
execlists->queue_priority_hint = queue_prio(execlists);
if (submit) {
*port = execlists_schedule_in(last, port - execlists->pending);
execlists->switch_priority_hint =
switch_prio(engine, *execlists->pending);
/*
* Skip if we ended up with exactly the same set of requests,
* e.g. trying to timeslice a pair of ordered contexts
*/
if (!memcmp(execlists->active, execlists->pending,
(port - execlists->pending + 1) * sizeof(*port))) {
do
execlists_schedule_out(fetch_and_zero(port));
while (port-- != execlists->pending);
goto skip_submit;
}
clear_ports(port + 1, last_port - port);
execlists_submit_ports(engine);
set_preempt_timeout(engine);
} else {
skip_submit:
ring_set_paused(engine, 0);
}
}
static void
cancel_port_requests(struct intel_engine_execlists * const execlists)
{
struct i915_request * const *port;
for (port = execlists->pending; *port; port++)
execlists_schedule_out(*port);
clear_ports(execlists->pending, ARRAY_SIZE(execlists->pending));
/* Mark the end of active before we overwrite *active */
for (port = xchg(&execlists->active, execlists->pending); *port; port++)
execlists_schedule_out(*port);
clear_ports(execlists->inflight, ARRAY_SIZE(execlists->inflight));
WRITE_ONCE(execlists->active, execlists->inflight);
}
static inline void
invalidate_csb_entries(const u32 *first, const u32 *last)
{
clflush((void *)first);
clflush((void *)last);
}
static inline bool
reset_in_progress(const struct intel_engine_execlists *execlists)
{
return unlikely(!__tasklet_is_enabled(&execlists->tasklet));
}
/*
* Starting with Gen12, the status has a new format:
*
* bit 0: switched to new queue
* bit 1: reserved
* bit 2: semaphore wait mode (poll or signal), only valid when
* switch detail is set to "wait on semaphore"
* bits 3-5: engine class
* bits 6-11: engine instance
* bits 12-14: reserved
* bits 15-25: sw context id of the lrc the GT switched to
* bits 26-31: sw counter of the lrc the GT switched to
* bits 32-35: context switch detail
* - 0: ctx complete
* - 1: wait on sync flip
* - 2: wait on vblank
* - 3: wait on scanline
* - 4: wait on semaphore
* - 5: context preempted (not on SEMAPHORE_WAIT or
* WAIT_FOR_EVENT)
* bit 36: reserved
* bits 37-43: wait detail (for switch detail 1 to 4)
* bits 44-46: reserved
* bits 47-57: sw context id of the lrc the GT switched away from
* bits 58-63: sw counter of the lrc the GT switched away from
*/
static inline bool
gen12_csb_parse(const struct intel_engine_execlists *execlists, const u32 *csb)
{
u32 lower_dw = csb[0];
u32 upper_dw = csb[1];
bool ctx_to_valid = GEN12_CSB_CTX_VALID(lower_dw);
bool ctx_away_valid = GEN12_CSB_CTX_VALID(upper_dw);
bool new_queue = lower_dw & GEN12_CTX_STATUS_SWITCHED_TO_NEW_QUEUE;
/*
* The context switch detail is not guaranteed to be 5 when a preemption
* occurs, so we can't just check for that. The check below works for
* all the cases we care about, including preemptions of WAIT
* instructions and lite-restore. Preempt-to-idle via the CTRL register
* would require some extra handling, but we don't support that.
*/
if (!ctx_away_valid || new_queue) {
GEM_BUG_ON(!ctx_to_valid);
return true;
}
/*
* switch detail = 5 is covered by the case above and we do not expect a
* context switch on an unsuccessful wait instruction since we always
* use polling mode.
*/
GEM_BUG_ON(GEN12_CTX_SWITCH_DETAIL(upper_dw));
return false;
}
static inline bool
gen8_csb_parse(const struct intel_engine_execlists *execlists, const u32 *csb)
{
return *csb & (GEN8_CTX_STATUS_IDLE_ACTIVE | GEN8_CTX_STATUS_PREEMPTED);
}
static void process_csb(struct intel_engine_cs *engine)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
const u32 * const buf = execlists->csb_status;
const u8 num_entries = execlists->csb_size;
u8 head, tail;
/*
* As we modify our execlists state tracking we require exclusive
* access. Either we are inside the tasklet, or the tasklet is disabled
* and we assume that is only inside the reset paths and so serialised.
*/
GEM_BUG_ON(!tasklet_is_locked(&execlists->tasklet) &&
!reset_in_progress(execlists));
GEM_BUG_ON(!intel_engine_in_execlists_submission_mode(engine));
/*
* Note that csb_write, csb_status may be either in HWSP or mmio.
* When reading from the csb_write mmio register, we have to be
* careful to only use the GEN8_CSB_WRITE_PTR portion, which is
* the low 4bits. As it happens we know the next 4bits are always
* zero and so we can simply masked off the low u8 of the register
* and treat it identically to reading from the HWSP (without having
* to use explicit shifting and masking, and probably bifurcating
* the code to handle the legacy mmio read).
*/
head = execlists->csb_head;
tail = READ_ONCE(*execlists->csb_write);
ENGINE_TRACE(engine, "cs-irq head=%d, tail=%d\n", head, tail);
if (unlikely(head == tail))
return;
/*
* Hopefully paired with a wmb() in HW!
*
* We must complete the read of the write pointer before any reads
* from the CSB, so that we do not see stale values. Without an rmb
* (lfence) the HW may speculatively perform the CSB[] reads *before*
* we perform the READ_ONCE(*csb_write).
*/
rmb();
do {
bool promote;
if (++head == num_entries)
head = 0;
/*
* We are flying near dragons again.
*
* We hold a reference to the request in execlist_port[]
* but no more than that. We are operating in softirq
* context and so cannot hold any mutex or sleep. That
* prevents us stopping the requests we are processing
* in port[] from being retired simultaneously (the
* breadcrumb will be complete before we see the
* context-switch). As we only hold the reference to the
* request, any pointer chasing underneath the request
* is subject to a potential use-after-free. Thus we
* store all of the bookkeeping within port[] as
* required, and avoid using unguarded pointers beneath
* request itself. The same applies to the atomic
* status notifier.
*/
ENGINE_TRACE(engine, "csb[%d]: status=0x%08x:0x%08x\n",
head, buf[2 * head + 0], buf[2 * head + 1]);
if (INTEL_GEN(engine->i915) >= 12)
promote = gen12_csb_parse(execlists, buf + 2 * head);
else
promote = gen8_csb_parse(execlists, buf + 2 * head);
if (promote) {
struct i915_request * const *old = execlists->active;
/* Point active to the new ELSP; prevent overwriting */
WRITE_ONCE(execlists->active, execlists->pending);
if (!inject_preempt_hang(execlists))
ring_set_paused(engine, 0);
/* cancel old inflight, prepare for switch */
trace_ports(execlists, "preempted", old);
while (*old)
execlists_schedule_out(*old++);
/* switch pending to inflight */
GEM_BUG_ON(!assert_pending_valid(execlists, "promote"));
WRITE_ONCE(execlists->active,
memcpy(execlists->inflight,
execlists->pending,
execlists_num_ports(execlists) *
sizeof(*execlists->pending)));
WRITE_ONCE(execlists->pending[0], NULL);
} else {
GEM_BUG_ON(!*execlists->active);
/* port0 completed, advanced to port1 */
trace_ports(execlists, "completed", execlists->active);
/*
* We rely on the hardware being strongly
* ordered, that the breadcrumb write is
* coherent (visible from the CPU) before the
* user interrupt and CSB is processed.
*/
GEM_BUG_ON(!i915_request_completed(*execlists->active) &&
!reset_in_progress(execlists));
execlists_schedule_out(*execlists->active++);
GEM_BUG_ON(execlists->active - execlists->inflight >
execlists_num_ports(execlists));
}
} while (head != tail);
execlists->csb_head = head;
set_timeslice(engine);
/*
* Gen11 has proven to fail wrt global observation point between
* entry and tail update, failing on the ordering and thus
* we see an old entry in the context status buffer.
*
* Forcibly evict out entries for the next gpu csb update,
* to increase the odds that we get a fresh entries with non
* working hardware. The cost for doing so comes out mostly with
* the wash as hardware, working or not, will need to do the
* invalidation before.
*/
invalidate_csb_entries(&buf[0], &buf[num_entries - 1]);
}
static void __execlists_submission_tasklet(struct intel_engine_cs *const engine)
{
lockdep_assert_held(&engine->active.lock);
if (!engine->execlists.pending[0]) {
rcu_read_lock(); /* protect peeking at execlists->active */
execlists_dequeue(engine);
rcu_read_unlock();
}
}
static void __execlists_hold(struct i915_request *rq)
{
LIST_HEAD(list);
do {
struct i915_dependency *p;
if (i915_request_is_active(rq))
__i915_request_unsubmit(rq);
RQ_TRACE(rq, "on hold\n");
clear_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags);
list_move_tail(&rq->sched.link, &rq->engine->active.hold);
i915_request_set_hold(rq);
list_for_each_entry(p, &rq->sched.waiters_list, wait_link) {
struct i915_request *w =
container_of(p->waiter, typeof(*w), sched);
/* Leave semaphores spinning on the other engines */
if (w->engine != rq->engine)
continue;
if (!i915_request_is_ready(w))
continue;
if (i915_request_completed(w))
continue;
if (i915_request_on_hold(rq))
continue;
list_move_tail(&w->sched.link, &list);
}
rq = list_first_entry_or_null(&list, typeof(*rq), sched.link);
} while (rq);
}
static bool execlists_hold(struct intel_engine_cs *engine,
struct i915_request *rq)
{
spin_lock_irq(&engine->active.lock);
if (i915_request_completed(rq)) { /* too late! */
rq = NULL;
goto unlock;
}
if (rq->engine != engine) { /* preempted virtual engine */
struct virtual_engine *ve = to_virtual_engine(rq->engine);
/*
* intel_context_inflight() is only protected by virtue
* of process_csb() being called only by the tasklet (or
* directly from inside reset while the tasklet is suspended).
* Assert that neither of those are allowed to run while we
* poke at the request queues.
*/
GEM_BUG_ON(!reset_in_progress(&engine->execlists));
/*
* An unsubmitted request along a virtual engine will
* remain on the active (this) engine until we are able
* to process the context switch away (and so mark the
* context as no longer in flight). That cannot have happened
* yet, otherwise we would not be hanging!
*/
spin_lock(&ve->base.active.lock);
GEM_BUG_ON(intel_context_inflight(rq->context) != engine);
GEM_BUG_ON(ve->request != rq);
ve->request = NULL;
spin_unlock(&ve->base.active.lock);
i915_request_put(rq);
rq->engine = engine;
}
/*
* Transfer this request onto the hold queue to prevent it
* being resumbitted to HW (and potentially completed) before we have
* released it. Since we may have already submitted following
* requests, we need to remove those as well.
*/
GEM_BUG_ON(i915_request_on_hold(rq));
GEM_BUG_ON(rq->engine != engine);
__execlists_hold(rq);
unlock:
spin_unlock_irq(&engine->active.lock);
return rq;
}
static bool hold_request(const struct i915_request *rq)
{
struct i915_dependency *p;
/*
* If one of our ancestors is on hold, we must also be on hold,
* otherwise we will bypass it and execute before it.
*/
list_for_each_entry(p, &rq->sched.signalers_list, signal_link) {
const struct i915_request *s =
container_of(p->signaler, typeof(*s), sched);
if (s->engine != rq->engine)
continue;
if (i915_request_on_hold(s))
return true;
}
return false;
}
static void __execlists_unhold(struct i915_request *rq)
{
LIST_HEAD(list);
do {
struct i915_dependency *p;
GEM_BUG_ON(!i915_request_on_hold(rq));
GEM_BUG_ON(!i915_sw_fence_signaled(&rq->submit));
i915_request_clear_hold(rq);
list_move_tail(&rq->sched.link,
i915_sched_lookup_priolist(rq->engine,
rq_prio(rq)));
set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags);
RQ_TRACE(rq, "hold release\n");
/* Also release any children on this engine that are ready */
list_for_each_entry(p, &rq->sched.waiters_list, wait_link) {
struct i915_request *w =
container_of(p->waiter, typeof(*w), sched);
if (w->engine != rq->engine)
continue;
if (!i915_request_on_hold(rq))
continue;
/* Check that no other parents are also on hold */
if (hold_request(rq))
continue;
list_move_tail(&w->sched.link, &list);
}
rq = list_first_entry_or_null(&list, typeof(*rq), sched.link);
} while (rq);
}
static void execlists_unhold(struct intel_engine_cs *engine,
struct i915_request *rq)
{
spin_lock_irq(&engine->active.lock);
/*
* Move this request back to the priority queue, and all of its
* children and grandchildren that were suspended along with it.
*/
__execlists_unhold(rq);
if (rq_prio(rq) > engine->execlists.queue_priority_hint) {
engine->execlists.queue_priority_hint = rq_prio(rq);
tasklet_hi_schedule(&engine->execlists.tasklet);
}
spin_unlock_irq(&engine->active.lock);
}
struct execlists_capture {
struct work_struct work;
struct i915_request *rq;
struct i915_gpu_coredump *error;
};
static void execlists_capture_work(struct work_struct *work)
{
struct execlists_capture *cap = container_of(work, typeof(*cap), work);
const gfp_t gfp = GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN;
struct intel_engine_cs *engine = cap->rq->engine;
struct intel_gt_coredump *gt = cap->error->gt;
struct intel_engine_capture_vma *vma;
/* Compress all the objects attached to the request, slow! */
vma = intel_engine_coredump_add_request(gt->engine, cap->rq, gfp);
if (vma) {
struct i915_vma_compress *compress =
i915_vma_capture_prepare(gt);
intel_engine_coredump_add_vma(gt->engine, vma, compress);
i915_vma_capture_finish(gt, compress);
}
gt->simulated = gt->engine->simulated;
cap->error->simulated = gt->simulated;
/* Publish the error state, and announce it to the world */
i915_error_state_store(cap->error);
i915_gpu_coredump_put(cap->error);
/* Return this request and all that depend upon it for signaling */
execlists_unhold(engine, cap->rq);
i915_request_put(cap->rq);
kfree(cap);
}
static struct execlists_capture *capture_regs(struct intel_engine_cs *engine)
{
const gfp_t gfp = GFP_ATOMIC | __GFP_NOWARN;
struct execlists_capture *cap;
cap = kmalloc(sizeof(*cap), gfp);
if (!cap)
return NULL;
cap->error = i915_gpu_coredump_alloc(engine->i915, gfp);
if (!cap->error)
goto err_cap;
cap->error->gt = intel_gt_coredump_alloc(engine->gt, gfp);
if (!cap->error->gt)
goto err_gpu;
cap->error->gt->engine = intel_engine_coredump_alloc(engine, gfp);
if (!cap->error->gt->engine)
goto err_gt;
return cap;
err_gt:
kfree(cap->error->gt);
err_gpu:
kfree(cap->error);
err_cap:
kfree(cap);
return NULL;
}
static bool execlists_capture(struct intel_engine_cs *engine)
{
struct execlists_capture *cap;
if (!IS_ENABLED(CONFIG_DRM_I915_CAPTURE_ERROR))
return true;
/*
* We need to _quickly_ capture the engine state before we reset.
* We are inside an atomic section (softirq) here and we are delaying
* the forced preemption event.
*/
cap = capture_regs(engine);
if (!cap)
return true;
cap->rq = execlists_active(&engine->execlists);
GEM_BUG_ON(!cap->rq);
rcu_read_lock();
cap->rq = active_request(cap->rq->context->timeline, cap->rq);
cap->rq = i915_request_get_rcu(cap->rq);
rcu_read_unlock();
if (!cap->rq)
goto err_free;
/*
* Remove the request from the execlists queue, and take ownership
* of the request. We pass it to our worker who will _slowly_ compress
* all the pages the _user_ requested for debugging their batch, after
* which we return it to the queue for signaling.
*
* By removing them from the execlists queue, we also remove the
* requests from being processed by __unwind_incomplete_requests()
* during the intel_engine_reset(), and so they will *not* be replayed
* afterwards.
*
* Note that because we have not yet reset the engine at this point,
* it is possible for the request that we have identified as being
* guilty, did in fact complete and we will then hit an arbitration
* point allowing the outstanding preemption to succeed. The likelihood
* of that is very low (as capturing of the engine registers should be
* fast enough to run inside an irq-off atomic section!), so we will
* simply hold that request accountable for being non-preemptible
* long enough to force the reset.
*/
if (!execlists_hold(engine, cap->rq))
goto err_rq;
INIT_WORK(&cap->work, execlists_capture_work);
schedule_work(&cap->work);
return true;
err_rq:
i915_request_put(cap->rq);
err_free:
i915_gpu_coredump_put(cap->error);
kfree(cap);
return false;
}
static noinline void preempt_reset(struct intel_engine_cs *engine)
{
const unsigned int bit = I915_RESET_ENGINE + engine->id;
unsigned long *lock = &engine->gt->reset.flags;
if (i915_modparams.reset < 3)
return;
if (test_and_set_bit(bit, lock))
return;
/* Mark this tasklet as disabled to avoid waiting for it to complete */
tasklet_disable_nosync(&engine->execlists.tasklet);
ENGINE_TRACE(engine, "preempt timeout %lu+%ums\n",
READ_ONCE(engine->props.preempt_timeout_ms),
jiffies_to_msecs(jiffies - engine->execlists.preempt.expires));
ring_set_paused(engine, 1); /* Freeze the current request in place */
if (execlists_capture(engine))
intel_engine_reset(engine, "preemption time out");
else
ring_set_paused(engine, 0);
tasklet_enable(&engine->execlists.tasklet);
clear_and_wake_up_bit(bit, lock);
}
static bool preempt_timeout(const struct intel_engine_cs *const engine)
{
const struct timer_list *t = &engine->execlists.preempt;
if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT)
return false;
if (!timer_expired(t))
return false;
return READ_ONCE(engine->execlists.pending[0]);
}
/*
* Check the unread Context Status Buffers and manage the submission of new
* contexts to the ELSP accordingly.
*/
static void execlists_submission_tasklet(unsigned long data)
{
struct intel_engine_cs * const engine = (struct intel_engine_cs *)data;
bool timeout = preempt_timeout(engine);
process_csb(engine);
if (!READ_ONCE(engine->execlists.pending[0]) || timeout) {
unsigned long flags;
spin_lock_irqsave(&engine->active.lock, flags);
__execlists_submission_tasklet(engine);
spin_unlock_irqrestore(&engine->active.lock, flags);
/* Recheck after serialising with direct-submission */
if (timeout && preempt_timeout(engine))
preempt_reset(engine);
}
}
static void __execlists_kick(struct intel_engine_execlists *execlists)
{
/* Kick the tasklet for some interrupt coalescing and reset handling */
tasklet_hi_schedule(&execlists->tasklet);
}
#define execlists_kick(t, member) \
__execlists_kick(container_of(t, struct intel_engine_execlists, member))
static void execlists_timeslice(struct timer_list *timer)
{
execlists_kick(timer, timer);
}
static void execlists_preempt(struct timer_list *timer)
{
execlists_kick(timer, preempt);
}
static void queue_request(struct intel_engine_cs *engine,
struct i915_request *rq)
{
GEM_BUG_ON(!list_empty(&rq->sched.link));
list_add_tail(&rq->sched.link,
i915_sched_lookup_priolist(engine, rq_prio(rq)));
set_bit(I915_FENCE_FLAG_PQUEUE, &rq->fence.flags);
}
static void __submit_queue_imm(struct intel_engine_cs *engine)
{
struct intel_engine_execlists * const execlists = &engine->execlists;
if (reset_in_progress(execlists))
return; /* defer until we restart the engine following reset */
if (execlists->tasklet.func == execlists_submission_tasklet)
__execlists_submission_tasklet(engine);
else
tasklet_hi_schedule(&execlists->tasklet);
}
static void submit_queue(struct intel_engine_cs *engine,
const struct i915_request *rq)
{
struct intel_engine_execlists *execlists = &engine->execlists;
if (rq_prio(rq) <= execlists->queue_priority_hint)
return;
execlists->queue_priority_hint = rq_prio(rq);
__submit_queue_imm(engine);
}
static bool ancestor_on_hold(const struct intel_engine_cs *engine,
const struct i915_request *rq)
{
GEM_BUG_ON(i915_request_on_hold(rq));
return !list_empty(&engine->active.hold) && hold_request(rq);
}
static void execlists_submit_request(struct i915_request *request)
{
struct intel_engine_cs *engine = request->engine;
unsigned long flags;
/* Will be called from irq-context when using foreign fences. */
spin_lock_irqsave(&engine->active.lock, flags);
if (unlikely(ancestor_on_hold(engine, request))) {
list_add_tail(&request->sched.link, &engine->active.hold);
i915_request_set_hold(request);
} else {
queue_request(engine, request);
GEM_BUG_ON(RB_EMPTY_ROOT(&engine->execlists.queue.rb_root));
GEM_BUG_ON(list_empty(&request->sched.link));
submit_queue(engine, request);
}
spin_unlock_irqrestore(&engine->active.lock, flags);
}
static void __execlists_context_fini(struct intel_context *ce)
{
intel_ring_put(ce->ring);
i915_vma_put(ce->state);
}
static void execlists_context_destroy(struct kref *kref)
{
struct intel_context *ce = container_of(kref, typeof(*ce), ref);
GEM_BUG_ON(!i915_active_is_idle(&ce->active));
GEM_BUG_ON(intel_context_is_pinned(ce));
if (ce->state)
__execlists_context_fini(ce);
intel_context_fini(ce);
intel_context_free(ce);
}
static void
set_redzone(void *vaddr, const struct intel_engine_cs *engine)
{
if (!IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
return;
vaddr += engine->context_size;
memset(vaddr, CONTEXT_REDZONE, I915_GTT_PAGE_SIZE);
}
static void
check_redzone(const void *vaddr, const struct intel_engine_cs *engine)
{
if (!IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
return;
vaddr += engine->context_size;
if (memchr_inv(vaddr, CONTEXT_REDZONE, I915_GTT_PAGE_SIZE))
dev_err_once(engine->i915->drm.dev,
"%s context redzone overwritten!\n",
engine->name);
}
static void execlists_context_unpin(struct intel_context *ce)
{
check_redzone((void *)ce->lrc_reg_state - LRC_STATE_PN * PAGE_SIZE,
ce->engine);
i915_gem_object_unpin_map(ce->state->obj);
}
static void
__execlists_update_reg_state(const struct intel_context *ce,
const struct intel_engine_cs *engine,
u32 head)
{
struct intel_ring *ring = ce->ring;
u32 *regs = ce->lrc_reg_state;
GEM_BUG_ON(!intel_ring_offset_valid(ring, head));
GEM_BUG_ON(!intel_ring_offset_valid(ring, ring->tail));
regs[CTX_RING_START] = i915_ggtt_offset(ring->vma);
regs[CTX_RING_HEAD] = head;
regs[CTX_RING_TAIL] = ring->tail;
/* RPCS */
if (engine->class == RENDER_CLASS) {
regs[CTX_R_PWR_CLK_STATE] =
intel_sseu_make_rpcs(engine->i915, &ce->sseu);
i915_oa_init_reg_state(ce, engine);
}
}
static int
__execlists_context_pin(struct intel_context *ce,
struct intel_engine_cs *engine)
{
void *vaddr;
GEM_BUG_ON(!ce->state);
GEM_BUG_ON(!i915_vma_is_pinned(ce->state));
vaddr = i915_gem_object_pin_map(ce