Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 520CCCD6E5D for ; Tue, 2 Jun 2026 18:14:55 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 1119310F2CC; Tue, 2 Jun 2026 18:14:55 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=lankhorst.se header.i=@lankhorst.se header.b="EhJ8dyRX"; dkim-atps=neutral Received: from lankhorst.se (unknown [141.105.120.124]) by gabe.freedesktop.org (Postfix) with ESMTPS id D3A4F10F2CC for ; Tue, 2 Jun 2026 18:14:53 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=lankhorst.se; s=default; t=1780424092; bh=Okj6ok18xhaaV+n4HuMAOe2XKQfep9xs61U+jbGFa54=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=EhJ8dyRXarpGbhvdj3iLytEj11ol+8GRVNiHfpuMfwypsOxbnXiHD5NdDzx5iSztm 38BcH8j9DBIMPhdD3ROv2WbNq8YLytsYATBxP8wijHH+/xXHIzTBoY0Seck1U624gZ ZwDvSm3b/mH2vXugcwb6xyhfvKOaDc0aJWI1PFZiYojEIrJ34LI1JP+cK5uXNpWISK vLrOcI5glvCJQOXETXBpH5eUWFXcqrZ/2TOXIFc1SaZB580N/Z8ZqEFymw23YDO1Wu YaOTfJqqweTMqXmSYaPR65i1ri9V7usyiTLVeHqntKcZdu8sjQprAkOV8iPHdjtV4f 8whoNxmIhb/xg== From: Maarten Lankhorst To: intel-xe@lists.freedesktop.org Cc: Maarten Lankhorst Subject: [RFC 1/2] drm/xe: Convert job preparation in 2 stages Date: Tue, 2 Jun 2026 20:14:46 +0200 Message-ID: <20260602181447.131216-2-dev@lankhorst.se> X-Mailer: git-send-email 2.53.0 In-Reply-To: <20260602181447.131216-1-dev@lankhorst.se> References: <20260602181447.131216-1-dev@lankhorst.se> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-BeenThere: intel-xe@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel Xe graphics driver List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-xe-bounces@lists.freedesktop.org Sender: "Intel-xe" Instead of writing the entire job at the time of emission, prepare it in advance. This allows us to calculate the size before we submit, in preparation of allowing more jobs to run simultaneously. Signed-off-by: Maarten Lankhorst --- drivers/gpu/drm/xe/xe_ring_ops.c | 150 ++++++++++++++---------- drivers/gpu/drm/xe/xe_ring_ops_types.h | 3 + drivers/gpu/drm/xe/xe_sched_job.c | 3 + drivers/gpu/drm/xe/xe_sched_job_types.h | 5 + 4 files changed, 96 insertions(+), 65 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_ring_ops.c b/drivers/gpu/drm/xe/xe_ring_ops.c index 39a670e91ba72..9f1b0ac30d1aa 100644 --- a/drivers/gpu/drm/xe/xe_ring_ops.c +++ b/drivers/gpu/drm/xe/xe_ring_ops.c @@ -320,15 +320,12 @@ static int emit_fake_watchdog(struct xe_lrc *lrc, u32 *dw, int i) } /* for engines that don't require any special HW handling (no EUs, no aux inval, etc) */ -static void __emit_job_gen12_simple(struct xe_sched_job *job, struct xe_lrc *lrc, - u64 batch_addr, u32 *head, u32 seqno) +static u32 __prepare_job_gen12_simple(struct xe_sched_job *job, struct xe_lrc *lrc, + u32 *dw, u64 batch_addr, u32 seqno) { - u32 dw[MAX_JOB_SIZE_DW], i = 0; - u32 ppgtt_flag = get_ppgtt_flag(job); + u32 i = 0, ppgtt_flag = get_ppgtt_flag(job); struct xe_gt *gt = job->q->gt; - *head = lrc->ring.tail; - if (job->ring_ops_force_reset) i = emit_fake_watchdog(lrc, dw, i); @@ -362,7 +359,7 @@ static void __emit_job_gen12_simple(struct xe_sched_job *job, struct xe_lrc *lrc xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW); - xe_lrc_write_ring(lrc, dw, i * sizeof(*dw)); + return i * sizeof(*dw); } static bool has_aux_ccs(struct xe_device *xe) @@ -379,16 +376,13 @@ static bool has_aux_ccs(struct xe_device *xe) return !xe->info.has_flat_ccs; } -static void __emit_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc, - u64 batch_addr, u32 *head, u32 seqno) +static u32 __prepare_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc, + u32 *dw, u64 batch_addr, u32 seqno) { - u32 dw[MAX_JOB_SIZE_DW], i = 0; - u32 ppgtt_flag = get_ppgtt_flag(job); + u32 i = 0, ppgtt_flag = get_ppgtt_flag(job); struct xe_gt *gt = job->q->gt; struct xe_device *xe = gt_to_xe(gt); - *head = lrc->ring.tail; - if (job->ring_ops_force_reset) i = emit_fake_watchdog(lrc, dw, i); @@ -427,23 +421,20 @@ static void __emit_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc, xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW); - xe_lrc_write_ring(lrc, dw, i * sizeof(*dw)); + return i * sizeof(*dw); } -static void __emit_job_gen12_render_compute(struct xe_sched_job *job, - struct xe_lrc *lrc, - u64 batch_addr, u32 *head, - u32 seqno) +static u32 __prepare_job_gen12_render_compute(struct xe_sched_job *job, + struct xe_lrc *lrc, + u32 *dw, u64 batch_addr, + u32 seqno) { - u32 dw[MAX_JOB_SIZE_DW], i = 0; - u32 ppgtt_flag = get_ppgtt_flag(job); + u32 i = 0, ppgtt_flag = get_ppgtt_flag(job); struct xe_gt *gt = job->q->gt; struct xe_device *xe = gt_to_xe(gt); bool lacks_render = !(gt->info.engine_mask & XE_HW_ENGINE_RCS_MASK); u32 mask_flags = 0; - *head = lrc->ring.tail; - if (job->ring_ops_force_reset) i = emit_fake_watchdog(lrc, dw, i); @@ -491,19 +482,17 @@ static void __emit_job_gen12_render_compute(struct xe_sched_job *job, xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW); - xe_lrc_write_ring(lrc, dw, i * sizeof(*dw)); + return i * sizeof(*dw); } -static void emit_migration_job_gen12(struct xe_sched_job *job, - struct xe_lrc *lrc, u32 *head, - u32 seqno) +static u32 prepare_migration_job_gen12(struct xe_sched_job *job, + struct xe_lrc *lrc, u32 *dw, + u32 seqno) { struct xe_gt *gt = job->q->gt; struct xe_device *xe = gt_to_xe(gt); u32 saddr = xe_lrc_start_seqno_ggtt_addr(lrc); - u32 dw[MAX_JOB_SIZE_DW], i = 0; - - *head = lrc->ring.tail; + u32 i = 0; xe_gt_assert(gt, !job->ring_ops_force_reset); @@ -529,94 +518,125 @@ static void emit_migration_job_gen12(struct xe_sched_job *job, xe_gt_assert(job->q->gt, i <= MAX_JOB_SIZE_DW); - xe_lrc_write_ring(lrc, dw, i * sizeof(*dw)); + return i * sizeof(*dw); } -static void emit_job_gen12_gsc(struct xe_sched_job *job) +static void prepare_job_gen12_gsc(struct xe_sched_job *job) { struct xe_gt *gt = job->q->gt; - xe_gt_assert(gt, job->q->width <= 1); /* no parallel submission for GSCCS */ + xe_gt_assert(gt, job->q->width == 1); /* no parallel submission for GSCCS */ - __emit_job_gen12_simple(job, job->q->lrc[0], - job->ptrs[0].batch_addr, - &job->ptrs[0].head, - xe_sched_job_lrc_seqno(job)); + job->ptrs[0].job_size = + __prepare_job_gen12_simple(job, job->q->lrc[0], + job->ptrs[0].dw, + job->ptrs[0].batch_addr, + xe_sched_job_lrc_seqno(job)); } -static void emit_job_gen12_copy(struct xe_sched_job *job) +static void prepare_job_gen12_copy(struct xe_sched_job *job) { int i; if (xe_sched_job_is_migration(job->q)) { - emit_migration_job_gen12(job, job->q->lrc[0], - &job->ptrs[0].head, - xe_sched_job_lrc_seqno(job)); + xe_gt_assert(job->q->gt, job->q->width == 1); + + job->ptrs[0].job_size = + prepare_migration_job_gen12(job, job->q->lrc[0], + job->ptrs[0].dw, + xe_sched_job_lrc_seqno(job)); return; } - for (i = 0; i < job->q->width; ++i) - __emit_job_gen12_simple(job, job->q->lrc[i], - job->ptrs[i].batch_addr, - &job->ptrs[i].head, - xe_sched_job_lrc_seqno(job)); + for (i = 0; i < job->q->width; ++i) { + job->ptrs[i].job_size = + __prepare_job_gen12_simple(job, job->q->lrc[i], + job->ptrs[i].dw, + job->ptrs[i].batch_addr, + xe_sched_job_lrc_seqno(job)); + xe_gt_assert(job->q->gt, job->ptrs[i].job_size == job->ptrs[0].job_size); + } } -static void emit_job_gen12_video(struct xe_sched_job *job) +static void prepare_job_gen12_video(struct xe_sched_job *job) { int i; /* FIXME: Not doing parallel handshake for now */ - for (i = 0; i < job->q->width; ++i) - __emit_job_gen12_video(job, job->q->lrc[i], - job->ptrs[i].batch_addr, - &job->ptrs[i].head, - xe_sched_job_lrc_seqno(job)); + for (i = 0; i < job->q->width; ++i) { + job->ptrs[i].job_size = + __prepare_job_gen12_video(job, job->q->lrc[i], + job->ptrs[i].dw, + job->ptrs[i].batch_addr, + xe_sched_job_lrc_seqno(job)); + xe_gt_assert(job->q->gt, job->ptrs[i].job_size == job->ptrs[0].job_size); + } } -static void emit_job_gen12_render_compute(struct xe_sched_job *job) +static void prepare_job_gen12_render_compute(struct xe_sched_job *job) { int i; - for (i = 0; i < job->q->width; ++i) - __emit_job_gen12_render_compute(job, job->q->lrc[i], - job->ptrs[i].batch_addr, - &job->ptrs[i].head, - xe_sched_job_lrc_seqno(job)); + for (i = 0; i < job->q->width; ++i) { + job->ptrs[i].job_size = + __prepare_job_gen12_render_compute(job, job->q->lrc[i], + job->ptrs[i].dw, + job->ptrs[i].batch_addr, + xe_sched_job_lrc_seqno(job)); + xe_gt_assert(job->q->gt, job->ptrs[i].job_size == job->ptrs[0].job_size); + } +} + +static void emit_prepared_job(struct xe_sched_job *job) +{ + for (u32 i = 0; i < job->q->width; ++i) { + struct xe_lrc *lrc = job->q->lrc[i]; + + job->ptrs[i].head = lrc->ring.tail; + xe_lrc_write_ring(lrc, job->ptrs[i].dw, job->ptrs[i].job_size); + } } static const struct xe_ring_ops ring_ops_gen12_gsc = { - .emit_job = emit_job_gen12_gsc, + .prepare_job = prepare_job_gen12_gsc, + .emit_job = emit_prepared_job, }; static const struct xe_ring_ops ring_ops_gen12_copy = { - .emit_job = emit_job_gen12_copy, + .prepare_job = prepare_job_gen12_copy, + .emit_job = emit_prepared_job, }; static const struct xe_ring_ops ring_ops_gen12_video_decode = { - .emit_job = emit_job_gen12_video, + .prepare_job = prepare_job_gen12_video, + .emit_job = emit_prepared_job, }; static const struct xe_ring_ops ring_ops_gen12_video_enhance = { - .emit_job = emit_job_gen12_video, + .prepare_job = prepare_job_gen12_video, + .emit_job = emit_prepared_job, }; static const struct xe_ring_ops ring_ops_gen12_render_compute = { - .emit_job = emit_job_gen12_render_compute, + .prepare_job = prepare_job_gen12_render_compute, + .emit_job = emit_prepared_job, }; static const struct xe_ring_ops auxccs_ring_ops_gen12_video_decode = { - .emit_job = emit_job_gen12_video, + .prepare_job = prepare_job_gen12_video, + .emit_job = emit_prepared_job, .emit_aux_table_inv = emit_aux_table_inv_video_decode, }; static const struct xe_ring_ops auxccs_ring_ops_gen12_video_enhance = { - .emit_job = emit_job_gen12_video, + .prepare_job = prepare_job_gen12_video, + .emit_job = emit_prepared_job, .emit_aux_table_inv = emit_aux_table_inv_video_enhance, }; static const struct xe_ring_ops auxccs_ring_ops_gen12_render_compute = { - .emit_job = emit_job_gen12_render_compute, + .prepare_job = prepare_job_gen12_render_compute, + .emit_job = emit_prepared_job, .emit_aux_table_inv = emit_aux_table_inv_render_compute, }; diff --git a/drivers/gpu/drm/xe/xe_ring_ops_types.h b/drivers/gpu/drm/xe/xe_ring_ops_types.h index 52ff96bc41004..5774ef6f3a265 100644 --- a/drivers/gpu/drm/xe/xe_ring_ops_types.h +++ b/drivers/gpu/drm/xe/xe_ring_ops_types.h @@ -18,6 +18,9 @@ struct xe_sched_job; * struct xe_ring_ops - Ring operations */ struct xe_ring_ops { + /** @prepare_job: Write job to @job struct */ + void (*prepare_job)(struct xe_sched_job *job); + /** @emit_job: Write job to ring */ void (*emit_job)(struct xe_sched_job *job); diff --git a/drivers/gpu/drm/xe/xe_sched_job.c b/drivers/gpu/drm/xe/xe_sched_job.c index ae5b38b2a884a..23bfdc1e4ef9f 100644 --- a/drivers/gpu/drm/xe/xe_sched_job.c +++ b/drivers/gpu/drm/xe/xe_sched_job.c @@ -266,6 +266,9 @@ void xe_sched_job_arm(struct xe_sched_job *job) job->ring_ops_flush_tlb = true; } + if (!job->restore_replay) + q->ring_ops->prepare_job(job); + /* Arm the pre-allocated fences */ for (i = 0; i < q->width; prev = fence, ++i) { struct dma_fence_chain *chain; diff --git a/drivers/gpu/drm/xe/xe_sched_job_types.h b/drivers/gpu/drm/xe/xe_sched_job_types.h index 0490b1247a6e9..5d4ea9dd09912 100644 --- a/drivers/gpu/drm/xe/xe_sched_job_types.h +++ b/drivers/gpu/drm/xe/xe_sched_job_types.h @@ -9,6 +9,7 @@ #include #include +#include "xe_ring_ops_types.h" struct xe_exec_queue; struct dma_fence; @@ -29,6 +30,10 @@ struct xe_job_ptrs { * job was submitted */ u32 head; + /** @job_size: size of job in bytes */ + u32 job_size; + /** @dw: the raw bytes to be added to lrc */ + u32 dw[MAX_JOB_SIZE_DW]; }; /** -- 2.53.0