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 B1513C6FD18 for ; Wed, 29 Mar 2023 00:09:56 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 726B010E49F; Wed, 29 Mar 2023 00:09:56 +0000 (UTC) Received: from mga06.intel.com (mga06b.intel.com [134.134.136.31]) by gabe.freedesktop.org (Postfix) with ESMTPS id 35FFA10E49F for ; Wed, 29 Mar 2023 00:09:54 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1680048594; x=1711584594; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=wfxfgoxOw2kb1X6BNK/I/8K43RXd/yoRGIIPCVKy2aA=; b=lZACs20RbzINkPwhnnDhucld+Ih+NDcr95MStrj52YyHdgjeuEZ7jnuk OQsC4xD+wTQPN685YYPpNmwux37BFc6ZiHQ0w7Brlzfrj9L9EPSil0P9n +KQLXUx81IkZ5GGDAxS4sawvbpnnD+eRdd/zzp2j8rJC6Y3bxxKs9yQF8 fp+6ehPQvKoVgvkVhdz6WJKcfyY6Ap+4RdaepOXjOSizwoaKu4oIxyXMD lqUolCVrPqOmNVzoX34ErUerLaF3O5SDAgHJipropndsMino0vIeonTNE MNlFdaBVEknIv5FrL6IpD8XFIpxuHuDlltYUIGPygYaa49/kEfFEXHD69 w==; X-IronPort-AV: E=McAfee;i="6600,9927,10663"; a="403350269" X-IronPort-AV: E=Sophos;i="5.98,299,1673942400"; d="scan'208";a="403350269" Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by orsmga104.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 28 Mar 2023 17:09:53 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=McAfee;i="6600,9927,10663"; a="773346495" X-IronPort-AV: E=Sophos;i="5.98,299,1673942400"; d="scan'208";a="773346495" Received: from lstrano-desk.jf.intel.com ([10.24.89.184]) by fmsmga003-auth.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 28 Mar 2023 17:09:53 -0700 From: Matthew Brost To: Date: Tue, 28 Mar 2023 17:09:58 -0700 Message-Id: <20230329000958.2188321-2-matthew.brost@intel.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20230329000958.2188321-1-matthew.brost@intel.com> References: <20230329000958.2188321-1-matthew.brost@intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Subject: [Intel-xe] [RFC PATCH 1/1] drm/xe: Return NULL in run_job for long running jobs 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" Return NULL in run_job for long running jobs, now to flow control the ring use prepare_job. FIXME: Need to implement cleanup for long runnig engines that encounter errors. Signed-off-by: Matthew Brost --- drivers/gpu/drm/xe/xe_engine.c | 54 ++++++++++++++++++++++++- drivers/gpu/drm/xe/xe_engine.h | 6 +++ drivers/gpu/drm/xe/xe_engine_types.h | 3 ++ drivers/gpu/drm/xe/xe_guc_submit.c | 31 ++++++++++++-- drivers/gpu/drm/xe/xe_sched_job.c | 5 +++ drivers/gpu/drm/xe/xe_sched_job_types.h | 2 + 6 files changed, 96 insertions(+), 5 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_engine.c b/drivers/gpu/drm/xe/xe_engine.c index 37209b13bcd6..440f69ed3dc8 100644 --- a/drivers/gpu/drm/xe/xe_engine.c +++ b/drivers/gpu/drm/xe/xe_engine.c @@ -40,8 +40,11 @@ static struct xe_engine *__xe_engine_create(struct xe_device *xe, e->flags = flags; e->hwe = hwe; e->gt = gt; - if (vm) + if (vm) { e->vm = xe_vm_get(vm); + if (xe_vm_no_dma_fences(vm)) + xa_init_flags(&e->pending_lr_jobs, XA_FLAGS_ALLOC1); + } e->class = hwe->class; e->width = width; e->logical_mask = logical_mask; @@ -149,8 +152,11 @@ void xe_engine_fini(struct xe_engine *e) for (i = 0; i < e->width; ++i) xe_lrc_finish(e->lrc + i); - if (e->vm) + if (e->vm) { xe_vm_put(e->vm); + if (xe_vm_no_dma_fences(e->vm)) + xa_destroy(&e->pending_lr_jobs); + } kfree(e); } @@ -789,3 +795,47 @@ int xe_engine_set_property_ioctl(struct drm_device *dev, void *data, return ret; } + +int xe_engine_add_lr_job(struct xe_engine *e, struct xe_sched_job *job) +{ + void *ret; + + if (!xe_engine_is_lr(e)) + return 0; + + ret = xa_store(&e->pending_lr_jobs, job->fence->seqno, job, GFP_KERNEL); + if (IS_ERR(ret)) + return PTR_ERR(ret); + XE_WARN_ON(ret); + + return 0; +} + +void xe_engine_del_lr_job(struct xe_engine *e, struct xe_sched_job *job) +{ + if (!xe_engine_is_lr(e)) + return; + + xa_erase(&e->pending_lr_jobs, job->fence->seqno); +} + +struct xe_sched_job *xe_engine_get_lr_job(struct xe_engine *e, u32 seqno) +{ + struct xe_sched_job *job; + + if (!xe_engine_is_lr(e)) + return NULL; + + xa_lock(&e->pending_lr_jobs); + job = xa_load(&e->pending_lr_jobs, seqno); + if (job) + dma_fence_get(job->fence); + xa_unlock(&e->pending_lr_jobs); + + return job; +} + +bool xe_engine_is_lr(struct xe_engine *e) +{ + return e->vm && xe_vm_no_dma_fences(e->vm); +} diff --git a/drivers/gpu/drm/xe/xe_engine.h b/drivers/gpu/drm/xe/xe_engine.h index a49cf2ab405e..b7aa5a03adcd 100644 --- a/drivers/gpu/drm/xe/xe_engine.h +++ b/drivers/gpu/drm/xe/xe_engine.h @@ -12,6 +12,7 @@ struct drm_device; struct drm_file; struct xe_device; +struct xe_sched_job; struct xe_file; struct xe_engine *xe_engine_create(struct xe_device *xe, struct xe_vm *vm, @@ -24,6 +25,11 @@ struct xe_engine *xe_engine_create_class(struct xe_device *xe, struct xe_gt *gt, void xe_engine_fini(struct xe_engine *e); void xe_engine_destroy(struct kref *ref); +int xe_engine_add_lr_job(struct xe_engine *e, struct xe_sched_job *job); +void xe_engine_del_lr_job(struct xe_engine *e, struct xe_sched_job *job); +struct xe_sched_job *xe_engine_get_lr_job(struct xe_engine *e, u32 seqno); +bool xe_engine_is_lr(struct xe_engine *e); + struct xe_engine *xe_engine_lookup(struct xe_file *xef, u32 id); static inline struct xe_engine *xe_engine_get(struct xe_engine *engine) diff --git a/drivers/gpu/drm/xe/xe_engine_types.h b/drivers/gpu/drm/xe/xe_engine_types.h index 36bfaeec23f4..46b718343f79 100644 --- a/drivers/gpu/drm/xe/xe_engine_types.h +++ b/drivers/gpu/drm/xe/xe_engine_types.h @@ -143,6 +143,9 @@ struct xe_engine { u32 acc_granularity; } usm; + /** @pending_lr_jobs: Pending jobs for long running work loads */ + struct xarray pending_lr_jobs; + /** @ops: submission backend engine operations */ const struct xe_engine_ops *ops; diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c index e857013070b9..590e5481d063 100644 --- a/drivers/gpu/drm/xe/xe_guc_submit.c +++ b/drivers/gpu/drm/xe/xe_guc_submit.c @@ -377,6 +377,22 @@ static void set_min_preemption_timeout(struct xe_guc *guc, struct xe_engine *e) __guc_engine_policy_action_size(&policy), 0, 0); } +static struct dma_fence * +guc_engine_prepare_job(struct drm_sched_job *drm_job, + struct drm_sched_entity *s_entity) +{ + struct xe_sched_job *job = to_xe_sched_job(drm_job), *lr_job; + struct xe_engine *e = job->engine; + u32 seqno = (u32)(job->fence->seqno) - + (e->lrc[0].ring.size * 2) / MAX_JOB_SIZE_BYTES; + + lr_job = xe_engine_get_lr_job(job->engine, seqno); + if (lr_job) + return lr_job->fence; + else + return NULL; +} + #define PARALLEL_SCRATCH_SIZE 2048 #define WQ_SIZE (PARALLEL_SCRATCH_SIZE / 2) #define WQ_OFFSET (PARALLEL_SCRATCH_SIZE - WQ_SIZE) @@ -675,18 +691,26 @@ guc_engine_run_job(struct drm_sched_job *drm_job) submit_engine(e); } - if (test_and_set_bit(JOB_FLAG_SUBMIT, &job->fence->flags)) + if (xe_engine_is_lr(job->engine)) + return NULL; + else if (test_and_set_bit(JOB_FLAG_SUBMIT, &job->fence->flags)) return job->fence; else return dma_fence_get(job->fence); } +static void __guc_engine_free_job(struct xe_sched_job *job) +{ + trace_xe_sched_job_free(job); + xe_sched_job_put(job); +} + static void guc_engine_free_job(struct drm_sched_job *drm_job) { struct xe_sched_job *job = to_xe_sched_job(drm_job); - trace_xe_sched_job_free(job); - xe_sched_job_put(job); + if (xe_engine_is_lr(job->engine)) + __guc_engine_free_job(job); } static int guc_read_stopped(struct xe_guc *guc) @@ -1056,6 +1080,7 @@ static void guc_engine_process_msg(struct drm_sched_msg *msg) } static const struct drm_sched_backend_ops drm_sched_ops = { + .prepare_job = guc_engine_prepare_job, .run_job = guc_engine_run_job, .free_job = guc_engine_free_job, .timedout_job = guc_engine_timedout_job, diff --git a/drivers/gpu/drm/xe/xe_sched_job.c b/drivers/gpu/drm/xe/xe_sched_job.c index 2985caa6097b..ece3a5db3978 100644 --- a/drivers/gpu/drm/xe/xe_sched_job.c +++ b/drivers/gpu/drm/xe/xe_sched_job.c @@ -154,6 +154,10 @@ struct xe_sched_job *xe_sched_job_create(struct xe_engine *e, for (i = 0; i < width; ++i) job->batch_addr[i] = batch_addr[i]; + err = xe_engine_add_lr_job(e, job); + if (err) + goto err_fences; + /* All other jobs require a VM to be open which has a ref */ if (unlikely(e->flags & ENGINE_FLAG_KERNEL)) xe_device_mem_access_get(job_to_xe(job)); @@ -190,6 +194,7 @@ void xe_sched_job_destroy(struct kref *ref) if (unlikely(job->engine->flags & ENGINE_FLAG_KERNEL)) xe_device_mem_access_put(job_to_xe(job)); + xe_engine_del_lr_job(job->engine, job); xe_engine_put(job->engine); dma_fence_put(job->fence); drm_sched_job_cleanup(&job->drm); diff --git a/drivers/gpu/drm/xe/xe_sched_job_types.h b/drivers/gpu/drm/xe/xe_sched_job_types.h index 5534bfacaa16..07406d894b26 100644 --- a/drivers/gpu/drm/xe/xe_sched_job_types.h +++ b/drivers/gpu/drm/xe/xe_sched_job_types.h @@ -22,6 +22,8 @@ struct xe_sched_job { struct xe_engine *engine; /** @refcount: ref count of this job */ struct kref refcount; + /** @lr_job_cb: long running job callback */ + struct dma_fence_cb lr_job_cb; /** * @fence: dma fence to indicate completion. 1 way relationship - job * can safely reference fence, fence cannot safely reference job. -- 2.34.1