From nobody Fri Oct 3 08:53:15 2025 Received: from sender4-pp-f112.zoho.com (sender4-pp-f112.zoho.com [136.143.188.112]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 6BFF2502BE for ; Thu, 4 Sep 2025 00:11:42 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=pass smtp.client-ip=136.143.188.112 ARC-Seal: i=2; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1756944704; cv=pass; b=DclnktbkoRDmwTGR9bWeeZmEMe06hITVwstG4c6g2K0IuJxo486s8wADEKDNSmaH49oDJvcCdjzyofPFjDiuy4u5Njz4P+3fTW8CtNYl71jqfKincnZA05LiCMYlXyoCYIolUrfiHm3PUO03Wa1B1QeFtNTRQuu2A37EngotmnQ= ARC-Message-Signature: i=2; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1756944704; c=relaxed/simple; bh=n4og9cOJebbwkf7hWAvniiOyLqJGqS8IaAZkjJiTuWM=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=d+uwPegOqtmYA0OhZgn/2lGM7vllp+xf9I/KRCnBFIFbP4SKU6Uysl0x+qsBEyBHv2l/eM0FN1jjYtdHedrFTpM6vHdfhtbipe/2jLqJKjRkZeMdmNi1ttvFmMqjChIElQOs9tKKLb6pwu0Dmclxb4aQxxJy78otEAjZf+onNLI= ARC-Authentication-Results: i=2; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=collabora.com; spf=pass smtp.mailfrom=collabora.com; dkim=pass (1024-bit key) header.d=collabora.com header.i=adrian.larumbe@collabora.com header.b=k74wl4b0; arc=pass smtp.client-ip=136.143.188.112 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=collabora.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=collabora.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (1024-bit key) header.d=collabora.com header.i=adrian.larumbe@collabora.com header.b="k74wl4b0" ARC-Seal: i=1; a=rsa-sha256; t=1756944683; cv=none; d=zohomail.com; s=zohoarc; b=EZ6OQoFpVOKv403w/n03HuhNYjZzVkBSY4dkw785QHnmPYcGVRquwX0ZDWM1+jKKRteoqzYQytIgtycC9JJ+HPR4vqDyMtJIGMFHLtl9roFQvD5kK2PYacgdajYdhqOsSK2wJAlHQ6j7cR5knERmR6C9M2mJvhIhMYHJLycYsFA= ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=zohomail.com; s=zohoarc; t=1756944683; h=Content-Type:Content-Transfer-Encoding:Cc:Cc:Date:Date:From:From:In-Reply-To:MIME-Version:Message-ID:References:Subject:Subject:To:To:Message-Id:Reply-To; bh=5ESHoHeDS3sgI1D34jKriQd6dTyT+3g/FWPiSQzFi58=; b=GW19dTC1yVBA4nH1vmHjf+l1psiDitfdLBb/OIQiDn0IhRgEHUn5lDZo7pzFW0GlBV1y78jql24pSVN/+WfGflDQ4AtfePkQtL2KklqI0waUnmyslI+RLo1x9IPGCIYdTMFvBuaL9OX+8pb4/9Z3Sris/49XudZh7ikiJs6ue+s= ARC-Authentication-Results: i=1; mx.zohomail.com; dkim=pass header.i=collabora.com; spf=pass smtp.mailfrom=adrian.larumbe@collabora.com; dmarc=pass header.from= DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; t=1756944683; s=zohomail; d=collabora.com; i=adrian.larumbe@collabora.com; h=From:From:To:To:Cc:Cc:Subject:Subject:Date:Date:Message-ID:In-Reply-To:References:MIME-Version:Content-Type:Content-Transfer-Encoding:Message-Id:Reply-To; bh=5ESHoHeDS3sgI1D34jKriQd6dTyT+3g/FWPiSQzFi58=; b=k74wl4b0eW+rDVRCvt0NDCI7BR9TFaO7iPJXYmvwhotmcHBz+db+Tth6moH3cJiM LR2yJk83FAIS74RJ7jX5mAmUYO9Ujo+NV6ohi96ha29ZgyJ2B+B9LxnDixk8+NQsX4d jZqTiJ7UGriFy7d2jnpzTGrgBXsOpTpwpLV1HoGE= Received: by mx.zohomail.com with SMTPS id 1756944680752565.3135719019061; Wed, 3 Sep 2025 17:11:20 -0700 (PDT) From: =?UTF-8?q?Adri=C3=A1n=20Larumbe?= To: linux-kernel@vger.kernel.org Cc: dri-devel@lists.freedesktop.org, Steven Price , Boris Brezillon , kernel@collabora.com, =?UTF-8?q?Adri=C3=A1n=20Larumbe?= , Rob Herring , Maarten Lankhorst , Maxime Ripard , Thomas Zimmermann , David Airlie , Simona Vetter Subject: [PATCH v2 2/4] drm/panfrost: Introduce JM contexts for manging job resources Date: Thu, 4 Sep 2025 01:08:00 +0100 Message-ID: <20250904001054.147465-3-adrian.larumbe@collabora.com> X-Mailer: git-send-email 2.50.0 In-Reply-To: <20250904001054.147465-1-adrian.larumbe@collabora.com> References: <20250904001054.147465-1-adrian.larumbe@collabora.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable From: Boris Brezillon A JM context describes GPU HW resourdce allocation for the jobs bound to it. At the time of writing this, it only holds the JS and queue scheduler priorities. When a context is created, all the scheduler entities created for job slots will have the same priority. Until context creation and destruction and attaching a context ID to a job submission are exposed to UM, all jobs shall be bound to the default Panfrost file context, which has medium priority. Signed-off-by: Boris Brezillon Signed-off-by: Adri=C3=A1n Larumbe --- drivers/gpu/drm/panfrost/panfrost_device.h | 4 +- drivers/gpu/drm/panfrost/panfrost_drv.c | 19 +- drivers/gpu/drm/panfrost/panfrost_job.c | 203 +++++++++++++++++---- drivers/gpu/drm/panfrost/panfrost_job.h | 25 ++- 4 files changed, 210 insertions(+), 41 deletions(-) diff --git a/drivers/gpu/drm/panfrost/panfrost_device.h b/drivers/gpu/drm/p= anfrost/panfrost_device.h index 077525a3ad68..5b164871eb95 100644 --- a/drivers/gpu/drm/panfrost/panfrost_device.h +++ b/drivers/gpu/drm/panfrost/panfrost_device.h @@ -15,6 +15,7 @@ #include =20 #include "panfrost_devfreq.h" +#include "panfrost_job.h" =20 struct panfrost_device; struct panfrost_mmu; @@ -22,7 +23,6 @@ struct panfrost_job_slot; struct panfrost_job; struct panfrost_perfcnt; =20 -#define NUM_JOB_SLOTS 3 #define MAX_PM_DOMAINS 5 =20 enum panfrost_drv_comp_bits { @@ -206,7 +206,7 @@ struct panfrost_engine_usage { struct panfrost_file_priv { struct panfrost_device *pfdev; =20 - struct drm_sched_entity sched_entity[NUM_JOB_SLOTS]; + struct xarray jm_ctxs; =20 struct panfrost_mmu *mmu; =20 diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panf= rost/panfrost_drv.c index 1ea6c509a5d5..398c067457d9 100644 --- a/drivers/gpu/drm/panfrost/panfrost_drv.c +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c @@ -279,6 +279,7 @@ static int panfrost_ioctl_submit(struct drm_device *dev= , void *data, struct panfrost_file_priv *file_priv =3D file->driver_priv; struct drm_panfrost_submit *args =3D data; struct drm_syncobj *sync_out =3D NULL; + struct panfrost_jm_ctx *jm_ctx; struct panfrost_job *job; int ret =3D 0, slot; =20 @@ -294,10 +295,17 @@ static int panfrost_ioctl_submit(struct drm_device *d= ev, void *data, return -ENODEV; } =20 + /* TODO: Use the default JM context until ctx management IOCTLs are expos= ed */ + jm_ctx =3D panfrost_jm_ctx_from_handle(file, 0); + if (!jm_ctx) { + ret =3D -EINVAL; + goto out_put_syncout; + } + job =3D kzalloc(sizeof(*job), GFP_KERNEL); if (!job) { ret =3D -ENOMEM; - goto out_put_syncout; + goto out_put_jm_ctx; } =20 kref_init(&job->refcount); @@ -307,12 +315,13 @@ static int panfrost_ioctl_submit(struct drm_device *d= ev, void *data, job->requirements =3D args->requirements; job->flush_id =3D panfrost_gpu_get_latest_flush_id(pfdev); job->mmu =3D file_priv->mmu; + job->ctx =3D panfrost_jm_ctx_get(jm_ctx); job->engine_usage =3D &file_priv->engine_usage; =20 slot =3D panfrost_job_get_slot(job); =20 ret =3D drm_sched_job_init(&job->base, - &file_priv->sched_entity[slot], + &jm_ctx->slots[slot].sched_entity, 1, NULL, file->client_id); if (ret) goto out_put_job; @@ -338,6 +347,8 @@ static int panfrost_ioctl_submit(struct drm_device *dev= , void *data, drm_sched_job_cleanup(&job->base); out_put_job: panfrost_job_put(job); +out_put_jm_ctx: + panfrost_jm_ctx_put(jm_ctx); out_put_syncout: if (sync_out) drm_syncobj_put(sync_out); @@ -564,7 +575,7 @@ panfrost_open(struct drm_device *dev, struct drm_file *= file) goto err_free; } =20 - ret =3D panfrost_job_open(panfrost_priv); + ret =3D panfrost_job_open(file); if (ret) goto err_job; =20 @@ -583,7 +594,7 @@ panfrost_postclose(struct drm_device *dev, struct drm_f= ile *file) struct panfrost_file_priv *panfrost_priv =3D file->driver_priv; =20 panfrost_perfcnt_close(file); - panfrost_job_close(panfrost_priv); + panfrost_job_close(file); =20 panfrost_mmu_ctx_put(panfrost_priv->mmu); kfree(panfrost_priv); diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panf= rost/panfrost_job.c index 82acabb21b27..b6853add307c 100644 --- a/drivers/gpu/drm/panfrost/panfrost_job.c +++ b/drivers/gpu/drm/panfrost/panfrost_job.c @@ -8,6 +8,7 @@ #include #include #include +#include #include #include =20 @@ -22,6 +23,7 @@ #include "panfrost_mmu.h" #include "panfrost_dump.h" =20 +#define MAX_JM_CTX_PER_FILE 128 #define JOB_TIMEOUT_MS 500 =20 #define job_write(dev, reg, data) writel(data, dev->iomem + (reg)) @@ -222,7 +224,7 @@ static void panfrost_job_hw_submit(struct panfrost_job = *job, int js) =20 /* start MMU, medium priority, cache clean/flush on end, clean/flush on * start */ - cfg |=3D JS_CONFIG_THREAD_PRI(8) | + cfg |=3D job->ctx->config | JS_CONFIG_START_FLUSH_CLEAN_INVALIDATE | JS_CONFIG_END_FLUSH_CLEAN_INVALIDATE | panfrost_get_job_chain_flag(job); @@ -359,6 +361,7 @@ static void panfrost_job_cleanup(struct kref *ref) kvfree(job->bos); } =20 + panfrost_jm_ctx_put(job->ctx); kfree(job); } =20 @@ -917,39 +920,184 @@ void panfrost_job_fini(struct panfrost_device *pfdev) destroy_workqueue(pfdev->reset.wq); } =20 -int panfrost_job_open(struct panfrost_file_priv *panfrost_priv) +int panfrost_job_open(struct drm_file *file) +{ + struct panfrost_file_priv *panfrost_priv =3D file->driver_priv; + int ret; + + struct drm_panfrost_jm_ctx_create default_jm_ctx =3D { + .priority =3D PANFROST_JM_CTX_PRIORITY_MEDIUM, + }; + + xa_init_flags(&panfrost_priv->jm_ctxs, XA_FLAGS_ALLOC); + + ret =3D panfrost_jm_ctx_create(file, &default_jm_ctx); + if (ret) + return ret; + + /* We expect the default context to be assigned handle 0. */ + if (WARN_ON(default_jm_ctx.handle)) + return -EINVAL; + + return 0; +} + +void panfrost_job_close(struct drm_file *file) +{ + struct panfrost_file_priv *panfrost_priv =3D file->driver_priv; + struct panfrost_jm_ctx *jm_ctx; + unsigned long i; + + xa_for_each(&panfrost_priv->jm_ctxs, i, jm_ctx) + panfrost_jm_ctx_destroy(file, i); + + xa_destroy(&panfrost_priv->jm_ctxs); +} + +int panfrost_job_is_idle(struct panfrost_device *pfdev) { - struct panfrost_device *pfdev =3D panfrost_priv->pfdev; struct panfrost_job_slot *js =3D pfdev->js; - struct drm_gpu_scheduler *sched; - int ret, i; + int i; =20 for (i =3D 0; i < NUM_JOB_SLOTS; i++) { - sched =3D &js->queue[i].sched; - ret =3D drm_sched_entity_init(&panfrost_priv->sched_entity[i], - DRM_SCHED_PRIORITY_NORMAL, &sched, - 1, NULL); - if (WARN_ON(ret)) - return ret; + /* If there are any jobs in the HW queue, we're not idle */ + if (atomic_read(&js->queue[i].sched.credit_count)) + return false; } + + return true; +} + +static void panfrost_jm_ctx_release(struct kref *kref) +{ + struct panfrost_jm_ctx *jm_ctx =3D container_of(kref, struct panfrost_jm_= ctx, refcnt); + + kfree(jm_ctx); +} + +void +panfrost_jm_ctx_put(struct panfrost_jm_ctx *jm_ctx) +{ + if (jm_ctx) + kref_put(&jm_ctx->refcnt, panfrost_jm_ctx_release); +} + +struct panfrost_jm_ctx * +panfrost_jm_ctx_get(struct panfrost_jm_ctx *jm_ctx) +{ + if (jm_ctx) + kref_get(&jm_ctx->refcnt); + + return jm_ctx; +} + +struct panfrost_jm_ctx * +panfrost_jm_ctx_from_handle(struct drm_file *file, u32 handle) +{ + struct panfrost_file_priv *priv =3D file->driver_priv; + struct panfrost_jm_ctx *jm_ctx; + + xa_lock(&priv->jm_ctxs); + jm_ctx =3D panfrost_jm_ctx_get(xa_load(&priv->jm_ctxs, handle)); + xa_unlock(&priv->jm_ctxs); + + return jm_ctx; +} + +static int jm_ctx_prio_to_drm_sched_prio(struct drm_file *file, + enum drm_panfrost_jm_ctx_priority in, + enum drm_sched_priority *out) +{ + switch (in) { + case PANFROST_JM_CTX_PRIORITY_LOW: + *out =3D DRM_SCHED_PRIORITY_LOW; + return 0; + case PANFROST_JM_CTX_PRIORITY_MEDIUM: + *out =3D DRM_SCHED_PRIORITY_NORMAL; + return 0; + case PANFROST_JM_CTX_PRIORITY_HIGH: + /* Higher priorities require CAP_SYS_NICE or DRM_MASTER */ + if (!capable(CAP_SYS_NICE) && !drm_is_current_master(file)) + return -EACCES; + + *out =3D DRM_SCHED_PRIORITY_HIGH; + return 0; + default: + return -EINVAL; + } +} + +int panfrost_jm_ctx_create(struct drm_file *file, + struct drm_panfrost_jm_ctx_create *args) +{ + struct panfrost_file_priv *priv =3D file->driver_priv; + struct panfrost_device *pfdev =3D priv->pfdev; + enum drm_sched_priority sched_prio; + struct panfrost_jm_ctx *jm_ctx; + + int ret; + + jm_ctx =3D kzalloc(sizeof(*jm_ctx), GFP_KERNEL); + if (!jm_ctx) + return -ENOMEM; + + kref_init(&jm_ctx->refcnt); + + /* Same priority for all JS within a single context */ + jm_ctx->config =3D JS_CONFIG_THREAD_PRI(args->priority); + + ret =3D jm_ctx_prio_to_drm_sched_prio(file, args->priority, &sched_prio); + if (ret) + goto err_put_jm_ctx; + + for (u32 i =3D 0; i < NUM_JOB_SLOTS - 1; i++) { + struct drm_gpu_scheduler *sched =3D &pfdev->js->queue[i].sched; + struct panfrost_js_ctx *js_ctx =3D &jm_ctx->slots[i]; + + ret =3D drm_sched_entity_init(&js_ctx->sched_entity, sched_prio, + &sched, 1, NULL); + if (ret) + goto err_put_jm_ctx; + + js_ctx->enabled =3D true; + } + + ret =3D xa_alloc(&priv->jm_ctxs, &args->handle, jm_ctx, + XA_LIMIT(0, MAX_JM_CTX_PER_FILE), GFP_KERNEL); + if (ret) + goto err_put_jm_ctx; + return 0; + +err_put_jm_ctx: + panfrost_jm_ctx_put(jm_ctx); + return ret; } =20 -void panfrost_job_close(struct panfrost_file_priv *panfrost_priv) +int panfrost_jm_ctx_destroy(struct drm_file *file, u32 handle) { - struct panfrost_device *pfdev =3D panfrost_priv->pfdev; - int i; + struct panfrost_file_priv *priv =3D file->driver_priv; + struct panfrost_device *pfdev =3D priv->pfdev; + struct panfrost_jm_ctx *jm_ctx; =20 - for (i =3D 0; i < NUM_JOB_SLOTS; i++) - drm_sched_entity_destroy(&panfrost_priv->sched_entity[i]); + jm_ctx =3D xa_erase(&priv->jm_ctxs, handle); + if (!jm_ctx) + return -EINVAL; + + for (u32 i =3D 0; i < ARRAY_SIZE(jm_ctx->slots); i++) { + if (jm_ctx->slots[i].enabled) + drm_sched_entity_destroy(&jm_ctx->slots[i].sched_entity); + } =20 /* Kill in-flight jobs */ spin_lock(&pfdev->js->job_lock); - for (i =3D 0; i < NUM_JOB_SLOTS; i++) { - struct drm_sched_entity *entity =3D &panfrost_priv->sched_entity[i]; - int j; + for (u32 i =3D 0; i < ARRAY_SIZE(jm_ctx->slots); i++) { + struct drm_sched_entity *entity =3D &jm_ctx->slots[i].sched_entity; + + if (!jm_ctx->slots[i].enabled) + continue; =20 - for (j =3D ARRAY_SIZE(pfdev->jobs[0]) - 1; j >=3D 0; j--) { + for (int j =3D ARRAY_SIZE(pfdev->jobs[0]) - 1; j >=3D 0; j--) { struct panfrost_job *job =3D pfdev->jobs[i][j]; u32 cmd; =20 @@ -980,18 +1128,7 @@ void panfrost_job_close(struct panfrost_file_priv *pa= nfrost_priv) } } spin_unlock(&pfdev->js->job_lock); -} =20 -int panfrost_job_is_idle(struct panfrost_device *pfdev) -{ - struct panfrost_job_slot *js =3D pfdev->js; - int i; - - for (i =3D 0; i < NUM_JOB_SLOTS; i++) { - /* If there are any jobs in the HW queue, we're not idle */ - if (atomic_read(&js->queue[i].sched.credit_count)) - return false; - } - - return true; + panfrost_jm_ctx_put(jm_ctx); + return 0; } diff --git a/drivers/gpu/drm/panfrost/panfrost_job.h b/drivers/gpu/drm/panf= rost/panfrost_job.h index ec581b97852b..f3e5010e6cf5 100644 --- a/drivers/gpu/drm/panfrost/panfrost_job.h +++ b/drivers/gpu/drm/panfrost/panfrost_job.h @@ -18,6 +18,7 @@ struct panfrost_job { =20 struct panfrost_device *pfdev; struct panfrost_mmu *mmu; + struct panfrost_jm_ctx *ctx; =20 /* Fence to be signaled by IRQ handler when the job is complete. */ struct dma_fence *done_fence; @@ -39,10 +40,30 @@ struct panfrost_job { u64 start_cycles; }; =20 +struct panfrost_js_ctx { + struct drm_sched_entity sched_entity; + bool enabled; +}; + +#define NUM_JOB_SLOTS 3 + +struct panfrost_jm_ctx { + struct kref refcnt; + u32 config; + struct panfrost_js_ctx slots[NUM_JOB_SLOTS]; +}; + +int panfrost_jm_ctx_create(struct drm_file *file, + struct drm_panfrost_jm_ctx_create *args); +int panfrost_jm_ctx_destroy(struct drm_file *file, u32 handle); +void panfrost_jm_ctx_put(struct panfrost_jm_ctx *jm_ctx); +struct panfrost_jm_ctx *panfrost_jm_ctx_get(struct panfrost_jm_ctx *jm_ctx= ); +struct panfrost_jm_ctx *panfrost_jm_ctx_from_handle(struct drm_file *file,= u32 handle); + int panfrost_job_init(struct panfrost_device *pfdev); void panfrost_job_fini(struct panfrost_device *pfdev); -int panfrost_job_open(struct panfrost_file_priv *panfrost_priv); -void panfrost_job_close(struct panfrost_file_priv *panfrost_priv); +int panfrost_job_open(struct drm_file *file); +void panfrost_job_close(struct drm_file *file); int panfrost_job_get_slot(struct panfrost_job *job); int panfrost_job_push(struct panfrost_job *job); void panfrost_job_put(struct panfrost_job *job); --=20 2.50.0