linux-kernel - Re: [PATCH 2/5] drm/nouveau/uvmm: Allow larger pages

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <DDBISJ2DUDF6.150HCB14ZRPH3@kernel.org>
Date: Mon, 06 Oct 2025 22:26:24 +0200
From: "Danilo Krummrich" <dakr@...nel.org>
To: "Mohamed Ahmed" <mohamedahmedegypt2001@...il.com>
Cc: <linux-kernel@...r.kernel.org>, <dri-devel@...ts.freedesktop.org>, "Mary
 Guillemard" <mary@...y.zone>, "Faith Ekstrand"
 <faith.ekstrand@...labora.com>, "Lyude Paul" <lyude@...hat.com>, "Maarten
 Lankhorst" <maarten.lankhorst@...ux.intel.com>, "Maxime Ripard"
 <mripard@...nel.org>, "Thomas Zimmermann" <tzimmermann@...e.de>, "David
 Airlie" <airlied@...il.com>, "Simona Vetter" <simona@...ll.ch>,
 <nouveau@...ts.freedesktop.org>
Subject: Re: [PATCH 2/5] drm/nouveau/uvmm: Allow larger pages

On Mon Oct 6, 2025 at 9:13 PM CEST, Mohamed Ahmed wrote:
> From: Mary Guillemard <mary@...y.zone>
>
> Now that everything in UVMM knows about the variable page shift, we can
> select larger values.
>
> The proposed approach rely on nouveau_bo::page unless it would cause
> alignment issues (in which case we fall back to searching an appropriate
> shift)
>
> Co-developed-by: Mohamed Ahmed <mohamedahmedegypt2001@...il.com>
> Signed-off-by: Mohamed Ahmed <mohamedahmedegypt2001@...il.com>
> Signed-off-by: Mary Guillemard <mary@...y.zone>

NIT: Both of your tags should come after Mary's tag. The same applied to some of
the other patches.

> ---
>  drivers/gpu/drm/nouveau/nouveau_uvmm.c | 55 +++++++++++++++++++++++++-
>  1 file changed, 53 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/nouveau/nouveau_uvmm.c b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
> index a92c729600d6..c336a121e320 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_uvmm.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_uvmm.c
> @@ -454,6 +454,56 @@ op_unmap_prepare_unwind(struct drm_gpuva *va)
>  	drm_gpuva_insert(va->vm, va);
>  }
>  
> +static bool
> +op_map_aligned_to_page_shift(const struct drm_gpuva_op_map *op, u8 page_shift)
> +{
> +	u64 page_size = 1ULL << page_shift;
> +
> +	return op->va.addr % page_size == 0 && op->va.range % page_size == 0 &&
> +		   op->gem.offset % page_size == 0;
> +}
> +
> +static u8
> +select_page_shift(struct nouveau_uvmm *uvmm, struct drm_gpuva_op_map *op)
> +{
> +	struct nouveau_bo *nvbo = nouveau_gem_object(op->gem.obj);
> +
> +	if (nvbo) {

In nouveau a struct drm_gpuva_op_map always has a valid GEM object set; we bail
out if userspace gives us an invalid GEM handle.

> +		/* If the BO preferred page shift already fits, use it. */
> +		if (op_map_aligned_to_page_shift(op, nvbo->page))
> +			return nvbo->page;
> +
> +		struct nouveau_mem *mem = nouveau_mem(nvbo->bo.resource);
> +		struct nvif_vmm *vmm = &uvmm->vmm.vmm;
> +		int i;
> +
> +		/* Otherwise let's find a granuality that will fit. */

Do we ever run into the "otherwise" case? nouveau_bo_fixup_align() seems to
already ensure that your previous call will never fail?

> +		for (i = 0; i < vmm->page_nr; i++) {
> +			/* Ignore anything that is bigger or identical to the BO preference. */
> +			if (vmm->page[i].shift >= nvbo->page)
> +				continue;
> +
> +			/* Skip incompatible domains. */
> +			if ((mem->mem.type & NVIF_MEM_VRAM) && !vmm->page[i].vram)
> +				continue;
> +			if ((mem->mem.type & NVIF_MEM_HOST) &&
> +			    (!vmm->page[i].host || vmm->page[i].shift > PAGE_SHIFT))
> +				continue;
> +
> +			/* If it fits, return the proposed shift. */
> +			if (op_map_aligned_to_page_shift(op, vmm->page[i].shift))
> +				return vmm->page[i].shift;
> +		}
> +
> +		/* If we get here then nothing can reconcile the requirements. This should never
> +		 * happen.
> +		 */
> +		WARN_ON(1);
> +	}
> +
> +	return PAGE_SHIFT;
> +}
> +
>  static void
>  nouveau_uvmm_sm_prepare_unwind(struct nouveau_uvmm *uvmm,
>  			       struct nouveau_uvma_prealloc *new,
> @@ -506,7 +556,7 @@ nouveau_uvmm_sm_prepare_unwind(struct nouveau_uvmm *uvmm,
>  			if (vmm_get_range)
>  				nouveau_uvmm_vmm_put(uvmm, vmm_get_start,
>  						     vmm_get_range,
> -						     PAGE_SHIFT);
> +						     select_page_shift(uvmm, &op->map));
>  			break;
>  		}
>  		case DRM_GPUVA_OP_REMAP: {
> @@ -636,7 +686,8 @@ nouveau_uvmm_sm_prepare(struct nouveau_uvmm *uvmm,
>  		case DRM_GPUVA_OP_MAP: {
>  			u64 vmm_get_range = vmm_get_end - vmm_get_start;
>  
> -			ret = op_map_prepare(uvmm, &new->map, &op->map, args, PAGE_SHIFT);
> +			ret = op_map_prepare(uvmm, &new->map, &op->map, args,
> +					     select_page_shift(uvmm, &op->map));

Let's move the call to select_page_shift() into op_map_prepare().

>  			if (ret)
>  				goto unwind;
>  
> -- 
> 2.51.0