Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     linux-kernel@vger.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        stable@vger.kernel.org, Felix Kuehling <Felix.Kuehling@amd.com>,
        =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>,
        David Yat Sin <david.yatsin@amd.com>,
        Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>,
        Alex Deucher <alexander.deucher@amd.com>,
        Sasha Levin <sashal@kernel.org>
Subject: [PATCH 5.15 595/846] drm/amdgpu: Dont inherit GEM object VMAs in child process
Date:   Mon, 24 Jan 2022 19:41:52 +0100
Message-Id: <20220124184121.566094072@linuxfoundation.org>
In-Reply-To: <20220124184100.867127425@linuxfoundation.org>
References: <20220124184100.867127425@linuxfoundation.org>
User-Agent: quilt/0.66
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>

[ Upstream commit fbcdbfde87509d523132b59f661a355c731139d0 ]

When an application having open file access to a node forks, its shared
mappings also get reflected in the address space of child process even
though it cannot access them with the object permissions applied. With the
existing permission checks on the gem objects, it might be reasonable to
also create the VMAs with VM_DONTCOPY flag so a user space application
doesn't need to explicitly call the madvise(addr, len, MADV_DONTFORK)
system call to prevent the pages in the mapped range to appear in the
address space of the child process. It also prevents the memory leaks
due to additional reference counts on the mapped BOs in the child
process that prevented freeing the memory in the parent for which we had
worked around earlier in the user space inside the thunk library.

Additionally, we faced this issue when using CRIU to checkpoint restore
an application that had such inherited mappings in the child which
confuse CRIU when it mmaps on restore. Having this flag set for the
render node VMAs helps. VMAs mapped via KFD already take care of this so
this is needed only for the render nodes.

To limit the impact of the change to user space consumers such as OpenGL
etc, limit it to KFD BOs only.

Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: David Yat Sin <david.yatsin@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index a1e63ba4c54a5..630dc99e49086 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -264,6 +264,9 @@ static int amdgpu_gem_object_mmap(struct drm_gem_object *obj, struct vm_area_str
 	    !(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
 		vma->vm_flags &= ~VM_MAYWRITE;
 
+	if (bo->kfd_bo)
+		vma->vm_flags |= VM_DONTCOPY;
+
 	return drm_gem_ttm_mmap(obj, vma);
 }
 
-- 
2.34.1