Attachment 369419 Details for Bug 197682 – Patch

[patch] Patch

bug-197682-20190508134803.patch (text/plain), 1.94 KB, created by Justin Fan on 2019-05-08 13:48:03 PDT

(hide)

Description:

Filename:

MIME Type:

Creator: Justin Fan

Created: 2019-05-08 13:48:03 PDT

Size: 1.94 KB

patch

obsolete

>Subversion Revision: 244966
>diff --git a/Source/WebCore/ChangeLog b/Source/WebCore/ChangeLog
>index c6c1f7679f545d94a40166b2a4b7d364ff85b7b1..6e5fe3d3918b568632f0e64795214536c3554c04 100644
>--- a/Source/WebCore/ChangeLog
>+++ b/Source/WebCore/ChangeLog
>@@ -1,3 +1,18 @@
>+2019-05-07  Justin Fan  <justin_fan@apple.com>
>+
>+        [WebGPU] Limit threadsPerThreadGroup for GPUComputePassEncoder.dispatch to prevent DOS on macOS
>+        https://bugs.webkit.org/show_bug.cgi?id=197682
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        Add temporary validation to prevent macOS systems from lock-up via extremely large GPU compute dispatches.
>+        Will be adjusted after WebGPU WG decides on the specification for GPUComputePassEncoder.dispatch.
>+
>+        No new tests as error handling is not hooked up yet.
>+
>+        * platform/graphics/gpu/cocoa/GPUComputePassEncoderMetal.mm:
>+        (WebCore::GPUComputePassEncoder::dispatch):
>+
> 2019-05-06  Zan Dobersek  <zdobersek@igalia.com>
> 
>         [GLib] WebCore::MainThreadSharedTimer should use the appropriate GSource priority, name
>diff --git a/Source/WebCore/platform/graphics/gpu/cocoa/GPUComputePassEncoderMetal.mm b/Source/WebCore/platform/graphics/gpu/cocoa/GPUComputePassEncoderMetal.mm
>index 713c1b791075a086b90b31309914a405d6afe682..9f92901b538a53d07034485c6351c9f8cbed877a 100644
>--- a/Source/WebCore/platform/graphics/gpu/cocoa/GPUComputePassEncoderMetal.mm
>+++ b/Source/WebCore/platform/graphics/gpu/cocoa/GPUComputePassEncoderMetal.mm
>@@ -92,6 +92,12 @@ void GPUComputePassEncoder::dispatch(unsigned x, unsigned y, unsigned z)
>         return;
>     }
> 
>+    // FIXME: Restrict the maximum number of threadgroups to this volume for now, to prevent extended DOS on macOS.
>+    if (x * y * z > 64000000) {
>+        LOG(WebGPU, "GPUComputePassEncoder::dispatch(): Threadgroup dispatch size too large!");
>+        return;
>+    }
>+
>     auto pipelineState = m_pipeline->platformComputePipeline();
>     ASSERT(pipelineState);
>

Flags:

mmaxfield: review-

Actions: View | Formatted Diff | Diff

Attachments on bug 197682: 369419