Skip to content

[DeviceASAN] Simplify shadow mapping logic for GPU device #2274

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 4 commits into from
Closed

[DeviceASAN] Simplify shadow mapping logic for GPU device #2274

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
27e3d37
[DeviceASAN] Simplify shadow mapping logic for GPU device
zhaomaosu Nov 1, 2024
d319540
Merge branch 'sycl' into simplify-shadow-mapping
zhaomaosu Nov 4, 2024
39b7bc2
Update MemToShadow for DG2 device
zhaomaosu Nov 4, 2024
7f2c1d9
Merge branch 'sycl' into simplify-shadow-mapping
zhaomaosu Nov 6, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 3 additions & 6 deletions source/loader/layers/sanitizer/asan_shadow.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -256,12 +256,9 @@ ur_result_t ShadowMemoryGPU::ReleaseShadow(std::shared_ptr<AllocInfo> AI) {
}

uptr ShadowMemoryPVC::MemToShadow(uptr Ptr) {
if (Ptr & 0xFF00000000000000ULL) { // Device USM
return ShadowBegin + 0x80000000000ULL +
((Ptr & 0xFFFFFFFFFFFFULL) >> ASAN_SHADOW_SCALE);
} else { // Only consider 47bit VA
return ShadowBegin + ((Ptr & 0x7FFFFFFFFFFFULL) >> ASAN_SHADOW_SCALE);
}
return ShadowBegin + (((((Ptr & 0x8000'0000'0000'0000) >> 16) + Ptr) &
0xf'ffff'ffff'ffff) >>
ASAN_SHADOW_SCALE);
}

uptr ShadowMemoryDG2::MemToShadow(uptr Ptr) {
Expand Down
6 changes: 6 additions & 0 deletions source/loader/layers/sanitizer/ur_sanddi.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,6 +73,12 @@ __urdlllocal ur_result_t UR_APICALL urAdapterGet(
return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
}

// Set env AllocateHostAllocationsInHeapExtendedHost to 0, then host and
// shared usm will have same allocation range on GPU device. This will
// simplify shadow memory mapping logic on GPU device.
setenv("NEOReadDebugKeys", "1", 1);
Copy link
Contributor

@pbalcer pbalcer Nov 7, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@nrspruit is it a good idea to set these env variables in the layer? If this is really required, wouldn't it be a better idea to make this into an L0 API?
Using a debug key for functionality doesn't seem like a good idea.

Copy link
Contributor

@pbalcer pbalcer Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't feel comfortable merging this patch. We need confirmation from the NEO folks that this makes sense.
@MichalMrozek can you help?

Copy link
Contributor

@MichalMrozek MichalMrozek Nov 26, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, this is not correct to set this.
Please do not merge it.
If you want to have specific release keys avaialable this needs discussion with driver team.

Copy link
Contributor

@MichalMrozek MichalMrozek Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

AllocateHostAllocationsInHeapExtendedHost is a debug key, debug keys cannot be used in production software. They are meant only for experiments. Driver team may remove or change this debug variable at any point of time without any notifications.

Copy link
Contributor

@AllanZyne AllanZyne Nov 27, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ok, I will send this request to driver team. Thanks.

setenv("AllocateHostAllocationsInHeapExtendedHost", "0", 1);

ur_result_t result = pfnAdapterGet(NumEntries, phAdapters, pNumAdapters);
if (result == UR_RESULT_SUCCESS && phAdapters) {
const uint32_t NumAdapters = pNumAdapters ? *pNumAdapters : NumEntries;
Expand Down
Loading