** Description changed:

  [ Impact ]
  
-  * The rocalution 10.0-0ubuntu1 test suite (rocalution-test, run via
-    debian/tests/upstream-binaries during dh_auto_test / autopkgtest) contains
-    a stack buffer overflow in testing_extract_coarse_mapping()
-    (clients/include/testing_local_vector.hpp). The test declares 3-element
-    `index` and `map` arrays but calls
-    `LocalVector::ExtractCoarseMapping(0, 5, index, 3, &size, map)`, which
-    reads/writes 5 elements (end - start = 5) of both arrays. This is
-    undefined behaviour: on amd64/arm64 it happens not to corrupt anything
-    observable, but on ppc64el the differing stack layout causes the
-    overflow to clobber adjacent stack data, crashing the rocalution-test
-    binary with a segfault during dh_auto_test and blocking the build/test
-    on that architecture, as can be seen on 
(https://launchpadlibrarian.net/878532915/buildlog_ubuntu-stonking-ppc64el.rocalution_10.0-0ubuntu1_BUILDING.txt.gz).
-  * Patch 0002-fix-testing_extract_coarse_mapping-stack-overflow.patch
-    enlarges `index` and `map` to 5 elements (matching the actual range
-    used) and populates `index` with the full 0..4 sequence, eliminating
-    the out-of-bounds access while preserving the test's intent. This is a
-    test-only fix; no library source, ABI, or public API is touched.
+  * The rocalution 10.0-0ubuntu1 test suite (rocalution-test, run via
+    debian/tests/upstream-binaries during dh_auto_test / autopkgtest) contains
+    a stack buffer overflow in testing_extract_coarse_mapping()
+    (clients/include/testing_local_vector.hpp). The test declares 3-element
+    `index` and `map` arrays but calls
+    `LocalVector::ExtractCoarseMapping(0, 5, index, 3, &size, map)`, which
+    reads/writes 5 elements (end - start = 5) of both arrays. This is
+    undefined behaviour: on amd64/arm64 it happens not to corrupt anything
+    observable, but on ppc64el the differing stack layout causes the
+    overflow to clobber adjacent stack data, crashing the rocalution-test
+    binary with a segfault during dh_auto_test and blocking the build/test
+    on that architecture, as can be seen on 
(https://launchpadlibrarian.net/878532915/buildlog_ubuntu-stonking-ppc64el.rocalution_10.0-0ubuntu1_BUILDING.txt.gz).
+  * Patch 0002-fix-testing_extract_coarse_mapping-stack-overflow.patch
+    enlarges `index` and `map` to 5 elements (matching the actual range
+    used) and populates `index` with the full 0..4 sequence, eliminating
+    the out-of-bounds access while preserving the test's intent. This is a
+    test-only fix; no library source, ABI, or public API is touched.
  
  [ Test Plan ]
  
  1. Build:
-    - dpkg-buildpackage / sbuild succeeds on amd64, arm64, and ppc64el.
-    - dh_auto_test completes without rocalution-test crashing/segfaulting
-      on ppc64el.
+    - dpkg-buildpackage / sbuild succeeds on amd64, arm64, and ppc64el.
+    - dh_auto_test completes without rocalution-test crashing/segfaulting
+      on ppc64el.
  2. Installability:
-    - apt install librocalution1 librocalution1-tests.
-    - Reverse dependencies remain installable without rebuild (no ABI
-      change).
+    - apt install librocalution1 librocalution1-tests.
+    - Reverse dependencies remain installable without rebuild (no ABI
+      change).
  3. Autopkgtest:
-    - Run the autopkgtest suite (Test-Command:
-      debian/tests/upstream-binaries librocalution1-tests) on amd64,
-      arm64, and ppc64el.
-    - Output: <TBD — paste autopkgtest run results here>
+    - Run the autopkgtest suite (Test-Command:
+      debian/tests/upstream-binaries librocalution1-tests) on amd64,
+      arm64, and ppc64el.
+    - Output: <TBD — paste autopkgtest run results here>
  
  [ Where problems could occur ]
  
-  * The change only widens two local test arrays and adds explicit
-    initializer values; it does not alter the ExtractCoarseMapping
-    implementation itself, so risk is limited to the test binary. A
-    plausible regression would be the test now passing a differently
-    shaped index array and silently exercising a different code path in
-    ExtractCoarseMapping, though the range (0..5) and count (5) passed to
-    the call are unchanged, so behaviour of the function under test is
-    equivalent.
-  * If a future upstream merge reintroduces the original undersized arrays
-    during a version bump, the same ppc64el crash could reappear; the
-    patch should be re-checked against upstream test sources at that time.
+  * The change only widens two local test arrays and adds explicit
+    initializer values; it does not alter the ExtractCoarseMapping
+    implementation itself, so risk is limited to the test binary. A
+    plausible regression would be the test now passing a differently
+    shaped index array and silently exercising a different code path in
+    ExtractCoarseMapping, though the range (0..5) and count (5) passed to
+    the call are unchanged, so behaviour of the function under test is
+    equivalent.
+  * If a future upstream merge reintroduces the original undersized arrays
+    during a version bump, the same ppc64el crash could reappear; the
+    patch should be re-checked against upstream test sources at that time.
  
  [ Other Info ]
  
-  * No ABI/symbols impact: this is a test-source-only change
-    (clients/include/testing_local_vector.hpp), not shipped in any binary
-    package's public interface.
-  * Forwarded: 
-  * Target: resolute 26.10
+  * No ABI/symbols impact: this is a test-source-only change
+    (clients/include/testing_local_vector.hpp), not shipped in any binary
+    package's public interface.
+  * Forwarded: https://github.com/ROCm/rocm-libraries/issues/12703
+  * Target: resolute 26.10

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2168844

Title:
  testing_extract_coarse_mapping test crash on ppc64el

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/rocalution/+bug/2168844/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to