Hello!
Here is v5 of the mremap start address optimization / fix for exec warning.
Description of patches ====================== These patches optimizes the start addresses in move_page_tables() and tests the changes. It addresses a warning [1] that occurs due to a downward, overlapping move on a mutually-aligned offset within a PMD during exec. By initiating the copy process at the PMD level when such alignment is present, we can prevent this warning and speed up the copying process at the same time. Linus Torvalds suggested this idea.
Please check the individual patches for more details.
[1] https://lore.kernel.org/all/ZB2GTBD%2FLWTrkOiO@dhcp22.suse.cz/
Link to v4: https://lore.kernel.org/all/20230531220807.2048037-1-joel@joelfernandes.org/
History of patches: v4->v5: 1. Rebased on mainline. 2. Several improvement suggestions from Lorenzo.
v3->v4: 1. Care to be taken to move purely within a VMA, in other words this check in call_align_down(): if (vma->vm_start != addr_masked) return false;
As an example of why this is needed: Consider the following range which is 2MB aligned and is a part of a larger 10MB range which is not shown. Each character is 256KB below making the source and destination 2MB each. The lower case letters are moved (s to d) and the upper case letters are not moved.
|DDDDddddSSSSssss|
If we align down 'ssss' to start from the 'SSSS', we will end up destroying SSSS. The above if statement prevents that and I verified it.
I also added a test for this in the last patch.
2. Handle the stack case separately. We do not care about #1 for stack movement because the 'SSSS' does not matter during this move. Further we need to do this to prevent the stack move warning.
if (!for_stack && vma->vm_start <= addr_masked) return false;
v2->v3: 1. Masked address was stored in int, fixed it to unsigned long to avoid truncation. 2. We now handle moves happening purely within a VMA, a new test is added to handle this. 3. More code comments.
v1->v2: 1. Trigger the optimization for mremaps smaller than a PMD. I tested by tracing that it works correctly.
2. Fix issue with bogus return value found by Linus if we broke out of the above loop for the first PMD itself.
v1: Initial RFC.
Joel Fernandes (1): selftests: mm: Add a test for moving from an offset from start of mapping
Joel Fernandes (Google) (6): mm/mremap: Optimize the start addresses in move_page_tables() mm/mremap: Allow moves within the same VMA selftests: mm: Fix failure case when new remap region was not found selftests: mm: Add a test for mutually aligned moves > PMD size selftests: mm: Add a test for remapping to area immediately after existing mapping selftests: mm: Add a test for remapping within a range
fs/exec.c | 2 +- include/linux/mm.h | 2 +- mm/mremap.c | 69 +++++- tools/testing/selftests/mm/mremap_test.c | 301 +++++++++++++++++++---- 4 files changed, 325 insertions(+), 49 deletions(-)
-- 2.42.0.rc1.204.g551eb34607-goog