mirror of
git://nv-tegra.nvidia.com/linux-nvgpu.git
synced 2025-12-22 17:36:20 +03:00
e05e655dd4
Bugs in current version are listed below: 1. Function alloc() or alloc_pte() allocate memory for len=0. 2. Function alloc() or alloc_pte() don't unlock nvgpu_allocator if pte_size is invalid. 3. Function alloc_pte() and alloc_fixed() set alloc_made flag unconditionally. 4. Function release_carveout() tries to acquire nvgpu_allocator lock twice causing unresponsive state. 5. If buddy allocation fails in balloc_do_alloc() or balloc_do_alloc_fixed() function, previously allocated buddies are not merged. This causes seg fault in ops->fini(). 6. With gva_space enabled and base=0, buddy_allocator updated base not checked for pde alignment. 7. In balloc_do_alloc_fixed(), align_order computed using __fls() results in one order higher than requested. 8. Initializing buddy allocator with size=0, initializes very large memory and will trigger seg fault with the changes in this patch. Setting size=1G so that further execution is successful. This patch fixes above listed bugs and updates following: 1. With gva_space enabled, BALLOC_PTE_SIZE_ANY is considered as BALLOC_PTE_SIZE_SMALL which allows alloc() to be used. 2. GPU_BALLOC_MAX_ORDER changed to 63U. Condition added to check that max_order is never greater than GPU_BALLOC_MAX_ORDER. 3. BUG() changed to nvgpu_do_assert(). JIRA NVGPU-3005 Change-Id: I20c28e20aa3404976d67f7884b4f8cbd5c908ba7 Signed-off-by: Vedashree Vidwans <vvidwans@nvidia.com> Reviewed-on: https://git-master.nvidia.com/r/2075646 Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com> GVS: Gerrit_Virtual_Submit Reviewed-by: Alex Waterman <alexw@nvidia.com> Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com> Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>