Hi Catalin and Will， The drivers/base/arch_numa.c is only shared by riscv and arm64, and the change from patch2 won't broke riscv. Could all patches be merged by arm64 tree? or any new comments? Many thanks. On 2021/8/9 17:37, Kefeng Wang wrote:
Percpu embedded first chunk allocator is the firstly option, but it could fails on ARM64, eg, "percpu: max_distance=0x5fcfdc640000 too large for vmalloc space 0x781fefff0000" "percpu: max_distance=0x600000540000 too large for vmalloc space 0x7dffb7ff0000" "percpu: max_distance=0x5fff9adb0000 too large for vmalloc space 0x5dffb7ff0000" then we could meet "WARNING: CPU: 15 PID: 461 at vmalloc.c:3087 pcpu_get_vm_areas+0x488/0x838", even the system could not boot successfully. Let's implement page mapping percpu first chunk allocator as a fallback to the embedding allocator to increase the robustness of the system. Also fix a crash when both NEED_PER_CPU_PAGE_FIRST_CHUNK and KASAN_VMALLOC enabled. Tested on ARM64 qemu with cmdline "percpu_alloc=page" based on v5.14-rc5. v3: - search for a range that fits instead of always picking the end from vmalloc area suggested by Catalin. - use NUMA_NO_NODE to avoid "virt_to_phys used for non-linear address:" issue in arm64 kasan_populate_early_vm_area_shadow(). - add Acked-by: Marco Elver <elver@xxxxxxxxxx> to patch v3 V2: - fix build error when CONFIG_KASAN disabled, found by lkp@xxxxxxxxx - drop wrong __weak comment from kasan_populate_early_vm_area_shadow(), found by Marco Elver <elver@xxxxxxxxxx> Kefeng Wang (3): vmalloc: Choose a better start address in vm_area_register_early() arm64: Support page mapping percpu first chunk allocator kasan: arm64: Fix pcpu_page_first_chunk crash with KASAN_VMALLOC arch/arm64/Kconfig | 4 ++ arch/arm64/mm/kasan_init.c | 16 ++++++++ drivers/base/arch_numa.c | 82 +++++++++++++++++++++++++++++++++----- include/linux/kasan.h | 6 +++ mm/kasan/init.c | 5 +++ mm/vmalloc.c | 17 +++++--- 6 files changed, 115 insertions(+), 15 deletions(-)
