On 07/11/18 at 05:42pm, Chao Fan wrote: > Hi all, > > I found there is a BUG about KASLR and ZONE_MOVABLE. > > When users use 'kernelcore=' parameter without 'movable_node', > movable memory is evenly distributed to all nodes. The size of > ZONE_MOVABLE depends on the kernel parameter 'kernelcore=' and > 'movablecore='. > But sometiomes, KASLR may put the uncompressed kernel to the > tail position of a node, which will cause the kernel memory > set as ZONE_MOVABLE. This region can not be offlined. > > Here is a very simple test in my qemu-kvm machine, there is > only one node: > > The command line: > [root@localhost ~]# cat /proc/cmdline > BOOT_IMAGE=/vmlinuz-4.18.0-rc3+ root=/dev/mapper/fedora_localhost--live-root > ro resume=/dev/mapper/fedora_localhost--live-swap > rd.lvm.lv=fedora_localhost-live/root rd.lvm.lv=fedora_localhost-live/swap > console=ttyS0 earlyprintk=ttyS0,115200n8 memblock=debug kernelcore=50% > > I use 'kernelcore=50%' here. > > Here is my early print result, I print the random_addr after KASLR chooses > physical memory: > early console in extract_kernel > input_data: 0x000000000266b3b1 > input_len: 0x00000000007d8802 > output: 0x0000000001000000 > output_len: 0x0000000001e15698 > kernel_total_size: 0x0000000001a8b000 > trampoline_32bit: 0x000000000009d000 > booted via startup_32() > Physical KASLR using RDRAND RDTSC... > random_addr: 0x000000012f000000 > Virtual KASLR using RDRAND RDTSC... > > The address for kernel is 0x000000012f000000 > > Here is the log of ZONE: > [ 0.000000] Zone ranges: > [ 0.000000] DMA [mem 0x0000000000001000-0x0000000000ffffff] > [ 0.000000] DMA32 [mem 0x0000000001000000-0x00000000ffffffff] > [ 0.000000] Normal [mem 0x0000000100000000-0x00000001f57fffff] > [ 0.000000] Device empty > [ 0.000000] Movable zone start for each node > [ 0.000000] Node 0: 0x000000011b000000 > [ 0.000000] Early memory node ranges > [ 0.000000] node 0: [mem 0x0000000000001000-0x000000000009efff] > [ 0.000000] node 0: [mem 0x0000000000100000-0x00000000bffd6fff] > [ 0.000000] node 0: [mem 0x0000000100000000-0x00000001f57fffff] > [ 0.000000] Initmem setup node 0 [mem > 0x0000000000001000-0x00000001f57fffff] > > Only one node in my machine, ZONE_MOVABLE begins from 0x000000011b000000, > which is lower than 0x000000012f000000. > So KASLR put the kernel to the ZONE_MOVABLE. > Try to solve this problem, I think there should be a new tactic in function > find_zone_movable_pfns_for_nodes() of mm/page_alloc.c. If kernel is uncompressed > in a tail position, then just set the memory after the kernel as ZONE_MOVABLE, > at the same time, memory in other nodes will be set as ZONE_MOVABLE. Hmm, it's an issue, worth fixing it. Otherwise the size of movable area will be smaller than we expect when add "kernel_core=" or "movable_core=". Add a check in find_zone_movable_pfns_for_nodes(), and use min() to get the starting address of movable area between aligned '_etext' and start_pfn. It will go to label 'restart' to calculate the 2nd round if not satisfiled. Hi Chao, Could you check if below patch works for you? >From ab6e47c6a78d1a4ccb577b995b7b386f3149732f Mon Sep 17 00:00:00 2001 From: Baoquan He <bhe@xxxxxxxxxx> Date: Wed, 11 Jul 2018 18:30:04 +0800 Subject: [PATCH] mm, page_alloc: find movable zone after kernel text In find_zone_movable_pfns_for_nodes(), when try to find the starting PFN movable zone begins in each node, kernel text position is not considered. KASLR may put kernel after which movable zone begins. Fix it by finding movable zone after kernel text. Signed-off-by: Baoquan He <bhe@xxxxxxxxxx> --- mm/page_alloc.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 1521100..fe346b4 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -6678,6 +6678,8 @@ static void __init find_zone_movable_pfns_for_nodes(void) unsigned long size_pages; start_pfn = max(start_pfn, zone_movable_pfn[nid]); + /* KASLR may put kernel after 'start_pfn', start after kernel */ + start_pfn = max(start_pfn, PAGE_ALIGN(_etext)); if (start_pfn >= end_pfn) continue; -- 2.1.0
