[RFC/PATCH v2] arm64: define MODULES_VADDR by module_alloc_base
Will Deacon
will.deacon at arm.com
Wed Aug 23 09:27:32 PDT 2017
On Tue, Aug 15, 2017 at 05:09:26PM +0800, Miles Chen wrote:
> After the kernel ASLR, the module virtual address is moved to
> [module_alloc_base, module_alloc_base + MODULES_VSIZE).
> However, the MODULES_VADDR is still defined as a constant and functions
> like is_vmalloc_or_module_addr() and dump function will not able to
> use correct module range information.
>
> Use module_alloc_base to define MODULES_VADDR. I tested the patch under
> three different conditions:
> 1.CONFIG_RANDOMIZE_BASE=y, seed=0, CONFIG_KASAN=n
> 2.CONFIG_RANDOMIZE_BASE=y, seed=0x2304909023333333, CONFIG_KASAN=n
> 3.CONFIG_RANDOMIZE_BASE=y, seed=0x2304909023333333, CONFIG_KASAN=y
>
> test log:
>
> 1.CONFIG_RANDOMIZE_BASE=y, seed=0, CONFIG_KASAN=n
>
> [ 0.000000] Virtual kernel memory layout:
> [ 0.000000] modules : 0xffffff8000550000 - 0xffffff8008550000 ( 128 MB)
Minor nit: but it might be worth putting this after the vmalloc line in
the case where they overlap.
> [ 0.000000] vmalloc : 0xffffff8008000000 - 0xffffffbebfff0000 ( 250 GB)
> [ 0.000000] .text : 0xffffff8008080000 - 0xffffff8008550000 ( 4928 KB)
> [ 0.000000] .rodata : 0xffffff8008550000 - 0xffffff80086a0000 ( 1344 KB)
> [ 0.000000] .init : 0xffffff80086a0000 - 0xffffff8008a30000 ( 3648 KB)
> [ 0.000000] .data : 0xffffff8008a30000 - 0xffffff8008ab9200 ( 549 KB)
> [ 0.000000] .bss : 0xffffff8008ab9200 - 0xffffff8008b0b238 ( 329 KB)
> [ 0.000000] fixed : 0xffffffbefe7fd000 - 0xffffffbefec00000 ( 4108 KB)
> [ 0.000000] PCI I/O : 0xffffffbefee00000 - 0xffffffbeffe00000 ( 16 MB)
> [ 0.000000] vmemmap : 0xffffffbf00000000 - 0xffffffc000000000 ( 4 GB maximum)
> [ 0.000000] 0xffffffbf00000000 - 0xffffffbf02000000 ( 32 MB actual)
> [ 0.000000] memory : 0xffffffc000000000 - 0xffffffc080000000 ( 2048 MB)
>
> \# cat kernel_page_tables
> ---[ Modules start ]---
> ---[ Modules end ]---
I still reckon we should just omit talking about the modules area at all
for the cases where they're mapped entirely within vmalloc here.
Will
More information about the linux-arm-kernel
mailing list