On Tue, Oct 15, 2024 at 12:56:57PM +0200, Ard Biesheuvel wrote: > On Mon, 14 Oct 2024 at 22:59, Kees Cook <kees@xxxxxxxxxx> wrote: > > > > On Wed, Oct 09, 2024 at 02:43:53PM +0200, Ard Biesheuvel wrote: > > > However, if a non-TLS definition of the symbol in question is visible in > > > the same compilation unit (which amounts to the whole of vmlinux if LTO > > > is enabled), it will drop the per-CPU prefix and emit a load from a > > > bogus address. > > > > I take this to mean that x86 32-bit kernels built with the stack > > protector and using Clang LTO will crash very quickly? > > > > Yeah. The linked issue is not quite clear, but it does suggest things > are pretty broken in that case. Yeah, i386_defconfig with CONFIG_LTO_CLANG_FULL=y explodes on boot for me without this change: [ 0.000000] Linux version 6.12.0-rc3-00044-g2f87d0916ce0 (nathan@thelio-3990X) (ClangBuiltLinux clang version 19.1.2 (https://github.com/llvm/llvm-project.git 7ba7d8e2f7b6445b60679da826210cdde29eaf8b), ClangBuiltLinux LLD 19.1.2 (https://github.com/llvm/llvm-project.git 7ba7d8e2f7b6445b60679da826210cdde29eaf8b)) #1 SMP PREEMPT_DYNAMIC Tue Oct 15 19:00:21 MST 2024 ... [ 0.631002] Freeing unused kernel image (initmem) memory: 936K [ 0.631613] Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: free_initmem+0x95/0x98 [ 0.632606] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.12.0-rc3-00044-g2f87d0916ce0 #1 [ 0.633467] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014 [ 0.634583] Call Trace: [ 0.634837] panic+0xd4/0x2cc [ 0.635146] ? _vdso_rng_data+0xd80/0xd80 [ 0.635550] ? _vdso_rng_data+0xd80/0xd80 [ 0.635965] ? rest_init+0xb0/0xb0 [ 0.636312] __stack_chk_fail+0x10/0x10 [ 0.636701] ? free_initmem+0x95/0x98 [ 0.637074] free_initmem+0x95/0x98 [ 0.637434] ? _vdso_rng_data+0xd80/0xd80 [ 0.637838] ? rest_init+0xb0/0xb0 [ 0.638196] kernel_init+0x42/0x1e4 [ 0.638558] ret_from_fork+0x2b/0x40 [ 0.638922] ret_from_fork_asm+0x12/0x18 [ 0.639331] entry_INT80_32+0x108/0x108 [ 0.639864] Kernel Offset: disabled [ 0.640224] ---[ end Kernel panic - not syncing: stack-protector: Kernel stack is corrupted in: free_initmem+0x95/0x98 ]--- I can confirm that this patch resolves that issue for me and LKDTM's REPORT_STACK_CANARY test passes with that configuration. Reviewed-by: Nathan Chancellor <nathan@xxxxxxxxxx> Tested-by: Nathan Chancellor <nathan@xxxxxxxxxx> I presume the '#ifndef CONFIG_X86_64' in arch/x86/entry/entry.S is present because only X86_32 uses '-mstack-protector-guard-reg='? I assume that will disappear when X86_64 supports this option (IIRC that was the plan)? Cheers, Nathan
