[PATCH] powerpc/crypto/crc32c-vpmsum: Fix missing preempt_disable()

Daniel Axtens dja at axtens.net
Fri Apr 7 11:00:06 AEST 2017


Hi Michael,

We'll also need a similar patch for the t10dif code that went in to
Herbert's next.

I will do a patch once I get my development environment up and going
again.

Regards,
Daniel

Michael Ellerman <mpe at ellerman.id.au> writes:

> In crc32c_vpmsum() we call enable_kernel_altivec() without first
> disabling preemption, which is not allowed:
>
>   WARNING: CPU: 9 PID: 2949 at ../arch/powerpc/kernel/process.c:277 enable_kernel_altivec+0x100/0x120
>   Modules linked in: dm_thin_pool dm_persistent_data dm_bio_prison dm_bufio libcrc32c vmx_crypto ...
>   CPU: 9 PID: 2949 Comm: docker Not tainted 4.11.0-rc5-compiler_gcc-6.3.1-00033-g308ac7563944 #381
>   ...
>   NIP [c00000000001e320] enable_kernel_altivec+0x100/0x120
>   LR [d000000003df0910] crc32c_vpmsum+0x108/0x150 [crc32c_vpmsum]
>   Call Trace:
>     0xc138fd09 (unreliable)
>     crc32c_vpmsum+0x108/0x150 [crc32c_vpmsum]
>     crc32c_vpmsum_update+0x3c/0x60 [crc32c_vpmsum]
>     crypto_shash_update+0x88/0x1c0
>     crc32c+0x64/0x90 [libcrc32c]
>     dm_bm_checksum+0x48/0x80 [dm_persistent_data]
>     sb_check+0x84/0x120 [dm_thin_pool]
>     dm_bm_validate_buffer.isra.0+0xc0/0x1b0 [dm_persistent_data]
>     dm_bm_read_lock+0x80/0xf0 [dm_persistent_data]
>     __create_persistent_data_objects+0x16c/0x810 [dm_thin_pool]
>     dm_pool_metadata_open+0xb0/0x1a0 [dm_thin_pool]
>     pool_ctr+0x4cc/0xb60 [dm_thin_pool]
>     dm_table_add_target+0x16c/0x3c0
>     table_load+0x184/0x400
>     ctl_ioctl+0x2f0/0x560
>     dm_ctl_ioctl+0x38/0x50
>     do_vfs_ioctl+0xd8/0x920
>     SyS_ioctl+0x68/0xc0
>     system_call+0x38/0xfc
>
> It used to be sufficient just to call pagefault_disable(), because that
> also disabled preemption. But the two were decoupled in commit 8222dbe21e79
> ("sched/preempt, mm/fault: Decouple preemption from the page fault
> logic") in mid 2015.
>
> So add the missing preempt_disable/enable(). We should also call
> disable_kernel_fp(), although it does nothing by default, there is a
> debug switch to make it active and all enables should be paired with
> disables.
>
> Fixes: 6dd7a82cc54e ("crypto: powerpc - Add POWER8 optimised crc32c")
> Cc: stable at vger.kernel.org # v4.8+
> Signed-off-by: Michael Ellerman <mpe at ellerman.id.au>
> ---
>  arch/powerpc/crypto/crc32c-vpmsum_glue.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/arch/powerpc/crypto/crc32c-vpmsum_glue.c b/arch/powerpc/crypto/crc32c-vpmsum_glue.c
> index 411994551afc..f058e0c3e4d4 100644
> --- a/arch/powerpc/crypto/crc32c-vpmsum_glue.c
> +++ b/arch/powerpc/crypto/crc32c-vpmsum_glue.c
> @@ -33,10 +33,13 @@ static u32 crc32c_vpmsum(u32 crc, unsigned char const *p, size_t len)
>  	}
>  
>  	if (len & ~VMX_ALIGN_MASK) {
> +		preempt_disable();
>  		pagefault_disable();
>  		enable_kernel_altivec();
>  		crc = __crc32c_vpmsum(crc, p, len & ~VMX_ALIGN_MASK);
> +		disable_kernel_altivec();
>  		pagefault_enable();
> +		preempt_enable();
>  	}
>  
>  	tail = len & VMX_ALIGN_MASK;
> -- 
> 2.7.4


More information about the Linuxppc-dev mailing list