crypto: sha256-mb - fix ctx pointer and digest copy

1. fix ctx pointer
Use req_ctx which is the ctx for the next job that have
been completed in the lanes instead of the first
completed job rctx, whose completion could have been
called and released.
2. fix digest copy
Use XMM register to copy another 16 bytes sha256 digest
instead of a regular register.

Signed-off-by: Xiaodong Liu <xiaodong.liu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/sha256-mb/sha256_mb.c

@@ -485,10 +485,10 @@ static int sha_complete_job(struct mcryptd_hash_request_ctx *rctx,
 
 			req = cast_mcryptd_ctx_to_req(req_ctx);
 			if (irqs_disabled())
-				rctx->complete(&req->base, ret);
+				req_ctx->complete(&req->base, ret);
 			else {
 				local_bh_disable();
-				rctx->complete(&req->base, ret);
+				req_ctx->complete(&req->base, ret);
 				local_bh_enable();
 			}
 		}

arch/x86/crypto/sha256-mb/sha256_mb_mgr_flush_avx2.S

@@ -265,13 +265,14 @@ ENTRY(sha256_mb_mgr_get_comp_job_avx2)
 	vpinsrd	$1, _args_digest+1*32(state, idx, 4), %xmm0, %xmm0
 	vpinsrd	$2, _args_digest+2*32(state, idx, 4), %xmm0, %xmm0
 	vpinsrd	$3, _args_digest+3*32(state, idx, 4), %xmm0, %xmm0
-	movl	_args_digest+4*32(state, idx, 4), tmp2_w
+	vmovd   _args_digest(state , idx, 4) , %xmm0
 	vpinsrd	$1, _args_digest+5*32(state, idx, 4), %xmm1, %xmm1
 	vpinsrd	$2, _args_digest+6*32(state, idx, 4), %xmm1, %xmm1
 	vpinsrd	$3, _args_digest+7*32(state, idx, 4), %xmm1, %xmm1
 
-	vmovdqu	%xmm0, _result_digest(job_rax)
-	movl	tmp2_w, _result_digest+1*16(job_rax)
+        vmovdqu %xmm0, _result_digest(job_rax)
+        offset =  (_result_digest + 1*16)
+        vmovdqu %xmm1, offset(job_rax)
 
 	pop	%rbx