Re: [PATCH] CRYPTO: Fix initialize 'psp_ret' to avoid uninitialized usage in error paths

[Gary R Hook <gary.hook@amd.com>]

On 11/27/19 6:41 PM, Haiwei Li wrote:
>  From 842cac9822aafd3cfe2da154b92b033fa1ed0d2d Mon Sep 17 00:00:00 2001
> From: Haiwei Li <lihaiwei@tencent.com>
> Date: Thu, 28 Nov 2019 08:25:16 +0800
> Subject: [PATCH] fix: initialize @psp_ret to avoid uninitialized usage 
> in error paths
> 
> Initialize @psp_ret to -1 to avoid uninitialized usage in error paths.
> Such as the function 'sev_flush_asides' in file 'arch/x86/kvm/svm.c'.

There is no uninitialized usage in error paths.

> 
> Signed-off-by: Haiwei Li <lihaiwei@tencent.com>
> Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
> ---
>   drivers/crypto/ccp/psp-dev.c | 3 +++
>   1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/crypto/ccp/psp-dev.c b/drivers/crypto/ccp/psp-dev.c
> index 39fdd06..3501562 100644
> --- a/drivers/crypto/ccp/psp-dev.c
> +++ b/drivers/crypto/ccp/psp-dev.c
> @@ -155,6 +155,9 @@ static int __sev_do_cmd_locked(int cmd, void *data, 
> int *psp_ret)
>       unsigned int phys_lsb, phys_msb;
>       unsigned int reg, ret = 0;
> 
> +    if (psp_ret)
> +        *psp_ret = -1;
> +

This function is not responsible for initializing memory that comes from 
elsewhere. Much like the use of errno, we should not modify memory if an 
error path causes __sev_do_cmd_locked() to return before any work is done.

Since this function can return two values (the return code, and the 
psp_ret argument), it has been defined to use the return value of the 
function to first indicate success or failure. Only in the case of a 
failure should the memory pointed to by psp_ret contain any useful 
information. In every other case, that memory should remain unmodified.

The return value that is stored in *psp_ret only represents information 
from the PSP. Therefore, it should only be modified when the PSP is 
called. Additionally, there is no "-1" return value from the PSP, and we 
will not be defining an default value at this time.

While I am somewhat sympathetic to the static checker's complaints, the 
proper solution for that problem is to initialize memory when it is 
allocated. Not here.

Therefore:

Nacked-by: Gary R Hook <gary.hook@amd.com>


>       if (!psp)
>           return -ENODEV;
> 
> -- 
> 1.8.3.1