2017-07-31 02:17:59

by Stephen Rothwell

Subject: linux-next: build warning after merge of the crypto tree

Hi Herbert,

After merging the crypto tree, today's linux-next build (x86_64
allmodconfig) produced this warning:

drivers/crypto/ccp/ccp-ops.c: In function 'ccp_run_rsa_cmd':
drivers/crypto/ccp/ccp-ops.c:1856:3: warning: 'sb_count' may be used uninitialized in this function [-Wmaybe-uninitialized]
cmd_q->ccp->vdata->perform->sbfree(cmd_q, op.sb_key, sb_count);
^

Maybe introduced by commit

6ba46c7d4d7e ("crypto: ccp - Fix base RSA function for version 5 CCPs")

--
Cheers,
Stephen Rothwell

2017-07-31 13:51:13

by Gary R Hook

Subject: Re: linux-next: build warning after merge of the crypto tree

On 07/30/2017 09:17 PM, Stephen Rothwell wrote:
> Hi Herbert,
>
> After merging the crypto tree, today's linux-next build (x86_64
> allmodconfig) produced this warning:
>
> drivers/crypto/ccp/ccp-ops.c: In function 'ccp_run_rsa_cmd':
> drivers/crypto/ccp/ccp-ops.c:1856:3: warning: 'sb_count' may be used uninitialized in this function [-Wmaybe-uninitialized]
> cmd_q->ccp->vdata->perform->sbfree(cmd_q, op.sb_key, sb_count);
> ^
>
> Maybe introduced by commit
>
> 6ba46c7d4d7e ("crypto: ccp - Fix base RSA function for version 5 CCPs")
>

Actually, it's been this way for a while: 956ee21 should be the commit
that introduced the spurious warning.

If you can explain to me a way to get the compiler to either (a) be
smarter, or (b) shut up about this, I'd like to hear it. I see this with
gcc 4.8/4.9/50.

IMO the compiler is incorrect. The code path where the variable is used
is complete, and the path where it is unused is complete. I don't see a
way for a logic error (and thus, use an uninitialized variable) to occur.

If someone can enlighten me, I'd appreciate it. Or, I can change the
code to recompute the value when it's needed.