Re: [PATCH v2 06/11] tpm: Add full HMAC and encrypt/decrypt session handling code

From: Yujie Liu <yujie.liu@intel.com>
To: Jarkko Sakkinen <jarkko@kernel.org>,
	James Bottomley <James.Bottomley@hansenpartnership.com>
Cc: kernel test robot <lkp@intel.com>,
	<linux-integrity@vger.kernel.org>,
	<oe-kbuild-all@lists.linux.dev>, <keyrings@vger.kernel.org>,
	Ard Biesheuvel <ardb@kernel.org>
Subject: Re: [PATCH v2 06/11] tpm: Add full HMAC and encrypt/decrypt session handling code
Date: Mon, 13 Feb 2023 17:31:58 +0800	[thread overview]
Message-ID: <Y+oDjm3dpqtXx1Jv@yujie-X299> (raw)
In-Reply-To: <Y+nqpLm2YyYkcZ+H@kernel.org>

On Mon, Feb 13, 2023 at 09:45:40AM +0200, Jarkko Sakkinen wrote:
> On Fri, Feb 10, 2023 at 09:48:15AM -0500, James Bottomley wrote:
> > On Wed, 2023-02-08 at 04:49 +0200, Jarkko Sakkinen wrote:
> > > On Fri, Feb 03, 2023 at 02:06:48PM +0800, Yujie Liu wrote:
> > > > Hi James,
> > > > 
> > > > On Wed, Jan 25, 2023 at 07:59:09AM -0500, James Bottomley wrote:
> > > > > On Wed, 2023-01-25 at 07:11 +0800, kernel test robot wrote:
> > > > > > 
> > > > > > If you fix the issue, kindly add following tag where applicable
> > > > > > > Reported-by: kernel test robot <lkp@intel.com>
> > > > > > 
> > > > > > All warnings (new ones prefixed by >>):
> > > > > > 
> > > > > >    drivers/char/tpm/tpm2-sessions.c:1184:5: warning: no
> > > > > > previous
> > > > > > prototype for 'tpm2_create_null_primary' [-Wmissing-prototypes]
> > > > > >     1184 | int tpm2_create_null_primary(struct tpm_chip *chip)
> > > > > > {
> > > > > >          |     ^~~~~~~~~~~~~~~~~~~~~~~~
> > > > > >    drivers/char/tpm/tpm2-sessions.c: In function
> > > > > > 'tpm_buf_check_hmac_response':
> > > > > > > > drivers/char/tpm/tpm2-sessions.c:831:1: warning: the frame
> > > > > > > > size
> > > > > > > > of 1132 bytes is larger than 1024 bytes [-Wframe-larger-
> > > > > > > > than=]
> > > > > >      831 | }
> > > > > >          | ^
> > > > > >    drivers/char/tpm/tpm2-sessions.c: In function
> > > > > > 'tpm_buf_fill_hmac_session':
> > > > > >    drivers/char/tpm/tpm2-sessions.c:579:1: warning: the frame
> > > > > > size of
> > > > > > 1132 bytes is larger than 1024 bytes [-Wframe-larger-than=]
> > > > > >      579 | }
> > > > > >          | ^
> > > > > 
> > > > > Is this a test problem?  I can't see why the code would only blow
> > > > > the
> > > > > stack on the arc architecture and not on any other ... does it
> > > > > have
> > > > > something funny with on stack crypto structures?
> > > > 
> > > > This warning is controlled by the value of CONFIG_FRAME_WARN.
> > > > 
> > > > For "make ARCH=arc allyesconfig", the default value is 1024, so
> > > > this frame warning shows up during the build.
> > > > 
> > > > For other arch such as "make ARCH=x86_64 allyesconfig", the default
> > > > value would be 2048 and won't have this warning.
> > > > 
> > > > Not sure if this is a real problem that need to be fixed, here just
> > > > providing above information for your reference.
> > > > 
> > > > --
> > > > Best Regards,
> > > > Yujie
> > > 
> > > *Must* be fixed given that it is how the default value is set now.
> > > This is wrong place to reconsider.
> > >
> > > 
> > > And we do not want to add functions that bloat the stack this way.
> > > 
> > > Shash just needs to be allocated from heap instead of stack.
> > 
> > On x86_64 the stack usage is measured at 984 bytes, so rather than
> > jumping to conclusions let's root cause why this is a problem only on
> > the arc architecture.  I suspect it's something to do with the
> > alignment constraints of shash.  I've also noted it shouldn't actually
> > warn on arc because the default stack warning size there should be 2048
> > (like x86_64).

The stack usage varies on different architectures, so does the default
value of CONFIG_FRAME_WARN. The warning shows up when the stack usage
exceeds the default warning size.

For arc arch, I reconfirmed that the default stack warning size is 1024
no matter allyesconfig or defconfig, while the usage could reach 1132
bytes.

$ make ARCH=arc allyesconfig
$ grep FRAME_WARN .config
CONFIG_FRAME_WARN=1024

$ make ARCH=arc defconfig
$ grep FRAME_WARN .config
CONFIG_FRAME_WARN=1024

> Would it such a big deal to allocate shash from heap? That would
> be IMHO more robust in the end.

Thanks Jarkko for the suggestion. This would be a faster and better fix
without having to look into this arc-specific problem.

Best Regards,
Yujie

> BR, Jarkko