Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 2 Jun 2020 13:29:51 +0300
From:   Petteri Aimonen <jpa@git.mail.kapsi.fi>
To:     Borislav Petkov <bp@alien8.de>
Cc:     Andy Lutomirski <luto@kernel.org>,
        Dave Hansen <dave.hansen@intel.com>,
        "H. Peter Anvin" <hpa@zytor.com>, x86-ml <x86@kernel.org>,
        lkml <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] arch/x86: reset MXCSR to default in kernel_fpu_begin()
Message-ID: <20200602102951.GE17423@lakka.kapsi.fi>
References: <20200602101119.GA11634@zn.tnic>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200602101119.GA11634@zn.tnic>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi,

> Instead of adding that kernel module which is x86-specific
> to a generic lib/ directory, it should be in, say,
> tools/testing/selftests/x86/test_fpu_module.c or so and instead of

The kernel module is not actually x86-specific, even though it is 
currently only enabled for x86. amdgpu driver already does kernel mode 
floating point operations on PPC64 also, and the same module could be 
used to test the same thing there.

> reading /proc/sys/debug/test_fpu, the user portion of the code would
> simply modprobe it.

To deterministically trigger the bug, the syscall has to come from the 
same thread that has modified MXCSR. Going through /usr/sbin/modprobe 
won't work, and manually doing the necessary syscalls for module loading 
seems too complicated.

> What is also missing is the user portion doing ldmxcsr before running
> the test and the test should be causing at least one exception which
> MXCSR masks off.

The fesetround() and feenableexcept() are the portable ways to modify 
MXCSR. The test module does cause Precision Exception and Denormal 
Exception if those exceptions are unmasked.

--
Petteri