[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <6754330a-bbbb-aa29-7800-f2d16216ad8c@intel.com>
Date: Tue, 29 Jun 2021 10:54:06 -0700
From: Dave Hansen <dave.hansen@...el.com>
To: "Bae, Chang Seok" <chang.seok.bae@...el.com>
Cc: "Lutomirski, Andy" <luto@...nel.org>, Borislav Petkov <bp@...e.de>,
Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...nel.org>, X86 ML <x86@...nel.org>,
"Brown, Len" <len.brown@...el.com>,
"Liu, Jing2" <jing2.liu@...el.com>,
"Shankar, Ravi V" <ravi.v.shankar@...el.com>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v5 14/28] x86/fpu/xstate: Prevent unauthorised use of
dynamic user state
On 6/29/21 10:43 AM, Bae, Chang Seok wrote:
> On Jun 16, 2021, at 12:01, Dave Hansen <dave.hansen@...el.com> wrote:
>> On 6/16/21 11:47 AM, Bae, Chang Seok wrote:
>>> Reading XINUSE via XGETBV is cheap but not free. I don't know spending a
>>> hundred cycles for this WARN is big deal but this is one of the most
>>> performance-critical paths.
>> Is XGETBV(1) really a hundred cycles? That seems absurdly high for a
>> non-serializing register read.
> My apologies -- I neglect not to check the log immediately. XGETBV(1) took
> about 26 cycles on my test.
How about we pay this 26-cycle cost, but only when XFD is in use? We
could either look at the shadowed value of the XFD MSR, or flip a global
variable the first time XFD gets used.
Powered by blists - more mailing lists