[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 2/3] x86: introduce xstate_zero


  • To: Fouad Hilly <fouad.hilly@xxxxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Tue, 5 Mar 2024 10:06:17 +0100
  • Autocrypt: addr=jbeulich@xxxxxxxx; keydata= xsDiBFk3nEQRBADAEaSw6zC/EJkiwGPXbWtPxl2xCdSoeepS07jW8UgcHNurfHvUzogEq5xk hu507c3BarVjyWCJOylMNR98Yd8VqD9UfmX0Hb8/BrA+Hl6/DB/eqGptrf4BSRwcZQM32aZK 7Pj2XbGWIUrZrd70x1eAP9QE3P79Y2oLrsCgbZJfEwCgvz9JjGmQqQkRiTVzlZVCJYcyGGsD /0tbFCzD2h20ahe8rC1gbb3K3qk+LpBtvjBu1RY9drYk0NymiGbJWZgab6t1jM7sk2vuf0Py O9Hf9XBmK0uE9IgMaiCpc32XV9oASz6UJebwkX+zF2jG5I1BfnO9g7KlotcA/v5ClMjgo6Gl MDY4HxoSRu3i1cqqSDtVlt+AOVBJBACrZcnHAUSuCXBPy0jOlBhxPqRWv6ND4c9PH1xjQ3NP nxJuMBS8rnNg22uyfAgmBKNLpLgAGVRMZGaGoJObGf72s6TeIqKJo/LtggAS9qAUiuKVnygo 3wjfkS9A3DRO+SpU7JqWdsveeIQyeyEJ/8PTowmSQLakF+3fote9ybzd880fSmFuIEJldWxp Y2ggPGpiZXVsaWNoQHN1c2UuY29tPsJgBBMRAgAgBQJZN5xEAhsDBgsJCAcDAgQVAggDBBYC AwECHgECF4AACgkQoDSui/t3IH4J+wCfQ5jHdEjCRHj23O/5ttg9r9OIruwAn3103WUITZee e7Sbg12UgcQ5lv7SzsFNBFk3nEQQCACCuTjCjFOUdi5Nm244F+78kLghRcin/awv+IrTcIWF hUpSs1Y91iQQ7KItirz5uwCPlwejSJDQJLIS+QtJHaXDXeV6NI0Uef1hP20+y8qydDiVkv6l IreXjTb7DvksRgJNvCkWtYnlS3mYvQ9NzS9PhyALWbXnH6sIJd2O9lKS1Mrfq+y0IXCP10eS FFGg+Av3IQeFatkJAyju0PPthyTqxSI4lZYuJVPknzgaeuJv/2NccrPvmeDg6Coe7ZIeQ8Yj t0ARxu2xytAkkLCel1Lz1WLmwLstV30g80nkgZf/wr+/BXJW/oIvRlonUkxv+IbBM3dX2OV8 AmRv1ySWPTP7AAMFB/9PQK/VtlNUJvg8GXj9ootzrteGfVZVVT4XBJkfwBcpC/XcPzldjv+3 HYudvpdNK3lLujXeA5fLOH+Z/G9WBc5pFVSMocI71I8bT8lIAzreg0WvkWg5V2WZsUMlnDL9 mpwIGFhlbM3gfDMs7MPMu8YQRFVdUvtSpaAs8OFfGQ0ia3LGZcjA6Ik2+xcqscEJzNH+qh8V m5jjp28yZgaqTaRbg3M/+MTbMpicpZuqF4rnB0AQD12/3BNWDR6bmh+EkYSMcEIpQmBM51qM EKYTQGybRCjpnKHGOxG0rfFY1085mBDZCH5Kx0cl0HVJuQKC+dV2ZY5AqjcKwAxpE75MLFkr wkkEGBECAAkFAlk3nEQCGwwACgkQoDSui/t3IH7nnwCfcJWUDUFKdCsBH/E5d+0ZnMQi+G0A nAuWpQkjM1ASeQwSHEeAWPgskBQL
  • Cc: Wei Liu <wei.liu2@xxxxxxxxxx>, Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Wei Liu <wl@xxxxxxx>, Xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>
  • Delivery-date: Tue, 05 Mar 2024 09:06:36 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 04.03.2024 10:13, Fouad Hilly wrote:
> +void xstate_zero(uint64_t mask)
> +{
> +    bool ok;
> +    struct xsave_struct tmp;
> +
> +    tmp.fpu_sse.mxcsr = MXCSR_DEFAULT;

This is a clear indication that the function name is wrong. Perhaps it is
"reset" that was meant?

> +    tmp.xsave_hdr.xstate_bv = 0;
> +    tmp.xsave_hdr.xcomp_bv = 0;
> +    memset(tmp.xsave_hdr.reserved, 0, sizeof(tmp.xsave_hdr.reserved));
> +
> +    ok = xrstor__(&tmp, mask, mask);

There's a lot of "tmp" that is left uninitialized, which would introduce
a leak of (stack) data. I think "tmp" instead wants to have an initializer
(consisting of just the explicit setting of MXCSR, leaving everything else
to be default, i.e. zero-initialized).

> +    ASSERT(ok);
>  }
>  
>  bool xsave_enabled(const struct vcpu *v)
> @@ -731,6 +753,9 @@ int handle_xsetbv(u32 index, u64 new_bv)
>      if ( (new_bv & ~xcr0_max) || !valid_xcr0(new_bv) )
>          return -EINVAL;
>  
> +    /* Zero state components before writing new XCR0 */
> +    xstate_zero(get_xcr0());

This change isn't explained in the description, doesn't fit the subject
(i.e. mechanical and functional change likely want splitting), and is
imo wrong: Why would XSETBV gain this kind of side effect, when processed
(emulated) in Xen? What I can see may want clearing are state components
which were never enabled before by a vCPU. That would then need doing
after writing the new XCR0 value, though. And it looks like that's
already in place - is there something wrong with that code?

Or is this about clearing in hardware state components about to be turned
off? If so, it would again need to be only the delta of components that's
reset here, and their state would need saving first. Upon re-enabling of
a component, that state would then need to be available for restoring.
This need for saving of state would then also explain why what's presently
named xstate_zero() can't very well use the vCPU's ->arch.xsave_area, but
needs to have a (relatively big) on-stack variable instead.

Jan



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.