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

Re: [XEN PATCH 1/2] x86: Enable BLD and handle #DB traps


  • To: Matthew Barnes <matthew.barnes@xxxxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Mon, 25 Mar 2024 17:12:50 +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: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>
  • Delivery-date: Mon, 25 Mar 2024 16:13:13 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 15.03.2024 18:52, Matthew Barnes wrote:
> --- a/xen/arch/x86/cpu/common.c
> +++ b/xen/arch/x86/cpu/common.c
> @@ -623,6 +623,11 @@ void identify_cpu(struct cpuinfo_x86 *c)
>       }
>  
>       setup_doitm();
> +
> +     if (cpu_has(c, X86_FEATURE_BLD)) {
> +             host_msr_debugctl |= IA32_DEBUGCTLMSR_BLD;
> +             wrmsrl(MSR_IA32_DEBUGCTLMSR, host_msr_debugctl);
> +     }

With the exception only occurring for CPL > 0, is this of any use in !PV
builds?

I'm also unconvinced of the use of cpu_has() here: We expect symmetry
(as can also be seen by you not using a per-CPU variable to hold the
designated register value). Perhaps boot_cpu_has() would be better here.

Together with the change to percpu_traps_init(), perhaps worth introducing
something like set_in_debugctl()?

> --- a/xen/arch/x86/hvm/vmx/entry.S
> +++ b/xen/arch/x86/hvm/vmx/entry.S
> @@ -46,8 +46,8 @@ ENTRY(vmx_asm_vmexit_handler)
>          /* WARNING! `ret`, `call *`, `jmp *` not safe before this point. */
>  
>          /* Hardware clears MSR_DEBUGCTL on VMExit.  Reinstate it if 
> debugging Xen. */
> +        mov host_msr_debugctl(%rip), %eax
>          .macro restore_lbr
> -            mov $IA32_DEBUGCTLMSR_LBR, %eax
>              mov $MSR_IA32_DEBUGCTLMSR, %ecx
>              xor %edx, %edx
>              wrmsr

The alternative just out of context is

        ALTERNATIVE "", restore_lbr, X86_FEATURE_XEN_LBR

i.e. the restore won't happen if XEN_LBR isn't active.

Together with the !PV question above I'd then like to ask whether playing
with the BLD bit is necessary at all while running a HVM vCPU. The bit
could be turned back on from the PV context-switch-in path instead. Which
would in turn remove the need for e.g. the wrmsrl() in identify_cpu(), I
believe.

In fact with us not using the "load debug controls" VM entry control I'm
having a hard time seeing how the carefully established VMCS field would
ever make it into the MSR. Instead we look to be running the guest with
the value we put there last. That wouldn't be quite right with the BLD
bit set in there unconditionally (the LBR one at least would only be set
when the respective command line option was used).

> --- a/xen/arch/x86/include/asm/debugreg.h
> +++ b/xen/arch/x86/include/asm/debugreg.h
> @@ -19,6 +19,7 @@
>  #define DR_TRAP1        (0x2)           /* db1 */
>  #define DR_TRAP2        (0x4)           /* db2 */
>  #define DR_TRAP3        (0x8)           /* db3 */
> +#define DR_TRAP11       (0x800)         /* db11 */

This isn't the flag for %dr11, so wants naming differently. Perhaps simply
DR_BLD.

> --- a/xen/arch/x86/msr.c
> +++ b/xen/arch/x86/msr.c
> @@ -24,6 +24,8 @@
>  
>  #include <public/hvm/params.h>
>  
> +uint32_t host_msr_debugctl;

This wants to be __ro_after_init, with appropriate care applied in
identify_cpu().

> --- a/xen/arch/x86/traps.c
> +++ b/xen/arch/x86/traps.c
> @@ -1936,9 +1936,12 @@ void asmlinkage do_debug(struct cpu_user_regs *regs)
>       */
>      write_debugreg(6, X86_DR6_DEFAULT);
>  
> +    if ( !( dr6 & DR_TRAP11 ) )
> +        return;

Nit: Stray blanks immediately inside the inner parentheses.

Jan



 


Rackspace

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