Re: [Minios-devel] [UNIKRAFT PATCH v2 9/9] plat/common: Add a notice regarding trap handling

[Yuri Volchkov <yuri.volchkov@xxxxxxxxx>]

Reviewed-by: Yuri Volchkov <yuri.volchkov@xxxxxxxxx>

Florian Schmidt <florian.schmidt@xxxxxxxxx> writes:

> Signed-off-by: Florian Schmidt <florian.schmidt@xxxxxxxxx>
> ---
>  plat/common/x86/traps.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
>
> diff --git a/plat/common/x86/traps.c b/plat/common/x86/traps.c
> index c4c520b4..cfb120f6 100644
> --- a/plat/common/x86/traps.c
> +++ b/plat/common/x86/traps.c
> @@ -40,6 +40,18 @@
>  #include <uk/print.h>
>  #include <uk/assert.h>
>  
> +/* A general word of caution when writing trap handlers. The platform trap
> + * entry code is set up to properly save general-purpose registers (e.g., 
> rsi,
> + * rdi, rax, r8, ...), but it does NOT save any floating-point or SSE/AVX
> + * registers. (This would require figuring out in the trap handler code 
> whether
> + * these are available to not risk a #UD trap inside the trap handler 
> itself.)
> + * Hence, you need to be extra careful not to do anything that clobbers these
> + * registers if you intend to return from the handler. This includes calling
> + * other functions, which may clobber those registers.
> + * Of course, if you end your trap handler with a UK_CRASH, knock yourself 
> out,
> + * it's not like the function you came from will ever have the chance to 
> notice.
> + */
> +
>  /* Traps handled on both Xen and KVM */
>  
>  DECLARE_TRAP_EC(divide_error,      "divide error")
> -- 
> 2.19.2
>

-- 
Yuri Volchkov
Software Specialist

NEC Europe Ltd
Kurfürsten-Anlage 36
D-69115 Heidelberg

_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/minios-devel