On Mon, Apr 8, 2024 at 10:16 AM Jan Beulich <jbeul...@suse.com> wrote:
>
> On 05.04.2024 14:11, Fouad Hilly wrote:
> > @@ -708,11 +712,13 @@ static long cf_check microcode_update_helper(void
> > *data)
> > return ret;
> > }
> >
> > -int microcode_update(XEN_GUEST_HANDLE(const_void) buf, unsigned long len)
> > +int microcode_update(XEN_GUEST_HANDLE(const_void) buf, unsigned long len,
> > unsigned int flags)
> > {
> > int ret;
> > struct ucode_buf *buffer;
> >
> > + ucode_force_flag = (flags == XENPF_UCODE_FLAG_FORCE_SET)? 1: 0;
>
> No need for ?: when the lhs has type bool.
>
> But - do we really need to resort to parameter passing via static variables
> here? If it's unavoidable, its setting needs to move inside a locked region
> (with that region covering everything up to all consumption of the value).
There are many function calls and checks of the firmware between
microcode_update() and the actual update, which makes static variable
the viable option.
In V2 I broke it down between the actual update_flags (static) and
force_flag (local to firmware update function), I understand that
might not be enough, I will look into further improvement for
microcode_update flags in V3.
>
> Further, to avoid the same issue again when another flag wants adding, you
> want to check that all other bits in the flags field are clear.
The above check is checking all bits in the flags field. Are you
referring to flag per bit where multiple flags can be set
simultaneously?
>
> > --- a/xen/arch/x86/include/asm/microcode.h
> > +++ b/xen/arch/x86/include/asm/microcode.h
> > @@ -22,7 +22,7 @@ struct cpu_signature {
> > DECLARE_PER_CPU(struct cpu_signature, cpu_sig);
> >
> > void microcode_set_module(unsigned int idx);
> > -int microcode_update(XEN_GUEST_HANDLE(const_void) buf, unsigned long len);
> > +int microcode_update(XEN_GUEST_HANDLE(const_void) buf, unsigned long len,
> > unsigned int flags);
>
> Nit: Too long line.
>
> > --- a/xen/include/public/platform.h
> > +++ b/xen/include/public/platform.h
> > @@ -99,6 +99,10 @@ struct xenpf_microcode_update {
> > /* IN variables. */
> > XEN_GUEST_HANDLE(const_void) data;/* Pointer to microcode data */
> > uint32_t length; /* Length of microcode data. */
> > + uint32_t flags; /* Flags to be passed with ucode. */
> > +/* Force to skip microcode version check when set */
> > +#define XENPF_UCODE_FLAG_FORCE_NOT_SET 0
> > +#define XENPF_UCODE_FLAG_FORCE_SET 1
> > };
>
> The safety of this growing of an existing stable ABI struct wants at least
> briefly mentioning in the description.
>
> > @@ -624,6 +628,10 @@ struct xenpf_ucode_revision {
> > typedef struct xenpf_ucode_revision xenpf_ucode_revision_t;
> > DEFINE_XEN_GUEST_HANDLE(xenpf_ucode_revision_t);
> >
> > +/* Hypercall to microcode_update with flags */
> > +#define XENPF_microcode_update2 66
> > +
> > +
>
> No double blank lines please.
>
> Jan
Thanks,
Fouad
