Re: [PATCH 2/2] x86: guard synthetic feature and bug enumerators

[Jason Andryuk <jason.andryuk@xxxxxxx>]

On 2025-10-08 01:56, Jan Beulich wrote:
> On 07.10.2025 21:38, Jason Andryuk wrote:
> > On 2025-10-07 08:22, Jan Beulich wrote:
> > > On 30.09.2025 01:36, Jason Andryuk wrote:
> > > > On 2025-09-25 06:48, Jan Beulich wrote:
> > > > > --- a/xen/arch/x86/include/asm/spec_ctrl.h
> > > > > +++ b/xen/arch/x86/include/asm/spec_ctrl.h
> > > > > @@ -73,7 +73,7 @@ static always_inline void spec_ctrl_new_
> > > > >     
> > > > >         /* (ab)use alternative_input() to specify clobbers. */
> > > > >         alternative_input("", "DO_OVERWRITE_RSB xu=%=", X86_BUG_IBPB_NO_RET,
> > > > > -                      : "rax", "rcx");
> > > > > +                      "i" (0) : "rax", "rcx");
> > > > "i" (0) is to work around the trailing comma in alternative_input() and
> > > > does nothing?
> > > Yes. If more such "uses" appeared, we may want to introduce some kind of
> > > abstraction.
> > Thanks for confirming.
> >
> > Reviewed-by: Jason Andryuk <jason.andryuk@xxxxxxx>
> Thanks.
>
> > Though I also wondered if just #define X86_BUG_MAX/X86_SYNTH_MAX
> > combined with a BUILD_BUG_ON might be good enough.  Your approach avoids
> > the extra define but is more complicated.  Anyway, just a thought.
> How would that end up simplifying things? IOW what would the BUILD_BUG_ON()
> look like that you're thinking about? After all X86_{SYNTH,BUG}_MAX aren't
> meaningfully different from X86_NR_{SYNTH,BUG}.
Originally, I was thinking something like
 XEN_CPUFEATURE(PDX_COMPRESSION,   X86_SYNTH(31)) /* PDX compression */
+#define X86_SYNTH_MAX 31 /* Bump when adding flags */

and:
BUILD_BUG_ON( ((X86_SYNTH_MAX / 32) + 1) > X86_NR_SYNTH )

Not automated, but adding a new flag should make it obvious 
X86_SYNTH_MAX should increase.
But as you point out the redundancy of X86_{SYNTH,BUG}_MAX and 
X86_NR_{SYNTH,BUG}.  But we could re-arrange to make X86_NR_{SYNTH,BUG} 
calculated from X86_{SYNTH,BUG}_MAX like below.
Again, it's not automated, but it should make it harder to miss 
increasing the value.
Regards,
Jason

diff --git i/xen/arch/x86/include/asm/cpufeatures.h 
w/xen/arch/x86/include/asm/cpufeatures.h
index 0a98676c16..724eb1599f 100644
--- i/xen/arch/x86/include/asm/cpufeatures.h
+++ w/xen/arch/x86/include/asm/cpufeatures.h
@@ -7,7 +7,6 @@
 #define FSCAPINTS FEATURESET_NR_ENTRIES

 /* Synthetic words follow the featureset words. */
-#define X86_NR_SYNTH 1
 #define X86_SYNTH(x) (FSCAPINTS * 32 + (x))

 /* Synthetic features */
@@ -43,9 +42,10 @@ XEN_CPUFEATURE(IBPB_ENTRY_PV,     X86_SYNTH(28)) /* 
MSR_PRED_CMD used by Xen for
 XEN_CPUFEATURE(IBPB_ENTRY_HVM,    X86_SYNTH(29)) /* MSR_PRED_CMD used 
by Xen for HVM */
 XEN_CPUFEATURE(USE_VMCALL,        X86_SYNTH(30)) /* Use VMCALL instead 
of VMMCALL */
 XEN_CPUFEATURE(PDX_COMPRESSION,   X86_SYNTH(31)) /* PDX compression */
+#define X86_SYNTH_MAX             31 /* Bump when adding new flags. */
+#define X86_NR_SYNTH              ((X86_SYNTH_MAX / 32) + 1)

 /* Bug words follow the synthetic words. */
-#define X86_NR_BUG 1
 #define X86_BUG(x) ((FSCAPINTS + X86_NR_SYNTH) * 32 + (x))

 #define X86_BUG_FPU_PTRS          X86_BUG( 0) /* (F)X{SAVE,RSTOR} 
doesn't save/restore FOP/FIP/FDP. */
@@ -62,6 +62,8 @@ XEN_CPUFEATURE(PDX_COMPRESSION,   X86_SYNTH(31)) /* 
PDX compression */
 #define X86_SPEC_BHB_TSX          X86_BUG(19) /* Use clear_bhb_tsx for 
BHI mitigation. */
 #define X86_SPEC_BHB_LOOPS        X86_BUG(20) /* Use clear_bhb_loops 
for BHI mitigation.*/
 #define X86_SPEC_BHB_LOOPS_LONG   X86_BUG(21) /* Upgrade 
clear_bhb_loops to the "long" sequence. */
+#define X86_BUX_MAX               21 /* Bump when adding new flags. */
+#define X86_NR_BUG                ((X86_BUG_MAX / 32) + 1)

 /* Total number of capability words, inc synth and bug words. */
 #define NCAPINTS (FSCAPINTS + X86_NR_SYNTH + X86_NR_BUG) /* N 32-bit 
words worth of info */