xen/arch/x86/arch.mk | 2 -- xen/arch/x86/include/asm/random.h | 2 +- xen/arch/x86/x86_emulate/0fc7.c | 15 +++++---------- 3 files changed, 6 insertions(+), 13 deletions(-)
From: Denis Mukhin <dmukhin@ford.com>
The new toolchain baseline knows the rdrand/rdseed instructions,
no need to carry the workaround in the code.
Resolves: https://gitlab.com/xen-project/xen/-/work_items/208
Signed-off-by: Denis Mukhin <dmukhin@ford.com>
---
Changes since v2:
- updated arch_get_random()
- Link to v2: https://lore.kernel.org/xen-devel/20250403182250.3329498-6-dmukhin@ford.com/
---
xen/arch/x86/arch.mk | 2 --
xen/arch/x86/include/asm/random.h | 2 +-
xen/arch/x86/x86_emulate/0fc7.c | 15 +++++----------
3 files changed, 6 insertions(+), 13 deletions(-)
diff --git a/xen/arch/x86/arch.mk b/xen/arch/x86/arch.mk
index 3bbaee2a44..5577bf6241 100644
--- a/xen/arch/x86/arch.mk
+++ b/xen/arch/x86/arch.mk
@@ -10,9 +10,7 @@ CFLAGS += -msoft-float
$(call cc-options-add,CFLAGS,CC,$(EMBEDDED_EXTRA_CFLAGS))
$(call cc-option-add,CFLAGS,CC,-Wnested-externs)
-$(call as-option-add,CFLAGS,CC,"rdrand %eax",-DHAVE_AS_RDRAND)
$(call as-option-add,CFLAGS,CC,"xsaveopt (%rax)",-DHAVE_AS_XSAVEOPT)
-$(call as-option-add,CFLAGS,CC,"rdseed %eax",-DHAVE_AS_RDSEED)
$(call as-option-add,CFLAGS,CC,"clwb (%rax)",-DHAVE_AS_CLWB)
$(call as-option-add,CFLAGS,CC,".equ \"x\"$(comma)1",-DHAVE_AS_QUOTED_SYM)
$(call as-option-add,CFLAGS,CC,"movdiri %rax$(comma)(%rax)",-DHAVE_AS_MOVDIR)
diff --git a/xen/arch/x86/include/asm/random.h b/xen/arch/x86/include/asm/random.h
index 9e1fe0bc1d..e1c1c765e1 100644
--- a/xen/arch/x86/include/asm/random.h
+++ b/xen/arch/x86/include/asm/random.h
@@ -8,7 +8,7 @@ static inline unsigned int arch_get_random(void)
unsigned int val = 0;
if ( cpu_has(¤t_cpu_data, X86_FEATURE_RDRAND) )
- asm volatile ( ".byte 0x0f,0xc7,0xf0" : "+a" (val) );
+ asm volatile ( "rdrand %0" : "=a" (val) );
return val;
}
diff --git a/xen/arch/x86/x86_emulate/0fc7.c b/xen/arch/x86/x86_emulate/0fc7.c
index 5268d5cafd..58c8f79501 100644
--- a/xen/arch/x86/x86_emulate/0fc7.c
+++ b/xen/arch/x86/x86_emulate/0fc7.c
@@ -32,7 +32,6 @@ int x86emul_0fc7(struct x86_emulate_state *s,
return X86EMUL_UNRECOGNIZED;
case 6: /* rdrand */
-#ifdef HAVE_AS_RDRAND
generate_exception_if(s->vex.pfx >= vex_f3, X86_EXC_UD);
host_and_vcpu_must_have(rdrand);
*dst = s->ea;
@@ -43,12 +42,12 @@ int x86emul_0fc7(struct x86_emulate_state *s,
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
default:
-# ifdef __x86_64__
+#ifdef __x86_64__
asm ( "rdrand %k0" ASM_FLAG_OUT(, "; setc %1")
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
case 8:
-# endif
+#endif
asm ( "rdrand %0" ASM_FLAG_OUT(, "; setc %1")
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
@@ -57,9 +56,6 @@ int x86emul_0fc7(struct x86_emulate_state *s,
if ( carry )
regs->eflags |= X86_EFLAGS_CF;
break;
-#else
- return X86EMUL_UNIMPLEMENTED;
-#endif
case 7: /* rdseed / rdpid */
if ( s->vex.pfx == vex_f3 ) /* rdpid */
@@ -77,7 +73,7 @@ int x86emul_0fc7(struct x86_emulate_state *s,
dst->bytes = 4;
break;
}
-#ifdef HAVE_AS_RDSEED
+
generate_exception_if(s->vex.pfx >= vex_f3, X86_EXC_UD);
host_and_vcpu_must_have(rdseed);
*dst = s->ea;
@@ -88,12 +84,12 @@ int x86emul_0fc7(struct x86_emulate_state *s,
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
default:
-# ifdef __x86_64__
+#ifdef __x86_64__
asm ( "rdseed %k0" ASM_FLAG_OUT(, "; setc %1")
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
case 8:
-# endif
+#endif
asm ( "rdseed %0" ASM_FLAG_OUT(, "; setc %1")
: "=r" (dst->val), ASM_FLAG_OUT("=@ccc", "=qm") (carry) );
break;
@@ -102,7 +98,6 @@ int x86emul_0fc7(struct x86_emulate_state *s,
if ( carry )
regs->eflags |= X86_EFLAGS_CF;
break;
-#endif
}
}
else
--
2.34.1
On 05/04/2025 2:25 am, dmkhn@proton.me wrote: > From: Denis Mukhin <dmukhin@ford.com> > > The new toolchain baseline knows the rdrand/rdseed instructions, > no need to carry the workaround in the code. A few minor notes. Instructions in CAPITALS please to make them easier to parse in context. The comma ought to be a semi-colon. You should note that arch_get_random() is adjusted too. Also, it's useful to state "No functional change." to help reviewers. (Or to help archaeologies figure out the intent of the patch if they subsequently find a bug in it.) > diff --git a/xen/arch/x86/include/asm/random.h b/xen/arch/x86/include/asm/random.h > index 9e1fe0bc1d..e1c1c765e1 100644 > --- a/xen/arch/x86/include/asm/random.h > +++ b/xen/arch/x86/include/asm/random.h > @@ -8,7 +8,7 @@ static inline unsigned int arch_get_random(void) > unsigned int val = 0; > > if ( cpu_has(¤t_cpu_data, X86_FEATURE_RDRAND) ) > - asm volatile ( ".byte 0x0f,0xc7,0xf0" : "+a" (val) ); > + asm volatile ( "rdrand %0" : "=a" (val) ); This was only tied to "a" because 0xf0 is the ModRM byte. Now that we can use the mnemonic, it can become "=r". I've fixed all on commit. ~Andrew
On Monday, April 7th, 2025 at 2:17 AM, Andrew Cooper <andrew.cooper3@citrix.com> wrote: > > > On 05/04/2025 2:25 am, dmkhn@proton.me wrote: > > > From: Denis Mukhin dmukhin@ford.com > > > > The new toolchain baseline knows the rdrand/rdseed instructions, > > no need to carry the workaround in the code. > > > A few minor notes. Instructions in CAPITALS please to make them easier > to parse in context. The comma ought to be a semi-colon. > > You should note that arch_get_random() is adjusted too. > > Also, it's useful to state "No functional change." to help reviewers. > (Or to help archaeologies figure out the intent of the patch if they > subsequently find a bug in it.) Noted, thanks for the feedback > > > diff --git a/xen/arch/x86/include/asm/random.h b/xen/arch/x86/include/asm/random.h > > index 9e1fe0bc1d..e1c1c765e1 100644 > > --- a/xen/arch/x86/include/asm/random.h > > +++ b/xen/arch/x86/include/asm/random.h > > @@ -8,7 +8,7 @@ static inline unsigned int arch_get_random(void) > > unsigned int val = 0; > > > > if ( cpu_has(¤t_cpu_data, X86_FEATURE_RDRAND) ) > > - asm volatile ( ".byte 0x0f,0xc7,0xf0" : "+a" (val) ); > > + asm volatile ( "rdrand %0" : "=a" (val) ); > > > This was only tied to "a" because 0xf0 is the ModRM byte. Now that we > can use the mnemonic, it can become "=r". > > I've fixed all on commit. Thank you! > > ~Andrew
© 2016 - 2025 Red Hat, Inc.