Re: [PATCH 07/10] Support Intel WRMSRNS

["H.J. Lu" <hjl.tools@gmail.com>]

On Sun, Oct 23, 2022 at 10:56 PM Jiang, Haochen <haochen.jiang@intel.com> wrote:
>
> > -----Original Message-----
> > From: Jan Beulich <jbeulich@suse.com>
> > Sent: Monday, October 17, 2022 3:18 PM
> > To: Jiang, Haochen <haochen.jiang@intel.com>
> > Cc: hjl.tools@gmail.com; Hu, Lin1 <lin1.hu@intel.com>;
> > binutils@sourceware.org
> > Subject: Re: [PATCH 07/10] Support Intel WRMSRNS
> >
> > On 14.10.2022 11:12, Haochen Jiang wrote:
> > > --- a/gas/config/tc-i386.c
> > > +++ b/gas/config/tc-i386.c
> > > @@ -1099,6 +1099,7 @@ static const arch_entry cpu_arch[] =
> > >    SUBARCH (avx_ne_convert, AVX_NE_CONVERT, ANY_AVX_NE_CONVERT,
> > false),
> > >    SUBARCH (cmpccxadd, CMPCCXADD, ANY_CMPCCXADD, false),
> > >    SUBARCH (raoint, RAOINT, ANY_RAOINT, false),
> > > +  SUBARCH (wrmsrns, WRMSRNS, ANY_WRMSRNS, false),
> >
> > As for some of the earlier patches - no need for ANY_WRMSRNS afaics.
>
> Done.

Please also remove CPU_ANY_WRMSRNS_FLAGS since it isn't used.

> >
> > > --- /dev/null
> > > +++ b/gas/testsuite/gas/i386/wrmsrns.s
> > > @@ -0,0 +1,9 @@
> > > +# Check WRMSRNS instructions
> > > +
> > > +   .allow_index_reg
> >
> > Nit: Why?
>
> Removed since there is no register usage.
>
> >
> > > --- a/gas/testsuite/gas/i386/x86-64-lockbad-1.l
> > > +++ b/gas/testsuite/gas/i386/x86-64-lockbad-1.l
> > > @@ -36,9 +36,9 @@
> > >  .*:41: Error: .*
> > >  .*:42: Error: .*
> > >  .*:43: Error: .*
> > > -.*:46: Error: .*
> > > +.*:44: Error: .*
> > >  .*:47: Error: .*
> > > -.*:49: Error: .*
> > > +.*:48: Error: .*
> > >  .*:50: Error: .*
> > >  .*:51: Error: .*
> > >  .*:52: Error: .*
> > > @@ -66,13 +66,15 @@
> > >  .*:74: Error: .*
> > >  .*:75: Error: .*
> > >  .*:76: Error: .*
> > > -.*:78: Error: .*
> > > +.*:77: Error: .*
> > >  .*:79: Error: .*
> > >  .*:80: Error: .*
> > >  .*:81: Error: .*
> > >  .*:82: Error: .*
> > >  .*:83: Error: .*
> > >  .*:84: Error: .*
> > > +.*:85: Error: .*
> > > +.*:86: Error: .*
> > >  GAS LISTING .*
> >
> > While for the diagnostics line numbers matter, ...
> >
> > > @@ -119,47 +121,49 @@ GAS LISTING .*
> > >  [  ]*41[   ]+lock sbb \(%rbx\), %eax
> > >  [  ]*42[   ]+lock sub \(%rbx\), %eax
> > >  [  ]*43[   ]+lock xor \(%rbx\), %eax
> > > -[  ]*44[   ]+
> > > -[  ]*45[   ]+\.intel_syntax noprefix
> > > -[  ]*46[   ]+lock mov eax,ebx
> > > -[  ]*47[   ]+lock mov eax,DWORD PTR \[rbx\]
> > > -[  ]*48[   ]+
> > > -[  ]*49[   ]+lock add eax,ebx
> > > -[  ]*50[   ]+lock add ebx,0x64
> > > -[  ]*51[   ]+lock adc eax,ebx
> > > -[  ]*52[   ]+lock adc ebx,0x64
> > > -[  ]*53[   ]+lock and eax,ebx
> > > -[  ]*54[   ]+lock and ebx,0x64
> > > -[  ]*55[   ]+lock btc ebx,eax
> > > -[  ]*56[   ]+lock btc ebx,0x64
> > > -[  ]*57[   ]+lock btr ebx,eax
> > > +[  ]*44[   ]+lock wrmsrns
> > > +[  ]*45[   ]+
> > > +[  ]*46[   ]+\.intel_syntax noprefix
> > > +[  ]*47[   ]+lock mov eax,ebx
> > > +[  ]*48[   ]+lock mov eax,DWORD PTR \[rbx\]
> > > +[  ]*49[   ]+
> > > +[  ]*50[   ]+lock add eax,ebx
> > > +[  ]*51[   ]+lock add ebx,0x64
> > > +[  ]*52[   ]+lock adc eax,ebx
> > > +[  ]*53[   ]+lock adc ebx,0x64
> > > +[  ]*54[   ]+lock and eax,ebx
> > > +[  ]*55[   ]+lock and ebx,0x64
> > > +[  ]*56[   ]+lock btc ebx,eax
> > > +[  ]*57[   ]+lock btc ebx,0x64
> > >
> >
> > GAS LISTING .*
> >
> > .. please abstract away line numbers (see many other testcases) rather than
> > updating them here.
> >
> > > --- a/gas/testsuite/gas/i386/x86-64-lockbad-1.s
> > > +++ b/gas/testsuite/gas/i386/x86-64-lockbad-1.s
> > > @@ -41,6 +41,7 @@ foo:
> > >     lock sbb (%rbx), %eax
> > >     lock sub (%rbx), %eax
> > >     lock xor (%rbx), %eax
> > > +   lock wrmsrns
> >
> > I wonder whether this is really necessary. If you limited yourself to ...
> >
> > >     .intel_syntax noprefix
> > >     lock mov eax,ebx
> > > @@ -82,3 +83,4 @@ foo:
> > >     lock sbb eax,DWORD PTR [rbx]
> > >     lock sub eax,DWORD PTR [rbx]
> > >     lock xor eax,DWORD PTR [rbx]
> > > +   lock wrmsrns
> >
> > ... this addition (which already seems excessive, as we don't test the majority of
> > insns here anyway), the overall diff to the testcase would end up much smaller.
>
> We removed the lockbad testcases here since most of insts are lockbad.
>
> >
> > > --- /dev/null
> > > +++ b/gas/testsuite/gas/i386/x86-64-wrmsrns-intel.d
> > > @@ -0,0 +1,12 @@
> > > +#as:
> > > +#objdump: -dw -Mintel
> > > +#name: x86_64 WRMSRNS insns (Intel disassembly)
> > > +#source: wrmsrns.s
> >
> > It's not just the source which can be shared here, but also the output
> > expectations.
>
> I get your point. But how to share here?
>
> >
> > > --- a/opcodes/i386-opc.tbl
> > > +++ b/opcodes/i386-opc.tbl
> > > @@ -3326,3 +3326,9 @@ aor, 0xf20f38fc, None, CpuRAOINT,
> > > Modrm|IgnoreSize|No_bSuf|No_wSuf|No_sSuf|No_ld
> > >  axor, 0xf30f38fc, None, CpuRAOINT,
> > > Modrm|IgnoreSize|No_bSuf|No_wSuf|No_sSuf|No_ldSuf,{ Reg32|Reg64,
> > > Dword|Qword|Unspecified|BaseIndex}
> > >
> > >  // RAOINT instructions end.
> > > +
> > > +// WRMSRNS instructions.
> > > +
> > > +wrmsrns, 0x0f01c6, None, CpuWRMSRNS,
> > > +No_bSuf|No_wSuf|No_lSuf|No_sSuf|No_qSuf|No_ldSuf, {}
> > > +
> > > +// WRMSRNS instructions end.
> >
> > Nit: Use singular in the comments?
>
> Done.
>
> Haochen
> >
> > Jan



-- 
H.J.