[PATCH V3 1/5] powerpc, perf: Add new BHRB related instructions for POWER8

Michael Neuling mikey at neuling.org
Mon Apr 22 12:50:03 EST 2013


Michael Ellerman <michael at ellerman.id.au> wrote:

> On Mon, Apr 22, 2013 at 11:13:43AM +1000, Michael Neuling wrote:
> > Michael Ellerman <michael at ellerman.id.au> wrote:
> > 
> > > On Thu, Apr 18, 2013 at 05:56:12PM +0530, Anshuman Khandual wrote:
> > > > This patch adds new POWER8 instruction encoding for reading
> > > > the BHRB buffer entries and also clearing it. Encoding for
> > > > "clrbhrb" instruction is straight forward.
> > > 
> > > Which is "clear branch history rolling buffer" ?
> > > 
> > > > But "mfbhrbe"
> > > > encoding involves reading a certain index of BHRB buffer
> > > > into a particular GPR register.
> > > 
> > > And "Move from branch history rolling buffer entry" ?
> > > 
> > > > diff --git a/arch/powerpc/include/asm/ppc-opcode.h b/arch/powerpc/include/asm/ppc-opcode.h
> > > > index 8752bc8..93ae5a1 100644
> > > > --- a/arch/powerpc/include/asm/ppc-opcode.h
> > > > +++ b/arch/powerpc/include/asm/ppc-opcode.h
> > > > @@ -82,6 +82,7 @@
> > > >  #define	__REGA0_R31	31
> > > >  
> > > >  /* sorted alphabetically */
> > > > +#define PPC_INST_BHRBE			0x7c00025c
> > > 
> > > I don't think you really need this, just use the literal value below.
> > 
> > The rest of the defines in this file do this, so Anshuman's right. 
> 
> I don't see the point, but sure let's be consistent. Though in that case
> he should do the same for PPC_CLRBHRB below.

Agreed.

Mikey

> 
> > > > @@ -297,6 +298,12 @@
> > > >  #define PPC_NAP			stringify_in_c(.long PPC_INST_NAP)
> > > >  #define PPC_SLEEP		stringify_in_c(.long PPC_INST_SLEEP)
> > > >  
> > > > +/* BHRB instructions */
> > > > +#define PPC_CLRBHRB		stringify_in_c(.long 0x7c00035c)
> > > > +#define PPC_MFBHRBE(r, n)	stringify_in_c(.long PPC_INST_BHRBE | \
> > > > +						__PPC_RS(r) | \
> > > > +							(((n) & 0x1f) << 11))
> > > 
> > > Why are you not using ___PPC_RB(n) here ?
> > 
> > Actually, this is wrong.  The number field should be 10 bits (0x3ff),
> > not 5 (0x1f)  Anshuman please fix.
> 
> ACK.
> 
> cheers
> 


More information about the Linuxppc-dev mailing list