On Fri, 2006-11-10 at 21:54 +0800, Li Yi (Adam) wrote: > Hi Jan and Philippe, > > I have tested the adeos-ipipe-bf53x-CVS-20061008-1.5-00.patch in > Xenomai-2.2.5 with latest Blackfin uclinux toolchain 06r2 release: > http://blackfin.uclinux.org/frs/download.php/1136/blackfin-toolchain-06r2-5.i386.rpm. > Since the toolchain changed a little for the "mcount" call, the ipipe-tracer > patch need to be updated a little. > > Also I add trace for IRQ entry and exit as Jan suggested. Here is the > details of the patch (with my explanation) > > diff -Nur linux-2.6.x-ipipe-orig/arch/blackfin/kernel/ipipe-mcount.S > linux-2.6.x-ipipe/arch/blackfin/kernel/ipipe-mcount.S > --- linux-2.6.x-ipipe-orig/arch/blackfin/kernel/ipipe-mcount.S > 2006-11-10 21:04:21.000000000 +0800 > +++ linux-2.6.x-ipipe/arch/blackfin/kernel/ipipe-mcount.S > 2006-11-10 16:41:11.000000000 +0800 > @@ -12,7 +12,7 @@ > > .align 4 /* just in case */ > > -ENTRY(_mcount) > +ENTRY(__mcount) > [--sp] = ( r7:0, p5:0 ); > [--sp] = ASTAT; > p1.L = _ipipe_trace_enable; > diff -Nur linux-2.6.x-ipipe-orig/arch/blackfin/kernel/ipipe-root.c > linux-2.6.x-ipipe/arch/blackfin/kernel/ipipe-root.c > --- linux-2.6.x-ipipe-orig/arch/blackfin/kernel/ipipe- root.c > 2006-11-10 21:04:21.000000000 +0800 > +++ linux-2.6.x-ipipe/arch/blackfin/kernel/ipipe-root.c 2006-11-10 > 16:41:59.000000000 +0800 > @@ -281,6 +281,6 @@ > EXPORT_SYMBOL(show_stack); > > #ifdef CONFIG_IPIPE_TRACE_MCOUNT > -void notrace mcount(void); > -EXPORT_SYMBOL(mcount); > +void notrace _mcount(void); > +EXPORT_SYMBOL(_mcount); > #endif /* CONFIG_IPIPE_TRACE_MCOUNT */ > > <Yi's note>: The gcc will add "Call _ _mcount" to each function > entry when there is "-pg" option
Ok. > > diff -Nur linux-2.6.x-ipipe-orig/arch/blackfin/mach-common/interrupt.S > linux-2.6.x-ipipe/arch/blackfin/mach-common/interrupt.S > --- linux-2.6.x-ipipe-orig/arch/blackfin/mach-common/interrupt.S > 2006-11-10 21:04: 21.000000000 +0800 > +++ linux-2.6.x-ipipe/arch/blackfin/mach-common/interrupt.S > 2006-11-10 16:43:55.000000000 +0800 > @@ -304,7 +304,10 @@ > p0.l = _irq_flags; > p0.h = _irq_flags; > r0 = [p0]; > - sti r0 > + sti r0; > + /* Reenable interrupts */ > + [--sp] = reti; > + r0 = [sp++]; > rts; > > #endif /* CONFIG_IPIPE */ > > > <Yi's Note>: I used the ipipe-trace to trace the latency test case: > "latency -t2 -p200 -f". And found that "__ipipe_call_irqtail()" will > disable global interrupts, putting the Blackfin to a state like: > "IMASK: 0x0000FFFF, ILAT: 0x00000040, IPEND: 0x0000c010" - that is, > Core Timer irq latched, but since IPEND[4] is set, the irq handler > cannot be invoked - thus longer latency. > Good catch, I forgot the fact that triggering an IRQ to run the scheduling tail would raise the global disable bit. Applied, thanks. [...] -- Philippe. _______________________________________________ Xenomai-core mailing list Xenomai-core@gna.org https://mail.gna.org/listinfo/xenomai-core
