On Fri, Feb 19, 2016 at 08:49:33AM +0100, Andrew Jones wrote:
> On Fri, Feb 19, 2016 at 02:40:53PM +1100, David Gibson wrote:
> > On Wed, Feb 17, 2016 at 07:40:54PM +0100, Andrew Jones wrote:
> > > Copy arm's setup code (also DT based) over to powerpc, adapting
> > > it a bit. Also bring over arm's setup selftest, giving powerpc
> > > its first test.
> > >
> > > The largest change from arm's setup.c is that instead of using a
> > > hardcoded SMP_CACHE_BYTES, cpu_set() is extended to extract the
> > > line size from the cpu DT nodes. That change also requires that
> > > we call cpu_init() before mem_init() in setup().
> > >
> > > Signed-off-by: Andrew Jones <drjones@xxxxxxxxxx>
> > > Reviewed-by: Thomas Huth <thuth@xxxxxxxxxx>
> > > Tested-by: Laurent Vivier <lvivier@xxxxxxxxxx>
> > > ---
> > > lib/powerpc/asm/setup.h | 30 +++++++++++
> > > lib/powerpc/setup.c | 141 ++++++++++++++++++++++++++++++++++++++++++++++++
> > > lib/ppc64/asm/setup.h | 1 +
> > > powerpc/Makefile.common | 1 +
> > > powerpc/cstart64.S | 16 +++++-
> > > powerpc/selftest.c | 63 ++++++++++++++++++++--
> > > 6 files changed, 248 insertions(+), 4 deletions(-)
> > > create mode 100644 lib/powerpc/asm/setup.h
> > > create mode 100644 lib/powerpc/setup.c
> > > create mode 100644 lib/ppc64/asm/setup.h
> > >
> > > diff --git a/lib/powerpc/asm/setup.h b/lib/powerpc/asm/setup.h
> > > new file mode 100644
> > > index 0000000000000..8288ff37ff1b9
> > > --- /dev/null
> > > +++ b/lib/powerpc/asm/setup.h
> > > @@ -0,0 +1,30 @@
> > > +#ifndef _ASMPOWERPC_SETUP_H_
> > > +#define _ASMPOWERPC_SETUP_H_
> > > +/*
> > > + * Copyright (C) 2016, Red Hat Inc, Andrew Jones <drjones@xxxxxxxxxx>
> > > + *
> > > + * This work is licensed under the terms of the GNU LGPL, version 2.
> > > + */
> > > +#include <libcflat.h>
> > > +#include <alloc.h> /* phys_addr_t */
> > > +
> > > +#define NR_CPUS 8 /* arbitrarily set for now */
> >
> > We probably want to set this larger sooner rather than later. The
> > test machines the virt dev team uses have 160 logical CPUs, and
> > they're not particularly large as Power systems go.
>
> One of the changes that the "Add SMP support" follow-on series will
> make is to change this to whatever qemu-kvm/spapr supports.
Ok, great.
> > > +extern u32 cpus[NR_CPUS];
> > > +extern int nr_cpus;
> > > +
> > > +#define NR_MEM_REGIONS 8
> > > +#define MR_F_PRIMARY (1U << 0)
> > > +struct mem_region {
> > > + phys_addr_t start;
> > > + phys_addr_t end;
> > > + unsigned int flags;
> > > +};
> > > +extern struct mem_region mem_regions[NR_MEM_REGIONS];
> > > +extern phys_addr_t __physical_start, __physical_end;
> > > +extern unsigned __smp_cache_bytes;
> > > +
> > > +#define PHYSICAL_START (__physical_start)
> > > +#define PHYSICAL_END (__physical_end)
> > > +#define SMP_CACHE_BYTES (__smp_cache_bytes)
> > > +
> > > +#endif /* _ASMPOWERPC_SETUP_H_ */
> > > diff --git a/lib/powerpc/setup.c b/lib/powerpc/setup.c
> > > new file mode 100644
> > > index 0000000000000..70f50deb9bacf
> > > --- /dev/null
> > > +++ b/lib/powerpc/setup.c
> > > @@ -0,0 +1,141 @@
> > > +/*
> > > + * Initialize machine setup information and I/O.
> > > + *
> > > + * After running setup() unit tests may query how many cpus they have
> > > + * (nr_cpus), how much memory they have (PHYSICAL_END - PHYSICAL_START),
> > > + * may use dynamic memory allocation (malloc, etc.), printf, and exit.
> > > + * Finally, argc and argv are also ready to be passed to main().
> > > + *
> > > + * Copyright (C) 2016, Red Hat Inc, Andrew Jones <drjones@xxxxxxxxxx>
> > > + *
> > > + * This work is licensed under the terms of the GNU LGPL, version 2.
> > > + */
> > > +#include <libcflat.h>
> > > +#include <libfdt/libfdt.h>
> > > +#include <devicetree.h>
> > > +#include <alloc.h>
> > > +#include <asm/setup.h>
> > > +#include <asm/page.h>
> > > +
> > > +extern unsigned long stacktop;
> > > +extern void io_init(void);
> > > +extern void setup_args(const char *args);
> > > +
> > > +u32 cpus[NR_CPUS] = { [0 ... NR_CPUS-1] = (~0U) };
> > > +int nr_cpus;
> > > +
> > > +struct mem_region mem_regions[NR_MEM_REGIONS];
> > > +phys_addr_t __physical_start, __physical_end;
> > > +unsigned __smp_cache_bytes;
> > > +
> > > +static void cpu_set(int fdtnode, u32 regval, void *info)
> > > +{
> > > + const struct fdt_property *prop;
> > > + unsigned *max_linesz = (unsigned *)info, sz;
> > > + int cpu = nr_cpus++;
> > > + u32 *data;
> > > +
> > > + if (cpu >= NR_CPUS) {
> > > + printf("Number cpus exceeds maximum supported (%d).\n",
> > > + NR_CPUS);
> > > + assert(0);
> > > + }
> > > + cpus[cpu] = regval;
> > > +
> > > + prop = fdt_get_property(dt_fdt(), fdtnode, "i-cache-line-size", NULL);
> > > + assert(prop != NULL);
> > > + data = (u32 *)prop->data;
> > > + sz = fdt32_to_cpu(*data);
> > > +
> > > + if (*max_linesz < sz)
> > > + *max_linesz = sz;
> > > +
> > > + prop = fdt_get_property(dt_fdt(), fdtnode, "d-cache-line-size", NULL);
> > > + assert(prop != NULL);
> > > + data = (u32 *)prop->data;
> > > + sz = fdt32_to_cpu(*data);
> > > +
> > > + if (*max_linesz < sz)
> > > + *max_linesz = sz;
> >
> > Hmm.. what is __smp_cache_bytes used for, exactly?
> >
> > If you're just trying to get cacheline alignment, then this is
> > correct.
> >
> > However, if you're using this to control loops which step over each
> > cacheline to invalidate / flush / whatever, then you'd need the
> > minimum cacheline size instead.
>
> OK, I can just save both to respective icache, dcache variables,
> deferring the decision on which, or max, or min, to use.
Or... you could just abort if they're not the same and worry about it
when we actually get a machine where they're different (i.e. probably
never).
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
Attachment:
signature.asc
Description: PGP signature
