Re: [Qemu-riscv] [PATCH v8 18/32] riscv: sifive_u: Set the minimum number of cpus to 2

[Jonathan Behrens]

Has there been testing with "-smp 2"? A while back I thought I read that the included uboot firmware was using a hard-coded device tree that indicated 4+1 CPUs, which I would have expected to cause Linux boot issues?

Jonathan

On Sun, Sep 15, 2019 at 1:31 PM Palmer Dabbelt <address@hidden> wrote:

On Sun, 15 Sep 2019 06:07:18 PDT (-0700), address@hidden wrote:
> Hi Palmer,
>
> On Sun, Sep 15, 2019 at 3:00 AM Palmer Dabbelt <address@hidden> wrote:
>>
>> On Fri, 13 Sep 2019 08:25:21 PDT (-0700), address@hidden wrote:
>> > Hi Palmer,
>> >
>> > On Fri, Sep 13, 2019 at 10:33 PM Palmer Dabbelt <address@hidden> wrote:
>> >>
>> >> On Fri, 06 Sep 2019 09:20:05 PDT (-0700), address@hidden wrote:
>> >> > It is not useful if we only have one management CPU.
>> >> >
>> >> > Signed-off-by: Bin Meng <address@hidden>
>> >> > Reviewed-by: Alistair Francis <address@hidden>
>> >> >
>> >> > ---
>> >> >
>> >> > Changes in v8: None
>> >> > Changes in v7: None
>> >> > Changes in v6: None
>> >> > Changes in v5: None
>> >> > Changes in v4: None
>> >> > Changes in v3:
>> >> > - use management cpu count + 1 for the min_cpus
>> >> >
>> >> > Changes in v2:
>> >> > - update the file header to indicate at least 2 harts are created
>> >> >
>> >> >  hw/riscv/sifive_u.c         | 4 +++-
>> >> >  include/hw/riscv/sifive_u.h | 2 ++
>> >> >  2 files changed, 5 insertions(+), 1 deletion(-)
>> >> >
>> >> > diff --git a/hw/riscv/sifive_u.c b/hw/riscv/sifive_u.c
>> >> > index 2947e06..2023b71 100644
>> >> > --- a/hw/riscv/sifive_u.c
>> >> > +++ b/hw/riscv/sifive_u.c
>> >> > @@ -10,7 +10,8 @@
>> >> >   * 1) CLINT (Core Level Interruptor)
>> >> >   * 2) PLIC (Platform Level Interrupt Controller)
>> >> >   *
>> >> > - * This board currently uses a hardcoded devicetree that indicates one hart.
>> >> > + * This board currently generates devicetree dynamically that indicates at least
>> >> > + * two harts.
>> >> >   *
>> >> >   * This program is free software; you can redistribute it and/or modify it
>> >> >   * under the terms and conditions of the GNU General Public License,
>> >> > @@ -433,6 +434,7 @@ static void riscv_sifive_u_machine_init(MachineClass *mc)
>> >> >       * management CPU.
>> >> >       */
>> >> >      mc->max_cpus = 4;
>> >> > +    mc->min_cpus = SIFIVE_U_MANAGEMENT_CPU_COUNT + 1;
>> >> >  }
>> >> >
>> >> >  DEFINE_MACHINE("sifive_u", riscv_sifive_u_machine_init)
>> >> > diff --git a/include/hw/riscv/sifive_u.h b/include/hw/riscv/sifive_u.h
>> >> > index f25bad8..6d22741 100644
>> >> > --- a/include/hw/riscv/sifive_u.h
>> >> > +++ b/include/hw/riscv/sifive_u.h
>> >> > @@ -69,6 +69,8 @@ enum {
>> >> >      SIFIVE_U_GEM_CLOCK_FREQ = 125000000
>> >> >  };
>> >> >
>> >> > +#define SIFIVE_U_MANAGEMENT_CPU_COUNT   1
>> >> > +
>> >> >  #define SIFIVE_U_PLIC_HART_CONFIG "MS"
>> >> >  #define SIFIVE_U_PLIC_NUM_SOURCES 54
>> >> >  #define SIFIVE_U_PLIC_NUM_PRIORITIES 7
>> >>
>> >> This fails "make check", so I'm going to squash in this
>> >>
>> >> diff --git a/hw/riscv/sifive_u.c b/hw/riscv/sifive_u.c
>> >> index ca9f7fea41..adecbf1dd9 100644
>> >> --- a/hw/riscv/sifive_u.c
>> >> +++ b/hw/riscv/sifive_u.c
>> >> @@ -528,6 +528,7 @@ static void riscv_sifive_u_machine_init(MachineClass *mc)
>> >>      mc->init = riscv_sifive_u_init;
>> >>      mc->max_cpus = SIFIVE_U_MANAGEMENT_CPU_COUNT + SIFIVE_U_COMPUTE_CPU_COUNT;
>> >>      mc->min_cpus = SIFIVE_U_MANAGEMENT_CPU_COUNT + 1;
>> >> +    mc->default_cpus = mc->max_cpus;
>> >
>> > Thank you for fixing the 'make check'. Shouldn't it be:
>> >
>> > mc->default_cpus = mc->min_cpus;
>>
>> We have 5 harts on the board that this matches, so I figured that'd be the
>> better default.
>>
>
> Per my understanding mc->default_cpus is used when invoking QEMU
> without passing '-smp n' (that's what 'make check' uses), and with the
> updated sifive_u machine, '-smp 2' is the actual useful configuration
> to boot Linux. For consistency with user experience on other machines,
> without '-smp' means we want a uni-processor machine hence I would
> suggest we set "mc->default_cpus = mc->min_cpus".

OK, I've spun a v3.  I never sent out v2 but I had tagged it, unless there's
any opposition I'll send this out when I'm back on normal internet.