Re: [Qemu-devel] [PATCH v1 1/2] xlnx-zynqmp: Properly support the smp command line option

[Eduardo Habkost <ehabkost@redhat.com>]

On Tue, Nov 07, 2017 at 12:04:18PM -0800, Alistair Francis wrote:
> On Tue, Nov 7, 2017 at 12:00 PM, Eduardo Habkost <ehabkost@redhat.com> wrote:
> > On Tue, Nov 07, 2017 at 10:59:39AM -0800, Alistair Francis wrote:
> >> Allow the -smp command line option to control the number of CPUs we
> >> create.
> >>
> >> Signed-off-by: Alistair Francis <alistair.francis@xilinx.com>
> >> ---
> >>
> >>  hw/arm/xlnx-zcu102.c |  3 ++-
> >>  hw/arm/xlnx-zynqmp.c | 26 ++++++++++++++++----------
> >>  2 files changed, 18 insertions(+), 11 deletions(-)
> >>
> >> diff --git a/hw/arm/xlnx-zcu102.c b/hw/arm/xlnx-zcu102.c
> >> index e2d15a1c9d..7ec03dad42 100644
> >> --- a/hw/arm/xlnx-zcu102.c
> >> +++ b/hw/arm/xlnx-zcu102.c
> >> @@ -235,7 +235,8 @@ static void xlnx_zcu102_machine_class_init(ObjectClass *oc, void *data)
> >>  {
> >>      MachineClass *mc = MACHINE_CLASS(oc);
> >>
> >> -    mc->desc = "Xilinx ZynqMP ZCU102 board";
> >> +    mc->desc = "Xilinx ZynqMP ZCU102 board with 4xA53s and 2xR5s based on " \
> >> +               "the value of smp";
> >>      mc->init = xlnx_zcu102_init;
> >>      mc->block_default_type = IF_IDE;
> >>      mc->units_per_default_bus = 1;
> >> diff --git a/hw/arm/xlnx-zynqmp.c b/hw/arm/xlnx-zynqmp.c
> >> index d4b6560194..c707c66322 100644
> >> --- a/hw/arm/xlnx-zynqmp.c
> >> +++ b/hw/arm/xlnx-zynqmp.c
> >> @@ -98,8 +98,9 @@ static void xlnx_zynqmp_create_rpu(XlnxZynqMPState *s, const char *boot_cpu,
> >>  {
> >>      Error *err = NULL;
> >>      int i;
> >> +    int num_rpus = MIN(smp_cpus - XLNX_ZYNQMP_NUM_APU_CPUS, XLNX_ZYNQMP_NUM_RPU_CPUS);
> >>
> >> -    for (i = 0; i < XLNX_ZYNQMP_NUM_RPU_CPUS; i++) {
> >> +    for (i = 0; i < num_rpus; i++) {
> >>          char *name;
> >>
> >>          object_initialize(&s->rpu_cpu[i], sizeof(s->rpu_cpu[i]),
> >> @@ -132,8 +133,9 @@ static void xlnx_zynqmp_init(Object *obj)
> >>  {
> >>      XlnxZynqMPState *s = XLNX_ZYNQMP(obj);
> >>      int i;
> >> +    int num_apus = MIN(smp_cpus, XLNX_ZYNQMP_NUM_APU_CPUS);
> >>
> >> -    for (i = 0; i < XLNX_ZYNQMP_NUM_APU_CPUS; i++) {
> >> +    for (i = 0; i < num_apus; i++) {
> >>          object_initialize(&s->apu_cpu[i], sizeof(s->apu_cpu[i]),
> >>                            "cortex-a53-" TYPE_ARM_CPU);
> >>          object_property_add_child(obj, "apu-cpu[*]", OBJECT(&s->apu_cpu[i]),
> >> @@ -182,6 +184,7 @@ static void xlnx_zynqmp_realize(DeviceState *dev, Error **errp)
> >>      MemoryRegion *system_memory = get_system_memory();
> >>      uint8_t i;
> >>      uint64_t ram_size;
> >> +    int num_apus = MIN(smp_cpus, XLNX_ZYNQMP_NUM_APU_CPUS);
> >>      const char *boot_cpu = s->boot_cpu ? s->boot_cpu : "apu-cpu[0]";
> >>      ram_addr_t ddr_low_size, ddr_high_size;
> >>      qemu_irq gic_spi[GIC_NUM_SPI_INTR];
> >> @@ -233,10 +236,10 @@ static void xlnx_zynqmp_realize(DeviceState *dev, Error **errp)
> >>
> >>      qdev_prop_set_uint32(DEVICE(&s->gic), "num-irq", GIC_NUM_SPI_INTR + 32);
> >>      qdev_prop_set_uint32(DEVICE(&s->gic), "revision", 2);
> >> -    qdev_prop_set_uint32(DEVICE(&s->gic), "num-cpu", XLNX_ZYNQMP_NUM_APU_CPUS);
> >> +    qdev_prop_set_uint32(DEVICE(&s->gic), "num-cpu", num_apus);
> >>
> >>      /* Realize APUs before realizing the GIC. KVM requires this.  */
> >> -    for (i = 0; i < XLNX_ZYNQMP_NUM_APU_CPUS; i++) {
> >> +    for (i = 0; i < num_apus; i++) {
> >>          char *name;
> >>
> >>          object_property_set_int(OBJECT(&s->apu_cpu[i]), QEMU_PSCI_CONDUIT_SMC,
> >> @@ -292,7 +295,7 @@ static void xlnx_zynqmp_realize(DeviceState *dev, Error **errp)
> >>          }
> >>      }
> >>
> >> -    for (i = 0; i < XLNX_ZYNQMP_NUM_APU_CPUS; i++) {
> >> +    for (i = 0; i < num_apus; i++) {
> >>          qemu_irq irq;
> >>
> >>          sysbus_connect_irq(SYS_BUS_DEVICE(&s->gic), i,
> >> @@ -307,11 +310,14 @@ static void xlnx_zynqmp_realize(DeviceState *dev, Error **errp)
> >>      }
> >>
> >>      if (s->has_rpu) {
> >> -        xlnx_zynqmp_create_rpu(s, boot_cpu, &err);
> >> -        if (err) {
> >> -            error_propagate(errp, err);
> >> -            return;
> >> -        }
> >> +        info_report("The 'has_rpu' property is no longer required, to use the "
> >> +                    "RPUs just use -smp 6.");
> >> +    }
> >
> > Is "-global driver=xlnx,,zynqmp,property=has_rpu,value=on"
> > without an explicit -smp option supposed to be a supported
> > configuration?
> >
> > 0) On current master, we have this:
> >
> >   $ ./aarch64-softmmu/qemu-system-aarch64 -machine xlnx-zcu102 -global driver=xlnx,,zynqmp,property=has_rpu,value=on
> >   **
> >   ERROR:/home/ehabkost/rh/proj/virt/qemu/tcg/tcg.c:538:tcg_register_thread: assertion failed: (n < max_cpus)
> >   Aborted (core dumped)
> >
> > 1) With your patch we have this:
> >
> >   $ ./aarch64-softmmu/qemu-system-aarch64 -machine xlnx-zcu102 -global driver=xlnx,,zynqmp,property=has_rpu,value=on -monitor stdio
> >   QEMU 2.10.50 monitor - type 'help' for more information
> >   (qemu) qemu-system-aarch64: info: The 'has_rpu' property is no longer required, to use the RPUs just use -smp 6.
> >   (qemu) info cpus
> >   * CPU #0: thread_id=1662
> >   (qemu)
> >
> > 2) With your patch plus Emilio's original min_cpus/default_cpus
> > proposal[1], we have this:
> >
> >   $ ./aarch64-softmmu/qemu-system-aarch64 -machine xlnx-zcu102 -global driver=xlnx,,zynqmp,property=has_rpu,value=on -monitor stdio
> >   QEMU 2.10.50 monitor - type 'help' for more information
> >   (qemu) qemu-system-aarch64: info: The 'has_rpu' property is no longer required, to use the RPUs just use -smp 6.
> >   (qemu) info cpus
> >   * CPU #0: thread_id=7112
> >     CPU #1: (halted) thread_id=7113
> >     CPU #2: (halted) thread_id=7114
> >     CPU #3: (halted) thread_id=7115
> >   (qemu)
> >
> > 3) With Emilio's max_additional_cpus proposal[2], we have this:
> >
> >   $ ./aarch64-softmmu/qemu-system-aarch64 -machine xlnx-zcu102 -global driver=xlnx,,zynqmp,property=has_rpu,value=on -monitor stdio
> >   QEMU 2.10.50 monitor - type 'help' for more information
> >   (qemu) info cpus
> >   * CPU #0: thread_id=4045
> >     CPU #1: (halted) thread_id=4046
> >     CPU #2: (halted) thread_id=4047
> >     CPU #3: (halted) thread_id=4048
> >     CPU #4: (halted) thread_id=4049
> >     CPU #5: (halted) thread_id=4050
> >   (qemu)
> >
> >
> > Which option is preferred?  I like option #2 because it's
> > simpler, but I would like to confirm this is really the intended
> > behavior.
> 
> After this patch (and the fix to TCG otherwise it seg faults) no one
> should use the has_rpu property. It is ignored and will print a
> message saying that it is ignored.
> 
> Users should only use the -smp option now. So option 2 is the way to go.

Sounds reasonable to me.

Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>

I just sent one small comment to the min_cpus/default_cpus patch
at [1], and I'm waiting for Richard's input on Emilio's TCG
fix[3].

[3] https://mid.mail-archive.com/20171106201332.GA2152@flamenco


> 
> Thanks,
> Alistair
> 
> >
> >
> > [1] https://mid.mail-archive.com/1509734853-3014-1-git-send-email-cota@braap.org
> > [2] https://mid.mail-archive.com/20171106215454.GB2152@flamenco
> >
> >
> >> +
> >> +    xlnx_zynqmp_create_rpu(s, boot_cpu, &err);
> >> +    if (err) {
> >> +        error_propagate(errp, err);
> >> +        return;
> >>      }
> >>
> >>      if (!s->boot_cpu_ptr) {
> >> --
> >> 2.11.0
> >>
> >
> > --
> > Eduardo

-- 
Eduardo