[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-ppc] [PATCH v3 03/10] ppc/pnv: add a core mask to PnvChip
From: |
Cédric Le Goater |
Subject: |
Re: [Qemu-ppc] [PATCH v3 03/10] ppc/pnv: add a core mask to PnvChip |
Date: |
Wed, 21 Sep 2016 09:57:50 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.3.0 |
On 09/20/2016 03:57 PM, David Gibson wrote:
> On Thu, Sep 15, 2016 at 02:45:53PM +0200, Cédric Le Goater wrote:
>> This will be used to build real HW ids for the cores and enforce some
>> limits on the available cores per chip.
>>
>> Signed-off-by: Cédric Le Goater <address@hidden>
>> ---
>>
>> Changes since v2 :
>>
>> - added POWER9 support
>> - removed cores_max
>> - introduces a pnv_chip_core_sanitize() helper to check the core
>> ids_mask and the maximum number of cores
>>
>> hw/ppc/pnv.c | 66
>> ++++++++++++++++++++++++++++++++++++++++++++++++++++
>> include/hw/ppc/pnv.h | 4 ++++
>> 2 files changed, 70 insertions(+)
>>
>> diff --git a/hw/ppc/pnv.c b/hw/ppc/pnv.c
>> index 2aa5be56c8dc..ec7dd6ac5ea1 100644
>> --- a/hw/ppc/pnv.c
>> +++ b/hw/ppc/pnv.c
>> @@ -226,11 +226,44 @@ static void ppc_powernv_init(MachineState *machine)
>> snprintf(chip_name, sizeof(chip_name), "chip[%d]", CHIP_HWID(i));
>> object_property_add_child(OBJECT(pnv), chip_name, chip,
>> &error_fatal);
>> object_property_set_int(chip, CHIP_HWID(i), "chip-id",
>> &error_fatal);
>> + object_property_set_int(chip, smp_cores, "nr-cores", &error_fatal);
>> + /*
>> + * We could customize cores_mask for the chip here. May be
>> + * using a powernv machine property, like 'num-chips'. Let the
>> + * chip choose the default for now.
>> + */
>> + object_property_set_int(chip, 0x0, "cores-mask", &error_fatal);
>
> If you're selecting the default, why set it at all?
To remind that we could add a machine property to define the layout of the
cores in each chip on the command line, something like :
-machine powernv,accel=tcg,num-chips=2,cores-mask="0xdeadbeef,0xbadc0ffee"
That would be nice to test different layouts.
>
>> object_property_set_bool(chip, true, "realized", &error_fatal);
>> }
>> g_free(chip_typename);
>> }
>>
>> +/* Allowed core identifiers on a POWER8 Processor Chip :
>> + *
>> + * <EX0 reserved>
>> + * EX1 - Venice only
>> + * EX2 - Venice only
>> + * EX3 - Venice only
>> + * EX4
>> + * EX5
>> + * EX6
>> + * <EX7,8 reserved> <reserved>
>> + * EX9 - Venice only
>> + * EX10 - Venice only
>> + * EX11 - Venice only
>> + * EX12
>> + * EX13
>> + * EX14
>> + * <EX15 reserved>
>> + */
>> +#define POWER8E_CORE_MASK (0x7070ull)
>> +#define POWER8_CORE_MASK (0x7e7eull)
>> +
>> +/*
>> + * POWER9 has 24 cores, ids starting at 0x20
>> + */
>> +#define POWER9_CORE_MASK (0xffffff00000000ull)
>> +
>> static void pnv_chip_power8e_class_init(ObjectClass *klass, void *data)
>> {
>> DeviceClass *dc = DEVICE_CLASS(klass);
>> @@ -239,6 +272,7 @@ static void pnv_chip_power8e_class_init(ObjectClass
>> *klass, void *data)
>> k->cpu_model = "POWER8E";
>> k->chip_type = PNV_CHIP_POWER8E;
>> k->chip_cfam_id = 0x221ef04980000000ull; /* P8 Murano DD2.1 */
>> + k->cores_mask = POWER8E_CORE_MASK;
>> dc->desc = "PowerNV Chip POWER8E";
>> }
>>
>> @@ -257,6 +291,7 @@ static void pnv_chip_power8_class_init(ObjectClass
>> *klass, void *data)
>> k->cpu_model = "POWER8";
>> k->chip_type = PNV_CHIP_POWER8;
>> k->chip_cfam_id = 0x220ea04980000000ull; /* P8 Venice DD2.0 */
>> + k->cores_mask = POWER8_CORE_MASK;
>> dc->desc = "PowerNV Chip POWER8";
>> }
>>
>> @@ -275,6 +310,7 @@ static void pnv_chip_power8nvl_class_init(ObjectClass
>> *klass, void *data)
>> k->cpu_model = "POWER8NVL";
>> k->chip_type = PNV_CHIP_POWER8NVL;
>> k->chip_cfam_id = 0x120d304980000000ull; /* P8 Naples DD1.0 */
>> + k->cores_mask = POWER8_CORE_MASK;
>> dc->desc = "PowerNV Chip POWER8NVL";
>> }
>>
>> @@ -293,6 +329,7 @@ static void pnv_chip_power9_class_init(ObjectClass
>> *klass, void *data)
>> k->cpu_model = "POWER9";
>> k->chip_type = PNV_CHIP_POWER9;
>> k->chip_cfam_id = 0x100d104980000000ull; /* P9 Nimbus DD1.0 */
>> + k->cores_mask = POWER9_CORE_MASK;
>> dc->desc = "PowerNV Chip POWER9";
>> }
>>
>> @@ -303,11 +340,38 @@ static const TypeInfo pnv_chip_power9_info = {
>> .class_init = pnv_chip_power9_class_init,
>> };
>>
>> +static void pnv_chip_core_sanitize(PnvChip *chip)
>> +{
>> + PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip);
>> + int cores_max = hweight_long(pcc->cores_mask);
>> +
>> + if (chip->nr_cores > cores_max) {
>> + error_report("warning: too many cores for chip ! Limiting to %d",
>> + cores_max);
>> + chip->nr_cores = cores_max;
>
> This is called from realize() which takes an errp argument. It would
> be better to pass that in and actually report the error up the chain
> here, rather than assuming a warning is the right answer.
yes. I forgot that.
> Also.. shouldn't you actually check nr_cores against the chip local
> mask instead of the class one?
I need to rework a bit that part because I want to use the mask for
the "cpu hwid" -> "icp index" mapping table. So I would need to modify
chip->cores_mask depending on chip->nr_cores in the case we ask for
less cores than there is in the mask.
Thanks,
C.
>> + }
>> +
>> + /* no custom mask for this chip, let's use the default one from
>> + * the chip class */
>> + if (!chip->cores_mask) {
>> + chip->cores_mask = pcc->cores_mask;
>> + }
>> +
>> + /* filter alien core ids ! some are reserved */
>> + if ((chip->cores_mask & pcc->cores_mask) != chip->cores_mask) {
>> + error_report("warning: invalid core mask for chip !");
>
> Likewise here you should propagate the error.
>
>> + }
>> + chip->cores_mask &= pcc->cores_mask;
>> +}
>> +
>> static void pnv_chip_realize(DeviceState *dev, Error **errp)
>> {
>> PnvChip *chip = PNV_CHIP(dev);
>> PnvChipClass *pcc = PNV_CHIP_GET_CLASS(chip);
>>
>> + /* Early checks on the core settings */
>> + pnv_chip_core_sanitize(chip);
>> +
>> if (pcc->realize) {
>> pcc->realize(chip, errp);
>> }
>> @@ -315,6 +379,8 @@ static void pnv_chip_realize(DeviceState *dev, Error
>> **errp)
>>
>> static Property pnv_chip_properties[] = {
>> DEFINE_PROP_UINT32("chip-id", PnvChip, chip_id, 0),
>> + DEFINE_PROP_UINT32("nr-cores", PnvChip, nr_cores, 1),
>> + DEFINE_PROP_UINT64("cores-mask", PnvChip, cores_mask, 0x0),
>> DEFINE_PROP_END_OF_LIST(),
>> };
>>
>> diff --git a/include/hw/ppc/pnv.h b/include/hw/ppc/pnv.h
>> index 6e6628edcf6a..cfc32586320f 100644
>> --- a/include/hw/ppc/pnv.h
>> +++ b/include/hw/ppc/pnv.h
>> @@ -42,6 +42,9 @@ typedef struct PnvChip {
>>
>> /*< public >*/
>> uint32_t chip_id;
>> +
>> + uint32_t nr_cores;
>> + uint64_t cores_mask;
>> } PnvChip;
>>
>> typedef struct PnvChipClass {
>> @@ -52,6 +55,7 @@ typedef struct PnvChipClass {
>> const char *cpu_model;
>> PnvChipType chip_type;
>> uint64_t chip_cfam_id;
>> + uint64_t cores_mask;
>>
>> void (*realize)(PnvChip *dev, Error **errp);
>> } PnvChipClass;
>
- [Qemu-ppc] [PATCH v3 00/10] ppc/pnv: loading skiboot and booting the kernel, Cédric Le Goater, 2016/09/15
- [Qemu-ppc] [PATCH v3 02/10] ppc/pnv: add a PnvChip object, Cédric Le Goater, 2016/09/15
- [Qemu-ppc] [PATCH v3 01/10] ppc/pnv: add skeleton PowerNV platform, Cédric Le Goater, 2016/09/15
- [Qemu-ppc] [PATCH v3 03/10] ppc/pnv: add a core mask to PnvChip, Cédric Le Goater, 2016/09/15
- [Qemu-ppc] [PATCH v3 04/10] ppc/pnv: add a PIR handler to PnvChip, Cédric Le Goater, 2016/09/15
- [Qemu-ppc] [PATCH v3 05/10] ppc/pnv: add a PnvCore object, Cédric Le Goater, 2016/09/15