Re: [RFC PATCH v3 14/27] hw/pci-host: Add ls7a1000 PCIe Host bridge support for Loongson3 Platform

[yangxiaojuan]

Hi, Mark

On 12/18/2021 07:39 AM, Mark Cave-Ayland wrote:
> On 04/12/2021 12:07, Xiaojuan Yang wrote:
> 
>> This is a model of the PCIe Host Bridge found on a Loongson-5000
>> processor. It includes a interrupt controller, some interface for
>> pci and nonpci devices. Mainly emulate part of it that is not
>> exactly the same as the host and only use part devices for
>> tcg mode. It support for MSI and MSIX interrupt sources.
>>
>> For more detailed info about ls7a1000 you can see the doc at
>> https://github.com/loongson/LoongArch-Documentation/releases/latest/
>> download/Loongson-7A1000-usermanual-2.00-EN.pdf
>>
>> Signed-off-by: Xiaojuan Yang <yangxiaojuan@loongson.cn>
>> Signed-off-by: Song Gao <gaosong@loongson.cn>
>> ---
>>   hw/pci-host/Kconfig        |   4 +
>>   hw/pci-host/ls7a.c         | 174 +++++++++++++++++++++++++++++++++++++
>>   hw/pci-host/meson.build    |   1 +
>>   include/hw/pci-host/ls7a.h |  51 +++++++++++
>>   4 files changed, 230 insertions(+)
>>   create mode 100644 hw/pci-host/ls7a.c
>>   create mode 100644 include/hw/pci-host/ls7a.h
>>
>> diff --git a/hw/pci-host/Kconfig b/hw/pci-host/Kconfig
>> index 2b5f7d58cc..b02a9d1454 100644
>> --- a/hw/pci-host/Kconfig
>> +++ b/hw/pci-host/Kconfig
>> @@ -77,3 +77,7 @@ config MV64361
>>       bool
>>       select PCI
>>       select I8259
>> +
>> +config PCI_EXPRESS_7A
>> +    bool
>> +    select PCI_EXPRESS
>> diff --git a/hw/pci-host/ls7a.c b/hw/pci-host/ls7a.c
>> new file mode 100644
>> index 0000000000..a783fb2eda
>> --- /dev/null
>> +++ b/hw/pci-host/ls7a.c
>> @@ -0,0 +1,174 @@
>> +/* SPDX-License-Identifier: GPL-2.0-or-later */
>> +/*
>> + * QEMU Loongson 7A1000 North Bridge Emulation
>> + *
>> + * Copyright (C) 2021 Loongson Technology Corporation Limited
>> + */
>> +
>> +#include "qemu/osdep.h"
>> +
>> +#include "hw/pci/pci.h"
>> +#include "hw/pci/pcie_host.h"
>> +#include "hw/qdev-properties.h"
>> +#include "qapi/error.h"
>> +#include "hw/irq.h"
>> +#include "hw/pci/pci_bridge.h"
>> +#include "hw/pci/pci_bus.h"
>> +#include "sysemu/reset.h"
>> +#include "hw/pci-host/ls7a.h"
>> +#include "migration/vmstate.h"
>> +
>> +static const VMStateDescription vmstate_ls7a_pcie = {
>> +    .name = "LS7A_PCIE",
>> +    .version_id = 1,
>> +    .minimum_version_id = 1,
>> +    .fields = (VMStateField[]) {
>> +        VMSTATE_PCI_DEVICE(parent_obj, LS7APCIState),
>> +        VMSTATE_END_OF_LIST()
>> +    }
>> +};
>> +
>> +static void pci_ls7a_config_write(void *opaque, hwaddr addr,
>> +                                  uint64_t val, unsigned size)
>> +{
>> +    pci_data_write(opaque, addr, val, size);
>> +}
>> +
>> +static uint64_t pci_ls7a_config_read(void *opaque,
>> +                                     hwaddr addr, unsigned size)
>> +{
>> +    uint64_t val;
>> +
>> +    val = pci_data_read(opaque, addr, size);
>> +
>> +    return val;
>> +}
>> +
>> +static const MemoryRegionOps pci_ls7a_config_ops = {
>> +    .read = pci_ls7a_config_read,
>> +    .write = pci_ls7a_config_write,
>> +    .valid = {
>> +        .min_access_size = 1,
>> +        .max_access_size = 4,
>> +    },
>> +    .impl = {
>> +        .min_access_size = 1,
>> +        .max_access_size = 4,
>> +    },
>> +    .endianness = DEVICE_LITTLE_ENDIAN,
>> +};
>> +
>> +static void ls7a_pciehost_realize(DeviceState *dev, Error **errp)
>> +{
>> +    PCIHostState *pci = PCI_HOST_BRIDGE(dev);
>> +    LS7APCIEHost *s = LS7A_HOST_DEVICE(dev);
>> +    PCIExpressHost *pex = PCIE_HOST_BRIDGE(dev);
> 
> SysbusDevice *sbd = SYS_BUS_DEVICE(dev) will be needed for later use.
> 
>> +    pci->bus = pci_register_root_bus(dev, "pcie.0", NULL, NULL, s,
>> +                                     get_system_memory(), get_system_io(),
>> +                                     PCI_DEVFN(1, 0), 128, TYPE_PCIE_BUS);
> 
> A device shouldn't map itself into an address space: that is the job of the 
> board. To achieve this LS7APCIEHost should have separate mmio and io memory 
> regions defined and pci_register_root_bus() configured to use these i.e.

OK, I get it and have modified all the region map. Thank you

> 
>     pci->bus = pci_register_root_bus(dev, "pcie.0", NULL, NULL, s,
>                                      &s->pci_mmio, &s->pci_io,
>                                      PCI_DEVFN(1, 0), 128, TYPE_PCIE_BUS);
> 
>> +    memory_region_init_io(&s->pci_conf, OBJECT(dev),
>> +                          &pci_ls7a_config_ops, pci->bus,
>> +                          "ls7a_pci_conf", HT1LO_PCICFG_SIZE);
>> +    memory_region_add_subregion(get_system_memory(), HT1LO_PCICFG_BASE,
>> +                                &s->pci_conf);
> 
> Here add sysbus_init_mmio(sbd, &s->pci_conf) and remove 
> memory_region_add_subregion().
> 
>> +    /* Add ls7a pci-io */
>> +    memory_region_init_alias(&s->pci_io, OBJECT(dev), "ls7a-pci-io",
>> +                             get_system_io(), 0, LS7A_PCI_IO_SIZE);
>> +    memory_region_add_subregion(get_system_memory(), LS7A_PCI_IO_BASE,
>> +                                &s->pci_io);
> 
> Remove the alias onto the system io memory region and instead use 
> sysbus_init_mmio(sbd, &s->pci_io).
> 
> You will also need to make the PCI mmio memory region availble to the board 
> using sysbus_init_mmio(sbd, &s->pci_mmio).
> 
>> +    pcie_host_mmcfg_update(pex, true, LS_PCIECFG_BASE, LS_PCIECFG_SIZE);
> 
> It looks like the pcie_host_mmcfg_*() functions are hardcoded to map the 
> device in system memory which is not recommended for new devices. The best 
> example I can find is in hw/pci-host/xilinx-pcie.c whereby 
> pcie_host_mmcfg_init() is used in xilinx_pcie_host_realize() to setup the 
> MMCFG memory region and then sysbus_init_mmio(sbd, &pex->mmio) is used to 
> make it available to the board.
> 
> Once all these sysbus memory regions have been defined they can be accessed 
> in the board code using sysbus_mmio_get_region().
> 
>> +    qdev_realize(DEVICE(&s->pci_dev), BUS(pci->bus), &error_fatal);
>> +}
>> +
>> +static void ls7a_pcie_realize(PCIDevice *d, Error **errp)
>> +{
>> +    /* pci status */
>> +    d->config[0x6] = 0x01;
>> +    /* base class code */
>> +    d->config[0xb] = 0x06;
>> +    /* header type */
>> +    d->config[0xe] = 0x80;
>> +    /* capabilities pointer */
>> +    d->config[0x34] = 0x40;
>> +    /* link status and control register 0 */
>> +    d->config[0x44] = 0x20;
>> +}
> 
> Should these definitely exist in realize() or should they belong in a reset() 
> function?

OK, I will put them into the reset function.

>  
>> +static void ls7a_pcie_class_init(ObjectClass *klass, void *data)
>> +{
>> +    DeviceClass *dc = DEVICE_CLASS(klass);
>> +    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
>> +
>> +    set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
>> +    dc->desc = "LS7A1000 PCIE Host bridge";
>> +    dc->vmsd = &vmstate_ls7a_pcie;
>> +    k->realize = ls7a_pcie_realize;
>> +    k->vendor_id = 0x0014;
>> +    k->device_id = 0x7a00;
> 
> Generally device and vendor ids are added as constants to 
> include/hw/pci/pci_ids.h.
> 
>> +    k->revision = 0x0;
>> +    k->class_id = PCI_CLASS_BRIDGE_HOST;
>> +    /*
>> +     * PCI-facing part of the host bridge, not usable without the
>> +     * host-facing part, which can't be device_add'ed, yet.
>> +     */
>> +    dc->user_creatable = false;
>> +}
>> +
>> +static const TypeInfo ls7a_pcie_device_info = {
>> +    .name          = TYPE_LS7A_PCIE,
> 
> Why TYPE_LS7A_PCIE? The normal convention would be to call the type 
> TYPE_LS7A_PCI_DEVICE to show that it is derived from a conventional PCI 
> device.

Here use TYPE_LS7A_PCIE means its connect to a pcie bridge, It seems 
TYPE_LS7A_PCI_DEVICE is more appropriate, for the bridge has show the pcie 
feature.
I will modify. Thank you for your advice.

> 
>> +    .parent        = TYPE_PCI_DEVICE,
>> +    .instance_size = sizeof(LS7APCIState),
>> +    .class_init    = ls7a_pcie_class_init,
>> +    .interfaces = (InterfaceInfo[]) {
>> +        { INTERFACE_CONVENTIONAL_PCI_DEVICE },
>> +        { },
>> +    },
>> +};
>> +
>> +static void ls7a_pciehost_initfn(Object *obj)
>> +{
>> +    LS7APCIEHost *s = LS7A_HOST_DEVICE(obj);
>> +    LS7APCIState *ls7a_pci = &s->pci_dev;
>> +
>> +    object_initialize_child(obj, "ls7a_pci", ls7a_pci, TYPE_LS7A_PCIE);
>> +    qdev_prop_set_int32(DEVICE(ls7a_pci), "addr", PCI_DEVFN(0, 0));
>> +    qdev_prop_set_bit(DEVICE(ls7a_pci), "multifunction", false);
>> +}
>> +
>> +static const char *ls7a_pciehost_root_bus_path(PCIHostState *host_bridge,
>> +                                          PCIBus *rootbus)
>> +{
>> +    return "0000:00";
>> +}
>> +
>> +static void ls7a_pciehost_class_init(ObjectClass *klass, void *data)
>> +{
>> +    DeviceClass *dc = DEVICE_CLASS(klass);
>> +    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
>> +
>> +    hc->root_bus_path = ls7a_pciehost_root_bus_path;
>> +    dc->realize = ls7a_pciehost_realize;
>> +    set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
>> +    dc->fw_name = "pci";
>> +    dc->user_creatable = false;
>> +}
>> +
>> +static const TypeInfo ls7a_pciehost_info = {
>> +    .name          = TYPE_LS7A_HOST_DEVICE,
> 
> TYPE_LS7A_PCIE_HOST_BRIDGE would be the normal convention.
> 
>> +    .parent        = TYPE_PCIE_HOST_BRIDGE,
>> +    .instance_size = sizeof(LS7APCIEHost),
>> +    .instance_init = ls7a_pciehost_initfn,
>> +    .class_init    = ls7a_pciehost_class_init,
>> +};
>> +
>> +static void ls7a_register_types(void)
>> +{
>> +    type_register_static(&ls7a_pciehost_info);
>> +    type_register_static(&ls7a_pcie_device_info);
>> +}
>> +
>> +type_init(ls7a_register_types)
>> diff --git a/hw/pci-host/meson.build b/hw/pci-host/meson.build
>> index 4c4f39c15c..c4955455fd 100644
>> --- a/hw/pci-host/meson.build
>> +++ b/hw/pci-host/meson.build
>> @@ -11,6 +11,7 @@ pci_ss.add(when: 'CONFIG_PCI_SABRE', if_true: 
>> files('sabre.c'))
>>   pci_ss.add(when: 'CONFIG_XEN_IGD_PASSTHROUGH', if_true: 
>> files('xen_igd_pt.c'))
>>   pci_ss.add(when: 'CONFIG_REMOTE_PCIHOST', if_true: files('remote.c'))
>>   pci_ss.add(when: 'CONFIG_SH_PCI', if_true: files('sh_pci.c'))
>> +pci_ss.add(when: 'CONFIG_PCI_EXPRESS_7A', if_true: files('ls7a.c'))
>>     # PPC devices
>>   pci_ss.add(when: 'CONFIG_RAVEN_PCI', if_true: files('raven.c'))
>> diff --git a/include/hw/pci-host/ls7a.h b/include/hw/pci-host/ls7a.h
>> new file mode 100644
>> index 0000000000..32d6f045dc
>> --- /dev/null
>> +++ b/include/hw/pci-host/ls7a.h
>> @@ -0,0 +1,51 @@
>> +/* SPDX-License-Identifier: GPL-2.0-or-later */
>> +/*
>> + * QEMU LoongArch CPU
>> + *
>> + * Copyright (c) 2021 Loongson Technology Corporation Limited
>> + */
>> +
>> +#ifndef HW_LS7A_H
>> +#define HW_LS7A_H
>> +
>> +#include "hw/pci/pci.h"
>> +#include "hw/pci/pcie_host.h"
>> +#include "hw/pci-host/pam.h"
>> +#include "qemu/units.h"
>> +#include "qemu/range.h"
>> +#include "qom/object.h"
>> +
>> +#define HT1LO_PCICFG_BASE        0x1a000000
>> +#define HT1LO_PCICFG_SIZE        0x02000000
>> +
>> +#define LS_PCIECFG_BASE          0x20000000
>> +#define LS_PCIECFG_SIZE          0x08000000
>> +
>> +#define LS7A_PCI_IO_BASE         0x18000000UL
>> +#define LS7A_PCI_IO_SIZE         0x00010000
>> +
>> +struct LS7APCIState {
>> +    /*< private >*/
>> +    PCIDevice parent_obj;
>> +    /*< public >*/
>> +};
>> +
>> +typedef struct LS7APCIState LS7APCIState;
>> +typedef struct LS7APCIEHost {
>> +    /*< private >*/
>> +    PCIExpressHost parent_obj;
>> +    /*< public >*/
>> +
>> +    LS7APCIState pci_dev;
>> +
>> +    MemoryRegion pci_conf;
>> +    MemoryRegion pci_io;
>> +} LS7APCIEHost;
>> +
>> +#define TYPE_LS7A_HOST_DEVICE "ls7a1000-pciehost"
>> +OBJECT_DECLARE_SIMPLE_TYPE(LS7APCIEHost, LS7A_HOST_DEVICE)
>> +
>> +#define TYPE_LS7A_PCIE "ls7a1000_pcie"
>> +OBJECT_DECLARE_SIMPLE_TYPE(LS7APCIState, LS7A_PCIE)
>> +
>> +#endif /* HW_LS7A_H */
> 
> 
> ATB,
> 
> Mark.