Re: [Qemu-devel] [PATCH v4 16/17] spapr_pci: enable basic hotplug operations

[Michael Roth <mdroth@linux.vnet.ibm.com>]

Quoting David Gibson (2015-01-18 23:58:28)
> On Tue, Dec 23, 2014 at 06:30:30AM -0600, Michael Roth wrote:
> > This enables hotplug for PHB bridges. Upon hotplug we generate the
> > OF-nodes required by PAPR specification and IEEE 1275-1994
> > "PCI Bus Binding to Open Firmware" for the device.
> > 
> > We associate the corresponding FDT for these nodes with the DrcEntry
> > corresponding to the slot, which will be fetched via
> > ibm,configure-connector RTAS calls by the guest as described by PAPR
> > specification. The FDT is cleaned up in the case of unplug.
> > 
> > Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
> > ---
> >  hw/ppc/spapr_pci.c | 268 +++++++++++++++++++++++++++++++++++++++++++++++++----
> >  1 file changed, 249 insertions(+), 19 deletions(-)
> > 
> > diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
> > index a5d7791..94e33b4 100644
> > --- a/hw/ppc/spapr_pci.c
> > +++ b/hw/ppc/spapr_pci.c
> > @@ -33,6 +33,7 @@
> >  #include <libfdt.h>
> >  #include "trace.h"
> >  #include "qemu/error-report.h"
> > +#include "qapi/qmp/qerror.h"
> >  
> >  #include "hw/pci/pci_bus.h"
> >  
> > @@ -51,6 +52,15 @@
> >  
> >  #include "hw/ppc/spapr_drc.h"
> >  
> > +#define FDT_MAX_SIZE            0x10000
> > +#define _FDT(exp) \
> > +    do { \
> > +        int ret = (exp);                                           \
> > +        if (ret < 0) {                                             \
> > +            return ret;                                            \
> > +        }                                                          \
> > +    } while (0)
> > +
> >  static sPAPRPHBState *find_phb(sPAPREnvironment *spapr, uint64_t buid)
> >  {
> >      sPAPRPHBState *sphb;
> > @@ -483,6 +493,237 @@ static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
> >      return &phb->iommu_as;
> >  }
> >  
> > +/* Macros to operate with address in OF binding to PCI */
> > +#define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
> > +#define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
> > +#define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
> > +#define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
> > +#define b_ss(x)         b_x((x), 24, 2) /* the space code */
> > +#define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
> > +#define b_ddddd(x)      b_x((x), 11, 5) /* device number */
> > +#define b_fff(x)        b_x((x), 8, 3)  /* function number */
> > +#define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
> > +
> > +/* for 'reg'/'assigned-addresses' OF properties */
> > +#define RESOURCE_CELLS_SIZE 2
> > +#define RESOURCE_CELLS_ADDRESS 3
> > +#define RESOURCE_CELLS_TOTAL \
> > +    (RESOURCE_CELLS_SIZE + RESOURCE_CELLS_ADDRESS)
> > +
> > +static void fill_resource_props(PCIDevice *d, int bus_num,
> > +                                uint32_t *reg, int *reg_size,
> > +                                uint32_t *assigned, int *assigned_size)
> 
> This is another interface which writes to a buffer without any size
> limit information being passed through, which makes me nervous.
> 
> > +{
> > +    uint32_t *reg_row, *assigned_row;
> > +    uint32_t dev_id = (b_bbbbbbbb(bus_num) |
> > +                       b_ddddd(PCI_SLOT(d->devfn)) |
> > +                       b_fff(PCI_FUNC(d->devfn)));
> > +    int i, idx = 0;
> > +
> > +    reg[0] = cpu_to_be32(dev_id);
> > +
> > +    for (i = 0; i < PCI_NUM_REGIONS; i++) {
> > +        if (!d->io_regions[i].size) {
> > +            continue;
> > +        }
> > +        reg_row = &reg[(idx + 1) * RESOURCE_CELLS_TOTAL];
> > +        assigned_row = &assigned[idx * RESOURCE_CELLS_TOTAL];
> > +        reg_row[0] = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i)));
> > +        if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) {
> > +            reg_row[0] |= cpu_to_be32(b_ss(1));
> > +        } else {
> > +            reg_row[0] |= cpu_to_be32(b_ss(2));
> > +        }
> > +        assigned_row[0] = cpu_to_be32(reg_row[0] | b_n(1));
> > +        assigned_row[3] = reg_row[3] = cpu_to_be32(d->io_regions[i].size >> 32);
> > +        assigned_row[4] = reg_row[4] = cpu_to_be32(d->io_regions[i].size);
> > +        assigned_row[1] = cpu_to_be32(d->io_regions[i].addr >> 32);
> > +        assigned_row[2] = cpu_to_be32(d->io_regions[i].addr);
> 
> You don't appear to ever fill in reg_row[1] and reg_row[2].
> 
> > +        idx++;
> > +    }
> > +
> > +    *reg_size = (idx + 1) * RESOURCE_CELLS_TOTAL * sizeof(uint32_t);
> > +    *assigned_size = idx * RESOURCE_CELLS_TOTAL * sizeof(uint32_t);
> > +}
> > +
> > +static int spapr_populate_pci_child_dt(PCIDevice *dev, void *fdt, int offset,
> > +                                       int phb_index, int drc_index)
> > +{
> > +    int slot = PCI_SLOT(dev->devfn);
> > +    char slotname[16];
> > +    bool is_bridge = 1;
> 
> Should use the true and false macros for a bool type, not 0 and 1.
> 
> > +    uint32_t reg[RESOURCE_CELLS_TOTAL * 8] = { 0 };
> > +    uint32_t assigned[RESOURCE_CELLS_TOTAL * 8] = { 0 };
> > +    int pci_status, reg_size, assigned_size;
> > +
> > +    if (pci_default_read_config(dev, PCI_HEADER_TYPE, 1) ==
> > +        PCI_HEADER_TYPE_NORMAL) {
> > +        is_bridge = 0;
> > +    }
> > +
> > +    _FDT(fdt_setprop_cell(fdt, offset, "vendor-id",
> > +                          pci_default_read_config(dev, PCI_VENDOR_ID, 2)));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "device-id",
> > +                          pci_default_read_config(dev, PCI_DEVICE_ID, 2)));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "revision-id",
> > +                          pci_default_read_config(dev, PCI_REVISION_ID, 1)));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "class-code",
> > +                          pci_default_read_config(dev, PCI_CLASS_DEVICE, 2) << 8));
> > +
> > +    _FDT(fdt_setprop_cell(fdt, offset, "interrupts",
> > +                          pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1)));
> > +
> > +    /* if this device is NOT a bridge */
> > +    if (!is_bridge) {
> > +        _FDT(fdt_setprop_cell(fdt, offset, "min-grant",
> > +            pci_default_read_config(dev, PCI_MIN_GNT, 1)));
> > +        _FDT(fdt_setprop_cell(fdt, offset, "max-latency",
> > +            pci_default_read_config(dev, PCI_MAX_LAT, 1)));
> > +        _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id",
> > +            pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2)));
> > +        _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id",
> > +            pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2)));
> > +    }
> > +
> > +    _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size",
> > +        pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1)));
> > +
> > +    /* the following fdt cells are masked off the pci status register */
> > +    pci_status = pci_default_read_config(dev, PCI_STATUS, 2);
> > +    _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed",
> > +                          PCI_STATUS_DEVSEL_MASK & pci_status));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "fast-back-to-back",
> > +                          PCI_STATUS_FAST_BACK & pci_status));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "66mhz-capable",
> > +                          PCI_STATUS_66MHZ & pci_status));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "udf-supported",
> > +                          PCI_STATUS_UDF & pci_status));
> 
> These aren't quite right.  According to the OF PCI binding these are
> boolean properties encoded in the usual way, which is to say absent
> for false and present-but-empty for true.   They shouldn't contain an
> actual value.
> 
> > +
> > +    _FDT(fdt_setprop_string(fdt, offset, "name", "pci"));
> > +    sprintf(slotname, "Slot %d", slot + phb_index * PCI_SLOT_MAX);
> > +    _FDT(fdt_setprop(fdt, offset, "ibm,loc-code", slotname, strlen(slotname)));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index));
> > +
> > +    _FDT(fdt_setprop_cell(fdt, offset, "#address-cells",
> > +                          RESOURCE_CELLS_ADDRESS));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "#size-cells",
> > +                          RESOURCE_CELLS_SIZE));
> > +    _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x",
> > +                          RESOURCE_CELLS_SIZE));
> > +    fill_resource_props(dev, phb_index, reg, &reg_size,
> > +                        assigned, &assigned_size);
> > +    _FDT(fdt_setprop(fdt, offset, "reg", reg, reg_size));
> > +    _FDT(fdt_setprop(fdt, offset, "assigned-addresses",
> > +                     assigned, assigned_size));
> > +
> > +    return 0;
> > +}
> > +
> > +/* create OF node for pci device and required OF DT properties */
> > +static void *spapr_create_pci_child_dt(sPAPRPHBState *phb, PCIDevice *dev,
> > +                                       int drc_index, int *dt_offset)
> > +{
> > +    void *fdt_orig, *fdt;
> > +    int offset, ret;
> > +    int slot = PCI_SLOT(dev->devfn);
> > +    char nodename[512];
> > +
> > +    fdt_orig = g_malloc0(FDT_MAX_SIZE);
> > +    offset = fdt_create(fdt_orig, FDT_MAX_SIZE);
> > +    fdt_begin_node(fdt_orig, "");
> > +    fdt_end_node(fdt_orig);
> > +    fdt_finish(fdt_orig);
> 
> Recent versions of libfdt have an fdt_create_empty_tree() function to
> simplify that standard idiom.

Hmm, it doesn't seem to be in the source that qemu.git/dtc points to, so I'm
hesitant to rely on it. Would it be viable to get the QEMU submodule
updated to v1.4.0?

> 
> > +    fdt = g_malloc0(FDT_MAX_SIZE);
> > +    fdt_open_into(fdt_orig, fdt, FDT_MAX_SIZE);
> 
> There's no need for a second malloc here - fdt_open_into() may be used
> in place.
> 
> > +    sprintf(nodename, "pci@%d", slot);
> > +    offset = fdt_add_subnode(fdt, 0, nodename);
> > +    ret = spapr_populate_pci_child_dt(dev, fdt, offset, phb->index, drc_index);
> > +    g_assert(!ret);
> > +    g_free(fdt_orig);
> > +
> > +    *dt_offset = offset;
> > +    return fdt;
> > +}
> > +
> > +static void spapr_device_hotplug_add(sPAPRDRConnector *drc,
> > +                                     sPAPRPHBState *phb,
> > +                                     PCIDevice *pdev)
> > +{
> > +    sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
> > +    DeviceState *dev = DEVICE(pdev);
> > +    int drc_index = drck->get_index(drc);
> > +    void *fdt = NULL;
> > +    int fdt_start_offset = 0;
> > +
> > +    /* boot-time devices get their device tree node created by SLOF, but for
> > +     * hotplugged devices we need QEMU to generate it so the guest can fetch
> > +     * it via RTAS
> 
> Now that we have to have this code in qemu for the hotplug case we may
> want to consider using it for boot-time devices as well, and removing
> the corresponding code from SLOF, but that's a problem for another day.

Makes sense, since we do this for PHBs already. Can look into it as a follow-up.

> 
> > +     */
> > +    if (dev->hotplugged) {
> > +        fdt = spapr_create_pci_child_dt(phb, pdev, drc_index,
> > +                                        &fdt_start_offset);
> > +    }
> > +    drck->attach(drc, DEVICE(pdev), fdt, fdt_start_offset, !dev->hotplugged);
> > +}
> > +
> > +static void spapr_device_hotplug_remove_cb(DeviceState *dev, void *opaque)
> > +{
> > +    object_unparent(OBJECT(dev));
> > +}
> > +
> > +static void spapr_device_hotplug_remove(sPAPRDRConnector *drc,
> > +                                        sPAPRPHBState *phb,
> > +                                        PCIDevice *pdev)
> > +{
> > +    sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
> > +
> > +    drck->detach(drc, DEVICE(pdev), spapr_device_hotplug_remove_cb, phb);
> > +}
> > +
> > +static void spapr_phb_hot_plug(HotplugHandler *plug_handler,
> > +                               DeviceState *plugged_dev, Error **errp)
> 
> So, this function is hotplugging a PCI device into an existing PHB,
> rather than hotplugging a PHB itself.  Since the DR protocol does
> support both operations, I could see this name becoming confusing.
> 
> > +{
> > +    sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
> > +    PCIDevice *pdev = PCI_DEVICE(plugged_dev);
> > +    sPAPRDRConnector *drc =
> > +        spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_PCI, pdev->devfn);
> 
> Is it safe to call this before checking phb->dr_enabled?

It will be NULL if the DRC wasn't created, so the assertion below the check
should catch any misuse before it happens.

> 
> > +    /* if DR is disabled we don't need to do anything in the case of
> > +     * hotplug or coldplug callbacks
> > +     */
> > +    if (!phb->dr_enabled) {
> > +        /* if this is a hotplug operation initiated by the user
> > +         * we need to let them know it's not enabled
> > +         */
> > +        if (plugged_dev->hotplugged) {
> > +            error_set(errp, QERR_BUS_NO_HOTPLUG,
> > +                      object_get_typename(OBJECT(phb)));
> > +        }
> > +        return;
> > +    }
> > +
> > +    g_assert(drc);
> > +    spapr_device_hotplug_add(drc, phb, pdev);
> > +}
> > +
> > +static void spapr_phb_hot_unplug(HotplugHandler *plug_handler,
> > +                                 DeviceState *plugged_dev, Error **errp)
> > +{
> > +    sPAPRPHBState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler));
> > +    PCIDevice *pdev = PCI_DEVICE(plugged_dev);
> > +    sPAPRDRConnector *drc =
> > +        spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_PCI, pdev->devfn);
> > +
> > +    if (!phb->dr_enabled) {
> > +        error_set(errp, QERR_BUS_NO_HOTPLUG,
> > +                  object_get_typename(OBJECT(phb)));
> > +        return;
> > +    }
> > +
> > +    spapr_device_hotplug_remove(drc, phb, pdev);
> > +}
> > +
> >  static void spapr_phb_realize(DeviceState *dev, Error **errp)
> >  {
> >      SysBusDevice *s = SYS_BUS_DEVICE(dev);
> > @@ -570,6 +811,7 @@ static void spapr_phb_realize(DeviceState *dev, Error **errp)
> >                             &sphb->memspace, &sphb->iospace,
> >                             PCI_DEVFN(0, 0), PCI_NUM_PINS, TYPE_PCI_BUS);
> >      phb->bus = bus;
> > +    qbus_set_hotplug_handler(BUS(phb->bus), DEVICE(sphb), NULL);
> >  
> >      /*
> >       * Initialize PHB address space.
> > @@ -806,6 +1048,7 @@ static void spapr_phb_class_init(ObjectClass *klass, void *data)
> >      PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass);
> >      DeviceClass *dc = DEVICE_CLASS(klass);
> >      sPAPRPHBClass *spc = SPAPR_PCI_HOST_BRIDGE_CLASS(klass);
> > +    HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass);
> >  
> >      hc->root_bus_path = spapr_phb_root_bus_path;
> >      dc->realize = spapr_phb_realize;
> > @@ -815,6 +1058,8 @@ static void spapr_phb_class_init(ObjectClass *klass, void *data)
> >      set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
> >      dc->cannot_instantiate_with_device_add_yet = false;
> >      spc->finish_realize = spapr_phb_finish_realize;
> > +    hp->plug = spapr_phb_hot_plug;
> > +    hp->unplug = spapr_phb_hot_unplug;
> >  }
> >  
> >  static const TypeInfo spapr_phb_info = {
> > @@ -823,6 +1068,10 @@ static const TypeInfo spapr_phb_info = {
> >      .instance_size = sizeof(sPAPRPHBState),
> >      .class_init    = spapr_phb_class_init,
> >      .class_size    = sizeof(sPAPRPHBClass),
> > +    .interfaces    = (InterfaceInfo[]) {
> > +        { TYPE_HOTPLUG_HANDLER },
> > +        { }
> > +    }
> >  };
> >  
> >  PCIHostState *spapr_create_phb(sPAPREnvironment *spapr, int index)
> > @@ -836,17 +1085,6 @@ PCIHostState *spapr_create_phb(sPAPREnvironment *spapr, int index)
> >      return PCI_HOST_BRIDGE(dev);
> >  }
> >  
> > -/* Macros to operate with address in OF binding to PCI */
> > -#define b_x(x, p, l)    (((x) & ((1<<(l))-1)) << (p))
> > -#define b_n(x)          b_x((x), 31, 1) /* 0 if relocatable */
> > -#define b_p(x)          b_x((x), 30, 1) /* 1 if prefetchable */
> > -#define b_t(x)          b_x((x), 29, 1) /* 1 if the address is aliased */
> > -#define b_ss(x)         b_x((x), 24, 2) /* the space code */
> > -#define b_bbbbbbbb(x)   b_x((x), 16, 8) /* bus number */
> > -#define b_ddddd(x)      b_x((x), 11, 5) /* device number */
> > -#define b_fff(x)        b_x((x), 8, 3)  /* function number */
> > -#define b_rrrrrrrr(x)   b_x((x), 0, 8)  /* register number */
> > -
> >  typedef struct sPAPRTCEDT {
> >      void *fdt;
> >      int node_off;
> > @@ -906,14 +1144,6 @@ int spapr_populate_pci_dt(sPAPRPHBState *phb,
> >          return bus_off;
> >      }
> >  
> > -#define _FDT(exp) \
> > -    do { \
> > -        int ret = (exp);                                           \
> > -        if (ret < 0) {                                             \
> > -            return ret;                                            \
> > -        }                                                          \
> > -    } while (0)
> > -
> >      /* Write PHB properties */
> >      _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci"));
> >      _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB"));
> 
> -- 
> 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