Re: [PATCH v2 3/9] ppc/pnv: use dev->parent_bus->parent to get the PHB

From: Daniel Henrique Barboza <danielhb413@gmail.com>
To: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>, qemu-devel@nongnu.org
Cc: qemu-ppc@nongnu.org, clg@kaod.org, fbarrat@linux.ibm.com
Subject: Re: [PATCH v2 3/9] ppc/pnv: use dev->parent_bus->parent to get the PHB
Date: Mon, 20 Jun 2022 16:05:03 -0300	[thread overview]
Message-ID: <675e09f8-ab0c-d2c7-e0f0-1f3c57bb7dd6@gmail.com> (raw)
In-Reply-To: <d36775a8-15f7-ab03-fc08-3fdcfc72d293@ilande.co.uk>

On 6/20/22 04:27, Mark Cave-Ayland wrote:
> On 18/06/2022 12:01, Daniel Henrique Barboza wrote:
> 
>> It is not advisable to execute an object_dynamic_cast() to poke into
>> bus->qbus.parent and follow it up with a C cast into the PnvPHB type we
>> think we got.
>>
>> A better way is to access the PnvPHB object via a QOM macro accessing
>> the existing parent links of the DeviceState. For a given
>> pnv-phb3/4-root-port 'dev', dev->parent_bus will give us the PHB bus,
>> and dev->parent_bus->parent is the PHB. Use the adequate QOM macro to
>> assert the type, and keep the NULL check in case we didn't get the
>> object we were expecting.
>>
>> Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
>> ---
>>   hw/pci-host/pnv_phb3.c | 10 +++++++---
>>   hw/pci-host/pnv_phb4.c | 10 +++++++---
>>   2 files changed, 14 insertions(+), 6 deletions(-)
>>
>> diff --git a/hw/pci-host/pnv_phb3.c b/hw/pci-host/pnv_phb3.c
>> index 4ba660f8b9..5e7f827415 100644
>> --- a/hw/pci-host/pnv_phb3.c
>> +++ b/hw/pci-host/pnv_phb3.c
>> @@ -1139,12 +1139,16 @@ static void pnv_phb3_root_port_realize(DeviceState *dev, Error **errp)
>>   {
>>       PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
>>       PCIDevice *pci = PCI_DEVICE(dev);
>> -    PCIBus *bus = pci_get_bus(pci);
>>       PnvPHB3 *phb = NULL;
>>       Error *local_err = NULL;
>> -    phb = (PnvPHB3 *) object_dynamic_cast(OBJECT(bus->qbus.parent),
>> -                                          TYPE_PNV_PHB3);
>> +    /*
>> +     * dev->parent_bus gives access to the pnv-phb-root bus.
>> +     * The PnvPHB3 is the owner (parent) of the bus.
>> +     */
>> +    if (dev->parent_bus) {
> 
> Here dev->parent_bus shouldn't be accessed directly: you should use qdev_get_parent_bus() instead.
> 
>> +        phb = PNV_PHB3(dev->parent_bus->parent);
>> +    }
> 
> This one is a bit trickier, since part of the qdev design is that devices should only be aware of their immediate bus, and not the device parenting that bus i.e. dev->parent_bus->parent shouldn't be allowed.
> 
> What is really needed here is to use QOM links (or embed the device as a suitable QOM child) to get the PHB reference which I imagine will be changed as part of the follow-up series. So I think this can be left as-is for now, and fixed later.

In the previous patch (2) I've put the root port as a child of the bus,
giving us this hierarchy:

     /pnv-phb3[0] (pnv-phb3)  <====== PHB
       /lsi (ics)
       /msi (phb3-msi)
       /msi32[0] (memory-region)
       /msi64[0] (memory-region)
       /pbcq (pnv-pbcq)
     (...)
       /phb3_iommu[0] (pnv-phb3-iommu-memory-region)
       /pnv-phb3-root.0 (pnv-phb3-root)  <=== bus
         /pnv-phb3-root-port[0] (pnv-phb3-root-port) <==== root-port
           /bus master container[0] (memory-region)
           /bus master[0] (memory-region)
           /pci_bridge_io[0] (memory-region)
           /pci_bridge_io[1] (memory-region)
           /pci_bridge_mem[0] (memory-region)
           /pci_bridge_pci[0] (memory-region)
           /pci_bridge_pref_mem[0] (memory-region)
           /pci_bridge_vga_io_hi[0] (memory-region)
           /pci_bridge_vga_io_lo[0] (memory-region)
           /pci_bridge_vga_mem[0] (memory-region)
           /pcie.0 (PCIE)

I did it like this instead of the PHB for no particular reason. If the root port of
other PHBs are located as a direct child of the PHB I can change it.

All that said, thinking more about it, since I need to access the PHB just
to set "chassis" and "slot" of the device, and I'm already setting a QOM
parent for it, I guess I'll just set that before root_port_realize() and spare us
from having to accessing the parent of the parent bus of the root_port.

Thanks,

Daniel

> 
>>       if (!phb) {
>>           error_setg(errp,
>> diff --git a/hw/pci-host/pnv_phb4.c b/hw/pci-host/pnv_phb4.c
>> index ffd9d8a947..a0ee52e820 100644
>> --- a/hw/pci-host/pnv_phb4.c
>> +++ b/hw/pci-host/pnv_phb4.c
>> @@ -1782,12 +1782,16 @@ static void pnv_phb4_root_port_realize(DeviceState *dev, Error **errp)
>>   {
>>       PCIERootPortClass *rpc = PCIE_ROOT_PORT_GET_CLASS(dev);
>>       PCIDevice *pci = PCI_DEVICE(dev);
>> -    PCIBus *bus = pci_get_bus(pci);
>>       PnvPHB4 *phb = NULL;
>>       Error *local_err = NULL;
>> -    phb = (PnvPHB4 *) object_dynamic_cast(OBJECT(bus->qbus.parent),
>> -                                          TYPE_PNV_PHB4);
>> +    /*
>> +     * dev->parent_bus gives access to the pnv-phb-root bus.
>> +     * The PnvPHB4 is the owner (parent) of the bus.
>> +     */
>> +    if (dev->parent_bus) {
>> +        phb = PNV_PHB4(dev->parent_bus->parent);
>> +    }
>>       if (!phb) {
>>           error_setg(errp, "%s must be connected to pnv-phb4 buses", dev->id);
> 
> I've had a quick look over the rest of the series and from what I can see this is definitely heading in the right direction :)
> 
> 
> ATB,
> 
> Mark.