On Tue, Nov 19, 2024 at 09:30:09PM +0100, Paolo Bonzini wrote:
> >
> > Do we have known places that we care a lot on object[_class]_dynamic_cast()
> > performance?
>
> The easiest way to check is probably to print the type of every successful
> object_dynamic_cast and object_class_dynamic_cast. I suspect the result
> will be virtio-blk-device and/or scsi-hd, but maybe those already do an
> unsafe cast (pointer type cast) instead of object_dynamic_cast.
Yes, it sounds more reasonable to me to optimize specific call sites so far
rather than provides something generic.. Though it could still be a
generic API so that devices can opt-in. Maybe still not as fast as an
unsafe cast, though.. I think I'll leave that to block experts when it may
needs some good measurements.
>
> I can give it some measurement if there is, otherwise I'm
> > guessing whatever changes could fall into the noise.
>
>
> Yes, probably. At most you can identify if there any heavy places out of
> the 34000 calls, and see if they can use an unsafe cast.
I can still trivially do this.
I traced qemu using bpf and interestingly in my case close to half (over
10000+) of the calls are about ahci_irq_lower() from different higher level
stack (yeah I used IDE in my setup.. with a split irqchi..), where it has:
PCIDevice *pci_dev = (PCIDevice *) object_dynamic_cast(OBJECT(dev_state),
TYPE_PCI_DEVICE);
So IIUC that can be open to a unsafe cast too, but considering IDE is ODD
FIXES stage, I'm not sure if I should send a patch at all. However I
copied John regardless.
Thanks,
--
Peter Xu
