On 2023-06-08 21:20:04 [-0300], Luis Claudio R. Goncalves wrote: > Sebastian, Steven, > > I don't know if this is the best solution for the problem, but surely the > simplest one. If this patch makes sense, it could probably go upstream as > the changes only affect RT. Lockdep should complain, too because the locking order here is raw_spinlock_t -> spinlock_t. Instead of making the lock raw, I would argue to remove the lock an safe a few cycles. The lock is only used by vmd_pci_read() and vmd_pci_write(). Both functions are invoked by pci_bus_read_config_*() or pci_bus_write_config_*() and both acquire pci_lock first. Therefore the lock does not add any benefit because vmd_pci_read() vs vmd_pci_write() is already serialized. If you make the patch feel free to reuse the wording above. >From a quick look, drivers/pci/controller/pci-tegra.c drivers/pci/controller/pcie-rcar-host.c could use some care ;) Sebastian
