Hi Andre, (sorry for the delay in reply) On 05/04/2019 16:31, Andre Przywara wrote: > On Thu, 7 Mar 2019 08:36:09 +0000 > Julien Thierry <julien.thierry@xxxxxxx> wrote: > > Hi, > >> Linux has this convention that the lower 0x1000 bytes of the IO space >> should not be used. (cf PCIBIOS_MIN_IO). >> >> Just allocate those bytes to prevent future allocation assigning it to >> devices. >> >> Signed-off-by: Julien Thierry <julien.thierry@xxxxxxx> >> --- >> arm/pci.c | 3 +++ >> 1 file changed, 3 insertions(+) >> >> diff --git a/arm/pci.c b/arm/pci.c >> index 83238ca..559e0cf 100644 >> --- a/arm/pci.c >> +++ b/arm/pci.c >> @@ -37,6 +37,9 @@ void pci__arm_init(struct kvm *kvm) >> >> /* Make PCI port allocation start at a properly aligned address */ >> pci_get_io_space_block(align_pad); >> + >> + /* Convention, don't allocate first 0x1000 bytes of PCI IO */ >> + pci_get_io_space_block(0x1000); > > Is this the same problem with mixing up I/O and MMIO space as in the other patch? > io_space means MMIO, right? > Oh yes, you're right. Thanks for catching that (and in the other patch as well). However, fixing it unveiled a bug which apparently requires me to change a bunch of things w.r.t. how we handle the configuration. At boot time, Linux (without probe only) reassigns bars without disabling the device response (it assumes that none of the devices it can configure are being used/accessed). This means that during the reassignment, bars from different or same devices can temporarily alias/overlap each other during boot time. And the current handling of PCI io/mmio region doesn't support that. I'll rework this to make things a little bit more flexible. Thanks, -- Julien Thierry
