On 9/8/20 12:30 PM, Philippe Mathieu-Daudé wrote:
>>> On 8/28/20 9:19 PM, Sai Pavan Boddu wrote:
>>>> This patch starts making the hcd-xhci.c pci free, as part of this
>>>> restructuring dma read/writes are handled without passing pci object.
>>>>
>>>> Signed-off-by: Sai Pavan Boddu <sai.pavan.bo...@xilinx.com>
>>>> ---
>>>> hw/usb/hcd-xhci.c | 24 +++++++++++------------- hw/usb/hcd-xhci.h |
>>>> 3 +++
>>>> 2 files changed, 14 insertions(+), 13 deletions(-)
>>>>
> [...]
>>>> --- a/hw/usb/hcd-xhci.h
>>>> +++ b/hw/usb/hcd-xhci.h
>>>> @@ -22,6 +22,8 @@
>>>> #ifndef HW_USB_HCD_XHCI_H
>>>> #define HW_USB_HCD_XHCI_H
>>>>
>>>> +#include "sysemu/dma.h"
>>>
>>> AddressSpace is forward-declared in "qemu/typedefs.h", so no need to
>>> include it here (yes in the sources including hcd-xhci.h).
>> [Sai Pavan Boddu] Yes you are right!, but without this " dma_addr_t " is
>> undefined.
>> At this point of the patch, hcd-xhci.h is compiled along with pci.h which
>> would provide 'dma_addr_t', but when we strip the pci wrapper around
>> hcd-xhci we would miss it. Let me know, if its good to add later in the
>> patch series when the split happens.
>
> OK :( I'd prefer to only include "sysemu/dma.h" in hw/usb/hcd-xhci.c.
>
> Cc'ing David who added dma_addr_t in commit e5332e6334f
> ("iommu: Introduce IOMMU emulation infrastructure"), it
> might be simpler to move its declaration to "qemu/typedefs.h"
> too.
Btw this type should be poisoned for user-mode.
>
> So no change needed for this patch.
>
>>
>> Regards,
>> Sai Pavan
>>>
>>> With that fixed:
>>> Reviewed-by: Philippe Mathieu-Daudé <phi...@redhat.com>
>>>
>>>> +
>>>> #define TYPE_XHCI "base-xhci"
>>>> #define TYPE_NEC_XHCI "nec-usb-xhci"
>>>> #define TYPE_QEMU_XHCI "qemu-xhci"
>>>> @@ -189,6 +191,7 @@ struct XHCIState {
>>>>
>>>> USBBus bus;
>>>> MemoryRegion mem;
>>>> + AddressSpace *as;
>>>> MemoryRegion mem_cap;
>>>> MemoryRegion mem_oper;
>>>> MemoryRegion mem_runtime;
>>>>
>>
>
