Hi Heinrich,
On Tue, 31 Dec 2019 at 09:59, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 12/31/19 4:49 PM, Sughosh Ganu wrote:
On Tue, 31 Dec 2019 at 20:38, Heinrich Schuchardt <xypron.glpk@gmx.de mailto:xypron.glpk@gmx.de> wrote:
On 12/31/19 2:31 PM, Sughosh Ganu wrote: > Scan the pci bus in board_init routine before scanning the virtio > devices. This enumerates all the virtio devices, including devices > found on the pci bus. > > Signed-off-by: Sughosh Ganu <sughosh.ganu@linaro.org <mailto:sughosh.ganu@linaro.org>> > --- > board/emulation/qemu-arm/qemu-arm.c | 7 +++++++ > 1 file changed, 7 insertions(+) > > diff --git a/board/emulation/qemu-arm/qemu-arm.c b/board/emulation/qemu-arm/qemu-arm.c > index 4e18733..6c5335c 100644 > --- a/board/emulation/qemu-arm/qemu-arm.c > +++ b/board/emulation/qemu-arm/qemu-arm.c > @@ -63,6 +63,13 @@ struct mm_region *mem_map = qemu_arm64_mem_map; > int board_init(void) > { > /* > + * Scan the pci bus before calling virtio_init. This > + * enumerates all virtio devices, including devices > + * on the pci bus. The last sentence is easy to misunderstand. You surely do not mean that pci_init() would enumerate all virtio devices including mmio virtio devices. I would suggest to rephrase it to "This enumerates all pci devices including virtio devices on the pci bus."Ok.
With the patch I no longer need to run `virtio scan` on qemu_arm64_defonfig before using the rng command supplied in patch [PATCH 1/1] cmd: add rng command https://lists.denx.de/pipermail/u-boot/2019-December/394539.html Unfortunately you only considered the ARM architecture. We need a solution that works on all QEMU architectures (MIPS, RISC-V, X86, ARM).That is fine with me, but currently, only risc-v and arm architectures are calling virtio_init.
How about eliminating board_init() for ARM and RISC-V and moving virtio_init() to common/board_r.c after initr_pci?That can be done, but currently, for some reason, initr_pci function calls pci_init only when CONFIG_DM_PCI is not defined. Not sure why that is the case, although initr_pci gets called after initr_dm.
-sughosh
This dates back to Simon's commit ff3e077bd23c ("dm: pci: Add a uclass for PCI").
I hope that Simon can tell us why he chose not to call pci_init() if DM_PCI is defined.
Driver model uses lazy init. We have discussed this before, and considered using a uclass flag to tell U-Boot that the uclass needs to be fully probed (i.e. probe all devices in that uclass, recursively).
But I'm not sure we got agreement. If you think it is a good idea you could send a patch. The only thing is that I think some platforms don't need PCI to boot so might not want this?
Regards, Simon