The issue was observed when virtio drivers failed to load for a Virtio network PCI device (1af4:1000).

The current implementation of virtio does not permit to have a PCI resource block starting at the address 0.

This can be observed in function pci_iomap_range, at line 28 of file ./lib/pci_iomap.c:
37 if (len <= offset || !start)
38    return NULL;

Because start variable value corresponds to the PCI resource start address:
33 resource_size_t start = pci_resource_start(dev, bar);

In function virtio_pci_legacy_probe (line 212), in file ./drivers/virtio/virtio_pci_legacy.c,
the returned NULL value results in returning the ENOMEM error:
246 rc = -ENOMEM;
247 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
248 if (!vp_dev->ioaddr)
249   goto err_iomap;

As it is already the case for other architectures (like ARM64), I suggest to add an offset the PCIBIOS_MIN_IO constant, it forces IO resource address to be non zero.

Not enough knowledge on PCI to evaluate possible issues resulting from PCIBIOS_MIN_IO value change.

Signed-off-by: Nicolas Peslerbe <redacted>
---
arch/riscv/include/asm/pci.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/riscv/include/asm/pci.h b/arch/riscv/include/asm/pci.h index 7fd52a30e605..15da9129aace 100644
--- a/arch/riscv/include/asm/pci.h
+++ b/arch/riscv/include/asm/pci.h

@@ -12,7 +12,7 @@

#include <asm/io.h>

-#define PCIBIOS_MIN_IO     0
+#define PCIBIOS_MIN_IO     0x1000
#define PCIBIOS_MIN_MEM    0

/* RISC-V shim does not initialize PCI bus */
--
2.32.0

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv