RISC-V: PCI: Avoid handing out address 0 to devices

For RISC-V platforms we permit assigning addresses from 0 to PCI devices,
both in the memory and the I/O bus space, and we happily do so if there
is no conflict, e.g.:

pci 0000:07:00.0: BAR 0: assigned [io  0x0000-0x0007]
pci 0000:07:00.1: BAR 0: assigned [io  0x0008-0x000f]
pci 0000:06:01.0: PCI bridge to [bus 07]
pci 0000:06:01.0:   bridge window [io  0x0000-0x0fff]

(with the SiFive HiFive Unmatched RISC-V board and a dual serial port
option card based on the OxSemi OXPCIe952 device wired for the legacy
UART mode).

Address 0 is treated specially however in many places, for example in
`pci_iomap_range' and `pci_iomap_wc_range' we require that the start
address is non-zero, and even if we let such an address through, then
individual device drivers could reject a request to handle a device at
such an address, such as in `uart_configure_port'.  Consequently given
devices configured as shown above only one is actually usable:

Serial: 8250/16550 driver, 4 ports, IRQ sharing disabled
serial 0000:07:00.0: enabling device (0000 -> 0001)
serial: probe of 0000:07:00.0 failed with error -12
serial 0000:07:00.1: enabling device (0000 -> 0001)
serial 0000:07:00.1: detected caps 00000700 should be 00000500
0000:07:00.1: ttyS0 at I/O 0x8 (irq = 39, base_baud = 15625000) is a 16C950/954

Therefore avoid handing out address 0, by bumping the lowest address
available to PCI via PCIBIOS_MIN_IO and PCIBIOS_MIN_MEM up by 4 and 16
respectively, which is the minimum allocation size for I/O and memory
BARs.

With this in place the system in question we have:

pci 0000:07:00.0: BAR 0: assigned [io  0x1000-0x1007]
pci 0000:07:00.1: BAR 0: assigned [io  0x1008-0x100f]
pci 0000:06:01.0: PCI bridge to [bus 07]
pci 0000:06:01.0:   bridge window [io  0x1000-0x1fff]

and then devices work correctly:

Serial: 8250/16550 driver, 4 ports, IRQ sharing disabled
serial 0000:07:00.0: enabling device (0000 -> 0001)
serial 0000:07:00.0: detected caps 00000700 should be 00000500
0000:07:00.0: ttyS0 at I/O 0x1000 (irq = 38, base_baud = 15625000) is a 16C950/954
serial 0000:07:00.1: enabling device (0000 -> 0001)
serial 0000:07:00.1: detected caps 00000700 should be 00000500
0000:07:00.1: ttyS1 at I/O 0x1008 (irq = 39, base_baud = 15625000) is a 16C950/954

Especially I/O space ranges are particularly valuable, because bridges
only decode bits from 12 up and consequently where 16-bit addressing is
in effect, as few as 16 separate ranges can be assigned to individual
buses only, however a generic change to avoid handing out address 0 only
has turned out controversial as per the discussion referred via the link
below.

Conversely sorting this out in platform code has been standard practice
since forever to avoid a clash with legacy devices subtractively decoded
by the southbridge where present.  This can be revised should a generic
solution be adopted sometime.

Signed-off-by: Maciej W. Rozycki <macro@orcam.me.uk>
Link: https://lore.kernel.org/r/alpine.DEB.2.21.2202260044180.25061@angie.orcam.me.uk
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>

arch/riscv/include/asm/pci.h

@@ -12,8 +12,8 @@
 
 #include <asm/io.h>
 
-#define PCIBIOS_MIN_IO		0
-#define PCIBIOS_MIN_MEM		0
+#define PCIBIOS_MIN_IO		4
+#define PCIBIOS_MIN_MEM		16
 
 /* RISC-V shim does not initialize PCI bus */
 #define pcibios_assign_all_busses() 1