On 04/01/16 08:05, Alexander Graf wrote:
On 02.01.16 21:44, Mark Cave-Ayland wrote:
Currently only PCI memory space is mapped in ob_pci_map_in(). By adding multi-space support to pci_bus_addr_to_host_addr(), it becomes possible to translate physical addresses for both memory and IO space.
Signed-off-by: Mark Cave-Ayland mark.cave-ayland@ilande.co.uk
openbios-devel/drivers/pci.c | 28 +++++++++++++++++++++------- 1 file changed, 21 insertions(+), 7 deletions(-)
diff --git a/openbios-devel/drivers/pci.c b/openbios-devel/drivers/pci.c index 6347118..97d3ef3 100644 --- a/openbios-devel/drivers/pci.c +++ b/openbios-devel/drivers/pci.c @@ -144,9 +144,13 @@ static void dump_reg_property(const char* description, int nreg, u32 *reg) } #endif
-static unsigned long pci_bus_addr_to_host_addr(uint32_t ba) +static unsigned long pci_bus_addr_to_host_addr(int space, uint32_t ba) {
- return arch->host_pci_base + (unsigned long)ba;
- if (space == IO_SPACE) {
return arch->io_base + (unsigned long)ba;- } else {
return arch->host_pci_base + (unsigned long)ba;- }
}
static void @@ -347,16 +351,20 @@ ob_pci_map_in(int *idx) { phys_addr_t phys; uint32_t ba;
- ucell size, virt;
ucell size, virt, tmp;
int space;
PCI_DPRINTF("ob_pci_bar_map_in idx=%p\n", idx);
size = POP();
- POP();
- tmp = POP(); POP(); ba = POP();
- phys = pci_bus_addr_to_host_addr(ba);
- /* Get the space from the pci-addr.hi */
- space = ((tmp & 0x03000000) >> 24);
You could try to reuse the defines we have in QEMU for the parameter bits:
#define FDT_PCI_RANGE_RELOCATABLE 0x80000000 #define FDT_PCI_RANGE_PREFETCHABLE 0x40000000 #define FDT_PCI_RANGE_ALIASED 0x20000000 #define FDT_PCI_RANGE_TYPE_MASK 0x03000000 #define FDT_PCI_RANGE_MMIO_64BIT 0x03000000 #define FDT_PCI_RANGE_MMIO 0x02000000 #define FDT_PCI_RANGE_IOPORT 0x01000000 #define FDT_PCI_RANGE_CONFIG 0x00000000
Yeah that seems a good idea - I'll add them into drivers/pci.h without the FDT_ prefix in the next respin.
While at it, you probably also want to double-check you're not mapping MMIO_64BIT or CONFIG by accident ;).
The default behaviour was just to return the address unaltered, so I'll add that in too. Not that it will help anything that tries to access an incorrect address, but at the very least it will appear unaltered in any properties which makes it quite easy to spot.
I also have another patchset which tidies up the OpenPIC mappings which I'll include in the next revision too, hopefully with the aim of committing tomorrow to help clear my patch backlog(!)
ATB,
Mark.