Hi Rafał, On Tue, Apr 7, 2015 at 4:51 AM, Rafał Miłecki <zajec5@xxxxxxxxx> wrote: > Hello, > > I own a home router Linksys WRT300N v1.0 with CardBus slot on PCB and > wireless card attached. > > This is MIPS SoC BCM4704 using following code for PCI controller: > 1) register_pci_controller > http://lxr.free-electrons.com/source/arch/mips/pci/pci.c?v=3.18#L167 > 2) ssb_pcicore_init_hostmode > http://lxr.free-electrons.com/source/drivers/ssb/driver_pcicore.c?v=3.18#L317 > > There are two PCI devices discoverable: > 1) 0000:00:00.0 (bridge) > 14e4:472d / class 0x068000 / hdr_type 0 > 2) 0000:00:01.0 (wireless device) > 14e4:4329 / class 0x028000 / hdr_type 0 > > My problem is that PCI subsystem assigns memory to the bridge device: > pci 0000:00:00.0: BAR 1: assigned [mem 0x40000000-0x47ffffff pref] > pci 0000:00:01.0: BAR 0: assigned [mem 0x48000000-0x48003fff] > pci 0000:00:00.0: BAR 0: assigned [mem 0x48004000-0x48005fff] > > This ancient & simple PCI controller allows assigning resources to the > wireless device only (slot 1). Trying to assign resources to the > bridge device (writing to slot 0 to registers PCI_BASE_ADDRESS_[0-5]) > results in overwriting wireless device (slot 0) configuration! > As you can guess from the above log, the last assignment (targeting > slot 0) will break (overwrite) wireless device (slot 1) configuration. > Trying to access any MMIO register of wireless device will cause "Data > bus error". > For more details please see my comment in OpenWrt Trac: > https://dev.openwrt.org/ticket/12682#comment:11 (there is a nice > configuration dump after every assignment). > > So I'm looking for a way to stop PCI subsystem from assigning any > memory to the bridge device (slot 0). I still need it to assign memory > for wireless device (slot 1) as its driver requires MMIO access. > > I'm wondering what is the best way to achieve that? > > MIPS arch code seems to respect PCI_PROBE_ONLY but this will stop > assigning memory to wireless device (slot 1) too. > > I was considering modifying my "struct pci_ops" write callback to > include something like this: > if (extpci_core->cardbusmode && dev == 0 && > off >= PCI_BASE_ADDRESS_0 && > off <= PCI_BASE_ADDRESS_5) > return -ENOTSUPP; > It seems to be working fine, all I get with above change is: > pci 0000:00:01.0: BAR 0: assigned [mem 0x40000000-0x40003fff] > , but I need an opinion if this is a correct solution. I'd have to understand the problem better to give you a good opinion. It sounds like you think there's a hardware defect in the bridge? If that's the case, we can often write a quirk to work around it. Could you open a report at bugzilla.kernel.org against drivers/PCI, and attach a complete dmesg log and output of "lspci -vv"? The newer the kernel you can try, the better. Bjorn
