On Sat, Apr 6, 2013 at 2:49 AM, James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote: > On Thu, 2013-04-04 at 12:06 -0600, Bjorn Helgaas wrote: >> On Thu, Mar 21, 2013 at 6:51 PM, Robert Hancock <hancockrwd@xxxxxxxxx> wrote: >> > -Reconsider whether supporting read/write on the resource files for IO port >> > regions like these makes any sense. Obviously mmap isn't very practical for >> > IO port access on x86 but you could even do something like an ioctl for this >> > purpose. Not very many pieces of software would need to access these files >> > so it's likely OK if the API is a bit ugly. That would prevent something >> > like grepping through sysfs from generating port accesses to random devices. >> >> This is the approach I'd like to push on for a kernel fix. > > Me too. I think the quirks approach is unsupportable. Worst case I > think we should have an ability for the *driver* to mark the region as > having strange access rules. > >> I'm not a >> VM person, but if it were possible to support .mmap() in such a way >> that a handler would be called for every access to the region, we >> could easily support I/O port access that way. > > We could ... the OS could trigger a page fault on every access using the > COW mechanism ... however, that mechanism wasn't intended for write > combining, so although it's theoretically possible to add it, I'd be a > bit wary. The current problem case is only I/O port accesses, so we only have to support 1/2/4-byte accesses, and I don't think write combining is an issue. But it still sounds like COW might be trying to force a square peg into a round hole. Maybe there are other approaches. I just thought of .mmap() because that's already used for MEM space accesses. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
