El 14/06/15 a les 14:05, Antti Kantee ha escrit:
On 14/06/15 11:54, Robert Millan wrote:
Thanks Antti. Here's the new patchset.
Almost there. Still small nits:
#ifdef RUMP_PCI_IOSPACE
- pba.pba_flags |= PCI_FLAGS_IO_OKAY;
+#ifdef rumpcomp_pci_initiopl
+ error = rumpcomp_pci_initiopl();
+ if (error == 0) {
+ pba.pba_flags |= PCI_FLAGS_IO_OKAY;
+ } else {
+ aprint_error("pci: unable to raise I/O privilege level (error %d), direct
I/O will be unavailable\n", error);
+ }
+#endif
#endif
The print still doesn't match the call. How do you know at that abstraction level that
initiopl() will exactly "raise I/O privilege level"?
Well that's what "iopl" stands for! Perhaps we should pick another name? :-)
Btw, iopl() is a Linux-ism. An x86-ism even, IIRC. For example the FreeBSD/x86 way is
open("/dev/io"). Maybe use something more generic like:
rumpcomp_pci_initio()
?
Also, do we still need RUMP_PCI_IOSPACE? Or can we say that everything
providing iospace access must now define rumpcomp_pci_initiopl?
A rumpcomp_pci_initiopl() dummy stub is cheap, seems reasonable to assume it's
present.
--
--
Robert Millan
