[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [BUG] x86/hvm/vioapic: 8-Bit IOREGSEL write does not work
>>> On 06.09.17 at 11:58, <andrew.cooper3@xxxxxxxxxx> wrote: > On 05/09/17 19:26, Christian Prochaska wrote: >> I've seen this problem with Xen 4.6.5 from the Xubuntu 16.04 >> distribution and from a quick look over the current vioapic code it >> seems to be still present. >> >> From the IOAPIC datasheet [1]: "To reference an IOAPIC register, a byte >> memory write that the PIIX3 decodes for the IOAPIC loads the IOREGSEL >> Register with an 8-bit value that specifies the IOAPIC register (address >> offset in Table 3.2) to be accessed." >> >> But the 'vioapic_write()' function does not consider the 'length' argument >> and always copies 4 Bytes from the unsigned long 'val' argument into the >> virtual 32-Bit IOREGSEL register. In the error case I've seen, 'length' >> was 1 and 'val' was 0xffff8300bb0cf801 and the IOAPIC version register >> with address offset 0x01 was not read correctly. > > That looks suspiciously like a Xen pointer, not a plausible val. Considering length is 1, I'd assume this is a Xen pointer with the low 8 bits overwritten with the actual value to be written out. > Irrespective, it is an error for the guest to issue anything other than > a 4 byte write into the IOREGSEL field, so we ought to be rejecting > length-1 accesses. With what the referred to data sheet text says in mind I'm not sure we shouldn't rather support 1-byte writes here. It specifically also says "The IOWIN Register must be accessed as a Dword quantity", so the same relaxation should not be done there. Even checking much newer data sheets (ICH10, C600) they still describe the register as an 8-bit one. Jan _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx https://lists.xen.org/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |