On 14/07/14 15:35, Peter Maydell wrote:
> On 14 July 2014 14:33, James Hogan <james.hogan@xxxxxxxxxx> wrote:
>> On 10/07/14 13:17, Peter Maydell wrote:
>>> More generally, there doesn't really seem to be provision in the
>>> KVM KVM_EXIT_MMIO API for returning "this access failed".
>>> I guess in theory userspace could do all the "figure out how
>>> to adjust CPU state to do exception entry and then run VCPU",
>>> but that seems like quite a lot of work which the kernel already
>>> knows how to do; is there some way to provide a simpler API
>>> that lets userspace just inform the kernel that it needs to
>>> fault the access?
>>
>> Indeed. Paolo's idea would work well I think. A data bus error exception
>> would likely be the only sensible error response required other than
>> ignoring writes or returning a garbage value for a read (which the
>> current KVM MMIO API already allows).
>
> I think we should make the API at least permit returning an
> (architecture-specific) error code to the kernel, rather than just
> a single boolean "failed" response. For instance on ARM the
> fault status registers include a single ExT bit for classifying
> the type of an external abort. (The meaning of the bit is
> IMPDEF; on the Cortex-A15 it can be used to distinguish
> AXI bus DECERR ("decode error", ie the interconnect doesn't
> have anything attached at that address) and SLVERR ("slave
> error", ie there was a device at that address but it chose to
> reject the transaction for some reason, eg unsupported
> transfer size, timeout, write to read-only location, FIFO
> overrun or just because the device was in a bad mood.)
Agreed (I wasn't suggesting a bool, just thinking out loud about how
mips would use that arch specific value :-) ).
Cheers
James
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
