What is an address translation in powerISA jarogn ?

Benjamin Herrenschmidt benh at kernel.crashing.org
Wed Oct 17 11:14:15 AEDT 2018


On Tue, 2018-10-16 at 20:58 +0300, Raz wrote:
> Section 5.7.3
> "Storage accesses in real, hypervisor real, and virtual real
> addressing modes are performed in a manner that depends on the
> contents of MSR HV , VPM, VRMASD, HRMOR, RMLS, RMOR (see Chapter 2),
> bit 0 of the
> effective address (EA0),"
> 
> Hello
> 1. If MSR_IR = 0 and MSR_DR = 0, does it mean that addresses are not
> translated by the MMU ?

It depends.

If HV=1 (hypervisor mode), then they are only translated to the extent
that HRMOR is applied if the MSB is 0, and untranslated if the MSB is
1.

If HV=0 (guest mode), then they *are* translated but using a different
mechanism than what's normally used when IR/DR=1. This mechanism
depends on whether you are using the Radix or the Hash MMU, and the top
2 bits are ignored.

With hash MMU, it's using things like VRMASD etc... (RMOR is deprecated
afaik) to lookup a "Virtual real mode" area in the hash table. It's
essentially a mapping of the guest "physical" space to real physical
space. It's usually initialized (and maintained) by the HV but the
guest can extend it using things like H_ENTER afaik.

With the radix MMU, it's the guest physical space as mapped by the 2nd
level page tables maintained by the hypervisor.

> 2. If EA0 is the 63-rd bit of the effective address e address ? Does
> this mean that the translation model is
>    derived from the address ? a non privileged context may access
> privileged memory.

Nope.

Cheers,
Ben.




More information about the Linuxppc-dev mailing list