[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] page fault during ins
From: |
Mark Williamson |
Subject: |
Re: [Qemu-devel] page fault during ins |
Date: |
Fri, 17 Dec 2004 00:07:15 +0000 |
User-agent: |
KMail/1.7.1 |
> Do we have any device where iomem read changes device state?
It seems like a plausible thing to do, despite seemingly slightly strange
semantically.
> Wouldn't such hardware cause problems with speculative execution
> (memory reads) on some real CPUs? Or am I missing something?
IO memory regions would have a flag(s) set in the relevant page tables telling
the CPU that it's not real memory. This flag would be set by the operating
system.
This allows the CPU to maintain correct semantics for such devices (not
reordering writes, not caching reads, etc), whilst still being able to
optimise accesses to physical RAM aggressively.
Cheers,
Mark