Skip to contentSkip to conversation

Patch to increase the number of tlb entries a guest uses

DISCUSSION Christian Ehrhardt 3 participants 4 messages Oct 20, 2008 – Oct 21, 2008

lore
Share & tools
lore
#0 lore raw

Hi Scott,
I just discussed with Hollis about shadow tlb management in the hypervisor and he mentioned that you already have a patch that lets the guest use more tlb entries than it would have on bare metal hardware and therefore allow to handle more tlb misses in the host because the shadow tlb already holds that information.

It would be nice if you could provide me and the kvmppc community that patch, because I wanted to modify our tlb handling a bit. If it would be easy enough I would like to test that with a different number of guest tlb entries too to see how big the impact of such a change would be to our shadow tlb management.

Signature · 3 lines
Grüsse / regards, 
Christian Ehrhardt
IBM Linux Technology Center, Open Virtualization
#1 lore raw
Quoted 7 lines · Christian Ehrhardt

Hi Scott,
I just discussed with Hollis about shadow tlb management in the hypervisor and he mentioned that you already have a patch that lets the guest use more tlb entries than it would have on bare metal hardware and therefore allow to handle more tlb misses in the host because the shadow tlb already holds that information.

Quoted 5 lines

It would be nice if you could provide me and the kvmppc community that patch, because I wanted to modify our tlb handling a bit. If it would be easy enough I would like to test that with a different number of guest tlb entries too to see how big the impact of such a change would be to our shadow tlb management.

It's not a patch to KVM, it's something I've done on Freescale's hypervisor (which isn't released yet).

-Scott

#2 lore raw
In reply to Christian Ehrhardt
Quoted 8 lines · Scott Wood
Quoted 7 lines · Christian Ehrhardt

Hi Scott,
I just discussed with Hollis about shadow tlb management in the hypervisor and he mentioned that you already have a patch that lets the guest use more tlb entries than it would have on bare metal hardware and therefore allow to handle more tlb misses in the host because the shadow tlb already holds that information.

Quoted 8 lines
Quoted 5 lines

It would be nice if you could provide me and the kvmppc community that patch, because I wanted to modify our tlb handling a bit. If it would be easy enough I would like to test that with a different number of guest tlb entries too to see how big the impact of such a change would be to our shadow tlb management.

It's not a patch to KVM, it's something I've done on Freescale's hypervisor (which isn't released yet).

I think Christian was referring to the Linux side of it. But I guess that since Freescale parts have TLBCFG registers advertising the size of the TLBs, there wouldn't necessarily be any Linux changes at all. (Hmm, that's pretty nice. :)

Signature · 2 lines
Hollis Blanchard
IBM Linux Technology Center
#3 lore raw
In reply to Christian Ehrhardt
Quoted 8 lines · Hollis Blanchard
Quoted 2 lines · Scott Wood

It's not a patch to KVM, it's something I've done on Freescale's hypervisor (which isn't released yet).

I think Christian was referring to the Linux side of it. But I guess that since Freescale parts have TLBCFG registers advertising the size of the TLBs, there wouldn't necessarily be any Linux changes at all. (Hmm, that's pretty nice. :)

Correct, there are no Linux changes for this. Actually, we can't quite advertise the size of the TLB we use in TLB0CFG, since there aren't enough bits -- but Linux doesn't care.

-Scott