Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On 27.11.2012, at 00:16, Paul Mackerras wrote: > On Mon, Nov 26, 2012 at 11:03:19PM +0100, Alexander Graf wrote: >> >> On 26.11.2012, at 22:48, Paul Mackerras wrote: >> >>> On Mon, Nov 26, 2012 at 02:10:33PM +0100, Alexander Graf wrote: On 23.11.2012, at 23:07, Paul Mackerras wrote: > On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: >> >> On 22.11.2012, at 10:28, Paul Mackerras wrote: >> >>> - With the possibility of the host paging out guest pages, the use of >>> H_LOCAL by an SMP guest is dangerous since the guest could possibly >>> retain and use a stale TLB entry pointing to a page that had been >>> removed from the guest. >> >> I don't understand this part. Don't we flush the TLB when the page gets >> evicted from the shadow HTAB? > > The H_LOCAL flag is something that we invented to allow the guest to > tell the host "I only ever used this translation (HPTE) on the current > vcpu" when it's removing or modifying an HPTE. The idea is that that > would then let the host use the tlbiel instruction (local TLB > invalidate) rather than the usual global tlbie instruction. Tlbiel is > faster because it doesn't need to go out on the fabric and get > processed by all cpus. In fact our guests don't use it at present, > but we put it in because we thought we should be able to get a > performance improvement, particularly on large machines. > > However, the catch is that the guest's setting of H_LOCAL might be > incorrect, in which case we could have a stale TLB entry on another > physical cpu. While the physical page that it refers to is still > owned by the guest, that stale entry doesn't matter from the host's > point of view. But if the host wants to take that page away from the > guest, the stale entry becomes a problem. That's exactly where my question lies. Does that mean we don't flush the TLB entry regardless when we take the page away from the guest? >>> >>> The question is how to find the TLB entry if the HPTE it came from is >>> no longer present. Flushing a TLB entry requires a virtual address. >>> When we're taking a page away from the guest we have the real address >>> of the page, not the virtual address. We can use the reverse-mapping >>> chains to loop through all the HPTEs that map the page, and from each >>> HPTE we can (and do) calculate a virtual address and do a TLBIE on >>> that virtual address (each HPTE could be at a different virtual >>> address). >>> >>> The difficulty comes when we no longer have the HPTE but we >>> potentially have a stale TLB entry, due to having used tlbiel when we >>> removed the HPTE. Without the HPTE the only way to get rid of the >>> stale TLB entry would be to completely flush all the TLB entries for >>> the guest's LPID on every physical CPU it had ever run on. Since I >>> don't want to go to that much effort, what I am proposing, and what >>> this patch implements, is to not ever use tlbiel when removing HPTEs >>> in SMP guests on POWER7. >>> >>> In other words, what this patch is about is making sure we don't get >>> these troublesome stale TLB entries. >> >> I see. You could keep a list of to-be-flushed VAs around that you could skim >> through when taking a page away from the guest. That way you make the fast >> case fast (add/remove of page from the guest) and the slow path slow >> (paging). > > Yes, I thought about that, but the problem is that the list of VAs > could get arbitrarily long and take up a lot of host memory. You can always cap it at an arbitrary number, similar to how the TLB itself is limited too. Alex -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On Mon, Nov 26, 2012 at 11:03:19PM +0100, Alexander Graf wrote: > > On 26.11.2012, at 22:48, Paul Mackerras wrote: > > > On Mon, Nov 26, 2012 at 02:10:33PM +0100, Alexander Graf wrote: > >> > >> On 23.11.2012, at 23:07, Paul Mackerras wrote: > >> > >>> On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: > > On 22.11.2012, at 10:28, Paul Mackerras wrote: > > > - With the possibility of the host paging out guest pages, the use of > > H_LOCAL by an SMP guest is dangerous since the guest could possibly > > retain and use a stale TLB entry pointing to a page that had been > > removed from the guest. > > I don't understand this part. Don't we flush the TLB when the page gets > evicted from the shadow HTAB? > >>> > >>> The H_LOCAL flag is something that we invented to allow the guest to > >>> tell the host "I only ever used this translation (HPTE) on the current > >>> vcpu" when it's removing or modifying an HPTE. The idea is that that > >>> would then let the host use the tlbiel instruction (local TLB > >>> invalidate) rather than the usual global tlbie instruction. Tlbiel is > >>> faster because it doesn't need to go out on the fabric and get > >>> processed by all cpus. In fact our guests don't use it at present, > >>> but we put it in because we thought we should be able to get a > >>> performance improvement, particularly on large machines. > >>> > >>> However, the catch is that the guest's setting of H_LOCAL might be > >>> incorrect, in which case we could have a stale TLB entry on another > >>> physical cpu. While the physical page that it refers to is still > >>> owned by the guest, that stale entry doesn't matter from the host's > >>> point of view. But if the host wants to take that page away from the > >>> guest, the stale entry becomes a problem. > >> > >> That's exactly where my question lies. Does that mean we don't flush the > >> TLB entry regardless when we take the page away from the guest? > > > > The question is how to find the TLB entry if the HPTE it came from is > > no longer present. Flushing a TLB entry requires a virtual address. > > When we're taking a page away from the guest we have the real address > > of the page, not the virtual address. We can use the reverse-mapping > > chains to loop through all the HPTEs that map the page, and from each > > HPTE we can (and do) calculate a virtual address and do a TLBIE on > > that virtual address (each HPTE could be at a different virtual > > address). > > > > The difficulty comes when we no longer have the HPTE but we > > potentially have a stale TLB entry, due to having used tlbiel when we > > removed the HPTE. Without the HPTE the only way to get rid of the > > stale TLB entry would be to completely flush all the TLB entries for > > the guest's LPID on every physical CPU it had ever run on. Since I > > don't want to go to that much effort, what I am proposing, and what > > this patch implements, is to not ever use tlbiel when removing HPTEs > > in SMP guests on POWER7. > > > > In other words, what this patch is about is making sure we don't get > > these troublesome stale TLB entries. > > I see. You could keep a list of to-be-flushed VAs around that you could skim > through when taking a page away from the guest. That way you make the fast > case fast (add/remove of page from the guest) and the slow path slow (paging). Yes, I thought about that, but the problem is that the list of VAs could get arbitrarily long and take up a lot of host memory. > But I'm fine with disallowing local flushes on remove completely for now. It > would be nice to get performance data on how much this would be a net win > though. There are certainly ways of keeping local flushes alive with the > scheme above. Yes, I definitely want to get some good performance data to see how much of a win it would be, and if there is a good win, work out some scheme to let us use the local flushes. > Thanks, applied to kvm-ppc-next. Thanks, Paul. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On 26.11.2012, at 22:48, Paul Mackerras wrote: > On Mon, Nov 26, 2012 at 02:10:33PM +0100, Alexander Graf wrote: >> >> On 23.11.2012, at 23:07, Paul Mackerras wrote: >> >>> On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: On 22.11.2012, at 10:28, Paul Mackerras wrote: > - With the possibility of the host paging out guest pages, the use of > H_LOCAL by an SMP guest is dangerous since the guest could possibly > retain and use a stale TLB entry pointing to a page that had been > removed from the guest. I don't understand this part. Don't we flush the TLB when the page gets evicted from the shadow HTAB? >>> >>> The H_LOCAL flag is something that we invented to allow the guest to >>> tell the host "I only ever used this translation (HPTE) on the current >>> vcpu" when it's removing or modifying an HPTE. The idea is that that >>> would then let the host use the tlbiel instruction (local TLB >>> invalidate) rather than the usual global tlbie instruction. Tlbiel is >>> faster because it doesn't need to go out on the fabric and get >>> processed by all cpus. In fact our guests don't use it at present, >>> but we put it in because we thought we should be able to get a >>> performance improvement, particularly on large machines. >>> >>> However, the catch is that the guest's setting of H_LOCAL might be >>> incorrect, in which case we could have a stale TLB entry on another >>> physical cpu. While the physical page that it refers to is still >>> owned by the guest, that stale entry doesn't matter from the host's >>> point of view. But if the host wants to take that page away from the >>> guest, the stale entry becomes a problem. >> >> That's exactly where my question lies. Does that mean we don't flush the TLB >> entry regardless when we take the page away from the guest? > > The question is how to find the TLB entry if the HPTE it came from is > no longer present. Flushing a TLB entry requires a virtual address. > When we're taking a page away from the guest we have the real address > of the page, not the virtual address. We can use the reverse-mapping > chains to loop through all the HPTEs that map the page, and from each > HPTE we can (and do) calculate a virtual address and do a TLBIE on > that virtual address (each HPTE could be at a different virtual > address). > > The difficulty comes when we no longer have the HPTE but we > potentially have a stale TLB entry, due to having used tlbiel when we > removed the HPTE. Without the HPTE the only way to get rid of the > stale TLB entry would be to completely flush all the TLB entries for > the guest's LPID on every physical CPU it had ever run on. Since I > don't want to go to that much effort, what I am proposing, and what > this patch implements, is to not ever use tlbiel when removing HPTEs > in SMP guests on POWER7. > > In other words, what this patch is about is making sure we don't get > these troublesome stale TLB entries. I see. You could keep a list of to-be-flushed VAs around that you could skim through when taking a page away from the guest. That way you make the fast case fast (add/remove of page from the guest) and the slow path slow (paging). But I'm fine with disallowing local flushes on remove completely for now. It would be nice to get performance data on how much this would be a net win though. There are certainly ways of keeping local flushes alive with the scheme above. Thanks, applied to kvm-ppc-next. Alex -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On Mon, Nov 26, 2012 at 02:10:33PM +0100, Alexander Graf wrote: > > On 23.11.2012, at 23:07, Paul Mackerras wrote: > > > On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: > >> > >> On 22.11.2012, at 10:28, Paul Mackerras wrote: > >> > >>> - With the possibility of the host paging out guest pages, the use of > >>> H_LOCAL by an SMP guest is dangerous since the guest could possibly > >>> retain and use a stale TLB entry pointing to a page that had been > >>> removed from the guest. > >> > >> I don't understand this part. Don't we flush the TLB when the page gets > >> evicted from the shadow HTAB? > > > > The H_LOCAL flag is something that we invented to allow the guest to > > tell the host "I only ever used this translation (HPTE) on the current > > vcpu" when it's removing or modifying an HPTE. The idea is that that > > would then let the host use the tlbiel instruction (local TLB > > invalidate) rather than the usual global tlbie instruction. Tlbiel is > > faster because it doesn't need to go out on the fabric and get > > processed by all cpus. In fact our guests don't use it at present, > > but we put it in because we thought we should be able to get a > > performance improvement, particularly on large machines. > > > > However, the catch is that the guest's setting of H_LOCAL might be > > incorrect, in which case we could have a stale TLB entry on another > > physical cpu. While the physical page that it refers to is still > > owned by the guest, that stale entry doesn't matter from the host's > > point of view. But if the host wants to take that page away from the > > guest, the stale entry becomes a problem. > > That's exactly where my question lies. Does that mean we don't flush the TLB > entry regardless when we take the page away from the guest? The question is how to find the TLB entry if the HPTE it came from is no longer present. Flushing a TLB entry requires a virtual address. When we're taking a page away from the guest we have the real address of the page, not the virtual address. We can use the reverse-mapping chains to loop through all the HPTEs that map the page, and from each HPTE we can (and do) calculate a virtual address and do a TLBIE on that virtual address (each HPTE could be at a different virtual address). The difficulty comes when we no longer have the HPTE but we potentially have a stale TLB entry, due to having used tlbiel when we removed the HPTE. Without the HPTE the only way to get rid of the stale TLB entry would be to completely flush all the TLB entries for the guest's LPID on every physical CPU it had ever run on. Since I don't want to go to that much effort, what I am proposing, and what this patch implements, is to not ever use tlbiel when removing HPTEs in SMP guests on POWER7. In other words, what this patch is about is making sure we don't get these troublesome stale TLB entries. Paul. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On 23.11.2012, at 23:07, Paul Mackerras wrote: > On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: >> >> On 22.11.2012, at 10:28, Paul Mackerras wrote: >> >>> - With the possibility of the host paging out guest pages, the use of >>> H_LOCAL by an SMP guest is dangerous since the guest could possibly >>> retain and use a stale TLB entry pointing to a page that had been >>> removed from the guest. >> >> I don't understand this part. Don't we flush the TLB when the page gets >> evicted from the shadow HTAB? > > The H_LOCAL flag is something that we invented to allow the guest to > tell the host "I only ever used this translation (HPTE) on the current > vcpu" when it's removing or modifying an HPTE. The idea is that that > would then let the host use the tlbiel instruction (local TLB > invalidate) rather than the usual global tlbie instruction. Tlbiel is > faster because it doesn't need to go out on the fabric and get > processed by all cpus. In fact our guests don't use it at present, > but we put it in because we thought we should be able to get a > performance improvement, particularly on large machines. > > However, the catch is that the guest's setting of H_LOCAL might be > incorrect, in which case we could have a stale TLB entry on another > physical cpu. While the physical page that it refers to is still > owned by the guest, that stale entry doesn't matter from the host's > point of view. But if the host wants to take that page away from the > guest, the stale entry becomes a problem. That's exactly where my question lies. Does that mean we don't flush the TLB entry regardless when we take the page away from the guest? Alex > > The solution I implement here is just not to use tlbiel in SMP guests. > UP guests are not so much of a problem because the potential attack > from the guest relies on having one cpu remove the HPTE and do tlbiel > while another cpu uses the stale TLB entry, which you can't do if you > only have one cpu. > > Paul. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On Fri, Nov 23, 2012 at 04:43:03PM +0100, Alexander Graf wrote: > > On 22.11.2012, at 10:28, Paul Mackerras wrote: > > > - With the possibility of the host paging out guest pages, the use of > > H_LOCAL by an SMP guest is dangerous since the guest could possibly > > retain and use a stale TLB entry pointing to a page that had been > > removed from the guest. > > I don't understand this part. Don't we flush the TLB when the page gets > evicted from the shadow HTAB? The H_LOCAL flag is something that we invented to allow the guest to tell the host "I only ever used this translation (HPTE) on the current vcpu" when it's removing or modifying an HPTE. The idea is that that would then let the host use the tlbiel instruction (local TLB invalidate) rather than the usual global tlbie instruction. Tlbiel is faster because it doesn't need to go out on the fabric and get processed by all cpus. In fact our guests don't use it at present, but we put it in because we thought we should be able to get a performance improvement, particularly on large machines. However, the catch is that the guest's setting of H_LOCAL might be incorrect, in which case we could have a stale TLB entry on another physical cpu. While the physical page that it refers to is still owned by the guest, that stale entry doesn't matter from the host's point of view. But if the host wants to take that page away from the guest, the stale entry becomes a problem. The solution I implement here is just not to use tlbiel in SMP guests. UP guests are not so much of a problem because the potential attack from the guest relies on having one cpu remove the HPTE and do tlbiel while another cpu uses the stale TLB entry, which you can't do if you only have one cpu. Paul. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/5] KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
On 22.11.2012, at 10:28, Paul Mackerras wrote:
> When we change or remove a HPT (hashed page table) entry, we can do
> either a global TLB invalidation (tlbie) that works across the whole
> machine, or a local invalidation (tlbiel) that only affects this core.
> Currently we do local invalidations if the VM has only one vcpu or if
> the guest requests it with the H_LOCAL flag, though the guest Linux
> kernel currently doesn't ever use H_LOCAL. Then, to cope with the
> possibility that vcpus moving around to different physical cores might
> expose stale TLB entries, there is some code in kvmppc_hv_entry to
> flush the whole TLB of entries for this VM if either this vcpu is now
> running on a different physical core from where it last ran, or if this
> physical core last ran a different vcpu.
>
> There are a number of problems on POWER7 with this as it stands:
>
> - The TLB invalidation is done per thread, whereas it only needs to be
> done per core, since the TLB is shared between the threads.
> - With the possibility of the host paging out guest pages, the use of
> H_LOCAL by an SMP guest is dangerous since the guest could possibly
> retain and use a stale TLB entry pointing to a page that had been
> removed from the guest.
I don't understand this part. Don't we flush the TLB when the page gets evicted
from the shadow HTAB?
Alex
> - The TLB invalidations that we do when a vcpu moves from one physical
> core to another are unnecessary in the case of an SMP guest that isn't
> using H_LOCAL.
> - The optimization of using local invalidations rather than global should
> apply to guests with one virtual core, not just one vcpu.
>
> (None of this applies on PPC970, since there we always have to
> invalidate the whole TLB when entering and leaving the guest, and we
> can't support paging out guest memory.)
>
> To fix these problems and simplify the code, we now maintain a simple
> cpumask of which cpus need to flush the TLB on entry to the guest.
> (This is indexed by cpu, though we only ever use the bits for thread
> 0 of each core.) Whenever we do a local TLB invalidation, we set the
> bits for every cpu except the bit for thread 0 of the core that we're
> currently running on. Whenever we enter a guest, we test and clear the
> bit for our core, and flush the TLB if it was set.
>
> On initial startup of the VM, and when resetting the HPT, we set all the
> bits in the need_tlb_flush cpumask, since any core could potentially have
> stale TLB entries from the previous VM to use the same LPID, or the
> previous contents of the HPT.
>
> Then, we maintain a count of the number of online virtual cores, and use
> that when deciding whether to use a local invalidation rather than the
> number of online vcpus. The code to make that decision is extracted out
> into a new function, global_invalidates(). For multi-core guests on
> POWER7 (i.e. when we are using mmu notifiers), we now never do local
> invalidations regardless of the H_LOCAL flag.
>
> Signed-off-by: Paul Mackerras
> ---
> arch/powerpc/include/asm/kvm_host.h |5 +--
> arch/powerpc/kernel/asm-offsets.c |4 +--
> arch/powerpc/kvm/book3s_64_mmu_hv.c |7 ++--
> arch/powerpc/kvm/book3s_hv.c|9 -
> arch/powerpc/kvm/book3s_hv_rm_mmu.c | 37 +---
> arch/powerpc/kvm/book3s_hv_rmhandlers.S | 56 ++-
> 6 files changed, 73 insertions(+), 45 deletions(-)
>
> diff --git a/arch/powerpc/include/asm/kvm_host.h
> b/arch/powerpc/include/asm/kvm_host.h
> index 58c7264..62fbd38 100644
> --- a/arch/powerpc/include/asm/kvm_host.h
> +++ b/arch/powerpc/include/asm/kvm_host.h
> @@ -246,11 +246,12 @@ struct kvm_arch {
> int using_mmu_notifiers;
> u32 hpt_order;
> atomic_t vcpus_running;
> + u32 online_vcores;
> unsigned long hpt_npte;
> unsigned long hpt_mask;
> atomic_t hpte_mod_interest;
> spinlock_t slot_phys_lock;
> - unsigned short last_vcpu[NR_CPUS];
> + cpumask_t need_tlb_flush;
> struct kvmppc_vcore *vcores[KVM_MAX_VCORES];
> struct kvmppc_linear_info *hpt_li;
> #endif /* CONFIG_KVM_BOOK3S_64_HV */
> @@ -275,6 +276,7 @@ struct kvmppc_vcore {
> int nap_count;
> int napping_threads;
> u16 pcpu;
> + u16 last_cpu;
> u8 vcore_state;
> u8 in_guest;
> struct list_head runnable_threads;
> @@ -523,7 +525,6 @@ struct kvm_vcpu_arch {
> u64 dec_jiffies;
> u64 dec_expires;
> unsigned long pending_exceptions;
> - u16 last_cpu;
> u8 ceded;
> u8 prodded;
> u32 last_inst;
> diff --git a/arch/powerpc/kernel/asm-offsets.c
> b/arch/powerpc/kernel/asm-offsets.c
> index 7523539..4e23ba2 100644
> --- a/arch/powerpc/kernel/asm-offsets.c
> +++ b/arch/powerpc/kernel/asm-offsets.c
> @@ -441,8 +441,7 @@ int main(void)
> DEFINE(KVM_HOST_LPCR, offsetof(struct kvm, arch.host_lpcr));
> DEFINE(KVM_HOST_SDR1, offsetof(struct kvm, arch.host
