Skip to content

Commit

Permalink
KVM: PPC: Book3S: Invalidate multiple TCEs at once
Browse files Browse the repository at this point in the history 
Invalidating a TCE cache entry for each updated TCE is quite expensive.
This makes use of the new iommu_table_ops::xchg_no_kill()/tce_kill()
callbacks to bring down the time spent in mapping a huge guest DMA window;
roughly 20s to 10s for each guest's 100GB of DMA space.

Signed-off-by: Alexey Kardashevskiy <[email protected]>
Acked-by: Paul Mackerras <[email protected]>
Signed-off-by: Michael Ellerman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
  • Loading branch information
@aik @mpe
aik authored and mpe committed Aug 29, 2019
1 parent 3587248 commit 01b7d12
Show file tree
Hide file tree
Showing 2 changed files with 48 additions and 19 deletions.
29 changes: 20 additions & 9 deletions arch/powerpc/kvm/book3s_64_vio.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -416,7 +416,7 @@ static void kvmppc_clear_tce(struct mm_struct *mm, struct iommu_table *tbl,
unsigned long hpa = 0;
enum dma_data_direction dir = DMA_NONE;

iommu_tce_xchg(mm, tbl, entry, &hpa, &dir);
iommu_tce_xchg_no_kill(mm, tbl, entry, &hpa, &dir);
}

static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
Expand Down Expand Up @@ -447,15 +447,16 @@ static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
unsigned long hpa = 0;
long ret;

if (WARN_ON_ONCE(iommu_tce_xchg(kvm->mm, tbl, entry, &hpa, &dir)))
if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
&dir)))
return H_TOO_HARD;

if (dir == DMA_NONE)
return H_SUCCESS;

ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
if (ret != H_SUCCESS)
iommu_tce_xchg(kvm->mm, tbl, entry, &hpa, &dir);
iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);

return ret;
}
Expand Down Expand Up @@ -501,7 +502,7 @@ long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
if (mm_iommu_mapped_inc(mem))
return H_TOO_HARD;

ret = iommu_tce_xchg(kvm->mm, tbl, entry, &hpa, &dir);
ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
if (WARN_ON_ONCE(ret)) {
mm_iommu_mapped_dec(mem);
return H_TOO_HARD;
Expand Down Expand Up @@ -579,6 +580,8 @@ long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
entry, ua, dir);

iommu_tce_kill(stit->tbl, entry, 1);

if (ret != H_SUCCESS) {
kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
goto unlock_exit;
Expand Down Expand Up @@ -656,13 +659,13 @@ long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
*/
if (get_user(tce, tces + i)) {
ret = H_TOO_HARD;
goto unlock_exit;
goto invalidate_exit;
}
tce = be64_to_cpu(tce);

if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
ret = H_PARAMETER;
goto unlock_exit;
goto invalidate_exit;
}

list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
Expand All @@ -673,13 +676,17 @@ long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
if (ret != H_SUCCESS) {
kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl,
entry);
goto unlock_exit;
goto invalidate_exit;
}
}

kvmppc_tce_put(stt, entry + i, tce);
}

invalidate_exit:
list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
iommu_tce_kill(stit->tbl, entry, npages);

unlock_exit:
srcu_read_unlock(&vcpu->kvm->srcu, idx);

Expand Down Expand Up @@ -718,7 +725,7 @@ long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
continue;

if (ret == H_TOO_HARD)
return ret;
goto invalidate_exit;

WARN_ON_ONCE(1);
kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
Expand All @@ -728,6 +735,10 @@ long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);

return H_SUCCESS;
invalidate_exit:
list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
iommu_tce_kill(stit->tbl, ioba >> stt->page_shift, npages);

return ret;
}
EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
38 changes: 28 additions & 10 deletions arch/powerpc/kvm/book3s_64_vio_hv.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -218,13 +218,14 @@ static long kvmppc_rm_ioba_validate(struct kvmppc_spapr_tce_table *stt,
return H_SUCCESS;
}

static long iommu_tce_xchg_rm(struct mm_struct *mm, struct iommu_table *tbl,
static long iommu_tce_xchg_no_kill_rm(struct mm_struct *mm,
struct iommu_table *tbl,
unsigned long entry, unsigned long *hpa,
enum dma_data_direction *direction)
{
long ret;

ret = tbl->it_ops->exchange_rm(tbl, entry, hpa, direction);
ret = tbl->it_ops->xchg_no_kill(tbl, entry, hpa, direction, true);

if (!ret && ((*direction == DMA_FROM_DEVICE) ||
(*direction == DMA_BIDIRECTIONAL))) {
Expand All @@ -240,13 +241,20 @@ static long iommu_tce_xchg_rm(struct mm_struct *mm, struct iommu_table *tbl,
return ret;
}

extern void iommu_tce_kill_rm(struct iommu_table *tbl,
unsigned long entry, unsigned long pages)
{
if (tbl->it_ops->tce_kill)
tbl->it_ops->tce_kill(tbl, entry, pages, true);
}

static void kvmppc_rm_clear_tce(struct kvm *kvm, struct iommu_table *tbl,
unsigned long entry)
{
unsigned long hpa = 0;
enum dma_data_direction dir = DMA_NONE;

iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);
}

static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
Expand Down Expand Up @@ -278,7 +286,7 @@ static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm,
unsigned long hpa = 0;
long ret;

if (iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir))
if (iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir))
/*
* real mode xchg can fail if struct page crosses
* a page boundary
Expand All @@ -290,7 +298,7 @@ static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm,

ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
if (ret)
iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);

return ret;
}
Expand Down Expand Up @@ -336,7 +344,7 @@ static long kvmppc_rm_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
return H_TOO_HARD;

ret = iommu_tce_xchg_rm(kvm->mm, tbl, entry, &hpa, &dir);
ret = iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);
if (ret) {
mm_iommu_mapped_dec(mem);
/*
Expand Down Expand Up @@ -417,6 +425,8 @@ long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
stit->tbl, entry, ua, dir);

iommu_tce_kill_rm(stit->tbl, entry, 1);

if (ret != H_SUCCESS) {
kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
return ret;
Expand Down Expand Up @@ -558,7 +568,7 @@ long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
ua = 0;
if (kvmppc_rm_tce_to_ua(vcpu->kvm, tce, &ua, NULL)) {
ret = H_PARAMETER;
goto unlock_exit;
goto invalidate_exit;
}

list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
Expand All @@ -569,13 +579,17 @@ long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
if (ret != H_SUCCESS) {
kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl,
entry);
goto unlock_exit;
goto invalidate_exit;
}
}

kvmppc_rm_tce_put(stt, entry + i, tce);
}

invalidate_exit:
list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
iommu_tce_kill_rm(stit->tbl, entry, npages);

unlock_exit:
if (rmap)
unlock_rmap(rmap);
Expand Down Expand Up @@ -618,7 +632,7 @@ long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
continue;

if (ret == H_TOO_HARD)
return ret;
goto invalidate_exit;

WARN_ON_ONCE_RM(1);
kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
Expand All @@ -628,7 +642,11 @@ long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
kvmppc_rm_tce_put(stt, ioba >> stt->page_shift, tce_value);

return H_SUCCESS;
invalidate_exit:
list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
iommu_tce_kill_rm(stit->tbl, ioba >> stt->page_shift, npages);

return ret;
}

/* This can be called in either virtual mode or real mode */
Expand Down

0 comments on commit 01b7d12

Please sign in to comment.