[PATCH kernel v3 0/5] powerpc/powernv/kvm: Invalidate multiple TCEs at once

[PATCH kernel v3 0/5] powerpc/powernv/kvm: Invalidate multiple TCEs at once

[Alexey Kardashevskiy]

So far TCE cache updates (IOMMU translation cache on POWER8/9
PHB/NPU units) were barely noticeable; however with 100+GB guests
we now see RCU stall warnings in guests because we spend too much
time in the host system firmware which does actual TCE cache
updates, hence this patchset.

This is a rework of
https://patchwork.ozlabs.org/patch/1149003/
https://patchwork.ozlabs.org/patch/1152985/ (cannot post link to the series
as it appears empty because of broken patchworks)

This depends on KVM-PPC's bugfix: https://patchwork.ozlabs.org/patch/1152937/

I expect 1/5 to go via the PPC tree, 2/5 via the KVM-PPC tree,
3/5 via the VFIO tree and the rest via the PPC tree.

Changes:
v3:
* added 4/5 to fix compile error from 5/5
* added "Book3S" to 2/5's subject line


This is based on sha1
42ac26d253eb Santosh Sivaraj "powerpc: add machine check safe copy_to_user".

Please comment. Thanks.



Alexey Kardashevskiy (5):
  powerpc/powernv/ioda: Split out TCE invalidation from TCE updates
  KVM: PPC: Book3S: Invalidate multiple TCEs at once
  vfio/spapr_tce: Invalidate multiple TCEs at once
  powerpc/pseries/iommu: Switch to xchg_no_kill
  powerpc/powernv/ioda: Remove obsolete iommu_table_ops::exchange
    callbacks

 arch/powerpc/include/asm/iommu.h          | 21 ++++++---
 arch/powerpc/kernel/iommu.c               | 23 ++++++----
 arch/powerpc/kvm/book3s_64_vio.c          | 29 ++++++++----
 arch/powerpc/kvm/book3s_64_vio_hv.c       | 38 +++++++++++----
 arch/powerpc/platforms/powernv/pci-ioda.c | 56 ++++-------------------
 arch/powerpc/platforms/pseries/iommu.c    |  5 +-
 drivers/vfio/vfio_iommu_spapr_tce.c       | 18 +++++---
 7 files changed, 99 insertions(+), 91 deletions(-)

-- 
2.17.1

[PATCH kernel v3 1/5] powerpc/powernv/ioda: Split out TCE invalidation from TCE updates

[Alexey Kardashevskiy]

At the moment updates in a TCE table are made by iommu_table_ops::exchange
which update one TCE and invalidates an entry in the PHB/NPU TCE cache
via set of registers called "TCE Kill" (hence the naming).
Writing a TCE is a simple xchg() but invalidating the TCE cache is
a relatively expensive OPAL call. Mapping a 100GB guest with PCI+NPU
passed through devices takes about 20s.

Thankfully we can do better. Since such big mappings happen at the boot
time and when memory is plugged/onlined (i.e. not often), these requests
come in 512 pages so we call call OPAL 512 times less which brings 20s
from the above to less than 10s. Also, since TCE caches can be flushed
entirely, calling OPAL for 512 TCEs helps skiboot [1] to decide whether
to flush the entire cache or not.

This implements 2 new iommu_table_ops callbacks:
- xchg_no_kill() to update a single TCE with no TCE invalidation;
- tce_kill() to invalidate multiple TCEs.
This uses the same xchg_no_kill() callback for IODA1/2.

This implements 2 new wrappers on top of the new callbacks similar to
the existing iommu_tce_xchg().

This does not use the new callbacks yet, the next patches will;
so this should not cause any behavioral change.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h          | 17 ++++++++++++++
 arch/powerpc/kernel/iommu.c               | 27 +++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci-ioda.c | 12 ++++++++++
 3 files changed, 56 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index d7bf1f104c15..837b5122f257 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -58,6 +58,17 @@ struct iommu_table_ops {
 			unsigned long *hpa,
 			enum dma_data_direction *direction);
 
+	int (*xchg_no_kill)(struct iommu_table *tbl,
+			long index,
+			unsigned long *hpa,
+			enum dma_data_direction *direction,
+			bool realmode);
+
+	void (*tce_kill)(struct iommu_table *tbl,
+			unsigned long index,
+			unsigned long pages,
+			bool realmode);
+
 	__be64 *(*useraddrptr)(struct iommu_table *tbl, long index, bool alloc);
 #endif
 	void (*clear)(struct iommu_table *tbl,
@@ -209,6 +220,12 @@ extern void iommu_del_device(struct device *dev);
 extern long iommu_tce_xchg(struct mm_struct *mm, struct iommu_table *tbl,
 		unsigned long entry, unsigned long *hpa,
 		enum dma_data_direction *direction);
+extern long iommu_tce_xchg_no_kill(struct mm_struct *mm,
+		struct iommu_table *tbl,
+		unsigned long entry, unsigned long *hpa,
+		enum dma_data_direction *direction);
+extern void iommu_tce_kill(struct iommu_table *tbl,
+		unsigned long entry, unsigned long pages);
 #else
 static inline void iommu_register_group(struct iommu_table_group *table_group,
 					int pci_domain_number,
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index e7a2b160d4c6..070492f9b46e 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -1037,6 +1037,33 @@ long iommu_tce_xchg(struct mm_struct *mm, struct iommu_table *tbl,
 }
 EXPORT_SYMBOL_GPL(iommu_tce_xchg);
 
+extern long iommu_tce_xchg_no_kill(struct mm_struct *mm,
+		struct iommu_table *tbl,
+		unsigned long entry, unsigned long *hpa,
+		enum dma_data_direction *direction)
+{
+	long ret;
+	unsigned long size = 0;
+
+	ret = tbl->it_ops->xchg_no_kill(tbl, entry, hpa, direction, false);
+	if (!ret && ((*direction == DMA_FROM_DEVICE) ||
+			(*direction == DMA_BIDIRECTIONAL)) &&
+			!mm_iommu_is_devmem(mm, *hpa, tbl->it_page_shift,
+					&size))
+		SetPageDirty(pfn_to_page(*hpa >> PAGE_SHIFT));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_tce_xchg_no_kill);
+
+void iommu_tce_kill(struct iommu_table *tbl,
+		unsigned long entry, unsigned long pages)
+{
+	if (tbl->it_ops->tce_kill)
+		tbl->it_ops->tce_kill(tbl, entry, pages, false);
+}
+EXPORT_SYMBOL_GPL(iommu_tce_kill);
+
 int iommu_take_ownership(struct iommu_table *tbl)
 {
 	unsigned long flags, i, sz = (tbl->it_size + 7) >> 3;
diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
index ec48ea25a674..4e56b2c620ec 100644
--- a/arch/powerpc/platforms/powernv/pci-ioda.c
+++ b/arch/powerpc/platforms/powernv/pci-ioda.c
@@ -1939,6 +1939,14 @@ static int pnv_ioda1_tce_build(struct iommu_table *tbl, long index,
 }
 
 #ifdef CONFIG_IOMMU_API
+/* Common for IODA1 and IODA2 */
+static int pnv_ioda_tce_xchg_no_kill(struct iommu_table *tbl, long index,
+		unsigned long *hpa, enum dma_data_direction *direction,
+		bool realmode)
+{
+	return pnv_tce_xchg(tbl, index, hpa, direction, !realmode);
+}
+
 static int pnv_ioda1_tce_xchg(struct iommu_table *tbl, long index,
 		unsigned long *hpa, enum dma_data_direction *direction)
 {
@@ -1975,6 +1983,8 @@ static struct iommu_table_ops pnv_ioda1_iommu_ops = {
 #ifdef CONFIG_IOMMU_API
 	.exchange = pnv_ioda1_tce_xchg,
 	.exchange_rm = pnv_ioda1_tce_xchg_rm,
+	.xchg_no_kill = pnv_ioda_tce_xchg_no_kill,
+	.tce_kill = pnv_pci_p7ioc_tce_invalidate,
 	.useraddrptr = pnv_tce_useraddrptr,
 #endif
 	.clear = pnv_ioda1_tce_free,
@@ -2140,6 +2150,8 @@ static struct iommu_table_ops pnv_ioda2_iommu_ops = {
 #ifdef CONFIG_IOMMU_API
 	.exchange = pnv_ioda2_tce_xchg,
 	.exchange_rm = pnv_ioda2_tce_xchg_rm,
+	.xchg_no_kill = pnv_ioda_tce_xchg_no_kill,
+	.tce_kill = pnv_pci_ioda2_tce_invalidate,
 	.useraddrptr = pnv_tce_useraddrptr,
 #endif
 	.clear = pnv_ioda2_tce_free,
-- 
2.17.1

[PATCH kernel v3 2/5] KVM: PPC: Book3S: Invalidate multiple TCEs at once

[Alexey Kardashevskiy]

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 <aik@ozlabs.ru>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
---
Changes:
v3:
* fixed the subject line to include "Book3S"
* added "ack" from Paul
---
 arch/powerpc/kvm/book3s_64_vio.c    | 29 +++++++++++++++-------
 arch/powerpc/kvm/book3s_64_vio_hv.c | 38 +++++++++++++++++++++--------
 2 files changed, 48 insertions(+), 19 deletions(-)

diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
index c4b606fe73eb..5834db0a54c6 100644
--- a/arch/powerpc/kvm/book3s_64_vio.c
+++ b/arch/powerpc/kvm/book3s_64_vio.c
@@ -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,
@@ -447,7 +447,8 @@ 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)
@@ -455,7 +456,7 @@ static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
 
 	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;
 }
@@ -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;
@@ -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;
@@ -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) {
@@ -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);
 
@@ -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);
@@ -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);
diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
index b4f20f13b860..ab6eeb8e753e 100644
--- a/arch/powerpc/kvm/book3s_64_vio_hv.c
+++ b/arch/powerpc/kvm/book3s_64_vio_hv.c
@@ -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))) {
@@ -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,
@@ -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
@@ -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;
 }
@@ -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);
 		/*
@@ -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;
@@ -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) {
@@ -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);
@@ -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);
@@ -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 */
-- 
2.17.1

[PATCH kernel v3 3/5] vfio/spapr_tce: Invalidate multiple TCEs at once

[Alexey Kardashevskiy]

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.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/vfio_iommu_spapr_tce.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
index babef8b00daf..3b18fa4d090a 100644
--- a/drivers/vfio/vfio_iommu_spapr_tce.c
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -435,7 +435,7 @@ static int tce_iommu_clear(struct tce_container *container,
 	unsigned long oldhpa;
 	long ret;
 	enum dma_data_direction direction;
-	unsigned long lastentry = entry + pages;
+	unsigned long lastentry = entry + pages, firstentry = entry;
 
 	for ( ; entry < lastentry; ++entry) {
 		if (tbl->it_indirect_levels && tbl->it_userspace) {
@@ -460,7 +460,7 @@ static int tce_iommu_clear(struct tce_container *container,
 
 		direction = DMA_NONE;
 		oldhpa = 0;
-		ret = iommu_tce_xchg(container->mm, tbl, entry, &oldhpa,
+		ret = iommu_tce_xchg_no_kill(container->mm, tbl, entry, &oldhpa,
 				&direction);
 		if (ret)
 			continue;
@@ -476,6 +476,8 @@ static int tce_iommu_clear(struct tce_container *container,
 		tce_iommu_unuse_page(container, oldhpa);
 	}
 
+	iommu_tce_kill(tbl, firstentry, pages);
+
 	return 0;
 }
 
@@ -518,8 +520,8 @@ static long tce_iommu_build(struct tce_container *container,
 
 		hpa |= offset;
 		dirtmp = direction;
-		ret = iommu_tce_xchg(container->mm, tbl, entry + i, &hpa,
-				&dirtmp);
+		ret = iommu_tce_xchg_no_kill(container->mm, tbl, entry + i,
+				&hpa, &dirtmp);
 		if (ret) {
 			tce_iommu_unuse_page(container, hpa);
 			pr_err("iommu_tce: %s failed ioba=%lx, tce=%lx, ret=%ld\n",
@@ -536,6 +538,8 @@ static long tce_iommu_build(struct tce_container *container,
 
 	if (ret)
 		tce_iommu_clear(container, tbl, entry, i);
+	else
+		iommu_tce_kill(tbl, entry, pages);
 
 	return ret;
 }
@@ -572,8 +576,8 @@ static long tce_iommu_build_v2(struct tce_container *container,
 		if (mm_iommu_mapped_inc(mem))
 			break;
 
-		ret = iommu_tce_xchg(container->mm, tbl, entry + i, &hpa,
-				&dirtmp);
+		ret = iommu_tce_xchg_no_kill(container->mm, tbl, entry + i,
+				&hpa, &dirtmp);
 		if (ret) {
 			/* dirtmp cannot be DMA_NONE here */
 			tce_iommu_unuse_page_v2(container, tbl, entry + i);
@@ -593,6 +597,8 @@ static long tce_iommu_build_v2(struct tce_container *container,
 
 	if (ret)
 		tce_iommu_clear(container, tbl, entry, i);
+	else
+		iommu_tce_kill(tbl, entry, pages);
 
 	return ret;
 }
-- 
2.17.1

[PATCH kernel v3 4/5] powerpc/pseries/iommu: Switch to xchg_no_kill

[Alexey Kardashevskiy]

This is the last implementation of iommu_table_ops::exchange() which
we are about to remove.

This implements xchg_no_kill() for pseries. Since it is paravirtual
platform, the hypervisor does TCE invalidations and we do not have
to deal with it here, hence no tce_kill() hook.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
Changes:
v3:
* new to the series and it fixes compile error from v2
---
 arch/powerpc/platforms/pseries/iommu.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c
index 42fb03253334..df7db33ca93b 100644
--- a/arch/powerpc/platforms/pseries/iommu.c
+++ b/arch/powerpc/platforms/pseries/iommu.c
@@ -621,7 +621,8 @@ static void pci_dma_bus_setup_pSeries(struct pci_bus *bus)
 
 #ifdef CONFIG_IOMMU_API
 static int tce_exchange_pseries(struct iommu_table *tbl, long index, unsigned
-				long *tce, enum dma_data_direction *direction)
+				long *tce, enum dma_data_direction *direction,
+				bool realmode)
 {
 	long rc;
 	unsigned long ioba = (unsigned long) index << tbl->it_page_shift;
@@ -649,7 +650,7 @@ static int tce_exchange_pseries(struct iommu_table *tbl, long index, unsigned
 struct iommu_table_ops iommu_table_lpar_multi_ops = {
 	.set = tce_buildmulti_pSeriesLP,
 #ifdef CONFIG_IOMMU_API
-	.exchange = tce_exchange_pseries,
+	.xchg_no_kill = tce_exchange_pseries,
 #endif
 	.clear = tce_freemulti_pSeriesLP,
 	.get = tce_get_pSeriesLP
-- 
2.17.1

[PATCH kernel v3 5/5] powerpc/powernv/ioda: Remove obsolete iommu_table_ops::exchange callbacks

[Alexey Kardashevskiy]

As now we have xchg_no_kill/tce_kill, these are not used anymore so
remove them.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h          | 10 -----
 arch/powerpc/kernel/iommu.c               | 26 +-----------
 arch/powerpc/platforms/powernv/pci-ioda.c | 50 -----------------------
 3 files changed, 1 insertion(+), 85 deletions(-)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index 837b5122f257..350101e11ddb 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -48,16 +48,6 @@ struct iommu_table_ops {
 	 * returns old TCE and DMA direction mask.
 	 * @tce is a physical address.
 	 */
-	int (*exchange)(struct iommu_table *tbl,
-			long index,
-			unsigned long *hpa,
-			enum dma_data_direction *direction);
-	/* Real mode */
-	int (*exchange_rm)(struct iommu_table *tbl,
-			long index,
-			unsigned long *hpa,
-			enum dma_data_direction *direction);
-
 	int (*xchg_no_kill)(struct iommu_table *tbl,
 			long index,
 			unsigned long *hpa,
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index 070492f9b46e..9704f3f76e63 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -1013,30 +1013,6 @@ int iommu_tce_check_gpa(unsigned long page_shift, unsigned long gpa)
 }
 EXPORT_SYMBOL_GPL(iommu_tce_check_gpa);
 
-long iommu_tce_xchg(struct mm_struct *mm, struct iommu_table *tbl,
-		unsigned long entry, unsigned long *hpa,
-		enum dma_data_direction *direction)
-{
-	long ret;
-	unsigned long size = 0;
-
-	ret = tbl->it_ops->exchange(tbl, entry, hpa, direction);
-
-	if (!ret && ((*direction == DMA_FROM_DEVICE) ||
-			(*direction == DMA_BIDIRECTIONAL)) &&
-			!mm_iommu_is_devmem(mm, *hpa, tbl->it_page_shift,
-					&size))
-		SetPageDirty(pfn_to_page(*hpa >> PAGE_SHIFT));
-
-	/* if (unlikely(ret))
-		pr_err("iommu_tce: %s failed on hwaddr=%lx ioba=%lx kva=%lx ret=%d\n",
-			__func__, hwaddr, entry << tbl->it_page_shift,
-				hwaddr, ret); */
-
-	return ret;
-}
-EXPORT_SYMBOL_GPL(iommu_tce_xchg);
-
 extern long iommu_tce_xchg_no_kill(struct mm_struct *mm,
 		struct iommu_table *tbl,
 		unsigned long entry, unsigned long *hpa,
@@ -1076,7 +1052,7 @@ int iommu_take_ownership(struct iommu_table *tbl)
 	 * requires exchange() callback defined so if it is not
 	 * implemented, we disallow taking ownership over the table.
 	 */
-	if (!tbl->it_ops->exchange)
+	if (!tbl->it_ops->xchg_no_kill)
 		return -EINVAL;
 
 	spin_lock_irqsave(&tbl->large_pool.lock, flags);
diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
index 4e56b2c620ec..c28d0d9b7ee0 100644
--- a/arch/powerpc/platforms/powernv/pci-ioda.c
+++ b/arch/powerpc/platforms/powernv/pci-ioda.c
@@ -1946,28 +1946,6 @@ static int pnv_ioda_tce_xchg_no_kill(struct iommu_table *tbl, long index,
 {
 	return pnv_tce_xchg(tbl, index, hpa, direction, !realmode);
 }
-
-static int pnv_ioda1_tce_xchg(struct iommu_table *tbl, long index,
-		unsigned long *hpa, enum dma_data_direction *direction)
-{
-	long ret = pnv_tce_xchg(tbl, index, hpa, direction, true);
-
-	if (!ret)
-		pnv_pci_p7ioc_tce_invalidate(tbl, index, 1, false);
-
-	return ret;
-}
-
-static int pnv_ioda1_tce_xchg_rm(struct iommu_table *tbl, long index,
-		unsigned long *hpa, enum dma_data_direction *direction)
-{
-	long ret = pnv_tce_xchg(tbl, index, hpa, direction, false);
-
-	if (!ret)
-		pnv_pci_p7ioc_tce_invalidate(tbl, index, 1, true);
-
-	return ret;
-}
 #endif
 
 static void pnv_ioda1_tce_free(struct iommu_table *tbl, long index,
@@ -1981,8 +1959,6 @@ static void pnv_ioda1_tce_free(struct iommu_table *tbl, long index,
 static struct iommu_table_ops pnv_ioda1_iommu_ops = {
 	.set = pnv_ioda1_tce_build,
 #ifdef CONFIG_IOMMU_API
-	.exchange = pnv_ioda1_tce_xchg,
-	.exchange_rm = pnv_ioda1_tce_xchg_rm,
 	.xchg_no_kill = pnv_ioda_tce_xchg_no_kill,
 	.tce_kill = pnv_pci_p7ioc_tce_invalidate,
 	.useraddrptr = pnv_tce_useraddrptr,
@@ -2113,30 +2089,6 @@ static int pnv_ioda2_tce_build(struct iommu_table *tbl, long index,
 	return ret;
 }
 
-#ifdef CONFIG_IOMMU_API
-static int pnv_ioda2_tce_xchg(struct iommu_table *tbl, long index,
-		unsigned long *hpa, enum dma_data_direction *direction)
-{
-	long ret = pnv_tce_xchg(tbl, index, hpa, direction, true);
-
-	if (!ret)
-		pnv_pci_ioda2_tce_invalidate(tbl, index, 1, false);
-
-	return ret;
-}
-
-static int pnv_ioda2_tce_xchg_rm(struct iommu_table *tbl, long index,
-		unsigned long *hpa, enum dma_data_direction *direction)
-{
-	long ret = pnv_tce_xchg(tbl, index, hpa, direction, false);
-
-	if (!ret)
-		pnv_pci_ioda2_tce_invalidate(tbl, index, 1, true);
-
-	return ret;
-}
-#endif
-
 static void pnv_ioda2_tce_free(struct iommu_table *tbl, long index,
 		long npages)
 {
@@ -2148,8 +2100,6 @@ static void pnv_ioda2_tce_free(struct iommu_table *tbl, long index,
 static struct iommu_table_ops pnv_ioda2_iommu_ops = {
 	.set = pnv_ioda2_tce_build,
 #ifdef CONFIG_IOMMU_API
-	.exchange = pnv_ioda2_tce_xchg,
-	.exchange_rm = pnv_ioda2_tce_xchg_rm,
 	.xchg_no_kill = pnv_ioda_tce_xchg_no_kill,
 	.tce_kill = pnv_pci_ioda2_tce_invalidate,
 	.useraddrptr = pnv_tce_useraddrptr,
-- 
2.17.1

Re: [PATCH kernel v3 1/5] powerpc/powernv/ioda: Split out TCE invalidation from TCE updates

[Michael Ellerman]

On Thu, 2019-08-29 at 08:52:48 UTC, Alexey Kardashevskiy wrote:
> At the moment updates in a TCE table are made by iommu_table_ops::exchange
> which update one TCE and invalidates an entry in the PHB/NPU TCE cache
> via set of registers called "TCE Kill" (hence the naming).
> Writing a TCE is a simple xchg() but invalidating the TCE cache is
> a relatively expensive OPAL call. Mapping a 100GB guest with PCI+NPU
> passed through devices takes about 20s.
> 
> Thankfully we can do better. Since such big mappings happen at the boot
> time and when memory is plugged/onlined (i.e. not often), these requests
> come in 512 pages so we call call OPAL 512 times less which brings 20s
> from the above to less than 10s. Also, since TCE caches can be flushed
> entirely, calling OPAL for 512 TCEs helps skiboot [1] to decide whether
> to flush the entire cache or not.
> 
> This implements 2 new iommu_table_ops callbacks:
> - xchg_no_kill() to update a single TCE with no TCE invalidation;
> - tce_kill() to invalidate multiple TCEs.
> This uses the same xchg_no_kill() callback for IODA1/2.
> 
> This implements 2 new wrappers on top of the new callbacks similar to
> the existing iommu_tce_xchg().
> 
> This does not use the new callbacks yet, the next patches will;
> so this should not cause any behavioral change.
> 
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>

Series applied to powerpc topic/ppc-kvm, thanks.

https://git.kernel.org/powerpc/c/35872480da47ec714fd9c4f2f3d2d83daf304851

cheers