KVM: PPC: Book3S HV: Migrate pinned pages out of CMA

From: Balbir Singh <[email protected]>
Subject: [RESEND][v2][PATCH] KVM: PPC: Book3S HV: Migrate pinned pages out of CMA

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end, but there might be bugs For example if after lru_add_drain(), the page is not isolated is this a BUG?

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Signed-off-by Balbir Singh <[email protected]>

---
 arch/powerpc/include/asm/mmu_context.h |  1 +
 arch/powerpc/mm/mmu_context_iommu.c    | 80 ++++++++++++++++++++++++++++++++--
 2 files changed, 77 insertions(+), 4 deletions(-)

arch/powerpc/include/asm/mmu_context.h +1 -0

index 9d2cd0c..475d1be 100644
@@ -18,6 +18,7 @@ extern void destroy_context(struct mm_struct *mm);
 #ifdef CONFIG_SPAPR_TCE_IOMMU
 struct mm_iommu_table_group_mem_t;
 
+extern int isolate_lru_page(struct page *page);	/* from internal.h */
 extern bool mm_iommu_preregistered(void);
 extern long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem);

arch/powerpc/mm/mmu_context_iommu.c +76 -4

index da6a216..c18f742 100644
@@ -15,6 +15,9 @@
 #include <linux/rculist.h>
 #include <linux/vmalloc.h>
 #include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
 #include <asm/mmu_context.h>
 
 static DEFINE_MUTEX(mem_list_mutex);

@@ -72,6 +75,54 @@ bool mm_iommu_preregistered(void)
 }
 EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
 
+/*
+ * Taken from alloc_migrate_target with changes to remove CMA allocations
+ */
+struct page *new_iommu_non_cma_page(struct page *page, unsigned long private,
+					int **resultp)
+{
+	gfp_t gfp_mask = GFP_USER;
+	struct page *new_page;
+
+	if (PageHuge(page) || PageTransHuge(page) || PageCompound(page))
+		return NULL;
+
+	if (PageHighMem(page))
+		gfp_mask |= __GFP_HIGHMEM;
+
+	/*
+	 * We don't want the allocation to force an OOM if possibe
+	 */
+	new_page = alloc_page(gfp_mask | __GFP_NORETRY | __GFP_NOWARN);
+	return new_page;
+}
+
+static int mm_iommu_move_page_from_cma(struct page *page)
+{
+	int ret;
+	LIST_HEAD(cma_migrate_pages);
+
+	/* Ignore huge pages for now */
+	if (PageHuge(page) || PageTransHuge(page) || PageCompound(page))
+		return -EBUSY;
+
+	lru_add_drain();
+	ret = isolate_lru_page(page);
+	if (ret)
+		get_page(page); /* Potential BUG? */
+
+	list_add(&page->lru, &cma_migrate_pages);
+	put_page(page); /* Drop the gup reference */
+
+	ret = migrate_pages(&cma_migrate_pages, new_iommu_non_cma_page,
+				NULL, 0, MIGRATE_SYNC, MR_CMA);
+	if (ret) {
+		if (!list_empty(&cma_migrate_pages))
+			putback_movable_pages(&cma_migrate_pages);
+	}
+	return 0;
+}
+
 long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem)
 {

@@ -124,15 +175,36 @@ long mm_iommu_get(unsigned long ua, unsigned long entries,
 	for (i = 0; i < entries; ++i) {
 		if (1 != get_user_pages_fast(ua + (i << PAGE_SHIFT),
 					1/* pages */, 1/* iswrite */, &page)) {
+			ret = -EFAULT;
 			for (j = 0; j < i; ++j)
-				put_page(pfn_to_page(
-						mem->hpas[j] >> PAGE_SHIFT));
+				put_page(pfn_to_page(mem->hpas[j] >>
+						PAGE_SHIFT));
 			vfree(mem->hpas);
 			kfree(mem);
-			ret = -EFAULT;
 			goto unlock_exit;
 		}
-
+		/*
+		 * If we get a page from the CMA zone, since we are going to
+		 * be pinning these entries, we might as well move them out
+		 * of the CMA zone if possible. NOTE: faulting in + migration
+		 * can be expensive. Batching can be considered later
+		 */
+		if (get_pageblock_migratetype(page) == MIGRATE_CMA) {
+			if (mm_iommu_move_page_from_cma(page))
+				goto populate;
+			if (1 != get_user_pages_fast(ua + (i << PAGE_SHIFT),
+						1/* pages */, 1/* iswrite */,
+						&page)) {
+				ret = -EFAULT;
+				for (j = 0; j < i; ++j)
+					put_page(pfn_to_page(mem->hpas[j] >>
+								PAGE_SHIFT));
+				vfree(mem->hpas);
+				kfree(mem);
+				goto unlock_exit;
+			}
+		}
+populate:
 		mem->hpas[i] = page_to_pfn(page) << PAGE_SHIFT;
 	}
 

Quoted 31 lines · Balbir Singh

From: Balbir Singh <[email protected]>
Subject: [RESEND][v2][PATCH] KVM: PPC: Book3S HV: Migrate pinned pages out of CMA

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end, but there might be bugs For example if after lru_add_drain(), the page is not isolated is this a BUG?

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Quoted 1 line

Signed-off-by Balbir Singh <[email protected]>

Acked-by Alexey Kardashevskiy <[email protected]>

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end. The full solution requires migration of THP pages in the CMA region. That work will be done incrementally on top of this.

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Signed-off-by Balbir Singh <[email protected]>

Acked-by Alexey Kardashevskiy <[email protected]>

---
 arch/powerpc/include/asm/mmu_context.h |  1 +
 arch/powerpc/mm/mmu_context_iommu.c    | 81 ++++++++++++++++++++++++++++++++--
 2 files changed, 78 insertions(+), 4 deletions(-)

arch/powerpc/include/asm/mmu_context.h +0 -0

index 9d2cd0c..475d1be 100644

arch/powerpc/include/asm/mmu_context.h +0 -0

 extern bool mm_iommu_preregistered(void);
 extern long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem);

arch/powerpc/mm/mmu_context_iommu.c +12 -0

index da6a216..e0f1c33 100644
@@ -15,6 +15,9 @@
 #include <linux/rculist.h>
 #include <linux/vmalloc.h>
 #include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
 #include <asm/mmu_context.h>
 
 static DEFINE_MUTEX(mem_list_mutex);

@@ -72,6 +75,55 @@ bool mm_iommu_preregistered(void)
 }
 EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
 
+/*
+ * Taken from alloc_migrate_target with changes to remove CMA allocations
+ */
+struct page *new_iommu_non_cma_page(struct page *page, unsigned long private,
+					int **resultp)
+{
+	gfp_t gfp_mask = GFP_USER;
+	struct page *new_page;
+

Quoted 43 lines · Balbir Singh

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end. The full solution requires migration of THP pages in the CMA region. That work will be done incrementally on top of this.

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Signed-off-by Balbir Singh <[email protected]>

Acked-by Alexey Kardashevskiy <[email protected]>

---
 arch/powerpc/include/asm/mmu_context.h |  1 +
 arch/powerpc/mm/mmu_context_iommu.c    | 81 ++++++++++++++++++++++++++++++++--
 2 files changed, 78 insertions(+), 4 deletions(-)

arch/powerpc/include/asm/mmu_context.h +0 -0

index 9d2cd0c..475d1be 100644
@@ -18,6 +18,7 @@ extern void destroy_context(struct mm_struct *mm);
 #ifdef CONFIG_SPAPR_TCE_IOMMU
 struct mm_iommu_table_group_mem_t;
 

arch/powerpc/include/asm/mmu_context.h extern void destroy_context(struct mm_struct *mm); · line 21

Quoted 1 line

arch/powerpc/include/asm/mmu_context.h +1 -0

+extern int isolate_lru_page(struct page *page);	/* from internal.h */

Small nit, cant we just add "mm/internal.h" header here with full path ?

Quoted 35 lines

arch/powerpc/include/asm/mmu_context.h +0 -0

 extern bool mm_iommu_preregistered(void);
 extern long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem);

arch/powerpc/mm/mmu_context_iommu.c +17 -0

index da6a216..e0f1c33 100644
@@ -15,6 +15,9 @@
 #include <linux/rculist.h>
 #include <linux/vmalloc.h>
 #include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
 #include <asm/mmu_context.h>
 
 static DEFINE_MUTEX(mem_list_mutex);

@@ -72,6 +75,55 @@ bool mm_iommu_preregistered(void)
 }
 EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
 
+/*
+ * Taken from alloc_migrate_target with changes to remove CMA allocations
+ */
+struct page *new_iommu_non_cma_page(struct page *page, unsigned long private,
+					int **resultp)
+{
+	gfp_t gfp_mask = GFP_USER;
+	struct page *new_page;
+
+	if (PageHuge(page) || PageTransHuge(page) || PageCompound(page))
+		return NULL;
+
+	if (PageHighMem(page))
+		gfp_mask |= __GFP_HIGHMEM;

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 96

Quoted 5 lines

arch/powerpc/mm/mmu_context_iommu.c +5 -0

+
+	/*
+	 * We don't want the allocation to force an OOM if possibe
+	 */
+	new_page = alloc_page(gfp_mask | __GFP_NORETRY | __GFP_NOWARN);

So what guarantees that the new page too wont come from MIGRATE_CMA page block ? Is absence of __GFP_MOVABLE flag enough.

we be checking that migrate type of the new allocated page is indeed not MIGRATE_CMA ?

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end. The full solution requires migration of THP pages in the CMA region. That work will be done incrementally on top of this.

Signed-off-by Balbir Singh <[email protected]>

Acked-by Alexey Kardashevskiy <[email protected]>

From: Balbir Singh <[email protected]>
Subject: [RESEND][v2][PATCH] KVM: PPC: Book3S HV: Migrate pinned pages out of CMA

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end, but there might be bugs For example if after lru_add_drain(), the page is not isolated is this a BUG?

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Signed-off-by Balbir Singh <[email protected]>

---
 arch/powerpc/include/asm/mmu_context.h |  1 +
 arch/powerpc/mm/mmu_context_iommu.c    | 80 ++++++++++++++++++++++++++++++++--
 2 files changed, 77 insertions(+), 4 deletions(-)

arch/powerpc/include/asm/mmu_context.h +1 -0

index 9d2cd0c..475d1be 100644
@@ -18,6 +18,7 @@ extern void destroy_context(struct mm_struct *mm);
 #ifdef CONFIG_SPAPR_TCE_IOMMU
 struct mm_iommu_table_group_mem_t;
 
+extern int isolate_lru_page(struct page *page);	/* from internal.h */
 extern bool mm_iommu_preregistered(void);
 extern long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem);

arch/powerpc/mm/mmu_context_iommu.c +4 -0

index da6a216..c18f742 100644
@@ -15,6 +15,9 @@
 #include <linux/rculist.h>
 #include <linux/vmalloc.h>
 #include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
 #include <asm/mmu_context.h>
 
 static DEFINE_MUTEX(mem_list_mutex);

@@ -72,6 +75,54 @@ bool mm_iommu_preregistered(void)
 }
 EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
 
+/*

arch/powerpc/mm/mmu_context_iommu.c +11 -0

+
+	if (PageHighMem(page))
+		gfp_mask |= __GFP_HIGHMEM;
+
+	/*
+	 * We don't want the allocation to force an OOM if possibe
+	 */
+	new_page = alloc_page(gfp_mask | __GFP_NORETRY | __GFP_NOWARN);
+	return new_page;
+}
+

arch/powerpc/mm/mmu_context_iommu.c +25 -4

 long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem)
 {

@@ -124,15 +175,36 @@ long mm_iommu_get(unsigned long ua, unsigned long entries,
 	for (i = 0; i < entries; ++i) {
 		if (1 != get_user_pages_fast(ua + (i << PAGE_SHIFT),
 					1/* pages */, 1/* iswrite */, &page)) {
+			ret = -EFAULT;
 			for (j = 0; j < i; ++j)
-				put_page(pfn_to_page(
-						mem->hpas[j] >> PAGE_SHIFT));
+				put_page(pfn_to_page(mem->hpas[j] >>
+						PAGE_SHIFT));
 			vfree(mem->hpas);
 			kfree(mem);
-			ret = -EFAULT;
 			goto unlock_exit;
 		}
-
+		/*
+		 * If we get a page from the CMA zone, since we are going to
+		 * be pinning these entries, we might as well move them out
+		 * of the CMA zone if possible. NOTE: faulting in + migration
+		 * can be expensive. Batching can be considered later
+		 */
+		if (get_pageblock_migratetype(page) == MIGRATE_CMA) {
+			if (mm_iommu_move_page_from_cma(page))
+				goto populate;
+			if (1 != get_user_pages_fast(ua + (i << PAGE_SHIFT),
+						1/* pages */, 1/* iswrite */,
+						&page)) {
+				ret = -EFAULT;
+				for (j = 0; j < i; ++j)
+					put_page(pfn_to_page(mem->hpas[j] >>
+								PAGE_SHIFT));
+				vfree(mem->hpas);
+				kfree(mem);
+				goto unlock_exit;
+			}
+		}
+populate:
 		mem->hpas[i] = page_to_pfn(page) << PAGE_SHIFT;
 	}
 

Signature · 7 lines

2.5.5

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Quoted 66 lines · Balbir Singh

From: Balbir Singh <[email protected]>
Subject: [RESEND][v2][PATCH] KVM: PPC: Book3S HV: Migrate pinned pages out of CMA

When PCI Device pass-through is enabled via VFIO, KVM-PPC will pin pages using get_user_pages_fast(). One of the downsides of the pinning is that the page could be in CMA region. The CMA region is used for other allocations like the hash page table. Ideally we want the pinned pages to be from non CMA region.

This patch (currently only for KVM PPC with VFIO) forcefully migrates the pages out (huge pages are omitted for the moment). There are more efficient ways of doing this, but that might be elaborate and might impact a larger audience beyond just the kvm ppc implementation.

The magic is in new_iommu_non_cma_page() which allocates the new page from a non CMA region.

I've tested the patches lightly at my end, but there might be bugs For example if after lru_add_drain(), the page is not isolated is this a BUG?

Previous discussion was at
http://permalink.gmane.org/gmane.linux.kernel.mm/136738

Cc Benjamin Herrenschmidt <[email protected]>

Cc Michael Ellerman <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Alexey Kardashevskiy <[email protected]>

Signed-off-by Balbir Singh <[email protected]>

---
 arch/powerpc/include/asm/mmu_context.h |  1 +
 arch/powerpc/mm/mmu_context_iommu.c    | 80 ++++++++++++++++++++++++++++++++--
 2 files changed, 77 insertions(+), 4 deletions(-)

arch/powerpc/include/asm/mmu_context.h +1 -0

index 9d2cd0c..475d1be 100644
@@ -18,6 +18,7 @@ extern void destroy_context(struct mm_struct *mm);
 #ifdef CONFIG_SPAPR_TCE_IOMMU
 struct mm_iommu_table_group_mem_t;
 
+extern int isolate_lru_page(struct page *page);	/* from internal.h */
 extern bool mm_iommu_preregistered(void);
 extern long mm_iommu_get(unsigned long ua, unsigned long entries,
 		struct mm_iommu_table_group_mem_t **pmem);

arch/powerpc/mm/mmu_context_iommu.c +3 -0

index da6a216..c18f742 100644
@@ -15,6 +15,9 @@
 #include <linux/rculist.h>
 #include <linux/vmalloc.h>
 #include <linux/mutex.h>
+#include <linux/migrate.h>
+#include <linux/hugetlb.h>
+#include <linux/swap.h>
 #include <asm/mmu_context.h>
 
 static DEFINE_MUTEX(mem_list_mutex);

@@ -72,6 +75,54 @@ bool mm_iommu_preregistered(void)
 }
 EXPORT_SYMBOL_GPL(mm_iommu_preregistered);
 

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 88

Quoted 11 lines

arch/powerpc/mm/mmu_context_iommu.c +11 -0

+/*
+ * Taken from alloc_migrate_target with changes to remove CMA allocations
+ */
+struct page *new_iommu_non_cma_page(struct page *page, unsigned long private,
+					int **resultp)
+{
+	gfp_t gfp_mask = GFP_USER;
+	struct page *new_page;
+
+	if (PageHuge(page) || PageTransHuge(page) || PageCompound(page))
+		return NULL;

Doesn't a PageCompound check cover all ?

Quoted 18 lines

arch/powerpc/mm/mmu_context_iommu.c +16 -0

+
+	if (PageHighMem(page))
+		gfp_mask |= __GFP_HIGHMEM;
+
+	/*
+	 * We don't want the allocation to force an OOM if possibe
+	 */
+	new_page = alloc_page(gfp_mask | __GFP_NORETRY | __GFP_NOWARN);
+	return new_page;
+}
+
+static int mm_iommu_move_page_from_cma(struct page *page)
+{
+	int ret;
+	LIST_HEAD(cma_migrate_pages);
+

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 109

Quoted 5 lines

arch/powerpc/mm/mmu_context_iommu.c +5 -0

+	/* Ignore huge pages for now */
+	if (PageHuge(page) || PageTransHuge(page) || PageCompound(page))
+		return -EBUSY;
+
+	lru_add_drain();

I guess I asked this last time. Shouldn't this be lru_add_drain_all() ? What if the page is in other cpu's pagevec ?

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 114

Quoted 7 lines

arch/powerpc/mm/mmu_context_iommu.c +5 -0

+	ret = isolate_lru_page(page);
+	if (ret)
+		get_page(page); /* Potential BUG? */
+
+	list_add(&page->lru, &cma_migrate_pages);

Is that correct ? if we failed the isolate_lru_page(), can we be sure we are not on lru at all ? ie, what if the page was on other cpu pagevec ?

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 116

Quoted 2 lines

arch/powerpc/mm/mmu_context_iommu.c +2 -0

+	put_page(page); /* Drop the gup reference */
+

Where is get user page (gup) here ? . I guess you mean drop the reference taken above ?

arch/powerpc/mm/mmu_context_iommu.c bool mm_iommu_preregistered(void) · line 125

Quoted 10 lines

arch/powerpc/mm/mmu_context_iommu.c +9 -0

+	ret = migrate_pages(&cma_migrate_pages, new_iommu_non_cma_page,
+				NULL, 0, MIGRATE_SYNC, MR_CMA);
+	if (ret) {
+		if (!list_empty(&cma_migrate_pages))
+			putback_movable_pages(&cma_migrate_pages);
+	}
+	return 0;
+}
+

I guess the plan was to not do it one page at a time and switch this to list of pages which we need to migrate. Any reason why that is not tried ?