Skip to contentSkip to conversation

Add MEMORY_DEVICE_COHERENT for coherent device memory mapping

PATCH Superseded Alex Sierra 5 participants 29 messages Jul 7, 2022 – Jul 15, 2022

Subsystems (24)

v8 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v8 cover Add MEMORY_DEVICE_COHERENT for coherent device memory mapping

This is our MEMORY_DEVICE_COHERENT patch series rebased and updated for current 5.19.0-rc5

Changes since the last version:
- Fixed problems with migration during long-term pinning in get_user_pages
- Open coded vm_normal_lru_pages as suggested in previous code review
- Update hmm_gup_test with more get_user_pages calls, include hmm_cow_in_device in hmm-test.

This patch series introduces MEMORY_DEVICE_COHERENT, a type of memory owned by a device that can be mapped into CPU page tables like MEMORY_DEVICE_GENERIC and can also be migrated like MEMORY_DEVICE_PRIVATE.

This patch series is mostly self-contained except for a few places where it needs to update other subsystems to handle the new memory type.

System stability and performance are not affected according to our ongoing testing, including xfstests.

How it works: The system BIOS advertises the GPU device memory (aka VRAM) as SPM (special purpose memory) in the UEFI system address map.

The amdgpu driver registers the memory with devmap as MEMORY_DEVICE_COHERENT using devm_memremap_pages. The initial user for this hardware page migration capability is the Frontier supercomputer project. This functionality is not AMD-specific. We expect other GPU vendors to find this functionality useful, and possibly other hardware types in the future.

Our test nodes in the lab are similar to the Frontier configuration, with .5 TB of system memory plus 256 GB of device memory split across 4 GPUs, all in a single coherent address space. Page migration is expected to improve application efficiency significantly. We will report empirical results as they become available.

Coherent device type pages at gup are now migrated back to system memory if they are being pinned long-term (FOLL_LONGTERM). The reason is, that long-term pinning would interfere with the device memory manager owning the device-coherent pages (e.g. evictions in TTM). These series incorporate Alistair Popple patches to do this migration from pin_user_pages() calls. hmm_gup_test has been added to hmm-test to test different get user pages calls.

This series includes handling of device-managed anonymous pages returned by vm_normal_pages. Although they behave like normal pages for purposes of mapping in CPU page tables and for COW, they do not support LRU lists, NUMA migration or THP.

We also introduced a FOLL_LRU flag that adds the same behaviour to follow_page and related APIs, to allow callers to specify that they expect to put pages on an LRU list.

v2:
- Rebase to latest 5.18-rc7.
- Drop patch "mm: add device coherent checker to remove migration pte" and modify try_to_migrate_one, to let DEVICE_COHERENT pages fall through to normal page path. Based on Alistair Popple's comment.
- Fix comment formatting.
- Reword comment in vm_normal_page about pte_devmap().
- Merge "drm/amdkfd: coherent type as sys mem on migration to ram" to "drm/amdkfd: add SPM support for SVM".

v3:
- Rebase to latest 5.18.0.
- Patch "mm: handling Non-LRU pages returned by vm_normal_pages" reordered.
- Add WARN_ON_ONCE for thp device coherent case.

v4:
- Rebase to latest 5.18.0
- Fix consitency between pages with FOLL_LRU flag set and pte_devmap at follow_page_pte.

v5:
- Remove unused zone_device_type from lib/test_hmm and selftest/vm/hmm-test.c.

v6:
- Rebase to 5.19.0-rc4
- Rename is_pinnable_page to is_longterm_pinnable_page and add a coherent device checker.
- Add a new gup test to hmm-test to cover fast pinnable case with FOLL_LONGTERM.

v7:
- Reorder patch series.
- Remove FOLL_LRU and check on each caller for LRU pages handling instead.

v8:
- Add "mm: move page zone helpers into new header-specific file" patch. The intention is to centralize all page zone helpers and keep them independent from mm.h and memremap.h.

Alex Sierra (13):
mm: rename is_pinnable_pages to is_longterm_pinnable_pages
mm: move page zone helpers into new header-specific file
mm: add zone device coherent type memory support
mm: handling Non-LRU pages returned by vm_normal_pages
mm: add device coherent vma selection for memory migration drm/amdkfd: add SPM support for SVM
lib: test_hmm add ioctl to get zone device type
lib: test_hmm add module param for zone device type
lib: add support for device coherent type in test_hmm
tools: update hmm-test to support device coherent type
tools: update test_hmm script to support SP config
tools: add hmm gup tests for device coherent type
tools: add selftests to hmm for COW in device memory

Alistair Popple (2):
mm: remove the vma check in migrate_vma_setup()
mm/gup: migrate device coherent pages when pinning instead of failing

 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c |  34 ++-
 drivers/infiniband/core/rw.c             |   1 -
 drivers/nvme/target/io-cmd-bdev.c        |   1 -
 fs/proc/task_mmu.c                       |   2 +-
 include/linux/memremap.h                 | 113 +-------
 include/linux/migrate.h                  |   1 +
 include/linux/mm.h                       |  91 +-----
 include/linux/page_zone.h                | 213 ++++++++++++++
 lib/test_hmm.c                           | 337 +++++++++++++++++------
 lib/test_hmm_uapi.h                      |  19 +-
 mm/gup.c                                 |  49 +++-
 mm/gup_test.c                            |   2 +-
 mm/huge_memory.c                         |   2 +-
 mm/hugetlb.c                             |   2 +-
 mm/internal.h                            |   1 +
 mm/khugepaged.c                          |   9 +-
 mm/ksm.c                                 |   6 +-
 mm/madvise.c                             |   4 +-
 mm/memcontrol.c                          |   8 +-
 mm/memory-failure.c                      |   8 +-
 mm/memory.c                              |  10 +-
 mm/mempolicy.c                           |   2 +-
 mm/memremap.c                            |  10 +
 mm/migrate.c                             |   4 +-
 mm/migrate_device.c                      | 115 ++++++--
 mm/mlock.c                               |   2 +-
 mm/mprotect.c                            |   2 +-
 mm/rmap.c                                |   5 +-
 tools/testing/selftests/vm/hmm-tests.c   | 311 +++++++++++++++++++--
 tools/testing/selftests/vm/test_hmm.sh   |  24 +-
 30 files changed, 1005 insertions(+), 383 deletions(-)
 create mode 100644 include/linux/page_zone.h
Signature · 1 line
2.32.0
#1 lore raw

PATCH v8 1/15 mm: rename is_pinnable_pages to is_longterm_pinnable_pages

is_pinnable_page() and folio_is_pinnable() were renamed to is_longterm_pinnable_page() and folio_is_longterm_pinnable() respectively. These functions are used in the FOLL_LONGTERM flag context.

---
 include/linux/mm.h | 8 ++++----
 mm/gup.c           | 4 ++--
 mm/gup_test.c      | 2 +-
 mm/hugetlb.c       | 2 +-
 4 files changed, 8 insertions(+), 8 deletions(-)
include/linux/mm.h +4 -4
index cf3d0d673f6b..3b31b33bd5be 100644
@@ -1592,7 +1592,7 @@ static inline bool page_needs_cow_for_dma(struct vm_area_struct *vma,
 
 /* MIGRATE_CMA and ZONE_MOVABLE do not allow pin pages */
 #ifdef CONFIG_MIGRATION
-static inline bool is_pinnable_page(struct page *page)
+static inline bool is_longterm_pinnable_page(struct page *page)
 {
 #ifdef CONFIG_CMA
 	int mt = get_pageblock_migratetype(page);
@@ -1603,15 +1603,15 @@ static inline bool is_pinnable_page(struct page *page)
 	return !is_zone_movable_page(page) || is_zero_pfn(page_to_pfn(page));
 }
 #else
-static inline bool is_pinnable_page(struct page *page)
+static inline bool is_longterm_pinnable_page(struct page *page)
 {
 	return true;
 }
 #endif
 
-static inline bool folio_is_pinnable(struct folio *folio)
+static inline bool folio_is_longterm_pinnable(struct folio *folio)
 {
-	return is_pinnable_page(&folio->page);
+	return is_longterm_pinnable_page(&folio->page);
 }
 
 static inline void set_page_zone(struct page *page, enum zone_type zone)
mm/gup.c +2 -2
index 551264407624..b65fe8bf5af4 100644
@@ -133,7 +133,7 @@ struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags)
 		 * path.
 		 */
 		if (unlikely((flags & FOLL_LONGTERM) &&
-			     !is_pinnable_page(page)))
+			     !is_longterm_pinnable_page(page)))
 			return NULL;
 
 		/*
@@ -1891,7 +1891,7 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
-		if (folio_is_pinnable(folio))
+		if (folio_is_longterm_pinnable(folio))
 			continue;
 
 		/*
mm/gup_test.c +1 -1
index d974dec19e1c..12b0a91767d3 100644
@@ -53,7 +53,7 @@ static void verify_dma_pinned(unsigned int cmd, struct page **pages,
 				dump_page(page, "gup_test failure");
 				break;
 			} else if (cmd == PIN_LONGTERM_BENCHMARK &&
-				WARN(!is_pinnable_page(page),
+				WARN(!is_longterm_pinnable_page(page),
 				     "pages[%lu] is NOT pinnable but pinned\n",
 				     i)) {
 				dump_page(page, "gup_test failure");
mm/hugetlb.c +1 -1
index a57e1be41401..368fd33787b0 100644
@@ -1135,7 +1135,7 @@ static struct page *dequeue_huge_page_node_exact(struct hstate *h, int nid)
 
 	lockdep_assert_held(&hugetlb_lock);
 	list_for_each_entry(page, &h->hugepage_freelists[nid], lru) {
-		if (pin && !is_pinnable_page(page))
+		if (pin && !is_longterm_pinnable_page(page))
 			continue;
 
 		if (PageHWPoison(page))
Signature · 1 line
2.32.0
Reviewed-by on v8 1/15
#2 lore raw

Re: [PATCH v8 01/15] mm: rename is_pinnable_pages to is_longterm_pinnable_pages

Quoted 5 lines · Alex Sierra

is_pinnable_page() and folio_is_pinnable() were renamed to is_longterm_pinnable_page() and folio_is_longterm_pinnable() respectively. These functions are used in the FOLL_LONGTERM flag context.

Quoted 8 lines
---
 include/linux/mm.h | 8 ++++----
 mm/gup.c           | 4 ++--
 mm/gup_test.c      | 2 +-
 mm/hugetlb.c       | 2 +-
 4 files changed, 8 insertions(+), 8 deletions(-)
Signature · 3 lines
Thanks,

David / dhildenb
#3 lore raw

PATCH v8 2/15 mm: move page zone helpers into new header-specific file

In reply to Alex Sierra

[WHY]
Have a cleaner way to expose all page zone helpers in one header file, rather than split them between mm.h and memremap.h files.

---
 drivers/infiniband/core/rw.c      |   1 -
 drivers/nvme/target/io-cmd-bdev.c |   1 -
 include/linux/memremap.h          | 113 +----------------
 include/linux/mm.h                |  79 +-----------
 include/linux/page_zone.h         | 194 ++++++++++++++++++++++++++++++
 mm/memcontrol.c                   |   1 -
 6 files changed, 196 insertions(+), 193 deletions(-)
 create mode 100644 include/linux/page_zone.h
drivers/infiniband/core/rw.c +0 -1
index 4d98f931a13d..5a3bd41b331c 100644
@@ -2,7 +2,6 @@
 /*
  * Copyright (c) 2016 HGST, a Western Digital Company.
  */
-#include <linux/memremap.h>
 #include <linux/moduleparam.h>
 #include <linux/slab.h>
 #include <linux/pci-p2pdma.h>
drivers/nvme/target/io-cmd-bdev.c +0 -1
index 27a72504d31c..16a8b7665fe4 100644
@@ -6,7 +6,6 @@
 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 #include <linux/blkdev.h>
 #include <linux/blk-integrity.h>
-#include <linux/memremap.h>
 #include <linux/module.h>
 #include "nvmet.h"
 
include/linux/memremap.h +1 -112

138 lines not on this page · raw message

include/linux/mm.h +1 -78

91 lines not on this page · raw message

include/linux/page_zone.h new file +194 -0

194 lines not on this page · raw message

mm/memcontrol.c +0 -1
index 618c366a2f07..a2df2f193f06 100644
@@ -53,7 +53,6 @@
 #include <linux/fs.h>
 #include <linux/seq_file.h>
 #include <linux/vmpressure.h>
-#include <linux/memremap.h>
 #include <linux/mm_inline.h>
 #include <linux/swap_cgroup.h>
 #include <linux/cpu.h>
Signature · 1 line
2.32.0
#4 lore raw

Re: [PATCH v8 02/15] mm: move page zone helpers into new header-specific file

Quoted 2 lines · Alex Sierra

[WHY]
Have a cleaner way to expose all page zone helpers in one header

What exactly is a "page zone"? Do you mean a buddy zone as in include/linux/mmzone.h ?

Signature · 3 lines
Thanks,

David / dhildenb
maintainer
#5 lore raw

Re: [PATCH v8 02/15] mm: move page zone helpers into new header-specific file

Quoted 6 lines · David Hildenbrand
Quoted 2 lines · Alex Sierra

[WHY]
Have a cleaner way to expose all page zone helpers in one header

What exactly is a "page zone"? Do you mean a buddy zone as in include/linux/mmzone.h ?

Zone as in ZONE_DEVICE. Maybe we could extend mmzone.h instead of creating page_zone.h? That should work as long as it's OK to include mmzone.h in memremap.h.

Regards,
  Felix

#6 lore raw

Re: [PATCH v8 02/15] mm: move page zone helpers into new header-specific file

Quoted 8 lines · Felix Kuehling
Quoted 6 lines · David Hildenbrand
Quoted 2 lines · Alex Sierra

[WHY]
Have a cleaner way to expose all page zone helpers in one header

What exactly is a "page zone"? Do you mean a buddy zone as in include/linux/mmzone.h ?

Zone as in ZONE_DEVICE. Maybe we could extend mmzone.h instead of

Yes, I think so. And try moving as little as possible in this patch.

Quoted 2 lines

creating page_zone.h? That should work as long as it's OK to include mmzone.h in memremap.h.

I think so.

Signature · 3 lines
Thanks,

David / dhildenb
#7 lore raw

PATCH 1/1 mm: move page zone helpers from mm.h to mmzone.h

In reply to Felix Kuehling

[WHY]
It makes more sense to have these helpers in zone specific header file, rather than the generic mm.h

---
 include/linux/memremap.h |  2 +-
 include/linux/mm.h       | 78 ---------------------------------------
 include/linux/mmzone.h   | 80 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 81 insertions(+), 79 deletions(-)
include/linux/memremap.h +1 -1
index 8af304f6b504..77229165c914 100644
@@ -2,7 +2,7 @@
 #ifndef _LINUX_MEMREMAP_H_
 #define _LINUX_MEMREMAP_H_
 
-#include <linux/mm.h>
+#include <linux/mmzone.h>
 #include <linux/range.h>
 #include <linux/ioport.h>
 #include <linux/percpu-refcount.h>
include/linux/mm.h +0 -78

84 lines not on this page · raw message

include/linux/mmzone.h +80 -0

86 lines not on this page · raw message

Signature · 1 line
2.32.0
#8 lore raw

PATCH v8 3/15 mm: add zone device coherent type memory support

In reply to Alex Sierra

Device memory that is cache coherent from device and CPU point of view. This is used on platforms that have an advanced system bus (like CAPI or CXL). Any page of a process can be migrated to such memory. However, no one should be allowed to pin such memory so that it can always be evicted.

[hch: rebased ontop of the refcount changes,
removed is_dev_private_or_coherent_page]

---
 include/linux/mm.h        |  4 +++-
 include/linux/page_zone.h | 19 +++++++++++++++++++
 mm/memcontrol.c           |  7 ++++---
 mm/memory-failure.c       |  8 ++++++--
 mm/memremap.c             | 10 ++++++++++
 mm/migrate_device.c       | 16 +++++++---------
 mm/rmap.c                 |  5 +++--
 7 files changed, 52 insertions(+), 17 deletions(-)
include/linux/mm.h +3 -1
index e551616cd208..83902102ffe6 100644
@@ -1523,7 +1523,9 @@ static inline bool is_longterm_pinnable_page(struct page *page)
 	if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE)
 		return false;
 #endif
-	return !is_zone_movable_page(page) || is_zero_pfn(page_to_pfn(page));
+	return !(is_device_coherent_page(page) ||
+		 is_zone_movable_page(page) ||
+		 is_zero_pfn(page_to_pfn(page)));
 }
 #else
 static inline bool is_longterm_pinnable_page(struct page *page)
include/linux/page_zone.h +19 -0
index 2a7a347173ee..a5d90ef13394 100644
@@ -59,6 +59,13 @@
  * A more complete discussion of unaddressable memory may be found in
  * include/linux/hmm.h and Documentation/vm/hmm.rst.
  *
+ * MEMORY_DEVICE_COHERENT:
+ * Device memory that is cache coherent from device and CPU point of view. This
+ * is used on platforms that have an advanced system bus (like CAPI or CXL). A
+ * driver can hotplug the device memory using ZONE_DEVICE and with that memory
+ * type. Any page of a process can be migrated to such memory. However no one
+ * should be allowed to pin such memory so that it can always be evicted.
+ *
  * MEMORY_DEVICE_FS_DAX:
  * Host memory that has similar access semantics as System RAM i.e. DMA
  * coherent and supports page pinning. In support of coordinating page
@@ -79,6 +86,7 @@
 enum memory_type {
 	/* 0 is reserved to catch uninitialized type fields */
 	MEMORY_DEVICE_PRIVATE = 1,
+	MEMORY_DEVICE_COHERENT,
 	MEMORY_DEVICE_FS_DAX,
 	MEMORY_DEVICE_GENERIC,
 	MEMORY_DEVICE_PCI_P2PDMA,
@@ -191,4 +199,15 @@ static inline bool is_pci_p2pdma_page(const struct page *page)
 		page->pgmap->type == MEMORY_DEVICE_PCI_P2PDMA;
 }
 
+static inline bool is_device_coherent_page(const struct page *page)
+{
+	return is_zone_device_page(page) &&
+		page->pgmap->type == MEMORY_DEVICE_COHERENT;
+}
+
+static inline bool folio_is_device_coherent(const struct folio *folio)
+{
+	return is_device_coherent_page(&folio->page);
+}
+
 #endif /* _PAGE_ZONE_H_ */
mm/memcontrol.c +4 -3
index a2df2f193f06..a9853f2b0d9b 100644
@@ -5664,8 +5664,8 @@ static int mem_cgroup_move_account(struct page *page,
  *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
  *     target for charge migration. if @target is not NULL, the entry is stored
  *     in target->ent.
- *   3(MC_TARGET_DEVICE): like MC_TARGET_PAGE  but page is MEMORY_DEVICE_PRIVATE
- *     (so ZONE_DEVICE page and thus not on the lru).
+ *   3(MC_TARGET_DEVICE): like MC_TARGET_PAGE  but page is device memory and
+ *   thus not on the lru.
  *     For now we such page is charge like a regular page would be as for all
  *     intent and purposes it is just special memory taking the place of a
  *     regular page.
@@ -5703,7 +5703,8 @@ static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
 		 */
 		if (page_memcg(page) == mc.from) {
 			ret = MC_TARGET_PAGE;
-			if (is_device_private_page(page))
+			if (is_device_private_page(page) ||
+			    is_device_coherent_page(page))
 				ret = MC_TARGET_DEVICE;
 			if (target)
 				target->page = page;
mm/memory-failure.c +6 -2
index da39ec8afca8..79f175eeb190 100644
@@ -1685,12 +1685,16 @@ static int memory_failure_dev_pagemap(unsigned long pfn, int flags,
 		goto unlock;
 	}
 
-	if (pgmap->type == MEMORY_DEVICE_PRIVATE) {
+	switch (pgmap->type) {
+	case MEMORY_DEVICE_PRIVATE:
+	case MEMORY_DEVICE_COHERENT:
 		/*
-		 * TODO: Handle HMM pages which may need coordination
+		 * TODO: Handle device pages which may need coordination
 		 * with device-side memory.
 		 */
 		goto unlock;
+	default:
+		break;
 	}
 
 	/*
mm/memremap.c +10 -0
index b870a659eee6..0f8f08f8a991 100644
@@ -315,6 +315,16 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
 			return ERR_PTR(-EINVAL);
 		}
 		break;
+	case MEMORY_DEVICE_COHERENT:
+		if (!pgmap->ops->page_free) {
+			WARN(1, "Missing page_free method\n");
+			return ERR_PTR(-EINVAL);
+		}
+		if (!pgmap->owner) {
+			WARN(1, "Missing owner\n");
+			return ERR_PTR(-EINVAL);
+		}
+		break;
 	case MEMORY_DEVICE_FS_DAX:
 		if (IS_ENABLED(CONFIG_FS_DAX_LIMITED)) {
 			WARN(1, "File system DAX not supported\n");
mm/migrate_device.c +7 -9
index 5052093d0262..a4847ad65da3 100644
@@ -518,7 +518,7 @@ EXPORT_SYMBOL(migrate_vma_setup);
  *     handle_pte_fault()
  *       do_anonymous_page()
  * to map in an anonymous zero page but the struct page will be a ZONE_DEVICE
- * private page.
+ * private or coherent page.
  */
 static void migrate_vma_insert_page(struct migrate_vma *migrate,
 				    unsigned long addr,
@@ -594,11 +594,8 @@ static void migrate_vma_insert_page(struct migrate_vma *migrate,
 						page_to_pfn(page));
 		entry = swp_entry_to_pte(swp_entry);
 	} else {
-		/*
-		 * For now we only support migrating to un-addressable device
-		 * memory.
-		 */
-		if (is_zone_device_page(page)) {
+		if (is_zone_device_page(page) &&
+		    !is_device_coherent_page(page)) {
 			pr_warn_once("Unsupported ZONE_DEVICE page type.\n");
 			goto abort;
 		}
@@ -701,10 +698,11 @@ void migrate_vma_pages(struct migrate_vma *migrate)
 
 		mapping = page_mapping(page);
 
-		if (is_device_private_page(newpage)) {
+		if (is_device_private_page(newpage) ||
+		    is_device_coherent_page(newpage)) {
 			/*
-			 * For now only support private anonymous when migrating
-			 * to un-addressable device memory.
+			 * For now only support anonymous memory migrating to
+			 * device private or coherent memory.
 			 */
 			if (mapping) {
 				migrate->src[i] &= ~MIGRATE_PFN_MIGRATE;
mm/rmap.c +3 -2
index 5bcb334cd6f2..04fac1af870b 100644
@@ -1957,7 +1957,7 @@ static bool try_to_migrate_one(struct folio *folio, struct vm_area_struct *vma,
 		/* Update high watermark before we lower rss */
 		update_hiwater_rss(mm);
 
-		if (folio_is_zone_device(folio)) {
+		if (folio_is_device_private(folio)) {
 			unsigned long pfn = folio_pfn(folio);
 			swp_entry_t entry;
 			pte_t swp_pte;
@@ -2131,7 +2131,8 @@ void try_to_migrate(struct folio *folio, enum ttu_flags flags)
 					TTU_SYNC)))
 		return;
 
-	if (folio_is_zone_device(folio) && !folio_is_device_private(folio))
+	if (folio_is_zone_device(folio) &&
+	    (!folio_is_device_private(folio) && !folio_is_device_coherent(folio)))
 		return;
 
 	/*
Signature · 1 line
2.32.0
#9 lore raw

PATCH v8 4/15 mm: handling Non-LRU pages returned by vm_normal_pages

In reply to Alex Sierra

With DEVICE_COHERENT, we'll soon have vm_normal_pages() return device-managed anonymous pages that are not LRU pages. Although they behave like normal pages for purposes of mapping in CPU page, and for COW. They do not support LRU lists, NUMA migration or THP.

Callers to follow_page() currently don't expect ZONE_DEVICE pages, however, with DEVICE_COHERENT we might now return ZONE_DEVICE. Check for ZONE_DEVICE pages in applicable users of follow_page() as well.

---
 fs/proc/task_mmu.c |  2 +-
 mm/huge_memory.c   |  2 +-
 mm/khugepaged.c    |  9 ++++++---
 mm/ksm.c           |  6 +++---
 mm/madvise.c       |  4 ++--
 mm/memory.c        | 10 +++++++++-
 mm/mempolicy.c     |  2 +-
 mm/migrate.c       |  4 ++--
 mm/mlock.c         |  2 +-
 mm/mprotect.c      |  2 +-
 10 files changed, 27 insertions(+), 16 deletions(-)
fs/proc/task_mmu.c +1 -1
index 2d04e3470d4c..2dd8c8a66924 100644
@@ -1792,7 +1792,7 @@ static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
 		return NULL;
 
 	page = vm_normal_page(vma, addr, pte);
-	if (!page)
+	if (!page || is_zone_device_page(page))
 		return NULL;
 
 	if (PageReserved(page))
mm/huge_memory.c +1 -1
index 834f288b3769..c47e95b02244 100644
@@ -2910,7 +2910,7 @@ static int split_huge_pages_pid(int pid, unsigned long vaddr_start,
 
 		if (IS_ERR(page))
 			continue;
-		if (!page)
+		if (!page || is_zone_device_page(page))
 			continue;
 
 		if (!is_transparent_hugepage(page))
mm/khugepaged.c +6 -3
index 16be62d493cd..671ac7800e53 100644
@@ -618,7 +618,7 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 			goto out;
 		}
 		page = vm_normal_page(vma, address, pteval);
-		if (unlikely(!page)) {
+		if (unlikely(!page) || unlikely(is_zone_device_page(page))) {
 			result = SCAN_PAGE_NULL;
 			goto out;
 		}
@@ -1267,7 +1267,7 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
 			writable = true;
 
 		page = vm_normal_page(vma, _address, pteval);
-		if (unlikely(!page)) {
+		if (unlikely(!page) || unlikely(is_zone_device_page(page))) {
 			result = SCAN_PAGE_NULL;
 			goto out_unmap;
 		}
@@ -1479,7 +1479,8 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 			goto abort;
 
 		page = vm_normal_page(vma, addr, *pte);
-
+		if (WARN_ON_ONCE(page && is_zone_device_page(page)))
+			page = NULL;
 		/*
 		 * Note that uprobe, debugger, or MAP_PRIVATE may change the
 		 * page table, but the new page will not be a subpage of hpage.
@@ -1497,6 +1498,8 @@ void collapse_pte_mapped_thp(struct mm_struct *mm, unsigned long addr)
 		if (pte_none(*pte))
 			continue;
 		page = vm_normal_page(vma, addr, *pte);
+		if (WARN_ON_ONCE(page && is_zone_device_page(page)))
+			goto abort;
 		page_remove_rmap(page, vma, false);
 	}
 
mm/ksm.c +3 -3
index 54f78c9eecae..831b18a7a50b 100644
@@ -475,7 +475,7 @@ static int break_ksm(struct vm_area_struct *vma, unsigned long addr)
 		cond_resched();
 		page = follow_page(vma, addr,
 				FOLL_GET | FOLL_MIGRATION | FOLL_REMOTE);
-		if (IS_ERR_OR_NULL(page))
+		if (IS_ERR_OR_NULL(page) || is_zone_device_page(page))
 			break;
 		if (PageKsm(page))
 			ret = handle_mm_fault(vma, addr,
@@ -560,7 +560,7 @@ static struct page *get_mergeable_page(struct rmap_item *rmap_item)
 		goto out;
 
 	page = follow_page(vma, addr, FOLL_GET);
-	if (IS_ERR_OR_NULL(page))
+	if (IS_ERR_OR_NULL(page) || is_zone_device_page(page))
 		goto out;
 	if (PageAnon(page)) {
 		flush_anon_page(vma, page, addr);
@@ -2308,7 +2308,7 @@ static struct rmap_item *scan_get_next_rmap_item(struct page **page)
 			if (ksm_test_exit(mm))
 				break;
 			*page = follow_page(vma, ksm_scan.address, FOLL_GET);
-			if (IS_ERR_OR_NULL(*page)) {
+			if (IS_ERR_OR_NULL(*page) || is_zone_device_page(*page)) {
 				ksm_scan.address += PAGE_SIZE;
 				cond_resched();
 				continue;
mm/madvise.c +2 -2
index 0316bbc6441b..e252635fe935 100644
@@ -421,7 +421,7 @@ static int madvise_cold_or_pageout_pte_range(pmd_t *pmd,
 			continue;
 
 		page = vm_normal_page(vma, addr, ptent);
-		if (!page)
+		if (!page || is_zone_device_page(page))
 			continue;
 
 		/*
@@ -639,7 +639,7 @@ static int madvise_free_pte_range(pmd_t *pmd, unsigned long addr,
 		}
 
 		page = vm_normal_page(vma, addr, ptent);
-		if (!page)
+		if (!page || is_zone_device_page(page))
 			continue;
 
 		/*
mm/memory.c +9 -1
index 7a089145cad4..4affde09cdd1 100644
@@ -624,6 +624,14 @@ struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
 		if (is_zero_pfn(pfn))
 			return NULL;
 		if (pte_devmap(pte))
+		/*
+		 * NOTE: New users of ZONE_DEVICE will not set pte_devmap()
+		 * and will have refcounts incremented on their struct pages
+		 * when they are inserted into PTEs, thus they are safe to
+		 * return here. Legacy ZONE_DEVICE pages that set pte_devmap()
+		 * do not have refcounts. Example of legacy ZONE_DEVICE is
+		 * MEMORY_DEVICE_FS_DAX type in pmem or virtio_fs drivers.
+		 */
 			return NULL;
 
 		print_bad_pte(vma, addr, pte, NULL);
@@ -4693,7 +4701,7 @@ static vm_fault_t do_numa_page(struct vm_fault *vmf)
 	pte = pte_modify(old_pte, vma->vm_page_prot);
 
 	page = vm_normal_page(vma, vmf->address, pte);
-	if (!page)
+	if (!page || is_zone_device_page(page))
 		goto out_map;
 
 	/* TODO: handle PTE-mapped THP */
mm/mempolicy.c +1 -1
index d39b01fd52fe..abc26890fc95 100644
@@ -523,7 +523,7 @@ static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
 		if (!pte_present(*pte))
 			continue;
 		page = vm_normal_page(vma, addr, *pte);
-		if (!page)
+		if (!page || is_zone_device_page(page))
 			continue;
 		/*
 		 * vm_normal_page() filters out zero pages, but there might
mm/migrate.c +2 -2
index 6c1ea61f39d8..a98a219d12ab 100644
@@ -1620,7 +1620,7 @@ static int add_page_for_migration(struct mm_struct *mm, unsigned long addr,
 		goto out;
 
 	err = -ENOENT;
-	if (!page)
+	if (!page || is_zone_device_page(page))
 		goto out;
 
 	err = 0;
@@ -1810,7 +1810,7 @@ static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
 		if (IS_ERR(page))
 			goto set_status;
 
-		if (page) {
+		if (page && !is_zone_device_page(page)) {
 			err = page_to_nid(page);
 			put_page(page);
 		} else {
mm/mlock.c +1 -1
index 716caf851043..b14e929084cc 100644
@@ -333,7 +333,7 @@ static int mlock_pte_range(pmd_t *pmd, unsigned long addr,
 		if (!pte_present(*pte))
 			continue;
 		page = vm_normal_page(vma, addr, *pte);
-		if (!page)
+		if (!page || is_zone_device_page(page))
 			continue;
 		if (PageTransCompound(page))
 			continue;
mm/mprotect.c +1 -1
index ba5592655ee3..e034aae2a98b 100644
@@ -95,7 +95,7 @@ static unsigned long change_pte_range(struct mmu_gather *tlb,
 					continue;
 
 				page = vm_normal_page(vma, addr, oldpte);
-				if (!page || PageKsm(page))
+				if (!page || is_zone_device_page(page) || PageKsm(page))
 					continue;
 
 				/* Also skip shared copy-on-write pages */
Signature · 1 line
2.32.0
#10 lore raw

PATCH v8 5/15 mm: add device coherent vma selection for memory migration

In reply to Alex Sierra

This case is used to migrate pages from device memory, back to system memory. Device coherent type memory is cache coherent from device and CPU point of view.

---
 include/linux/migrate.h |  1 +
 mm/migrate_device.c     | 12 +++++++++---
 2 files changed, 10 insertions(+), 3 deletions(-)
include/linux/migrate.h +1 -0
index 069a89e847f3..b84908debe5c 100644
@@ -148,6 +148,7 @@ static inline unsigned long migrate_pfn(unsigned long pfn)
 enum migrate_vma_direction {
 	MIGRATE_VMA_SELECT_SYSTEM = 1 << 0,
 	MIGRATE_VMA_SELECT_DEVICE_PRIVATE = 1 << 1,
+	MIGRATE_VMA_SELECT_DEVICE_COHERENT = 1 << 2,
 };
 
 struct migrate_vma {
mm/migrate_device.c +9 -3
index a4847ad65da3..18bc6483f63a 100644
@@ -148,15 +148,21 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp,
 			if (is_writable_device_private_entry(entry))
 				mpfn |= MIGRATE_PFN_WRITE;
 		} else {
-			if (!(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM))
-				goto next;
 			pfn = pte_pfn(pte);
-			if (is_zero_pfn(pfn)) {
+			if (is_zero_pfn(pfn) &&
+			    (migrate->flags & MIGRATE_VMA_SELECT_SYSTEM)) {
 				mpfn = MIGRATE_PFN_MIGRATE;
 				migrate->cpages++;
 				goto next;
 			}
 			page = vm_normal_page(migrate->vma, addr, pte);
+			if (page && !is_zone_device_page(page) &&
+			    !(migrate->flags & MIGRATE_VMA_SELECT_SYSTEM))
+				goto next;
+			else if (page && is_device_coherent_page(page) &&
+			    (!(migrate->flags & MIGRATE_VMA_SELECT_DEVICE_COHERENT) ||
+			     page->pgmap->owner != migrate->pgmap_owner))
+				goto next;
 			mpfn = migrate_pfn(pfn) | MIGRATE_PFN_MIGRATE;
 			mpfn |= pte_write(pte) ? MIGRATE_PFN_WRITE : 0;
 		}
Signature · 1 line
2.32.0
#11 lore raw

PATCH v8 6/15 mm: remove the vma check in migrate_vma_setup()

In reply to Alex Sierra

From: Alistair Popple <[email protected]>

migrate_vma_setup() checks that a valid vma is passed so that the page tables can be walked to find the pfns associated with a given address range. However in some cases the pfns are already known, such as when migrating device coherent pages during pin_user_pages() meaning a valid vma isn't required.

---
 mm/migrate_device.c | 34 +++++++++++++++++-----------------
 1 file changed, 17 insertions(+), 17 deletions(-)
mm/migrate_device.c +17 -17
index 18bc6483f63a..cf9668376c5a 100644
@@ -486,24 +486,24 @@ int migrate_vma_setup(struct migrate_vma *args)
 
 	args->start &= PAGE_MASK;
 	args->end &= PAGE_MASK;
-	if (!args->vma || is_vm_hugetlb_page(args->vma) ||
-	    (args->vma->vm_flags & VM_SPECIAL) || vma_is_dax(args->vma))
-		return -EINVAL;
-	if (nr_pages <= 0)
-		return -EINVAL;
-	if (args->start < args->vma->vm_start ||
-	    args->start >= args->vma->vm_end)
-		return -EINVAL;
-	if (args->end <= args->vma->vm_start || args->end > args->vma->vm_end)
-		return -EINVAL;
 	if (!args->src || !args->dst)
 		return -EINVAL;
-
-	memset(args->src, 0, sizeof(*args->src) * nr_pages);
-	args->cpages = 0;
-	args->npages = 0;
-
-	migrate_vma_collect(args);
+	if (args->vma) {
+		if (is_vm_hugetlb_page(args->vma) ||
+		    (args->vma->vm_flags & VM_SPECIAL) || vma_is_dax(args->vma))
+			return -EINVAL;
+		if (args->start < args->vma->vm_start ||
+		    args->start >= args->vma->vm_end)
+			return -EINVAL;
+		if (args->end <= args->vma->vm_start ||
+		    args->end > args->vma->vm_end)
+			return -EINVAL;
+		memset(args->src, 0, sizeof(*args->src) * nr_pages);
+		args->cpages = 0;
+		args->npages = 0;
+
+		migrate_vma_collect(args);
+	}
 
 	if (args->cpages)
 		migrate_vma_unmap(args);
@@ -685,7 +685,7 @@ void migrate_vma_pages(struct migrate_vma *migrate)
 			continue;
 		}
 
-		if (!page) {
+		if (!page && migrate->vma) {
 			if (!(migrate->src[i] & MIGRATE_PFN_MIGRATE))
 				continue;
 			if (!notified) {
Signature · 1 line
2.32.0
#12 lore raw

Re: [PATCH v8 06/15] mm: remove the vma check in migrate_vma_setup()

Quoted 7 lines · Alex Sierra

From: Alistair Popple <[email protected]>

migrate_vma_setup() checks that a valid vma is passed so that the page tables can be walked to find the pfns associated with a given address range. However in some cases the pfns are already known, such as when migrating device coherent pages during pin_user_pages() meaning a valid vma isn't required.

As raised in my other reply, without a VMA ... it feels odd to use a "migrate_vma" API. For an internal (mm/migrate_device.c) use case it is ok I guess, but it certainly adds a bit of confusion. For example, because migrate_vma_setup() will undo ref+lock not obtained by it.

I guess the interesting point is that

a) Besides migrate_vma_pages() and migrate_vma_setup(), the ->vma is unused.

b) migrate_vma_setup() does collect+unmap+cleanup if unmap failed.

c) With our source page in our hands, we cannot be processing a hole in a VMA.

Not sure if it's better. but I would

a) Enforce in migrate_vma_setup() that there is a VMA. Code outside of mm/migrate_device.c shouldn't be doing some hacks like this.

b) Don't call migrate_vma_setup() from migrate_device_page(), but directly migrate_vma_unmap() and add a comment.

That will leave a single change to this patch (migrate_vma_pages()). But is that even required? Because ....

mm/migrate_device.c void migrate_vma_pages(struct migrate_vma *migrate) · line 688

Quoted 6 lines
mm/migrate_device.c +1 -1
@@ -685,7 +685,7 @@ void migrate_vma_pages(struct migrate_vma *migrate)
 			continue;
 		}
 
-		if (!page) {
+		if (!page && migrate->vma) {

How could we ever have !page in case of migrate_device_page()?

Instead, I think a VM_BUG_ON(migrate->vma); should hold and you can just simplify.

Quoted 3 lines
mm/migrate_device.c +0 -0
 			if (!(migrate->src[i] & MIGRATE_PFN_MIGRATE))
 				continue;
 			if (!notified) {
Signature · 3 lines
Thanks,

David / dhildenb
#13 lore raw

Re: [PATCH v8 06/15] mm: remove the vma check in migrate_vma_setup()

Quoted 37 lines · David Hildenbrand
Quoted 7 lines · Alex Sierra

From: Alistair Popple <[email protected]>

migrate_vma_setup() checks that a valid vma is passed so that the page tables can be walked to find the pfns associated with a given address range. However in some cases the pfns are already known, such as when migrating device coherent pages during pin_user_pages() meaning a valid vma isn't required.

As raised in my other reply, without a VMA ... it feels odd to use a "migrate_vma" API. For an internal (mm/migrate_device.c) use case it is ok I guess, but it certainly adds a bit of confusion. For example, because migrate_vma_setup() will undo ref+lock not obtained by it.

I guess the interesting point is that

a) Besides migrate_vma_pages() and migrate_vma_setup(), the ->vma is unused.

b) migrate_vma_setup() does collect+unmap+cleanup if unmap failed.

c) With our source page in our hands, we cannot be processing a hole in a VMA.

Not sure if it's better. but I would

a) Enforce in migrate_vma_setup() that there is a VMA. Code outside of mm/migrate_device.c shouldn't be doing some hacks like this.

b) Don't call migrate_vma_setup() from migrate_device_page(), but directly migrate_vma_unmap() and add a comment.

That will leave a single change to this patch (migrate_vma_pages()). But is that even required? Because ....

Quoted 8 lines

mm/migrate_device.c void migrate_vma_pages(struct migrate_vma *migrate) · line 688

Quoted 6 lines
mm/migrate_device.c +1 -1
@@ -685,7 +685,7 @@ void migrate_vma_pages(struct migrate_vma *migrate)
 			continue;
 		}
 
-		if (!page) {
+		if (!page && migrate->vma) {

How could we ever have !page in case of migrate_device_page()?

Oh good point. This patch was originally part of a larger series I was working on at the time but you're right - for migrate_device_page() we should never hit this case. I will respin the next patch (number 7 in this series) to include this.

Quoted 6 lines

Instead, I think a VM_BUG_ON(migrate->vma); should hold and you can just simplify.

Quoted 3 lines
mm/migrate_device.c +0 -0
 			if (!(migrate->src[i] & MIGRATE_PFN_MIGRATE))
 				continue;
 			if (!notified) {
#14 lore raw

PATCH v8 7/15 mm/gup: migrate device coherent pages when pinning instead of failing

In reply to Alex Sierra

From: Alistair Popple <[email protected]>

Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE.

[hch: rebased to the split device memory checks,
moved migrate_device_page to migrate_device.c]

---
 mm/gup.c            | 47 +++++++++++++++++++++++++++++++++++-----
 mm/internal.h       |  1 +
 mm/migrate_device.c | 53 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 96 insertions(+), 5 deletions(-)
mm/gup.c +42 -5
index b65fe8bf5af4..9b6b9923d22d 100644
@@ -1891,9 +1891,43 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
-		if (folio_is_longterm_pinnable(folio))
+		/*
+		 * Device private pages will get faulted in during gup so it
+		 * shouldn't be possible to see one here.
+		 */
+		if (WARN_ON_ONCE(folio_is_device_private(folio))) {
+			ret = -EFAULT;
+			goto unpin_pages;
+		}
+
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			WARN_ON_ONCE(PageCompound(&folio->page));
+
+			/*
+			 * Migration will fail if the page is pinned, so convert
+			 * the pin on the source page to a normal reference.
+			 */
+			if (gup_flags & FOLL_PIN) {
+				get_page(&folio->page);
+				unpin_user_page(&folio->page);
+			}
+
+			pages[i] = migrate_device_page(&folio->page, gup_flags);
+			if (!pages[i]) {
+				ret = -EBUSY;
+				goto unpin_pages;
+			}
 			continue;
+		}
 
+		if (folio_is_longterm_pinnable(folio))
+			continue;
 		/*
 		 * Try to move out any movable page before pinning the range.
 		 */
@@ -1929,10 +1963,13 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	return nr_pages;
 
 unpin_pages:
-	if (gup_flags & FOLL_PIN) {
-		unpin_user_pages(pages, nr_pages);
-	} else {
-		for (i = 0; i < nr_pages; i++)
+	for (i = 0; i < nr_pages; i++) {
+		if (!pages[i])
+			continue;
+
+		if (gup_flags & FOLL_PIN)
+			unpin_user_page(pages[i]);
+		else
 			put_page(pages[i]);
 	}
 
mm/internal.h +1 -0
index c0f8fbe0445b..eeab4ee7a4a3 100644
@@ -853,6 +853,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);
 
 void free_zone_device_page(struct page *page);
+struct page *migrate_device_page(struct page *page, unsigned int gup_flags);
 
 /*
  * mm/gup.c
mm/migrate_device.c +53 -0

56 lines not on this page · raw message

Signature · 1 line
2.32.0
#15 lore raw

Re: [PATCH v8 07/15] mm/gup: migrate device coherent pages when pinning instead of failing

Quoted 29 lines · Alex Sierra

From: Alistair Popple <[email protected]>

Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE.

[hch: rebased to the split device memory checks,
moved migrate_device_page to migrate_device.c]

---
 mm/gup.c            | 47 +++++++++++++++++++++++++++++++++++-----
 mm/internal.h       |  1 +
 mm/migrate_device.c | 53 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 96 insertions(+), 5 deletions(-)
mm/gup.c +0 -0
index b65fe8bf5af4..9b6b9923d22d 100644
@@ -1891,9 +1891,43 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;

mm/gup.c static long check_and_migrate_movable_pages(unsigned long nr_pages, · line 1901

Quoted 10 lines
mm/gup.c +8 -1
 
-		if (folio_is_longterm_pinnable(folio))
+		/*
+		 * Device private pages will get faulted in during gup so it
+		 * shouldn't be possible to see one here.
+		 */
+		if (WARN_ON_ONCE(folio_is_device_private(folio))) {
+			ret = -EFAULT;
+			goto unpin_pages;
+		}

I'd just drop that. Device private pages are never part of a present PTE. So if we could actually get a grab of one via GUP we would be in bigger trouble ... already before this patch.

mm/gup.c static long check_and_migrate_movable_pages(unsigned long nr_pages, · line 1910

Quoted 9 lines
mm/gup.c +9 -0
+
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			WARN_ON_ONCE(PageCompound(&folio->page));

Maybe that belongs into migrate_device_page()?

Quoted 3 lines
mm/gup.c +3 -0
+
+			/*
+			 * Migration will fail if the page is pinned, so convert

[...]

Quoted 7 lines

/*
* mm/gup.c

mm/migrate_device.c +0 -0
index cf9668376c5a..5decd26dd551 100644

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 803

Quoted 10 lines
mm/migrate_device.c +7 -0
@@ -794,3 +794,56 @@ void migrate_vma_finalize(struct migrate_vma *migrate)
 	}
 }
 EXPORT_SYMBOL(migrate_vma_finalize);
+
+/*
+ * Migrate a device coherent page back to normal memory.  The caller should have
+ * a reference on page which will be copied to the new page if migration is
+ * successful or dropped on failure.
+ */
+struct page *migrate_device_page(struct page *page, unsigned int gup_flags)

Function name should most probably indicate that we're dealing with coherent pages here?

Quoted 5 lines
mm/migrate_device.c +5 -0
+{
+	unsigned long src_pfn, dst_pfn = 0;
+	struct migrate_vma args;
+	struct page *dpage;
+

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 818

Quoted 10 lines
mm/migrate_device.c +10 -0
+	lock_page(page);
+	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
+	args.src = &src_pfn;
+	args.dst = &dst_pfn;
+	args.cpages = 1;
+	args.npages = 1;
+	args.vma = NULL;
+	migrate_vma_setup(&args);
+	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
+		return NULL;

Wow, these refcount and page locking/unlocking rules with this migrate_* api are confusing now. And the usage here of sometimes returning and sometimes falling trough don't make it particularly easier to understand here.

I'm not 100% happy about reusing migrate_vma_setup() usage if there *is no VMA*. That's just absolutely confusing, because usually migrate_vma_setup() itself would do the collection step and ref+lock pages. :/

In general, I can see why/how we're reusing the migrate_vma_* API here, but there is absolutely no VMA ... not sure what to improve besides providing a second API that does a simple single-page migration. But that can be changed later ...

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 821

Quoted 3 lines
mm/migrate_device.c +3 -0
+
+	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
+

alloc_page()

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 827

Quoted 6 lines
mm/migrate_device.c +6 -0
+	/*
+	 * get/pin the new page now so we don't have to retry gup after
+	 * migrating. We already have a reference so this should never fail.
+	 */
+	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
+		__free_pages(dpage, 0);

__free_page()

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 829

Quoted 2 lines
mm/migrate_device.c +2 -0
+		dpage = NULL;
+	}

Hm, this means that we are not pinning via the PTE at hand, but via something we expect migration to put into the PTE. I'm not really happy about this.

Ideally, we'd make the pinning decision only on the actual GUP path, not in here. Just like in the migrate_pages() case, where we end up dropping all refs/pins and looking up again via GUP from the PTE.

For example, I wonder if something nasty could happen if the PTE got mapped R/O in the meantime and you're pinning R/W here ...

TBH, all this special casing on gup_flags here is nasty. Please, let's just do it like migrate_pages() and do another GUP walk. Absolutely no need to optimize.

[...]

I'd go with something like the following on top (which does not touch on the general semantic issue with migrate_vma_* ). Note that I most probably messed up some refcount/lock handling and that it's broken.
Just to give you an idea what I think could be cleaner.

mm/gup.c +26 -23
index 9b6b9923d22d..17041b3e605e 100644
@@ -1881,7 +1881,7 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	unsigned long isolation_error_count = 0, i;
 	struct folio *prev_folio = NULL;
 	LIST_HEAD(movable_page_list);
-	bool drain_allow = true;
+	bool drain_allow = true, any_device_coherent = false;
 	int ret = 0;
 
 	for (i = 0; i < nr_pages; i++) {
@@ -1891,15 +1891,6 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
-		/*
-		 * Device private pages will get faulted in during gup so it
-		 * shouldn't be possible to see one here.
-		 */
-		if (WARN_ON_ONCE(folio_is_device_private(folio))) {
-			ret = -EFAULT;
-			goto unpin_pages;
-		}
-
 		/*
 		 * Device coherent pages are managed by a driver and should not
 		 * be pinned indefinitely as it prevents the driver moving the
@@ -1907,7 +1898,12 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 		 * to migrate the page out of device memory.
 		 */
 		if (folio_is_device_coherent(folio)) {
-			WARN_ON_ONCE(PageCompound(&folio->page));
+			/*
+			 * We always want a new GUP lookup with device coherent
+			 * pages.
+			 */
+			any_device_coherent = true;
+			pages[i] = 0;
 
 			/*
 			 * Migration will fail if the page is pinned, so convert
@@ -1918,11 +1914,12 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 				unpin_user_page(&folio->page);
 			}
 
-			pages[i] = migrate_device_page(&folio->page, gup_flags);
-			if (!pages[i]) {
-				ret = -EBUSY;
+			ret = migrate_device_coherent_page(&folio->page);
+			if (ret)
 				goto unpin_pages;
-			}
+			/* The reference to our folio is stale now. */
+			prev_folio = NULL;
+			folio = NULL;
 			continue;
 		}
 
@@ -1953,7 +1950,8 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 				    folio_nr_pages(folio));
 	}
 
-	if (!list_empty(&movable_page_list) || isolation_error_count)
+	if (!list_empty(&movable_page_list) || isolation_error_count ||
+	    any_device_coherent)
 		goto unpin_pages;
 
 	/*
@@ -1963,14 +1961,19 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	return nr_pages;
 
 unpin_pages:
-	for (i = 0; i < nr_pages; i++) {
-		if (!pages[i])
-			continue;
+	/* We have to be careful if we stumbled over device coherent pages. */
+	if (unlikely(any_device_coherent || !(gup_flags & FOLL_PIN))) {
+		for (i = 0; i < nr_pages; i++) {
+			if (!pages[i])
+				continue;
 
-		if (gup_flags & FOLL_PIN)
-			unpin_user_page(pages[i]);
-		else
-			put_page(pages[i]);
+			if (gup_flags & FOLL_PIN)
+				unpin_user_page(pages[i]);
+			else
+				put_page(pages[i]);
+		}
+	} else {
+		unpin_user_pages(pages, nr_pages);
 	}
 
 	if (!list_empty(&movable_page_list)) {
mm/internal.h +1 -1
index eeab4ee7a4a3..899dab512c5a 100644
@@ -853,7 +853,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);
 
 void free_zone_device_page(struct page *page);
-struct page *migrate_device_page(struct page *page, unsigned int gup_flags);
+int migrate_device_coherent_page(struct page *page);
 
 /*
  * mm/gup.c
mm/migrate_device.c +21 -34

74 lines not on this page · raw message

Signature · 8 lines
2.35.3



-- 
Thanks,

David / dhildenb
#16 lore raw

Re: [PATCH v8 07/15] mm/gup: migrate device coherent pages when pinning instead of failing

Quoted 10 lines · David Hildenbrand

Review

Quoted 8 lines
(quoted diff) +8 -0
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			WARN_ON_ONCE(PageCompound(&folio->page));

Maybe that belongs into migrate_device_page()?

... and should be simply WARN_ON_ONCE(folio_test_large(folio)). No need to go converting it back into a page in order to test things that can't possibly be true.

#17 lore raw

Re: [PATCH v8 07/15] mm/gup: migrate device coherent pages when pinning instead of failing

Quoted 12 lines · Matthew Wilcox
Quoted 10 lines · David Hildenbrand

Review

Quoted 8 lines
(quoted diff) +8 -0
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			WARN_ON_ONCE(PageCompound(&folio->page));

Maybe that belongs into migrate_device_page()?

Quoted 4 lines

... and should be simply WARN_ON_ONCE(folio_test_large(folio)). No need to go converting it back into a page in order to test things that can't possibly be true.

Good point.

Signature · 3 lines
Thanks,

David / dhildenb
#18 lore raw

PATCH 1/1 mm/gup: migrate device coherent pages when pinning instead of failing

Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE.

[hch: rebased to the split device memory checks,
moved migrate_device_page to migrate_device.c]

---

This patch hopefully addresses all of David's comments. It replaces both my "mm:
remove the vma check in migrate_vma_setup()" and "mm/gup: migrate device coherent pages when pinning instead of failing" patches. I'm not sure what the best way of including this is, perhaps Alex can respin the series with this patch instead?

- Alistair

 mm/gup.c            | 50 +++++++++++++++++++++++++++++++++++++------
 mm/internal.h       |  1 +
 mm/migrate_device.c | 52 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 96 insertions(+), 7 deletions(-)
mm/gup.c +43 -7
index b65fe8bf5af4..22b97ab61cd9 100644
@@ -1881,7 +1881,7 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	unsigned long isolation_error_count = 0, i;
 	struct folio *prev_folio = NULL;
 	LIST_HEAD(movable_page_list);
-	bool drain_allow = true;
+	bool drain_allow = true, coherent_pages = false;
 	int ret = 0;
 
 	for (i = 0; i < nr_pages; i++) {
@@ -1891,9 +1891,38 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
-		if (folio_is_longterm_pinnable(folio))
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			/*
+			 * We always want a new GUP lookup with device coherent
+			 * pages.
+			 */
+			pages[i] = 0;
+			coherent_pages = true;
+
+			/*
+			 * Migration will fail if the page is pinned, so convert
+			 * the pin on the source page to a normal reference.
+			 */
+			if (gup_flags & FOLL_PIN) {
+				get_page(&folio->page);
+				unpin_user_page(&folio->page);
+			}
+
+			ret = migrate_device_coherent_page(&folio->page);
+			if (ret)
+				goto unpin_pages;
+
 			continue;
+		}
 
+		if (folio_is_longterm_pinnable(folio))
+			continue;
 		/*
 		 * Try to move out any movable page before pinning the range.
 		 */
@@ -1919,7 +1948,8 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 				    folio_nr_pages(folio));
 	}
 
-	if (!list_empty(&movable_page_list) || isolation_error_count)
+	if (!list_empty(&movable_page_list) || isolation_error_count
+		|| coherent_pages)
 		goto unpin_pages;
 
 	/*
@@ -1929,10 +1959,16 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	return nr_pages;
 
 unpin_pages:
-	if (gup_flags & FOLL_PIN) {
-		unpin_user_pages(pages, nr_pages);
-	} else {
-		for (i = 0; i < nr_pages; i++)
+	/*
+	 * pages[i] might be NULL if any device coherent pages were found.
+	 */
+	for (i = 0; i < nr_pages; i++) {
+		if (!pages[i])
+			continue;
+
+		if (gup_flags & FOLL_PIN)
+			unpin_user_page(pages[i]);
+		else
 			put_page(pages[i]);
 	}
 
mm/internal.h +1 -0
index c0f8fbe0445b..899dab512c5a 100644
@@ -853,6 +853,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);
 
 void free_zone_device_page(struct page *page);
+int migrate_device_coherent_page(struct page *page);
 
 /*
  * mm/gup.c
mm/migrate_device.c +52 -0

61 lines not on this page · raw message

Signature · 1 line
2.35.1
#19 lore raw

Re: [PATCH] mm/gup: migrate device coherent pages when pinning instead of failing

Quoted 15 lines · Alistair Popple

Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE.

[hch: rebased to the split device memory checks,
moved migrate_device_page to migrate_device.c]

Quoted 7 lines

---

This patch hopefully addresses all of David's comments. It replaces both my "mm:
remove the vma check in migrate_vma_setup()" and "mm/gup: migrate device coherent pages when pinning instead of failing" patches. I'm not sure what the best way of including this is, perhaps Alex can respin the series with this patch instead?

For sure Alistair. I'll include this in my next patch series version.

Thanks,
Alex Sierra

Quoted 171 lines

- Alistair

  mm/gup.c            | 50 +++++++++++++++++++++++++++++++++++++------
  mm/internal.h       |  1 +
  mm/migrate_device.c | 52 +++++++++++++++++++++++++++++++++++++++++++++
  3 files changed, 96 insertions(+), 7 deletions(-)
mm/gup.c +43 -7
index b65fe8bf5af4..22b97ab61cd9 100644
@@ -1881,7 +1881,7 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
  	unsigned long isolation_error_count = 0, i;
  	struct folio *prev_folio = NULL;
  	LIST_HEAD(movable_page_list);
-	bool drain_allow = true;
+	bool drain_allow = true, coherent_pages = false;
  	int ret = 0;
 
  	for (i = 0; i < nr_pages; i++) {
@@ -1891,9 +1891,38 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
  			continue;
  		prev_folio = folio;
 
-		if (folio_is_longterm_pinnable(folio))
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			/*
+			 * We always want a new GUP lookup with device coherent
+			 * pages.
+			 */
+			pages[i] = 0;
+			coherent_pages = true;
+
+			/*
+			 * Migration will fail if the page is pinned, so convert
+			 * the pin on the source page to a normal reference.
+			 */
+			if (gup_flags & FOLL_PIN) {
+				get_page(&folio->page);
+				unpin_user_page(&folio->page);
+			}
+
+			ret = migrate_device_coherent_page(&folio->page);
+			if (ret)
+				goto unpin_pages;
+
  			continue;
+		}
 
+		if (folio_is_longterm_pinnable(folio))
+			continue;
  		/*
  		 * Try to move out any movable page before pinning the range.
  		 */
@@ -1919,7 +1948,8 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
  				    folio_nr_pages(folio));
  	}
 
-	if (!list_empty(&movable_page_list) || isolation_error_count)
+	if (!list_empty(&movable_page_list) || isolation_error_count
+		|| coherent_pages)
  		goto unpin_pages;
 
  	/*
@@ -1929,10 +1959,16 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
  	return nr_pages;
 
  unpin_pages:
-	if (gup_flags & FOLL_PIN) {
-		unpin_user_pages(pages, nr_pages);
-	} else {
-		for (i = 0; i < nr_pages; i++)
+	/*
+	 * pages[i] might be NULL if any device coherent pages were found.
+	 */
+	for (i = 0; i < nr_pages; i++) {
+		if (!pages[i])
+			continue;
+
+		if (gup_flags & FOLL_PIN)
+			unpin_user_page(pages[i]);
+		else
  			put_page(pages[i]);
  	}
 
mm/internal.h +1 -0
index c0f8fbe0445b..899dab512c5a 100644
@@ -853,6 +853,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
  		      unsigned long addr, int page_nid, int *flags);
 
  void free_zone_device_page(struct page *page);
+int migrate_device_coherent_page(struct page *page);
 
  /*
   * mm/gup.c
mm/migrate_device.c +52 -0
index 18bc6483f63a..7feeb447e3b9 100644
@@ -686,6 +686,12 @@ void migrate_vma_pages(struct migrate_vma *migrate)
  		}
 
  		if (!page) {
+			/*
+			 * The only time there is no vma is when called from
+			 * migrate_device_coherent_page(). However this isn't
+			 * called if the page could not be unmapped.
+			 */
+			VM_BUG_ON(!migrate->vma);
  			if (!(migrate->src[i] & MIGRATE_PFN_MIGRATE))
  				continue;
  			if (!notified) {
@@ -794,3 +800,49 @@ void migrate_vma_finalize(struct migrate_vma *migrate)
  	}
  }
  EXPORT_SYMBOL(migrate_vma_finalize);
+
+/*
+ * Migrate a device coherent page back to normal memory. The caller should have
+ * a reference on page which will be copied to the new page if migration is
+ * successful or dropped on failure.
+ */
+int migrate_device_coherent_page(struct page *page)
+{
+	unsigned long src_pfn, dst_pfn = 0;
+	struct migrate_vma args;
+	struct page *dpage;
+
+	WARN_ON_ONCE(PageCompound(page));
+
+	lock_page(page);
+	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
+	args.src = &src_pfn;
+	args.dst = &dst_pfn;
+	args.cpages = 1;
+	args.npages = 1;
+	args.vma = NULL;
+
+	/*
+	 * We don't have a VMA and don't need to walk the page tables to find
+	 * the source page. So call migrate_vma_unmap() directly to unmap the
+	 * page as migrate_vma_setup() will fail if args.vma == NULL.
+	 */
+	migrate_vma_unmap(&args);
+	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
+		return -EBUSY;
+
+	dpage = alloc_page(GFP_USER | __GFP_NOWARN);
+	if (dpage) {
+		lock_page(dpage);
+		dst_pfn = migrate_pfn(page_to_pfn(dpage));
+	}
+
+	migrate_vma_pages(&args);
+	if (src_pfn & MIGRATE_PFN_MIGRATE)
+		copy_highpage(dpage, page);
+	migrate_vma_finalize(&args);
+
+	if (src_pfn & MIGRATE_PFN_MIGRATE)
+		return 0;
+	return -EBUSY;
+}
#20 lore raw

Re: [PATCH v8 07/15] mm/gup: migrate device coherent pages when pinning instead of failing

In reply to David Hildenbrand
Quoted 31 lines · David Hildenbrand
Quoted 30 lines · Alex Sierra

From: Alistair Popple <[email protected]>

Currently any attempts to pin a device coherent page will fail. This is because device coherent pages need to be managed by a device driver, and pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are coherent and accessible from the CPU so can be migrated just like pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin them first try migrating them out of ZONE_DEVICE.

[hch: rebased to the split device memory checks,
moved migrate_device_page to migrate_device.c]

---
 mm/gup.c            | 47 +++++++++++++++++++++++++++++++++++-----
 mm/internal.h       |  1 +
 mm/migrate_device.c | 53 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 96 insertions(+), 5 deletions(-)
mm/gup.c +0 -0
index b65fe8bf5af4..9b6b9923d22d 100644
@@ -1891,9 +1891,43 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
Quoted 13 lines

mm/gup.c static long check_and_migrate_movable_pages(unsigned long nr_pages, · line 1901

Quoted 9 lines
mm/gup.c +8 -1
-		if (folio_is_longterm_pinnable(folio))
+		/*
+		 * Device private pages will get faulted in during gup so it
+		 * shouldn't be possible to see one here.
+		 */
+		if (WARN_ON_ONCE(folio_is_device_private(folio))) {
+			ret = -EFAULT;
+			goto unpin_pages;
+		}

I'd just drop that. Device private pages are never part of a present PTE. So if we could actually get a grab of one via GUP we would be in bigger trouble ... already before this patch.

Fair.

Quoted 11 lines

mm/gup.c static long check_and_migrate_movable_pages(unsigned long nr_pages, · line 1910

Quoted 9 lines
mm/gup.c +9 -0
+
+		/*
+		 * Device coherent pages are managed by a driver and should not
+		 * be pinned indefinitely as it prevents the driver moving the
+		 * page. So when trying to pin with FOLL_LONGTERM instead try
+		 * to migrate the page out of device memory.
+		 */
+		if (folio_is_device_coherent(folio)) {
+			WARN_ON_ONCE(PageCompound(&folio->page));

Maybe that belongs into migrate_device_page()?

Ok (noting Matthew's comment there as well).

Quoted 17 lines
Quoted 3 lines
mm/gup.c +3 -0
+
+			/*
+			 * Migration will fail if the page is pinned, so convert

[...]

Quoted 11 lines

/*
* mm/gup.c

mm/migrate_device.c +1 -0
index cf9668376c5a..5decd26dd551 100644
@@ -794,3 +794,56 @@ void migrate_vma_finalize(struct migrate_vma *migrate)
 	}
 }
 EXPORT_SYMBOL(migrate_vma_finalize);
+
Quoted 8 lines

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 803

Quoted 6 lines
mm/migrate_device.c +6 -0
+/*
+ * Migrate a device coherent page back to normal memory.  The caller should have
+ * a reference on page which will be copied to the new page if migration is
+ * successful or dropped on failure.
+ */
+struct page *migrate_device_page(struct page *page, unsigned int gup_flags)

Function name should most probably indicate that we're dealing with coherent pages here?

Ok.

Quoted 20 lines
Quoted 5 lines
mm/migrate_device.c +5 -0
+{
+	unsigned long src_pfn, dst_pfn = 0;
+	struct migrate_vma args;
+	struct page *dpage;
+

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 818

Quoted 10 lines
mm/migrate_device.c +10 -0
+	lock_page(page);
+	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
+	args.src = &src_pfn;
+	args.dst = &dst_pfn;
+	args.cpages = 1;
+	args.npages = 1;
+	args.vma = NULL;
+	migrate_vma_setup(&args);
+	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
+		return NULL;

Wow, these refcount and page locking/unlocking rules with this migrate_* api are confusing now. And the usage here of sometimes returning and sometimes falling trough don't make it particularly easier to understand here.

Quoted 7 lines

I'm not 100% happy about reusing migrate_vma_setup() usage if there *is no VMA*. That's just absolutely confusing, because usually migrate_vma_setup() itself would do the collection step and ref+lock pages. :/

In general, I can see why/how we're reusing the migrate_vma_* API here, but there is absolutely no VMA ... not sure what to improve besides providing a second API that does a simple single-page migration. But that can be changed later ...

Yeah, as noted in your other response I think it should be ok to just call migrate_vma_unmap() directly from migrate_device_page() so I assume that would adequately deal with this.

Quoted 25 lines

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 821

Quoted 3 lines
mm/migrate_device.c +3 -0
+
+	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
+

alloc_page()

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 827

Quoted 6 lines
mm/migrate_device.c +6 -0
+	/*
+	 * get/pin the new page now so we don't have to retry gup after
+	 * migrating. We already have a reference so this should never fail.
+	 */
+	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
+		__free_pages(dpage, 0);

__free_page()

mm/migrate_device.c void migrate_vma_finalize(struct migrate_vma *migrate) · line 829

Quoted 2 lines
mm/migrate_device.c +2 -0
+		dpage = NULL;
+	}

Hm, this means that we are not pinning via the PTE at hand, but via something we expect migration to put into the PTE. I'm not really happy about this.

Ideally, we'd make the pinning decision only on the actual GUP path, not in here. Just like in the migrate_pages() case, where we end up dropping all refs/pins and looking up again via GUP from the PTE.

Quoted 5 lines

For example, I wonder if something nasty could happen if the PTE got mapped R/O in the meantime and you're pinning R/W here ...

TBH, all this special casing on gup_flags here is nasty. Please, let's just do it like migrate_pages() and do another GUP walk. Absolutely no need to optimize.

The only reason to pass gup_flags is to check FOLL_PIN vs. FOLL_GET so that we can do the right reference on the destination page. I did the optimisation because we already have the destination page with a reference and GUP/PUP does not make any guarantees about the current PTE state anyway.

However I noticed there might be a race here - during migration we replace present PTEs with migration entries. On fork these get copied via copy_nonpresent_pte() and made read-only. However we don't check if the page a migration entry points to is pinned or not. For an ordinary PTE copy_present_pte() would copy the page for a COW mapping, but this won't happen if the page happens to be undergoing migration (even though the migration will ultimately fail due to the pin).

Anyway I don't think this patch currently makes that any worse, but if we fix the above it will because there is a brief period during which the page we're pinning won't look like a pinned page.

So I will go with the suggestion to do another GUP walk.

Quoted 8 lines

[...]

I'd go with something like the following on top (which does not touch on the general semantic issue with migrate_vma_* ). Note that I most probably messed up some refcount/lock handling and that it's broken.
Just to give you an idea what I think could be cleaner.

Thanks! At a glance it looks roughly right but I will check and respin it to incorporate the comments.

Quoted 191 lines
mm/gup.c +26 -23
index 9b6b9923d22d..17041b3e605e 100644
@@ -1881,7 +1881,7 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	unsigned long isolation_error_count = 0, i;
 	struct folio *prev_folio = NULL;
 	LIST_HEAD(movable_page_list);
-	bool drain_allow = true;
+	bool drain_allow = true, any_device_coherent = false;
 	int ret = 0;
 
 	for (i = 0; i < nr_pages; i++) {
@@ -1891,15 +1891,6 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 			continue;
 		prev_folio = folio;
 
-		/*
-		 * Device private pages will get faulted in during gup so it
-		 * shouldn't be possible to see one here.
-		 */
-		if (WARN_ON_ONCE(folio_is_device_private(folio))) {
-			ret = -EFAULT;
-			goto unpin_pages;
-		}
-
 		/*
 		 * Device coherent pages are managed by a driver and should not
 		 * be pinned indefinitely as it prevents the driver moving the
@@ -1907,7 +1898,12 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 		 * to migrate the page out of device memory.
 		 */
 		if (folio_is_device_coherent(folio)) {
-			WARN_ON_ONCE(PageCompound(&folio->page));
+			/*
+			 * We always want a new GUP lookup with device coherent
+			 * pages.
+			 */
+			any_device_coherent = true;
+			pages[i] = 0;
 
 			/*
 			 * Migration will fail if the page is pinned, so convert
@@ -1918,11 +1914,12 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 				unpin_user_page(&folio->page);
 			}
 
-			pages[i] = migrate_device_page(&folio->page, gup_flags);
-			if (!pages[i]) {
-				ret = -EBUSY;
+			ret = migrate_device_coherent_page(&folio->page);
+			if (ret)
 				goto unpin_pages;
-			}
+			/* The reference to our folio is stale now. */
+			prev_folio = NULL;
+			folio = NULL;
 			continue;
 		}
 
@@ -1953,7 +1950,8 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 				    folio_nr_pages(folio));
 	}
 
-	if (!list_empty(&movable_page_list) || isolation_error_count)
+	if (!list_empty(&movable_page_list) || isolation_error_count ||
+	    any_device_coherent)
 		goto unpin_pages;
 
 	/*
@@ -1963,14 +1961,19 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
 	return nr_pages;
 
 unpin_pages:
-	for (i = 0; i < nr_pages; i++) {
-		if (!pages[i])
-			continue;
+	/* We have to be careful if we stumbled over device coherent pages. */
+	if (unlikely(any_device_coherent || !(gup_flags & FOLL_PIN))) {
+		for (i = 0; i < nr_pages; i++) {
+			if (!pages[i])
+				continue;
 
-		if (gup_flags & FOLL_PIN)
-			unpin_user_page(pages[i]);
-		else
-			put_page(pages[i]);
+			if (gup_flags & FOLL_PIN)
+				unpin_user_page(pages[i]);
+			else
+				put_page(pages[i]);
+		}
+	} else {
+		unpin_user_pages(pages, nr_pages);
 	}
 
 	if (!list_empty(&movable_page_list)) {
mm/internal.h +1 -1
index eeab4ee7a4a3..899dab512c5a 100644
@@ -853,7 +853,7 @@ int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);
 
 void free_zone_device_page(struct page *page);
-struct page *migrate_device_page(struct page *page, unsigned int gup_flags);
+int migrate_device_coherent_page(struct page *page);
 
 /*
  * mm/gup.c
mm/migrate_device.c +21 -35
index 5decd26dd551..dfb78ea3d326 100644
@@ -797,53 +797,40 @@ EXPORT_SYMBOL(migrate_vma_finalize);
 
 /*
  * Migrate a device coherent page back to normal memory.  The caller should have
- * a reference on page which will be copied to the new page if migration is
- * successful or dropped on failure.
+ * a reference on page, which will be dropped on return.
  */
-struct page *migrate_device_page(struct page *page, unsigned int gup_flags)
+int migrate_device_coherent_page(struct page *page)
 {
 	unsigned long src_pfn, dst_pfn = 0;
-	struct migrate_vma args;
+	struct migrate_vma args = {
+		.src = &src_pfn,
+		.dst = &dst_pfn,
+		.cpages = 1,
+		.npages = 1,
+		.vma = NULL,
+	};
 	struct page *dpage;
 
+	VM_WARN_ON_ONCE(PageCompound(page));
+
 	lock_page(page);
 	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
-	args.src = &src_pfn;
-	args.dst = &dst_pfn;
-	args.cpages = 1;
-	args.npages = 1;
-	args.vma = NULL;
-	migrate_vma_setup(&args);
-	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
-		return NULL;
-
-	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
-
-	/*
-	 * get/pin the new page now so we don't have to retry gup after
-	 * migrating. We already have a reference so this should never fail.
-	 */
-	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
-		__free_pages(dpage, 0);
-		dpage = NULL;
-	}
 
-	if (dpage) {
-		lock_page(dpage);
-		dst_pfn = migrate_pfn(page_to_pfn(dpage));
+	migrate_vma_setup(&args);
+	if (src_pfn & MIGRATE_PFN_MIGRATE) {
+		dpage = alloc_page(GFP_USER | __GFP_NOWARN);
+		if (dpage) {
+			dst_pfn = migrate_pfn(page_to_pfn(dpage));
+			lock_page(dpage);
+		}
 	}
 
 	migrate_vma_pages(&args);
 	if (src_pfn & MIGRATE_PFN_MIGRATE)
 		copy_highpage(dpage, page);
 	migrate_vma_finalize(&args);
-	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
-		if (gup_flags & FOLL_PIN)
-			unpin_user_page(dpage);
-		else
-			put_page(dpage);
-		dpage = NULL;
-	}
 
-	return dpage;
+	if (src_pfn & MIGRATE_PFN_MIGRATE)
+		return 0;
+	return -EBUSY;
 }
--

2.35.3

#21 lore raw

PATCH v8 8/15 drm/amdkfd: add SPM support for SVM

In reply to Alex Sierra

When CPU is connected throug XGMI, it has coherent access to VRAM resource. In this case that resource is taken from a table in the device gmc aperture base. This resource is used along with the device type, which could be DEVICE_PRIVATE or DEVICE_COHERENT to create the device page map region.
Also, MIGRATE_VMA_SELECT_DEVICE_COHERENT flag is selected for coherent type case during migration to device.

---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 34 +++++++++++++++---------
 1 file changed, 21 insertions(+), 13 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +21 -13
index e44376c2ecdc..f73e3e340413 100644
@@ -671,13 +671,15 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
 	migrate.vma = vma;
 	migrate.start = start;
 	migrate.end = end;
-	migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
 	migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev);
+	if (adev->gmc.xgmi.connected_to_cpu)
+		migrate.flags = MIGRATE_VMA_SELECT_DEVICE_COHERENT;
+	else
+		migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
 
 	buf = kvcalloc(npages,
 		       2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t),
 		       GFP_KERNEL);
-
 	if (!buf)
 		goto out;
 
@@ -947,7 +949,7 @@ int svm_migrate_init(struct amdgpu_device *adev)
 {
 	struct kfd_dev *kfddev = adev->kfd.dev;
 	struct dev_pagemap *pgmap;
-	struct resource *res;
+	struct resource *res = NULL;
 	unsigned long size;
 	void *r;
 
@@ -962,28 +964,34 @@ int svm_migrate_init(struct amdgpu_device *adev)
 	 * should remove reserved size
 	 */
 	size = ALIGN(adev->gmc.real_vram_size, 2ULL << 20);
-	res = devm_request_free_mem_region(adev->dev, &iomem_resource, size);
-	if (IS_ERR(res))
-		return -ENOMEM;
+	if (adev->gmc.xgmi.connected_to_cpu) {
+		pgmap->range.start = adev->gmc.aper_base;
+		pgmap->range.end = adev->gmc.aper_base + adev->gmc.aper_size - 1;
+		pgmap->type = MEMORY_DEVICE_COHERENT;
+	} else {
+		res = devm_request_free_mem_region(adev->dev, &iomem_resource, size);
+		if (IS_ERR(res))
+			return -ENOMEM;
+		pgmap->range.start = res->start;
+		pgmap->range.end = res->end;
+		pgmap->type = MEMORY_DEVICE_PRIVATE;
+	}
 
-	pgmap->type = MEMORY_DEVICE_PRIVATE;
 	pgmap->nr_range = 1;
-	pgmap->range.start = res->start;
-	pgmap->range.end = res->end;
 	pgmap->ops = &svm_migrate_pgmap_ops;
 	pgmap->owner = SVM_ADEV_PGMAP_OWNER(adev);
-	pgmap->flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
-
+	pgmap->flags = 0;
 	/* Device manager releases device-specific resources, memory region and
 	 * pgmap when driver disconnects from device.
 	 */
 	r = devm_memremap_pages(adev->dev, pgmap);
 	if (IS_ERR(r)) {
 		pr_err("failed to register HMM device memory\n");
-
 		/* Disable SVM support capability */
 		pgmap->type = 0;
-		devm_release_mem_region(adev->dev, res->start, resource_size(res));
+		if (pgmap->type == MEMORY_DEVICE_PRIVATE)
+			devm_release_mem_region(adev->dev, res->start,
+						res->end - res->start + 1);
 		return PTR_ERR(r);
 	}
 
Signature · 1 line
2.32.0
#22 lore raw

PATCH v8 9/15 lib: test_hmm add ioctl to get zone device type

In reply to Alex Sierra

new ioctl cmd added to query zone device type. This will be used once the test_hmm adds zone device coherent type.

---
 lib/test_hmm.c      | 11 +++++++++--
 lib/test_hmm_uapi.h | 14 ++++++++++----
 2 files changed, 19 insertions(+), 6 deletions(-)
lib/test_hmm.c +9 -2
index cfe632047839..915ef6b5b0d4 100644
@@ -87,6 +87,7 @@ struct dmirror_chunk {
 struct dmirror_device {
 	struct cdev		cdevice;
 	struct hmm_devmem	*devmem;
+	unsigned int            zone_device_type;
 
 	unsigned int		devmem_capacity;
 	unsigned int		devmem_count;
@@ -1260,14 +1261,20 @@ static void dmirror_device_remove(struct dmirror_device *mdevice)
 static int __init hmm_dmirror_init(void)
 {
 	int ret;
-	int id;
+	int id = 0;
+	int ndevices = 0;
 
 	ret = alloc_chrdev_region(&dmirror_dev, 0, DMIRROR_NDEVICES,
 				  "HMM_DMIRROR");
 	if (ret)
 		goto err_unreg;
 
-	for (id = 0; id < DMIRROR_NDEVICES; id++) {
+	memset(dmirror_devices, 0, DMIRROR_NDEVICES * sizeof(dmirror_devices[0]));
+	dmirror_devices[ndevices++].zone_device_type =
+				HMM_DMIRROR_MEMORY_DEVICE_PRIVATE;
+	dmirror_devices[ndevices++].zone_device_type =
+				HMM_DMIRROR_MEMORY_DEVICE_PRIVATE;
+	for (id = 0; id < ndevices; id++) {
 		ret = dmirror_device_init(dmirror_devices + id, id);
 		if (ret)
 			goto err_chrdev;
lib/test_hmm_uapi.h +10 -4
index f14dea5dcd06..0511af7464ee 100644
@@ -31,10 +31,11 @@ struct hmm_dmirror_cmd {
 /* Expose the address space of the calling process through hmm device file */
 #define HMM_DMIRROR_READ		_IOWR('H', 0x00, struct hmm_dmirror_cmd)
 #define HMM_DMIRROR_WRITE		_IOWR('H', 0x01, struct hmm_dmirror_cmd)
-#define HMM_DMIRROR_MIGRATE		_IOWR('H', 0x02, struct hmm_dmirror_cmd)
-#define HMM_DMIRROR_SNAPSHOT		_IOWR('H', 0x03, struct hmm_dmirror_cmd)
-#define HMM_DMIRROR_EXCLUSIVE		_IOWR('H', 0x04, struct hmm_dmirror_cmd)
-#define HMM_DMIRROR_CHECK_EXCLUSIVE	_IOWR('H', 0x05, struct hmm_dmirror_cmd)
+#define HMM_DMIRROR_MIGRATE_TO_DEV	_IOWR('H', 0x02, struct hmm_dmirror_cmd)
+#define HMM_DMIRROR_MIGRATE_TO_SYS	_IOWR('H', 0x03, struct hmm_dmirror_cmd)
+#define HMM_DMIRROR_SNAPSHOT		_IOWR('H', 0x04, struct hmm_dmirror_cmd)
+#define HMM_DMIRROR_EXCLUSIVE		_IOWR('H', 0x05, struct hmm_dmirror_cmd)
+#define HMM_DMIRROR_CHECK_EXCLUSIVE	_IOWR('H', 0x06, struct hmm_dmirror_cmd)
 
 /*
  * Values returned in hmm_dmirror_cmd.ptr for HMM_DMIRROR_SNAPSHOT.
@@ -62,4 +63,9 @@ enum {
 	HMM_DMIRROR_PROT_DEV_PRIVATE_REMOTE	= 0x30,
 };
 
+enum {
+	/* 0 is reserved to catch uninitialized type fields */
+	HMM_DMIRROR_MEMORY_DEVICE_PRIVATE = 1,
+};
+
 #endif /* _LIB_TEST_HMM_UAPI_H */
Signature · 1 line
2.32.0
#23 lore raw

PATCH v8 10/15 lib: test_hmm add module param for zone device type

In reply to Alex Sierra

In order to configure device coherent in test_hmm, two module parameters should be passed, which correspond to the SP start address of each device (2) spm_addr_dev0 & spm_addr_dev1. If no parameters are passed, private device type is configured.

---
 lib/test_hmm.c      | 73 ++++++++++++++++++++++++++++++++-------------
 lib/test_hmm_uapi.h |  1 +
 2 files changed, 53 insertions(+), 21 deletions(-)
lib/test_hmm.c +52 -21
index 915ef6b5b0d4..afb30af9f3ff 100644
@@ -37,6 +37,16 @@
 #define DEVMEM_CHUNK_SIZE		(256 * 1024 * 1024U)
 #define DEVMEM_CHUNKS_RESERVE		16
 
+static unsigned long spm_addr_dev0;
+module_param(spm_addr_dev0, long, 0644);
+MODULE_PARM_DESC(spm_addr_dev0,
+		"Specify start address for SPM (special purpose memory) used for device 0. By setting this Coherent device type will be used. Make sure spm_addr_dev1 is set too. Minimum SPM size should be DEVMEM_CHUNK_SIZE.");
+
+static unsigned long spm_addr_dev1;
+module_param(spm_addr_dev1, long, 0644);
+MODULE_PARM_DESC(spm_addr_dev1,
+		"Specify start address for SPM (special purpose memory) used for device 1. By setting this Coherent device type will be used. Make sure spm_addr_dev0 is set too. Minimum SPM size should be DEVMEM_CHUNK_SIZE.");
+
 static const struct dev_pagemap_ops dmirror_devmem_ops;
 static const struct mmu_interval_notifier_ops dmirror_min_ops;
 static dev_t dmirror_dev;
@@ -455,28 +465,44 @@ static int dmirror_write(struct dmirror *dmirror, struct hmm_dmirror_cmd *cmd)
 	return ret;
 }
 
-static bool dmirror_allocate_chunk(struct dmirror_device *mdevice,
+static int dmirror_allocate_chunk(struct dmirror_device *mdevice,
 				   struct page **ppage)
 {
 	struct dmirror_chunk *devmem;
-	struct resource *res;
+	struct resource *res = NULL;
 	unsigned long pfn;
 	unsigned long pfn_first;
 	unsigned long pfn_last;
 	void *ptr;
+	int ret = -ENOMEM;
 
 	devmem = kzalloc(sizeof(*devmem), GFP_KERNEL);
 	if (!devmem)
-		return false;
+		return ret;
 
-	res = request_free_mem_region(&iomem_resource, DEVMEM_CHUNK_SIZE,
-				      "hmm_dmirror");
-	if (IS_ERR(res))
+	switch (mdevice->zone_device_type) {
+	case HMM_DMIRROR_MEMORY_DEVICE_PRIVATE:
+		res = request_free_mem_region(&iomem_resource, DEVMEM_CHUNK_SIZE,
+					      "hmm_dmirror");
+		if (IS_ERR_OR_NULL(res))
+			goto err_devmem;
+		devmem->pagemap.range.start = res->start;
+		devmem->pagemap.range.end = res->end;
+		devmem->pagemap.type = MEMORY_DEVICE_PRIVATE;
+		break;
+	case HMM_DMIRROR_MEMORY_DEVICE_COHERENT:
+		devmem->pagemap.range.start = (MINOR(mdevice->cdevice.dev) - 2) ?
+							spm_addr_dev0 :
+							spm_addr_dev1;
+		devmem->pagemap.range.end = devmem->pagemap.range.start +
+					    DEVMEM_CHUNK_SIZE - 1;
+		devmem->pagemap.type = MEMORY_DEVICE_COHERENT;
+		break;
+	default:
+		ret = -EINVAL;
 		goto err_devmem;
+	}
 
-	devmem->pagemap.type = MEMORY_DEVICE_PRIVATE;
-	devmem->pagemap.range.start = res->start;
-	devmem->pagemap.range.end = res->end;
 	devmem->pagemap.nr_range = 1;
 	devmem->pagemap.ops = &dmirror_devmem_ops;
 	devmem->pagemap.owner = mdevice;
@@ -497,10 +523,14 @@ static bool dmirror_allocate_chunk(struct dmirror_device *mdevice,
 		mdevice->devmem_capacity = new_capacity;
 		mdevice->devmem_chunks = new_chunks;
 	}
-
 	ptr = memremap_pages(&devmem->pagemap, numa_node_id());
-	if (IS_ERR(ptr))
+	if (IS_ERR_OR_NULL(ptr)) {
+		if (ptr)
+			ret = PTR_ERR(ptr);
+		else
+			ret = -EFAULT;
 		goto err_release;
+	}
 
 	devmem->mdevice = mdevice;
 	pfn_first = devmem->pagemap.range.start >> PAGE_SHIFT;
@@ -529,15 +559,17 @@ static bool dmirror_allocate_chunk(struct dmirror_device *mdevice,
 	}
 	spin_unlock(&mdevice->lock);
 
-	return true;
+	return 0;
 
 err_release:
 	mutex_unlock(&mdevice->devmem_lock);
-	release_mem_region(devmem->pagemap.range.start, range_len(&devmem->pagemap.range));
+	if (res && devmem->pagemap.type == MEMORY_DEVICE_PRIVATE)
+		release_mem_region(devmem->pagemap.range.start,
+				   range_len(&devmem->pagemap.range));
 err_devmem:
 	kfree(devmem);
 
-	return false;
+	return ret;
 }
 
 static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice)
@@ -562,7 +594,7 @@ static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice)
 		spin_unlock(&mdevice->lock);
 	} else {
 		spin_unlock(&mdevice->lock);
-		if (!dmirror_allocate_chunk(mdevice, &dpage))
+		if (dmirror_allocate_chunk(mdevice, &dpage))
 			goto error;
 	}
 
@@ -1232,10 +1264,8 @@ static int dmirror_device_init(struct dmirror_device *mdevice, int id)
 	if (ret)
 		return ret;
 
-	/* Build a list of free ZONE_DEVICE private struct pages */
-	dmirror_allocate_chunk(mdevice, NULL);
-
-	return 0;
+	/* Build a list of free ZONE_DEVICE struct pages */
+	return dmirror_allocate_chunk(mdevice, NULL);
 }
 
 static void dmirror_device_remove(struct dmirror_device *mdevice)
@@ -1248,8 +1278,9 @@ static void dmirror_device_remove(struct dmirror_device *mdevice)
 				mdevice->devmem_chunks[i];
 
 			memunmap_pages(&devmem->pagemap);
-			release_mem_region(devmem->pagemap.range.start,
-					   range_len(&devmem->pagemap.range));
+			if (devmem->pagemap.type == MEMORY_DEVICE_PRIVATE)
+				release_mem_region(devmem->pagemap.range.start,
+						   range_len(&devmem->pagemap.range));
 			kfree(devmem);
 		}
 		kfree(mdevice->devmem_chunks);
lib/test_hmm_uapi.h +1 -0
index 0511af7464ee..f700da7807c1 100644
@@ -66,6 +66,7 @@ enum {
 enum {
 	/* 0 is reserved to catch uninitialized type fields */
 	HMM_DMIRROR_MEMORY_DEVICE_PRIVATE = 1,
+	HMM_DMIRROR_MEMORY_DEVICE_COHERENT,
 };
 
 #endif /* _LIB_TEST_HMM_UAPI_H */
Signature · 1 line
2.32.0
#24 lore raw

PATCH v8 11/15 lib: add support for device coherent type in test_hmm

In reply to Alex Sierra

Device Coherent type uses device memory that is coherently accesible by the CPU. This could be shown as SP (special purpose) memory range at the BIOS-e820 memory enumeration. If no SP memory is supported in system, this could be faked by setting CONFIG_EFI_FAKE_MEMMAP.

Currently, test_hmm only supports two different SP ranges of at least 256MB size. This could be specified in the kernel parameter variable efi_fake_mem. Ex. Two SP ranges of 1GB starting at 0x100000000 & 0x140000000 physical address. Ex. efi_fake_mem=1G@0x100000000:0x40000,1G@0x140000000:0x40000

Private and coherent device mirror instances can be created in the same probed. This is done by passing the module parameters spm_addr_dev0 & spm_addr_dev1. In this case, it will create four instances of device_mirror. The first two correspond to private device type, the last two to coherent type. Then, they can be easily accessed from user space through /dev/hmm_mirror<num_device>. Usually num_device 0 and 1 are for private, and 2 and 3 for coherent types. If no module parameters are passed, two instances of private type device_mirror will be created only.

---
 lib/test_hmm.c      | 253 +++++++++++++++++++++++++++++++++-----------
 lib/test_hmm_uapi.h |   4 +
 2 files changed, 196 insertions(+), 61 deletions(-)
lib/test_hmm.c +192 -61
index afb30af9f3ff..7930853e7fc5 100644
@@ -32,11 +32,22 @@
 
 #include "test_hmm_uapi.h"
 
-#define DMIRROR_NDEVICES		2
+#define DMIRROR_NDEVICES		4
 #define DMIRROR_RANGE_FAULT_TIMEOUT	1000
 #define DEVMEM_CHUNK_SIZE		(256 * 1024 * 1024U)
 #define DEVMEM_CHUNKS_RESERVE		16
 
+/*
+ * For device_private pages, dpage is just a dummy struct page
+ * representing a piece of device memory. dmirror_devmem_alloc_page
+ * allocates a real system memory page as backing storage to fake a
+ * real device. zone_device_data points to that backing page. But
+ * for device_coherent memory, the struct page represents real
+ * physical CPU-accessible memory that we can use directly.
+ */
+#define BACKING_PAGE(page) (is_device_private_page((page)) ? \
+			   (page)->zone_device_data : (page))
+
 static unsigned long spm_addr_dev0;
 module_param(spm_addr_dev0, long, 0644);
 MODULE_PARM_DESC(spm_addr_dev0,
@@ -125,6 +136,21 @@ static int dmirror_bounce_init(struct dmirror_bounce *bounce,
 	return 0;
 }
 
+static bool dmirror_is_private_zone(struct dmirror_device *mdevice)
+{
+	return (mdevice->zone_device_type ==
+		HMM_DMIRROR_MEMORY_DEVICE_PRIVATE) ? true : false;
+}
+
+static enum migrate_vma_direction
+dmirror_select_device(struct dmirror *dmirror)
+{
+	return (dmirror->mdevice->zone_device_type ==
+		HMM_DMIRROR_MEMORY_DEVICE_PRIVATE) ?
+		MIGRATE_VMA_SELECT_DEVICE_PRIVATE :
+		MIGRATE_VMA_SELECT_DEVICE_COHERENT;
+}
+
 static void dmirror_bounce_fini(struct dmirror_bounce *bounce)
 {
 	vfree(bounce->ptr);
@@ -575,16 +601,19 @@ static int dmirror_allocate_chunk(struct dmirror_device *mdevice,
 static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice)
 {
 	struct page *dpage = NULL;
-	struct page *rpage;
+	struct page *rpage = NULL;
 
 	/*
-	 * This is a fake device so we alloc real system memory to store
-	 * our device memory.
+	 * For ZONE_DEVICE private type, this is a fake device so we allocate
+	 * real system memory to store our device memory.
+	 * For ZONE_DEVICE coherent type we use the actual dpage to store the
+	 * data and ignore rpage.
 	 */
-	rpage = alloc_page(GFP_HIGHUSER);
-	if (!rpage)
-		return NULL;
-
+	if (dmirror_is_private_zone(mdevice)) {
+		rpage = alloc_page(GFP_HIGHUSER);
+		if (!rpage)
+			return NULL;
+	}
 	spin_lock(&mdevice->lock);
 
 	if (mdevice->free_pages) {
@@ -603,7 +632,8 @@ static struct page *dmirror_devmem_alloc_page(struct dmirror_device *mdevice)
 	return dpage;
 
 error:
-	__free_page(rpage);
+	if (rpage)
+		__free_page(rpage);
 	return NULL;
 }
 
@@ -629,12 +659,16 @@ static void dmirror_migrate_alloc_and_copy(struct migrate_vma *args,
 		 * unallocated pte_none() or read-only zero page.
 		 */
 		spage = migrate_pfn_to_page(*src);
+		if (WARN(spage && is_zone_device_page(spage),
+		     "page already in device spage pfn: 0x%lx\n",
+		     page_to_pfn(spage)))
+			continue;
 
 		dpage = dmirror_devmem_alloc_page(mdevice);
 		if (!dpage)
 			continue;
 
-		rpage = dpage->zone_device_data;
+		rpage = BACKING_PAGE(dpage);
 		if (spage)
 			copy_highpage(rpage, spage);
 		else
@@ -648,6 +682,8 @@ static void dmirror_migrate_alloc_and_copy(struct migrate_vma *args,
 		 */
 		rpage->zone_device_data = dmirror;
 
+		pr_debug("migrating from sys to dev pfn src: 0x%lx pfn dst: 0x%lx\n",
+			 page_to_pfn(spage), page_to_pfn(dpage));
 		*dst = migrate_pfn(page_to_pfn(dpage));
 		if ((*src & MIGRATE_PFN_WRITE) ||
 		    (!spage && args->vma->vm_flags & VM_WRITE))
@@ -725,11 +761,7 @@ static int dmirror_migrate_finalize_and_map(struct migrate_vma *args,
 		if (!dpage)
 			continue;
 
-		/*
-		 * Store the page that holds the data so the page table
-		 * doesn't have to deal with ZONE_DEVICE private pages.
-		 */
-		entry = dpage->zone_device_data;
+		entry = BACKING_PAGE(dpage);
 		if (*dst & MIGRATE_PFN_WRITE)
 			entry = xa_tag_pointer(entry, DPT_XA_TAG_WRITE);
 		entry = xa_store(&dmirror->pt, pfn, entry, GFP_ATOMIC);
@@ -809,15 +841,126 @@ static int dmirror_exclusive(struct dmirror *dmirror,
 	return ret;
 }
 
-static int dmirror_migrate(struct dmirror *dmirror,
-			   struct hmm_dmirror_cmd *cmd)
+static vm_fault_t dmirror_devmem_fault_alloc_and_copy(struct migrate_vma *args,
+						      struct dmirror *dmirror)
+{
+	const unsigned long *src = args->src;
+	unsigned long *dst = args->dst;
+	unsigned long start = args->start;
+	unsigned long end = args->end;
+	unsigned long addr;
+
+	for (addr = start; addr < end; addr += PAGE_SIZE,
+				       src++, dst++) {
+		struct page *dpage, *spage;
+
+		spage = migrate_pfn_to_page(*src);
+		if (!spage || !(*src & MIGRATE_PFN_MIGRATE))
+			continue;
+
+		if (WARN_ON(!is_device_private_page(spage) &&
+			    !is_device_coherent_page(spage)))
+			continue;
+		spage = BACKING_PAGE(spage);
+		dpage = alloc_page_vma(GFP_HIGHUSER_MOVABLE, args->vma, addr);
+		if (!dpage)
+			continue;
+		pr_debug("migrating from dev to sys pfn src: 0x%lx pfn dst: 0x%lx\n",
+			 page_to_pfn(spage), page_to_pfn(dpage));
+
+		lock_page(dpage);
+		xa_erase(&dmirror->pt, addr >> PAGE_SHIFT);
+		copy_highpage(dpage, spage);
+		*dst = migrate_pfn(page_to_pfn(dpage));
+		if (*src & MIGRATE_PFN_WRITE)
+			*dst |= MIGRATE_PFN_WRITE;
+	}
+	return 0;
+}
+
+static unsigned long
+dmirror_successful_migrated_pages(struct migrate_vma *migrate)
+{
+	unsigned long cpages = 0;
+	unsigned long i;
+
+	for (i = 0; i < migrate->npages; i++) {
+		if (migrate->src[i] & MIGRATE_PFN_VALID &&
+		    migrate->src[i] & MIGRATE_PFN_MIGRATE)
+			cpages++;
+	}
+	return cpages;
+}
+
+static int dmirror_migrate_to_system(struct dmirror *dmirror,
+				     struct hmm_dmirror_cmd *cmd)
 {
 	unsigned long start, end, addr;
 	unsigned long size = cmd->npages << PAGE_SHIFT;
 	struct mm_struct *mm = dmirror->notifier.mm;
 	struct vm_area_struct *vma;
-	unsigned long src_pfns[64];
-	unsigned long dst_pfns[64];
+	unsigned long src_pfns[64] = { 0 };
+	unsigned long dst_pfns[64] = { 0 };
+	struct migrate_vma args;
+	unsigned long next;
+	int ret;
+
+	start = cmd->addr;
+	end = start + size;
+	if (end < start)
+		return -EINVAL;
+
+	/* Since the mm is for the mirrored process, get a reference first. */
+	if (!mmget_not_zero(mm))
+		return -EINVAL;
+
+	cmd->cpages = 0;
+	mmap_read_lock(mm);
+	for (addr = start; addr < end; addr = next) {
+		vma = vma_lookup(mm, addr);
+		if (!vma || !(vma->vm_flags & VM_READ)) {
+			ret = -EINVAL;
+			goto out;
+		}
+		next = min(end, addr + (ARRAY_SIZE(src_pfns) << PAGE_SHIFT));
+		if (next > vma->vm_end)
+			next = vma->vm_end;
+
+		args.vma = vma;
+		args.src = src_pfns;
+		args.dst = dst_pfns;
+		args.start = addr;
+		args.end = next;
+		args.pgmap_owner = dmirror->mdevice;
+		args.flags = dmirror_select_device(dmirror);
+
+		ret = migrate_vma_setup(&args);
+		if (ret)
+			goto out;
+
+		pr_debug("Migrating from device mem to sys mem\n");
+		dmirror_devmem_fault_alloc_and_copy(&args, dmirror);
+
+		migrate_vma_pages(&args);
+		cmd->cpages += dmirror_successful_migrated_pages(&args);
+		migrate_vma_finalize(&args);
+	}
+out:
+	mmap_read_unlock(mm);
+	mmput(mm);
+
+	return ret;
+}
+
+static int dmirror_migrate_to_device(struct dmirror *dmirror,
+				struct hmm_dmirror_cmd *cmd)
+{
+	unsigned long start, end, addr;
+	unsigned long size = cmd->npages << PAGE_SHIFT;
+	struct mm_struct *mm = dmirror->notifier.mm;
+	struct vm_area_struct *vma;
+	unsigned long src_pfns[64] = { 0 };
+	unsigned long dst_pfns[64] = { 0 };
 	struct dmirror_bounce bounce;
 	struct migrate_vma args;
 	unsigned long next;
@@ -854,6 +997,7 @@ static int dmirror_migrate(struct dmirror *dmirror,
 		if (ret)
 			goto out;
 
+		pr_debug("Migrating from sys mem to device mem\n");
 		dmirror_migrate_alloc_and_copy(&args, dmirror);
 		migrate_vma_pages(&args);
 		dmirror_migrate_finalize_and_map(&args, dmirror);
@@ -862,7 +1006,10 @@ static int dmirror_migrate(struct dmirror *dmirror,
 	mmap_read_unlock(mm);
 	mmput(mm);
 
-	/* Return the migrated data for verification. */
+	/*
+	 * Return the migrated data for verification.
+	 * Only for pages in device zone
+	 */
 	ret = dmirror_bounce_init(&bounce, start, size);
 	if (ret)
 		return ret;
@@ -905,6 +1052,12 @@ static void dmirror_mkentry(struct dmirror *dmirror, struct hmm_range *range,
 			*perm = HMM_DMIRROR_PROT_DEV_PRIVATE_LOCAL;
 		else
 			*perm = HMM_DMIRROR_PROT_DEV_PRIVATE_REMOTE;
+	} else if (is_device_coherent_page(page)) {
+		/* Is the page migrated to this device or some other? */
+		if (dmirror->mdevice == dmirror_page_to_device(page))
+			*perm = HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL;
+		else
+			*perm = HMM_DMIRROR_PROT_DEV_COHERENT_REMOTE;
 	} else if (is_zero_pfn(page_to_pfn(page)))
 		*perm = HMM_DMIRROR_PROT_ZERO;
 	else
@@ -1092,8 +1245,12 @@ static long dmirror_fops_unlocked_ioctl(struct file *filp,
 		ret = dmirror_write(dmirror, &cmd);
 		break;
 
-	case HMM_DMIRROR_MIGRATE:
-		ret = dmirror_migrate(dmirror, &cmd);
+	case HMM_DMIRROR_MIGRATE_TO_DEV:
+		ret = dmirror_migrate_to_device(dmirror, &cmd);
+		break;
+
+	case HMM_DMIRROR_MIGRATE_TO_SYS:
+		ret = dmirror_migrate_to_system(dmirror, &cmd);
 		break;
 
 	case HMM_DMIRROR_EXCLUSIVE:
@@ -1155,14 +1312,13 @@ static const struct file_operations dmirror_fops = {
 
 static void dmirror_devmem_free(struct page *page)
 {
-	struct page *rpage = page->zone_device_data;
+	struct page *rpage = BACKING_PAGE(page);
 	struct dmirror_device *mdevice;
 
-	if (rpage)
+	if (rpage != page)
 		__free_page(rpage);
 
 	mdevice = dmirror_page_to_device(page);
-
 	spin_lock(&mdevice->lock);
 	mdevice->cfree++;
 	page->zone_device_data = mdevice->free_pages;
@@ -1170,43 +1326,11 @@ static void dmirror_devmem_free(struct page *page)
 	spin_unlock(&mdevice->lock);
 }
 
-static vm_fault_t dmirror_devmem_fault_alloc_and_copy(struct migrate_vma *args,
-						      struct dmirror *dmirror)
-{
-	const unsigned long *src = args->src;
-	unsigned long *dst = args->dst;
-	unsigned long start = args->start;
-	unsigned long end = args->end;
-	unsigned long addr;
-
-	for (addr = start; addr < end; addr += PAGE_SIZE,
-				       src++, dst++) {
-		struct page *dpage, *spage;
-
-		spage = migrate_pfn_to_page(*src);
-		if (!spage || !(*src & MIGRATE_PFN_MIGRATE))
-			continue;
-		spage = spage->zone_device_data;
-
-		dpage = alloc_page_vma(GFP_HIGHUSER_MOVABLE, args->vma, addr);
-		if (!dpage)
-			continue;
-
-		lock_page(dpage);
-		xa_erase(&dmirror->pt, addr >> PAGE_SHIFT);
-		copy_highpage(dpage, spage);
-		*dst = migrate_pfn(page_to_pfn(dpage));
-		if (*src & MIGRATE_PFN_WRITE)
-			*dst |= MIGRATE_PFN_WRITE;
-	}
-	return 0;
-}
-
 static vm_fault_t dmirror_devmem_fault(struct vm_fault *vmf)
 {
 	struct migrate_vma args;
-	unsigned long src_pfns;
-	unsigned long dst_pfns;
+	unsigned long src_pfns = 0;
+	unsigned long dst_pfns = 0;
 	struct page *rpage;
 	struct dmirror *dmirror;
 	vm_fault_t ret;
@@ -1226,7 +1350,7 @@ static vm_fault_t dmirror_devmem_fault(struct vm_fault *vmf)
 	args.src = &src_pfns;
 	args.dst = &dst_pfns;
 	args.pgmap_owner = dmirror->mdevice;
-	args.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE;
+	args.flags = dmirror_select_device(dmirror);
 
 	if (migrate_vma_setup(&args))
 		return VM_FAULT_SIGBUS;
@@ -1305,6 +1429,12 @@ static int __init hmm_dmirror_init(void)
 				HMM_DMIRROR_MEMORY_DEVICE_PRIVATE;
 	dmirror_devices[ndevices++].zone_device_type =
 				HMM_DMIRROR_MEMORY_DEVICE_PRIVATE;
+	if (spm_addr_dev0 && spm_addr_dev1) {
+		dmirror_devices[ndevices++].zone_device_type =
+					HMM_DMIRROR_MEMORY_DEVICE_COHERENT;
+		dmirror_devices[ndevices++].zone_device_type =
+					HMM_DMIRROR_MEMORY_DEVICE_COHERENT;
+	}
 	for (id = 0; id < ndevices; id++) {
 		ret = dmirror_device_init(dmirror_devices + id, id);
 		if (ret)
@@ -1327,7 +1457,8 @@ static void __exit hmm_dmirror_exit(void)
 	int id;
 
 	for (id = 0; id < DMIRROR_NDEVICES; id++)
-		dmirror_device_remove(dmirror_devices + id);
+		if (dmirror_devices[id].zone_device_type)
+			dmirror_device_remove(dmirror_devices + id);
 	unregister_chrdev_region(dmirror_dev, DMIRROR_NDEVICES);
 }
 
lib/test_hmm_uapi.h +4 -0
index f700da7807c1..e31d58c9034a 100644
@@ -50,6 +50,8 @@ struct hmm_dmirror_cmd {
  *					device the ioctl() is made
  * HMM_DMIRROR_PROT_DEV_PRIVATE_REMOTE: Migrated device private page on some
  *					other device
+ * HMM_DMIRROR_PROT_DEV_COHERENT: Migrate device coherent page on the device
+ *				  the ioctl() is made
  */
 enum {
 	HMM_DMIRROR_PROT_ERROR			= 0xFF,
@@ -61,6 +63,8 @@ enum {
 	HMM_DMIRROR_PROT_ZERO			= 0x10,
 	HMM_DMIRROR_PROT_DEV_PRIVATE_LOCAL	= 0x20,
 	HMM_DMIRROR_PROT_DEV_PRIVATE_REMOTE	= 0x30,
+	HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL	= 0x40,
+	HMM_DMIRROR_PROT_DEV_COHERENT_REMOTE	= 0x50,
 };
 
 enum {
Signature · 1 line
2.32.0
#25 lore raw

PATCH v8 12/15 tools: update hmm-test to support device coherent type

In reply to Alex Sierra

Test cases such as migrate_fault and migrate_multiple, were modified to explicit migrate from device to sys memory without the need of page faults, when using device coherent type.

Snapshot test case updated to read memory device type first and based on that, get the proper returned results migrate_ping_pong test case added to test explicit migration from device to sys memory for both private and coherent zone types.

Helpers to migrate from device to sys memory and vicerversa were also added.

---
 tools/testing/selftests/vm/hmm-tests.c | 121 ++++++++++++++++++++-----
 1 file changed, 100 insertions(+), 21 deletions(-)
tools/testing/selftests/vm/hmm-tests.c +100 -21
index 203323967b50..4b547188ec40 100644
@@ -46,6 +46,13 @@ struct hmm_buffer {
 	uint64_t	faults;
 };
 
+enum {
+	HMM_PRIVATE_DEVICE_ONE,
+	HMM_PRIVATE_DEVICE_TWO,
+	HMM_COHERENCE_DEVICE_ONE,
+	HMM_COHERENCE_DEVICE_TWO,
+};
+
 #define TWOMEG		(1 << 21)
 #define HMM_BUFFER_SIZE (1024 << 12)
 #define HMM_PATH_MAX    64
@@ -60,6 +67,21 @@ FIXTURE(hmm)
 	unsigned int	page_shift;
 };
 
+FIXTURE_VARIANT(hmm)
+{
+	int     device_number;
+};
+
+FIXTURE_VARIANT_ADD(hmm, hmm_device_private)
+{
+	.device_number = HMM_PRIVATE_DEVICE_ONE,
+};
+
+FIXTURE_VARIANT_ADD(hmm, hmm_device_coherent)
+{
+	.device_number = HMM_COHERENCE_DEVICE_ONE,
+};
+
 FIXTURE(hmm2)
 {
 	int		fd0;
@@ -68,6 +90,24 @@ FIXTURE(hmm2)
 	unsigned int	page_shift;
 };
 
+FIXTURE_VARIANT(hmm2)
+{
+	int     device_number0;
+	int     device_number1;
+};
+
+FIXTURE_VARIANT_ADD(hmm2, hmm2_device_private)
+{
+	.device_number0 = HMM_PRIVATE_DEVICE_ONE,
+	.device_number1 = HMM_PRIVATE_DEVICE_TWO,
+};
+
+FIXTURE_VARIANT_ADD(hmm2, hmm2_device_coherent)
+{
+	.device_number0 = HMM_COHERENCE_DEVICE_ONE,
+	.device_number1 = HMM_COHERENCE_DEVICE_TWO,
+};
+
 static int hmm_open(int unit)
 {
 	char pathname[HMM_PATH_MAX];
@@ -81,12 +121,19 @@ static int hmm_open(int unit)
 	return fd;
 }
 
+static bool hmm_is_coherent_type(int dev_num)
+{
+	return (dev_num >= HMM_COHERENCE_DEVICE_ONE);
+}
+
 FIXTURE_SETUP(hmm)
 {
 	self->page_size = sysconf(_SC_PAGE_SIZE);
 	self->page_shift = ffs(self->page_size) - 1;
 
-	self->fd = hmm_open(0);
+	self->fd = hmm_open(variant->device_number);
+	if (self->fd < 0 && hmm_is_coherent_type(variant->device_number))
+		SKIP(exit(0), "DEVICE_COHERENT not available");
 	ASSERT_GE(self->fd, 0);
 }
 
@@ -95,9 +142,11 @@ FIXTURE_SETUP(hmm2)
 	self->page_size = sysconf(_SC_PAGE_SIZE);
 	self->page_shift = ffs(self->page_size) - 1;
 
-	self->fd0 = hmm_open(0);
+	self->fd0 = hmm_open(variant->device_number0);
+	if (self->fd0 < 0 && hmm_is_coherent_type(variant->device_number0))
+		SKIP(exit(0), "DEVICE_COHERENT not available");
 	ASSERT_GE(self->fd0, 0);
-	self->fd1 = hmm_open(1);
+	self->fd1 = hmm_open(variant->device_number1);
 	ASSERT_GE(self->fd1, 0);
 }
 
@@ -211,6 +260,20 @@ static void hmm_nanosleep(unsigned int n)
 	nanosleep(&t, NULL);
 }
 
+static int hmm_migrate_sys_to_dev(int fd,
+				   struct hmm_buffer *buffer,
+				   unsigned long npages)
+{
+	return hmm_dmirror_cmd(fd, HMM_DMIRROR_MIGRATE_TO_DEV, buffer, npages);
+}
+
+static int hmm_migrate_dev_to_sys(int fd,
+				   struct hmm_buffer *buffer,
+				   unsigned long npages)
+{
+	return hmm_dmirror_cmd(fd, HMM_DMIRROR_MIGRATE_TO_SYS, buffer, npages);
+}
+
 /*
  * Simple NULL test of device open/close.
  */
@@ -875,7 +938,7 @@ TEST_F(hmm, migrate)
 		ptr[i] = i;
 
 	/* Migrate memory to device. */
-	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages);
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, npages);
 
@@ -923,7 +986,7 @@ TEST_F(hmm, migrate_fault)
 		ptr[i] = i;
 
 	/* Migrate memory to device. */
-	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages);
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, npages);
 
@@ -936,7 +999,7 @@ TEST_F(hmm, migrate_fault)
 		ASSERT_EQ(ptr[i], i);
 
 	/* Migrate memory to the device again. */
-	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages);
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, npages);
 
@@ -976,7 +1039,7 @@ TEST_F(hmm, migrate_shared)
 	ASSERT_NE(buffer->ptr, MAP_FAILED);
 
 	/* Migrate memory to device. */
-	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer, npages);
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
 	ASSERT_EQ(ret, -ENOENT);
 
 	hmm_buffer_free(buffer);
@@ -1015,7 +1078,7 @@ TEST_F(hmm2, migrate_mixed)
 	p = buffer->ptr;
 
 	/* Migrating a protected area should be an error. */
-	ret = hmm_dmirror_cmd(self->fd1, HMM_DMIRROR_MIGRATE, buffer, npages);
+	ret = hmm_migrate_sys_to_dev(self->fd1, buffer, npages);
 	ASSERT_EQ(ret, -EINVAL);
 
 	/* Punch a hole after the first page address. */
@@ -1023,7 +1086,7 @@ TEST_F(hmm2, migrate_mixed)
 	ASSERT_EQ(ret, 0);
 
 	/* We expect an error if the vma doesn't cover the range. */
-	ret = hmm_dmirror_cmd(self->fd1, HMM_DMIRROR_MIGRATE, buffer, 3);
+	ret = hmm_migrate_sys_to_dev(self->fd1, buffer, 3);
 	ASSERT_EQ(ret, -EINVAL);
 
 	/* Page 2 will be a read-only zero page. */
@@ -1055,13 +1118,13 @@ TEST_F(hmm2, migrate_mixed)
 
 	/* Now try to migrate pages 2-5 to device 1. */
 	buffer->ptr = p + 2 * self->page_size;
-	ret = hmm_dmirror_cmd(self->fd1, HMM_DMIRROR_MIGRATE, buffer, 4);
+	ret = hmm_migrate_sys_to_dev(self->fd1, buffer, 4);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, 4);
 
 	/* Page 5 won't be migrated to device 0 because it's on device 1. */
 	buffer->ptr = p + 5 * self->page_size;
-	ret = hmm_dmirror_cmd(self->fd0, HMM_DMIRROR_MIGRATE, buffer, 1);
+	ret = hmm_migrate_sys_to_dev(self->fd0, buffer, 1);
 	ASSERT_EQ(ret, -ENOENT);
 	buffer->ptr = p;
 
@@ -1070,8 +1133,12 @@ TEST_F(hmm2, migrate_mixed)
 }
 
 /*
- * Migrate anonymous memory to device private memory and fault it back to system
- * memory multiple times.
+ * Migrate anonymous memory to device memory and back to system memory
+ * multiple times. In case of private zone configuration, this is done
+ * through fault pages accessed by CPU. In case of coherent zone configuration,
+ * the pages from the device should be explicitly migrated back to system memory.
+ * The reason is Coherent device zone has coherent access by CPU, therefore
+ * it will not generate any page fault.
  */
 TEST_F(hmm, migrate_multiple)
 {
@@ -1107,8 +1174,7 @@ TEST_F(hmm, migrate_multiple)
 			ptr[i] = i;
 
 		/* Migrate memory to device. */
-		ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_MIGRATE, buffer,
-				      npages);
+		ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
 		ASSERT_EQ(ret, 0);
 		ASSERT_EQ(buffer->cpages, npages);
 
@@ -1116,7 +1182,13 @@ TEST_F(hmm, migrate_multiple)
 		for (i = 0, ptr = buffer->mirror; i < size / sizeof(*ptr); ++i)
 			ASSERT_EQ(ptr[i], i);
 
-		/* Fault pages back to system memory and check them. */
+		/* Migrate back to system memory and check them. */
+		if (hmm_is_coherent_type(variant->device_number)) {
+			ret = hmm_migrate_dev_to_sys(self->fd, buffer, npages);
+			ASSERT_EQ(ret, 0);
+			ASSERT_EQ(buffer->cpages, npages);
+		}
+
 		for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i)
 			ASSERT_EQ(ptr[i], i);
 
@@ -1354,13 +1426,13 @@ TEST_F(hmm2, snapshot)
 
 	/* Page 5 will be migrated to device 0. */
 	buffer->ptr = p + 5 * self->page_size;
-	ret = hmm_dmirror_cmd(self->fd0, HMM_DMIRROR_MIGRATE, buffer, 1);
+	ret = hmm_migrate_sys_to_dev(self->fd0, buffer, 1);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, 1);
 
 	/* Page 6 will be migrated to device 1. */
 	buffer->ptr = p + 6 * self->page_size;
-	ret = hmm_dmirror_cmd(self->fd1, HMM_DMIRROR_MIGRATE, buffer, 1);
+	ret = hmm_migrate_sys_to_dev(self->fd1, buffer, 1);
 	ASSERT_EQ(ret, 0);
 	ASSERT_EQ(buffer->cpages, 1);
 
@@ -1377,9 +1449,16 @@ TEST_F(hmm2, snapshot)
 	ASSERT_EQ(m[2], HMM_DMIRROR_PROT_ZERO | HMM_DMIRROR_PROT_READ);
 	ASSERT_EQ(m[3], HMM_DMIRROR_PROT_READ);
 	ASSERT_EQ(m[4], HMM_DMIRROR_PROT_WRITE);
-	ASSERT_EQ(m[5], HMM_DMIRROR_PROT_DEV_PRIVATE_LOCAL |
-			HMM_DMIRROR_PROT_WRITE);
-	ASSERT_EQ(m[6], HMM_DMIRROR_PROT_NONE);
+	if (!hmm_is_coherent_type(variant->device_number0)) {
+		ASSERT_EQ(m[5], HMM_DMIRROR_PROT_DEV_PRIVATE_LOCAL |
+				HMM_DMIRROR_PROT_WRITE);
+		ASSERT_EQ(m[6], HMM_DMIRROR_PROT_NONE);
+	} else {
+		ASSERT_EQ(m[5], HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL |
+				HMM_DMIRROR_PROT_WRITE);
+		ASSERT_EQ(m[6], HMM_DMIRROR_PROT_DEV_COHERENT_REMOTE |
+				HMM_DMIRROR_PROT_WRITE);
+	}
 
 	hmm_buffer_free(buffer);
 }
Signature · 1 line
2.32.0
#26 lore raw

PATCH v8 13/15 tools: update test_hmm script to support SP config

In reply to Alex Sierra

Add two more parameters to set spm_addr_dev0 & spm_addr_dev1 addresses. These two parameters configure the start SP addresses for each device in test_hmm driver. Consequently, this configures zone device type as coherent.

---
 tools/testing/selftests/vm/test_hmm.sh | 24 +++++++++++++++++++++---
 1 file changed, 21 insertions(+), 3 deletions(-)
tools/testing/selftests/vm/test_hmm.sh +21 -3
index 0647b525a625..539c9371e592 100755
@@ -40,11 +40,26 @@ check_test_requirements()
 
 load_driver()
 {
-	modprobe $DRIVER > /dev/null 2>&1
+	if [ $# -eq 0 ]; then
+		modprobe $DRIVER > /dev/null 2>&1
+	else
+		if [ $# -eq 2 ]; then
+			modprobe $DRIVER spm_addr_dev0=$1 spm_addr_dev1=$2
+				> /dev/null 2>&1
+		else
+			echo "Missing module parameters. Make sure pass"\
+			"spm_addr_dev0 and spm_addr_dev1"
+			usage
+		fi
+	fi
 	if [ $? == 0 ]; then
 		major=$(awk "\$2==\"HMM_DMIRROR\" {print \$1}" /proc/devices)
 		mknod /dev/hmm_dmirror0 c $major 0
 		mknod /dev/hmm_dmirror1 c $major 1
+		if [ $# -eq 2 ]; then
+			mknod /dev/hmm_dmirror2 c $major 2
+			mknod /dev/hmm_dmirror3 c $major 3
+		fi
 	fi
 }
 
@@ -58,7 +73,7 @@ run_smoke()
 {
 	echo "Running smoke test. Note, this test provides basic coverage."
 
-	load_driver
+	load_driver $1 $2
 	$(dirname "${BASH_SOURCE[0]}")/hmm-tests
 	unload_driver
 }
@@ -75,6 +90,9 @@ usage()
 	echo "# Smoke testing"
 	echo "./${TEST_NAME}.sh smoke"
 	echo
+	echo "# Smoke testing with SPM enabled"
+	echo "./${TEST_NAME}.sh smoke <spm_addr_dev0> <spm_addr_dev1>"
+	echo
 	exit 0
 }
 
@@ -84,7 +102,7 @@ function run_test()
 		usage
 	else
 		if [ "$1" = "smoke" ]; then
-			run_smoke
+			run_smoke $2 $3
 		else
 			usage
 		fi
Signature · 1 line
2.32.0
#27 lore raw

PATCH v8 14/15 tools: add hmm gup tests for device coherent type

In reply to Alex Sierra

The intention is to test hmm device coherent type under different get user pages paths. Also, test gup with FOLL_LONGTERM flag set in device coherent pages. These pages should get migrated back to system memory.

---
 tools/testing/selftests/vm/hmm-tests.c | 110 +++++++++++++++++++++++++
 1 file changed, 110 insertions(+)
tools/testing/selftests/vm/hmm-tests.c +110 -0
index 4b547188ec40..bb38b9777610 100644
@@ -36,6 +36,7 @@
  * in the usual include/uapi/... directory.
  */
 #include "../../../../lib/test_hmm_uapi.h"
+#include "../../../../mm/gup_test.h"
 
 struct hmm_buffer {
 	void		*ptr;
@@ -59,6 +60,9 @@ enum {
 #define NTIMES		10
 
 #define ALIGN(x, a) (((x) + (a - 1)) & (~((a) - 1)))
+/* Just the flags we need, copied from mm.h: */
+#define FOLL_WRITE	0x01	/* check pte is writable */
+#define FOLL_LONGTERM   0x10000 /* mapping lifetime is indefinite */
 
 FIXTURE(hmm)
 {
@@ -1764,4 +1768,110 @@ TEST_F(hmm, exclusive_cow)
 	hmm_buffer_free(buffer);
 }
 
+static int gup_test_exec(int gup_fd, unsigned long addr, int cmd,
+			 int npages, int size, int flags)
+{
+	struct gup_test gup = {
+		.nr_pages_per_call	= npages,
+		.addr			= addr,
+		.gup_flags		= FOLL_WRITE | flags,
+		.size			= size,
+	};
+
+	if (ioctl(gup_fd, cmd, &gup)) {
+		perror("ioctl on error\n");
+		return errno;
+	}
+
+	return 0;
+}
+
+/*
+ * Test get user device pages through gup_test. Setting PIN_LONGTERM flag.
+ * This should trigger a migration back to system memory for both, private
+ * and coherent type pages.
+ * This test makes use of gup_test module. Make sure GUP_TEST_CONFIG is added
+ * to your configuration before you run it.
+ */
+TEST_F(hmm, hmm_gup_test)
+{
+	struct hmm_buffer *buffer;
+	int gup_fd;
+	unsigned long npages;
+	unsigned long size;
+	unsigned long i;
+	int *ptr;
+	int ret;
+	unsigned char *m;
+
+	gup_fd = open("/sys/kernel/debug/gup_test", O_RDWR);
+	if (gup_fd == -1)
+		SKIP(return, "Skipping test, could not find gup_test driver");
+
+	npages = 4;
+	size = npages << self->page_shift;
+
+	buffer = malloc(sizeof(*buffer));
+	ASSERT_NE(buffer, NULL);
+
+	buffer->fd = -1;
+	buffer->size = size;
+	buffer->mirror = malloc(size);
+	ASSERT_NE(buffer->mirror, NULL);
+
+	buffer->ptr = mmap(NULL, size,
+			   PROT_READ | PROT_WRITE,
+			   MAP_PRIVATE | MAP_ANONYMOUS,
+			   buffer->fd, 0);
+	ASSERT_NE(buffer->ptr, MAP_FAILED);
+
+	/* Initialize buffer in system memory. */
+	for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i)
+		ptr[i] = i;
+
+	/* Migrate memory to device. */
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
+	ASSERT_EQ(ret, 0);
+	ASSERT_EQ(buffer->cpages, npages);
+	/* Check what the device read. */
+	for (i = 0, ptr = buffer->mirror; i < size / sizeof(*ptr); ++i)
+		ASSERT_EQ(ptr[i], i);
+
+	ASSERT_EQ(gup_test_exec(gup_fd,
+				(unsigned long)buffer->ptr,
+				GUP_BASIC_TEST, 1, self->page_size, 0), 0);
+	ASSERT_EQ(gup_test_exec(gup_fd,
+				(unsigned long)buffer->ptr + 1 * self->page_size,
+				GUP_FAST_BENCHMARK, 1, self->page_size, 0), 0);
+	ASSERT_EQ(gup_test_exec(gup_fd,
+				(unsigned long)buffer->ptr + 2 * self->page_size,
+				PIN_FAST_BENCHMARK, 1, self->page_size, FOLL_LONGTERM), 0);
+	ASSERT_EQ(gup_test_exec(gup_fd,
+				(unsigned long)buffer->ptr + 3 * self->page_size,
+				PIN_LONGTERM_BENCHMARK, 1, self->page_size, 0), 0);
+
+	/* Take snapshot to CPU pagetables */
+	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_SNAPSHOT, buffer, npages);
+	ASSERT_EQ(ret, 0);
+	ASSERT_EQ(buffer->cpages, npages);
+	m = buffer->mirror;
+	if (hmm_is_coherent_type(variant->device_number)) {
+		ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | HMM_DMIRROR_PROT_WRITE, m[0]);
+		ASSERT_EQ(HMM_DMIRROR_PROT_DEV_COHERENT_LOCAL | HMM_DMIRROR_PROT_WRITE, m[1]);
+	} else {
+		ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[0]);
+		ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[1]);
+	}
+	ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[2]);
+	ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[3]);
+	/*
+	 * Check again the content on the pages. Make sure there's no
+	 * corrupted data.
+	 */
+	for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i)
+		ASSERT_EQ(ptr[i], i);
+
+	close(gup_fd);
+	hmm_buffer_free(buffer);
+}
 TEST_HARNESS_MAIN
Signature · 1 line
2.32.0
#28 lore raw

PATCH v8 15/15 tools: add selftests to hmm for COW in device memory

In reply to Alex Sierra

The objective is to test device migration mechanism in pages marked as COW, for private and coherent device type. In case of writing to COW private page(s), a page fault will migrate pages back to system memory first. Then, these pages will be duplicated. In case of COW device coherent type, pages are duplicated directly from device memory.

---
 tools/testing/selftests/vm/hmm-tests.c | 80 ++++++++++++++++++++++++++
 1 file changed, 80 insertions(+)
tools/testing/selftests/vm/hmm-tests.c +80 -0
index bb38b9777610..716b62c05e3d 100644
@@ -1874,4 +1874,84 @@ TEST_F(hmm, hmm_gup_test)
 	close(gup_fd);
 	hmm_buffer_free(buffer);
 }
+
+/*
+ * Test copy-on-write in device pages.
+ * In case of writing to COW private page(s), a page fault will migrate pages
+ * back to system memory first. Then, these pages will be duplicated. In case
+ * of COW device coherent type, pages are duplicated directly from device
+ * memory.
+ */
+TEST_F(hmm, hmm_cow_in_device)
+{
+	struct hmm_buffer *buffer;
+	unsigned long npages;
+	unsigned long size;
+	unsigned long i;
+	int *ptr;
+	int ret;
+	unsigned char *m;
+	pid_t pid;
+	int status;
+
+	npages = 4;
+	size = npages << self->page_shift;
+
+	buffer = malloc(sizeof(*buffer));
+	ASSERT_NE(buffer, NULL);
+
+	buffer->fd = -1;
+	buffer->size = size;
+	buffer->mirror = malloc(size);
+	ASSERT_NE(buffer->mirror, NULL);
+
+	buffer->ptr = mmap(NULL, size,
+			   PROT_READ | PROT_WRITE,
+			   MAP_PRIVATE | MAP_ANONYMOUS,
+			   buffer->fd, 0);
+	ASSERT_NE(buffer->ptr, MAP_FAILED);
+
+	/* Initialize buffer in system memory. */
+	for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i)
+		ptr[i] = i;
+
+	/* Migrate memory to device. */
+
+	ret = hmm_migrate_sys_to_dev(self->fd, buffer, npages);
+	ASSERT_EQ(ret, 0);
+	ASSERT_EQ(buffer->cpages, npages);
+
+	pid = fork();
+	if (pid == -1)
+		ASSERT_EQ(pid, 0);
+	if (!pid) {
+		/* Child process waitd for SIGTERM from the parent. */
+		while (1) {
+		}
+		perror("Should not reach this\n");
+		exit(0);
+	}
+	/* Parent process writes to COW pages(s) and gets a
+	 * new copy in system. In case of device private pages,
+	 * this write causes a migration to system mem first.
+	 */
+	for (i = 0, ptr = buffer->ptr; i < size / sizeof(*ptr); ++i)
+		ptr[i] = i;
+
+	/* Terminate child and wait */
+	EXPECT_EQ(0, kill(pid, SIGTERM));
+	EXPECT_EQ(pid, waitpid(pid, &status, 0));
+	EXPECT_NE(0, WIFSIGNALED(status));
+	EXPECT_EQ(SIGTERM, WTERMSIG(status));
+
+	/* Take snapshot to CPU pagetables */
+	ret = hmm_dmirror_cmd(self->fd, HMM_DMIRROR_SNAPSHOT, buffer, npages);
+	ASSERT_EQ(ret, 0);
+	ASSERT_EQ(buffer->cpages, npages);
+	m = buffer->mirror;
+	for (i = 0; i < npages; i++)
+		ASSERT_EQ(HMM_DMIRROR_PROT_WRITE, m[i]);
+
+	hmm_buffer_free(buffer);
+}
 TEST_HARNESS_MAIN
Signature · 1 line
2.32.0