diff mbox series

[RFC,V2,5/9] mm: add support for copy offload for folio Migration

Message ID 20250319192211.10092-6-shivankg@amd.com (mailing list archive)
State New
Headers show
Series Enhancements to Page Migration with Multi-threading and Batch Offloading to DMA | expand

Commit Message

Shivank Garg March 19, 2025, 7:22 p.m. UTC 
From: Mike Day <michael.day@amd.com>

Offload-Copy drivers should implement following functions to enable folio
migration offloading:
migrate_offc() - This function takes src and dst folios list undergoing
migration. It is responsible for transfer of page content between the
src and dst folios.
can_migrate_offc() - It performs necessary checks if offload copying
migration is supported for the give src and dst folios.

Offload-Copy driver should include a mechanism to call start_offloading and
stop_offloading for enabling and disabling migration offload respectively.

[Shivank: Rename the APIs and files to generalize the original DMA-specific
offload implementation to support various copy offloading mechanisms such as
DMA engines, CPU multi-threading, or other
hardware accelerators.]

Signed-off-by: Mike Day <michael.day@amd.com>
Signed-off-by: Shivank Garg <shivankg@amd.com>
---
 include/linux/migrate_offc.h | 36 +++++++++++++++++++++++++
 mm/Kconfig                   |  8 ++++++
 mm/Makefile                  |  1 +
 mm/migrate.c                 | 43 ++++++++++++++++++++++++++++--
 mm/migrate_offc.c            | 51 ++++++++++++++++++++++++++++++++++++
 5 files changed, 137 insertions(+), 2 deletions(-)
 create mode 100644 include/linux/migrate_offc.h
 create mode 100644 mm/migrate_offc.c
diff mbox series

Patch

diff --git a/include/linux/migrate_offc.h b/include/linux/migrate_offc.h
new file mode 100644
index 000000000000..908f81ebd621
--- /dev/null
+++ b/include/linux/migrate_offc.h
@@ -0,0 +1,36 @@ 
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef _MIGRATE_OFFC_H
+#define _MIGRATE_OFFC_H
+#include <linux/migrate_mode.h>
+
+#define MIGRATOR_NAME_LEN 32
+struct migrator {
+	char name[MIGRATOR_NAME_LEN];
+	int (*migrate_offc)(struct list_head *dst_list, struct list_head *src_list, int folio_cnt);
+	bool (*can_migrate_offc)(struct folio *dst, struct folio *src);
+	struct rcu_head srcu_head;
+	struct module *owner;
+};
+
+extern struct migrator migrator;
+extern struct mutex migrator_mut;
+extern struct srcu_struct mig_srcu;
+
+#ifdef CONFIG_OFFC_MIGRATION
+void srcu_mig_cb(struct rcu_head *head);
+void offc_update_migrator(struct migrator *mig);
+unsigned char *get_active_migrator_name(void);
+bool can_offc_migrate(struct folio *dst, struct folio *src);
+void start_offloading(struct migrator *migrator);
+void stop_offloading(void);
+#else
+static inline void srcu_mig_cb(struct rcu_head *head) { };
+static inline void offc_update_migrator(struct migrator *mig) { };
+static inline unsigned char *get_active_migrator_name(void) { return NULL; };
+static inline bool can_offc_migrate(struct folio *dst, struct folio *src) {return true; };
+static inline void start_offloading(struct migrator *migrator) { };
+static inline void stop_offloading(void) { };
+#endif /* CONFIG_OFFC_MIGRATION */
+
+#endif /* _MIGRATE_OFFC_H */
diff --git a/mm/Kconfig b/mm/Kconfig
index 1b501db06417..7a0693c3be4e 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -722,6 +722,14 @@  config MIGRATION
 config DEVICE_MIGRATION
 	def_bool MIGRATION && ZONE_DEVICE
 
+config OFFC_MIGRATION
+	bool "Migrate Pages offloading copy"
+	def_bool n
+	depends on MIGRATION
+	help
+	 An interface allowing external modules or driver to offload
+	 page copying in page migration.
+
 config ARCH_ENABLE_HUGEPAGE_MIGRATION
 	bool
 
diff --git a/mm/Makefile b/mm/Makefile
index 850386a67b3e..010142414176 100644
--- a/mm/Makefile
+++ b/mm/Makefile
@@ -93,6 +93,7 @@  obj-$(CONFIG_FAILSLAB) += failslab.o
 obj-$(CONFIG_FAIL_PAGE_ALLOC) += fail_page_alloc.o
 obj-$(CONFIG_MEMTEST)		+= memtest.o
 obj-$(CONFIG_MIGRATION) += migrate.o
+obj-$(CONFIG_OFFC_MIGRATION) += migrate_offc.o
 obj-$(CONFIG_NUMA) += memory-tiers.o
 obj-$(CONFIG_DEVICE_MIGRATION) += migrate_device.o
 obj-$(CONFIG_TRANSPARENT_HUGEPAGE) += huge_memory.o khugepaged.o
diff --git a/mm/migrate.c b/mm/migrate.c
index 8b6cfb60087c..862a3d1eff60 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -44,6 +44,7 @@ 
 #include <linux/sched/sysctl.h>
 #include <linux/memory-tiers.h>
 #include <linux/pagewalk.h>
+#include <linux/migrate_offc.h>
 
 #include <asm/tlbflush.h>
 
@@ -743,6 +744,37 @@  void folio_migrate_flags(struct folio *newfolio, struct folio *folio)
 }
 EXPORT_SYMBOL(folio_migrate_flags);
 
+DEFINE_STATIC_CALL(_folios_copy, folios_mc_copy);
+DEFINE_STATIC_CALL(_can_offc_migrate, can_offc_migrate);
+
+#ifdef CONFIG_OFFC_MIGRATION
+void srcu_mig_cb(struct rcu_head *head)
+{
+	static_call_query(_folios_copy);
+}
+
+void offc_update_migrator(struct migrator *mig)
+{
+	int index;
+
+	mutex_lock(&migrator_mut);
+	index = srcu_read_lock(&mig_srcu);
+	strscpy(migrator.name, mig ? mig->name : "kernel", MIGRATOR_NAME_LEN);
+	static_call_update(_folios_copy, mig ? mig->migrate_offc : folios_mc_copy);
+	static_call_update(_can_offc_migrate, mig ? mig->can_migrate_offc : can_offc_migrate);
+	if (READ_ONCE(migrator.owner))
+		module_put(migrator.owner);
+	xchg(&migrator.owner, mig ? mig->owner : NULL);
+	if (READ_ONCE(migrator.owner))
+		try_module_get(migrator.owner);
+	srcu_read_unlock(&mig_srcu, index);
+	mutex_unlock(&migrator_mut);
+	call_srcu(&mig_srcu, &migrator.srcu_head, srcu_mig_cb);
+	srcu_barrier(&mig_srcu);
+}
+
+#endif /* CONFIG_OFFC_MIGRATION */
+
 /************************************************************
  *                    Migration functions
  ***********************************************************/
@@ -1028,11 +1060,15 @@  static int _move_to_new_folio_prep(struct folio *dst, struct folio *src,
 {
 	int rc = -EAGAIN;
 	bool is_lru = !__folio_test_movable(src);
+	bool can_migrate;
 
 	VM_BUG_ON_FOLIO(!folio_test_locked(src), src);
 	VM_BUG_ON_FOLIO(!folio_test_locked(dst), dst);
 
-	if (likely(is_lru)) {
+	can_migrate = static_call(_can_offc_migrate)(dst, src);
+	if (unlikely(!can_migrate))
+		rc = -EAGAIN;
+	else if (likely(is_lru)) {
 		struct address_space *mapping = folio_mapping(src);
 
 		if (!mapping)
@@ -1868,7 +1904,10 @@  static void migrate_folios_batch_move(struct list_head *src_folios,
 		goto out;
 
 	/* Batch copy the folios */
-	rc = folios_mc_copy(dst_folios, src_folios, nr_batched_folios);
+	rc = static_call(_folios_copy)(dst_folios, src_folios, nr_batched_folios);
+	/* TODO:  Is there a better way of handling the poison
+	 * recover for batch copy and falling back to serial copy?
+	 */
 
 	/* TODO:  Is there a better way of handling the poison
 	 * recover for batch copy, instead of falling back to serial copy?
diff --git a/mm/migrate_offc.c b/mm/migrate_offc.c
new file mode 100644
index 000000000000..c632928a7c27
--- /dev/null
+++ b/mm/migrate_offc.c
@@ -0,0 +1,51 @@ 
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/migrate.h>
+#include <linux/migrate_offc.h>
+#include <linux/rculist.h>
+#include <linux/static_call.h>
+
+atomic_t dispatch_to_offc = ATOMIC_INIT(0);
+EXPORT_SYMBOL_GPL(dispatch_to_offc);
+
+DEFINE_MUTEX(migrator_mut);
+DEFINE_SRCU(mig_srcu);
+
+struct migrator migrator = {
+	.name = "kernel",
+	.migrate_offc = folios_mc_copy,
+	.can_migrate_offc = can_offc_migrate,
+	.srcu_head.func = srcu_mig_cb,
+	.owner = NULL,
+};
+
+bool can_offc_migrate(struct folio *dst, struct folio *src)
+{
+	return true;
+}
+EXPORT_SYMBOL_GPL(can_offc_migrate);
+
+void start_offloading(struct migrator *m)
+{
+	int offloading = 0;
+
+	pr_info("starting migration offload by %s\n", m->name);
+	offc_update_migrator(m);
+	atomic_try_cmpxchg(&dispatch_to_offc, &offloading, 1);
+}
+EXPORT_SYMBOL_GPL(start_offloading);
+
+void stop_offloading(void)
+{
+	int offloading = 1;
+
+	pr_info("stopping migration offload by %s\n", migrator.name);
+	offc_update_migrator(NULL);
+	atomic_try_cmpxchg(&dispatch_to_offc, &offloading, 0);
+}
+EXPORT_SYMBOL_GPL(stop_offloading);
+
+unsigned char *get_active_migrator_name(void)
+{
+	return migrator.name;
+}
+EXPORT_SYMBOL_GPL(get_active_migrator_name);