[PATCH v3 0/3] sched: make migrate_enable/migrate_disable inline

[PATCH v3 0/3] sched: make migrate_enable/migrate_disable inline

[Menglong Dong]

In this series, we make migrate_enable/migrate_disable inline to obtain
better performance in some case.

In the first patch, we add the macro "COMPILE_OFFSETS" to all the
asm-offset.c to avoid circular dependency in the 2nd patch.

In the 2nd patch, we generate the offset of nr_pinned in "struct rq" with
rq-offsets.c, as the "struct rq" is defined internally and we need to
access the "nr_pinned" field in migrate_enable and migrate_disable. Then,
we move the definition of migrate_enable/migrate_disable from
kernel/sched/core.c to include/linux/sched.h.

In the 3rd patch, we fix some typos in include/linux/preempt.h.

One of the beneficiaries of this series is BPF trampoline. Without this
series, the migrate_enable/migrate_disable is hot when we run the
benchmark for FENTRY, FEXIT, MODIFY_RETURN, etc:

  54.63% bpf_prog_2dcccf652aac1793_bench_trigger_fentry [k]
                 bpf_prog_2dcccf652aac1793_bench_trigger_fentry
  10.43% [kernel] [k] migrate_enable
  10.07% bpf_trampoline_6442517037 [k] bpf_trampoline_6442517037
  8.06% [kernel] [k] __bpf_prog_exit_recur
  4.11% libc.so.6 [.] syscall
  2.15% [kernel] [k] entry_SYSCALL_64
  1.48% [kernel] [k] memchr_inv
  1.32% [kernel] [k] fput
  1.16% [kernel] [k] _copy_to_user
  0.73% [kernel] [k] bpf_prog_test_run_raw_tp

Before this patch, the performance of BPF FENTRY is:

  fentry         :  113.030 ± 0.149M/s
  fentry         :  112.501 ± 0.187M/s
  fentry         :  112.828 ± 0.267M/s
  fentry         :  115.287 ± 0.241M/s

After this patch, the performance of BPF FENTRY increases to:

  fentry         :  143.644 ± 0.670M/s
  fentry         :  149.764 ± 0.362M/s
  fentry         :  149.642 ± 0.156M/s
  fentry         :  145.263 ± 0.221M/s

Changes since V2:
* some modification on the 2nd patch, as Peter advised:
  - don't export runqueues, define migrate_enable and migrate_disable in
    kernel/sched/core.c and use them for kernel modules instead
  - define the macro this_rq_pinned()
  - add some comment for this_rq_raw()

Changes since V1:
* use PERCPU_PTR() for this_rq_raw() if !CONFIG_SMP in the 2nd patch

Menglong Dong (3):
  arch: add the macro COMPILE_OFFSETS to all the asm-offsets.c
  sched: make migrate_enable/migrate_disable inline
  sched: fix some typos in include/linux/preempt.h

 Kbuild                               |  13 +++-
 arch/alpha/kernel/asm-offsets.c      |   1 +
 arch/arc/kernel/asm-offsets.c        |   1 +
 arch/arm/kernel/asm-offsets.c        |   2 +
 arch/arm64/kernel/asm-offsets.c      |   1 +
 arch/csky/kernel/asm-offsets.c       |   1 +
 arch/hexagon/kernel/asm-offsets.c    |   1 +
 arch/loongarch/kernel/asm-offsets.c  |   2 +
 arch/m68k/kernel/asm-offsets.c       |   1 +
 arch/microblaze/kernel/asm-offsets.c |   1 +
 arch/mips/kernel/asm-offsets.c       |   2 +
 arch/nios2/kernel/asm-offsets.c      |   1 +
 arch/openrisc/kernel/asm-offsets.c   |   1 +
 arch/parisc/kernel/asm-offsets.c     |   1 +
 arch/powerpc/kernel/asm-offsets.c    |   1 +
 arch/riscv/kernel/asm-offsets.c      |   1 +
 arch/s390/kernel/asm-offsets.c       |   1 +
 arch/sh/kernel/asm-offsets.c         |   1 +
 arch/sparc/kernel/asm-offsets.c      |   1 +
 arch/um/kernel/asm-offsets.c         |   2 +
 arch/xtensa/kernel/asm-offsets.c     |   1 +
 include/linux/preempt.h              |  11 +--
 include/linux/sched.h                | 106 +++++++++++++++++++++++++++
 kernel/bpf/verifier.c                |   1 +
 kernel/sched/core.c                  |  63 ++++------------
 kernel/sched/rq-offsets.c            |  12 +++
 26 files changed, 173 insertions(+), 57 deletions(-)
 create mode 100644 kernel/sched/rq-offsets.c

-- 
2.50.1

[PATCH v3 1/3] arch: add the macro COMPILE_OFFSETS to all the asm-offsets.c

[Menglong Dong]

The include/generated/asm-offsets.h is generated in Kbuild during
compiling from arch/SRCARCH/kernel/asm-offsets.c. When we want to
generate another similar offset header file, circular dependency can
happen.

For example, we want to generate a offset file include/generated/test.h,
which is included in include/sched/sched.h. If we generate asm-offsets.h
first, it will fail, as include/sched/sched.h is included in asm-offsets.c
and include/generated/test.h doesn't exist; If we generate test.h first,
it can't success neither, as include/generated/asm-offsets.h is included
by it.

In x86_64, the macro COMPILE_OFFSETS is used to avoid such circular
dependency. We can generate asm-offsets.h first, and if the
COMPILE_OFFSETS is defined, we don't include the "generated/test.h".

And we define the macro COMPILE_OFFSETS for all the asm-offsets.c for this
purpose.

Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
---
 arch/alpha/kernel/asm-offsets.c      | 1 +
 arch/arc/kernel/asm-offsets.c        | 1 +
 arch/arm/kernel/asm-offsets.c        | 2 ++
 arch/arm64/kernel/asm-offsets.c      | 1 +
 arch/csky/kernel/asm-offsets.c       | 1 +
 arch/hexagon/kernel/asm-offsets.c    | 1 +
 arch/loongarch/kernel/asm-offsets.c  | 2 ++
 arch/m68k/kernel/asm-offsets.c       | 1 +
 arch/microblaze/kernel/asm-offsets.c | 1 +
 arch/mips/kernel/asm-offsets.c       | 2 ++
 arch/nios2/kernel/asm-offsets.c      | 1 +
 arch/openrisc/kernel/asm-offsets.c   | 1 +
 arch/parisc/kernel/asm-offsets.c     | 1 +
 arch/powerpc/kernel/asm-offsets.c    | 1 +
 arch/riscv/kernel/asm-offsets.c      | 1 +
 arch/s390/kernel/asm-offsets.c       | 1 +
 arch/sh/kernel/asm-offsets.c         | 1 +
 arch/sparc/kernel/asm-offsets.c      | 1 +
 arch/um/kernel/asm-offsets.c         | 2 ++
 arch/xtensa/kernel/asm-offsets.c     | 1 +
 20 files changed, 24 insertions(+)

diff --git a/arch/alpha/kernel/asm-offsets.c b/arch/alpha/kernel/asm-offsets.c
index e9dad60b147f..1ebb05890499 100644
--- a/arch/alpha/kernel/asm-offsets.c
+++ b/arch/alpha/kernel/asm-offsets.c
@@ -4,6 +4,7 @@
  * This code generates raw asm output which is post-processed to extract
  * and format the required data.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/types.h>
 #include <linux/stddef.h>
diff --git a/arch/arc/kernel/asm-offsets.c b/arch/arc/kernel/asm-offsets.c
index f77deb799175..2978da85fcb6 100644
--- a/arch/arc/kernel/asm-offsets.c
+++ b/arch/arc/kernel/asm-offsets.c
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
  */
+#define COMPILE_OFFSETS
 
 #include <linux/sched.h>
 #include <linux/mm.h>
diff --git a/arch/arm/kernel/asm-offsets.c b/arch/arm/kernel/asm-offsets.c
index 123f4a8ef446..2101938d27fc 100644
--- a/arch/arm/kernel/asm-offsets.c
+++ b/arch/arm/kernel/asm-offsets.c
@@ -7,6 +7,8 @@
  * This code generates raw asm output which is post-processed to extract
  * and format the required data.
  */
+#define COMPILE_OFFSETS
+
 #include <linux/compiler.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
diff --git a/arch/arm64/kernel/asm-offsets.c b/arch/arm64/kernel/asm-offsets.c
index 30d4bbe68661..b6367ff3a49c 100644
--- a/arch/arm64/kernel/asm-offsets.c
+++ b/arch/arm64/kernel/asm-offsets.c
@@ -6,6 +6,7 @@
  *               2001-2002 Keith Owens
  * Copyright (C) 2012 ARM Ltd.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/arm_sdei.h>
 #include <linux/sched.h>
diff --git a/arch/csky/kernel/asm-offsets.c b/arch/csky/kernel/asm-offsets.c
index d1e903579473..5525c8e7e1d9 100644
--- a/arch/csky/kernel/asm-offsets.c
+++ b/arch/csky/kernel/asm-offsets.c
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
 // Copyright (C) 2018 Hangzhou C-SKY Microsystems co.,ltd.
+#define COMPILE_OFFSETS
 
 #include <linux/sched.h>
 #include <linux/kernel_stat.h>
diff --git a/arch/hexagon/kernel/asm-offsets.c b/arch/hexagon/kernel/asm-offsets.c
index 03a7063f9456..50eea9fa6f13 100644
--- a/arch/hexagon/kernel/asm-offsets.c
+++ b/arch/hexagon/kernel/asm-offsets.c
@@ -8,6 +8,7 @@
  *
  * Copyright (c) 2010-2012, The Linux Foundation. All rights reserved.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/compat.h>
 #include <linux/types.h>
diff --git a/arch/loongarch/kernel/asm-offsets.c b/arch/loongarch/kernel/asm-offsets.c
index db1e4bb26b6a..3017c7157600 100644
--- a/arch/loongarch/kernel/asm-offsets.c
+++ b/arch/loongarch/kernel/asm-offsets.c
@@ -4,6 +4,8 @@
  *
  * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
  */
+#define COMPILE_OFFSETS
+
 #include <linux/types.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
diff --git a/arch/m68k/kernel/asm-offsets.c b/arch/m68k/kernel/asm-offsets.c
index 906d73230537..67a1990f9d74 100644
--- a/arch/m68k/kernel/asm-offsets.c
+++ b/arch/m68k/kernel/asm-offsets.c
@@ -9,6 +9,7 @@
  * #defines from the assembly-language output.
  */
 
+#define COMPILE_OFFSETS
 #define ASM_OFFSETS_C
 
 #include <linux/stddef.h>
diff --git a/arch/microblaze/kernel/asm-offsets.c b/arch/microblaze/kernel/asm-offsets.c
index 104c3ac5f30c..b4b67d58e7f6 100644
--- a/arch/microblaze/kernel/asm-offsets.c
+++ b/arch/microblaze/kernel/asm-offsets.c
@@ -7,6 +7,7 @@
  * License. See the file "COPYING" in the main directory of this archive
  * for more details.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/init.h>
 #include <linux/stddef.h>
diff --git a/arch/mips/kernel/asm-offsets.c b/arch/mips/kernel/asm-offsets.c
index 1e29efcba46e..5debd9a3854a 100644
--- a/arch/mips/kernel/asm-offsets.c
+++ b/arch/mips/kernel/asm-offsets.c
@@ -9,6 +9,8 @@
  * Kevin Kissell, kevink@mips.com and Carsten Langgaard, carstenl@mips.com
  * Copyright (C) 2000 MIPS Technologies, Inc.
  */
+#define COMPILE_OFFSETS
+
 #include <linux/compat.h>
 #include <linux/types.h>
 #include <linux/sched.h>
diff --git a/arch/nios2/kernel/asm-offsets.c b/arch/nios2/kernel/asm-offsets.c
index e3d9b7b6fb48..88190b503ce5 100644
--- a/arch/nios2/kernel/asm-offsets.c
+++ b/arch/nios2/kernel/asm-offsets.c
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2011 Tobias Klauser <tklauser@distanz.ch>
  */
+#define COMPILE_OFFSETS
 
 #include <linux/stddef.h>
 #include <linux/sched.h>
diff --git a/arch/openrisc/kernel/asm-offsets.c b/arch/openrisc/kernel/asm-offsets.c
index 710651d5aaae..3cc826f2216b 100644
--- a/arch/openrisc/kernel/asm-offsets.c
+++ b/arch/openrisc/kernel/asm-offsets.c
@@ -18,6 +18,7 @@
  * compile this file to assembler, and then extract the
  * #defines from the assembly-language output.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/signal.h>
 #include <linux/sched.h>
diff --git a/arch/parisc/kernel/asm-offsets.c b/arch/parisc/kernel/asm-offsets.c
index 757816a7bd4b..9abfe65492c6 100644
--- a/arch/parisc/kernel/asm-offsets.c
+++ b/arch/parisc/kernel/asm-offsets.c
@@ -13,6 +13,7 @@
  *    Copyright (C) 2002 Randolph Chung <tausq with parisc-linux.org>
  *    Copyright (C) 2003 James Bottomley <jejb at parisc-linux.org>
  */
+#define COMPILE_OFFSETS
 
 #include <linux/types.h>
 #include <linux/sched.h>
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index b3048f6d3822..a4bc80b30410 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -8,6 +8,7 @@
  * compile this file to assembler, and then extract the
  * #defines from the assembly-language output.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/compat.h>
 #include <linux/signal.h>
diff --git a/arch/riscv/kernel/asm-offsets.c b/arch/riscv/kernel/asm-offsets.c
index 6e8c0d6feae9..7d42d3b8a32a 100644
--- a/arch/riscv/kernel/asm-offsets.c
+++ b/arch/riscv/kernel/asm-offsets.c
@@ -3,6 +3,7 @@
  * Copyright (C) 2012 Regents of the University of California
  * Copyright (C) 2017 SiFive
  */
+#define COMPILE_OFFSETS
 
 #include <linux/kbuild.h>
 #include <linux/mm.h>
diff --git a/arch/s390/kernel/asm-offsets.c b/arch/s390/kernel/asm-offsets.c
index 95ecad9c7d7d..a8915663e917 100644
--- a/arch/s390/kernel/asm-offsets.c
+++ b/arch/s390/kernel/asm-offsets.c
@@ -4,6 +4,7 @@
  * This code generates raw asm output which is post-processed to extract
  * and format the required data.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/kbuild.h>
 #include <linux/sched.h>
diff --git a/arch/sh/kernel/asm-offsets.c b/arch/sh/kernel/asm-offsets.c
index a0322e832845..429b6a763146 100644
--- a/arch/sh/kernel/asm-offsets.c
+++ b/arch/sh/kernel/asm-offsets.c
@@ -8,6 +8,7 @@
  * compile this file to assembler, and then extract the
  * #defines from the assembly-language output.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/stddef.h>
 #include <linux/types.h>
diff --git a/arch/sparc/kernel/asm-offsets.c b/arch/sparc/kernel/asm-offsets.c
index 3d9b9855dce9..6e660bde48dd 100644
--- a/arch/sparc/kernel/asm-offsets.c
+++ b/arch/sparc/kernel/asm-offsets.c
@@ -10,6 +10,7 @@
  *
  * On sparc, thread_info data is static and TI_XXX offsets are computed by hand.
  */
+#define COMPILE_OFFSETS
 
 #include <linux/sched.h>
 #include <linux/mm_types.h>
diff --git a/arch/um/kernel/asm-offsets.c b/arch/um/kernel/asm-offsets.c
index 1fb12235ab9c..a69873aa697f 100644
--- a/arch/um/kernel/asm-offsets.c
+++ b/arch/um/kernel/asm-offsets.c
@@ -1 +1,3 @@
+#define COMPILE_OFFSETS
+
 #include <sysdep/kernel-offsets.h>
diff --git a/arch/xtensa/kernel/asm-offsets.c b/arch/xtensa/kernel/asm-offsets.c
index da38de20ae59..cfbced95e944 100644
--- a/arch/xtensa/kernel/asm-offsets.c
+++ b/arch/xtensa/kernel/asm-offsets.c
@@ -11,6 +11,7 @@
  *
  * Chris Zankel <chris@zankel.net>
  */
+#define COMPILE_OFFSETS
 
 #include <asm/processor.h>
 #include <asm/coprocessor.h>
-- 
2.50.1

[PATCH v3 2/3] sched: make migrate_enable/migrate_disable inline

[Menglong Dong]

For now, migrate_enable and migrate_disable are global, which makes them
become hotspots in some case. Take BPF for example, the function calling
to migrate_enable and migrate_disable in BPF trampoline can introduce
significant overhead, and following is the 'perf top' of FENTRY's
benchmark (./tools/testing/selftests/bpf/bench trig-fentry):

  54.63% bpf_prog_2dcccf652aac1793_bench_trigger_fentry [k]
                 bpf_prog_2dcccf652aac1793_bench_trigger_fentry
  10.43% [kernel] [k] migrate_enable
  10.07% bpf_trampoline_6442517037 [k] bpf_trampoline_6442517037
  8.06% [kernel] [k] __bpf_prog_exit_recur
  4.11% libc.so.6 [.] syscall
  2.15% [kernel] [k] entry_SYSCALL_64
  1.48% [kernel] [k] memchr_inv
  1.32% [kernel] [k] fput
  1.16% [kernel] [k] _copy_to_user
  0.73% [kernel] [k] bpf_prog_test_run_raw_tp

So in this commit, we make migrate_enable/migrate_disable inline to obtain
better performance. The struct rq is defined internally in
kernel/sched/sched.h, and the field "nr_pinned" is accessed in
migrate_enable/migrate_disable, which makes it hard to make them inline.

Alexei Starovoitov suggests to generate the offset of "nr_pinned" in [1],
so we can define the migrate_enable/migrate_disable in
include/linux/sched.h and access "this_rq()->nr_pinned" with
"(void *)this_rq() + RQ_nr_pinned".

The offset of "nr_pinned" is generated in include/generated/rq-offsets.h
by kernel/sched/rq-offsets.c.

Generally speaking, we move the definition of migrate_enable and
migrate_disable to include/linux/sched.h from kernel/sched/core.c. The
calling to __set_cpus_allowed_ptr() is leaved in ___migrate_enable().

The "struct rq" is not available in include/linux/sched.h, so we can't
access the "runqueues" with this_cpu_ptr(), as the compilation will fail
in this_cpu_ptr() -> raw_cpu_ptr() -> __verify_pcpu_ptr():
  typeof((ptr) + 0)

So we introduce the this_rq_raw() and access the runqueues with
arch_raw_cpu_ptr/PERCPU_PTR directly.

The variable "runqueues" is not visible in the kernel modules, and export
it is not a good idea. As Peter Zijlstra advised in [2], we define and
export migrate_enable/migrate_disable in kernel/sched/core.c too, and use
them for the modules.

Before this patch, the performance of BPF FENTRY is:

  fentry         :  113.030 ± 0.149M/s
  fentry         :  112.501 ± 0.187M/s
  fentry         :  112.828 ± 0.267M/s
  fentry         :  115.287 ± 0.241M/s

After this patch, the performance of BPF FENTRY increases to:

  fentry         :  143.644 ± 0.670M/s
  fentry         :  149.764 ± 0.362M/s
  fentry         :  149.642 ± 0.156M/s
  fentry         :  145.263 ± 0.221M/s

Link: https://lore.kernel.org/bpf/CAADnVQ+5sEDKHdsJY5ZsfGDO_1SEhhQWHrt2SMBG5SYyQ+jt7w@mail.gmail.com/ [1]
Link: https://lore.kernel.org/all/20250819123214.GH4067720@noisy.programming.kicks-ass.net/ [2]
Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
---
v3:
- don't export runqueues, define migrate_enable and migrate_disable in
  kernel/sched/core.c and use them for kernel modules instead
- define the macro this_rq_pinned()
- add some comment for this_rq_raw()

v2:
- use PERCPU_PTR() for this_rq_raw() if !CONFIG_SMP
---
 Kbuild                    |  13 ++++-
 include/linux/preempt.h   |   3 --
 include/linux/sched.h     | 106 ++++++++++++++++++++++++++++++++++++++
 kernel/bpf/verifier.c     |   1 +
 kernel/sched/core.c       |  63 +++++-----------------
 kernel/sched/rq-offsets.c |  12 +++++
 6 files changed, 145 insertions(+), 53 deletions(-)
 create mode 100644 kernel/sched/rq-offsets.c

diff --git a/Kbuild b/Kbuild
index f327ca86990c..13324b4bbe23 100644
--- a/Kbuild
+++ b/Kbuild
@@ -34,13 +34,24 @@ arch/$(SRCARCH)/kernel/asm-offsets.s: $(timeconst-file) $(bounds-file)
 $(offsets-file): arch/$(SRCARCH)/kernel/asm-offsets.s FORCE
 	$(call filechk,offsets,__ASM_OFFSETS_H__)
 
+# Generate rq-offsets.h
+
+rq-offsets-file := include/generated/rq-offsets.h
+
+targets += kernel/sched/rq-offsets.s
+
+kernel/sched/rq-offsets.s: $(offsets-file)
+
+$(rq-offsets-file): kernel/sched/rq-offsets.s FORCE
+	$(call filechk,offsets,__RQ_OFFSETS_H__)
+
 # Check for missing system calls
 
 quiet_cmd_syscalls = CALL    $<
       cmd_syscalls = $(CONFIG_SHELL) $< $(CC) $(c_flags) $(missing_syscalls_flags)
 
 PHONY += missing-syscalls
-missing-syscalls: scripts/checksyscalls.sh $(offsets-file)
+missing-syscalls: scripts/checksyscalls.sh $(rq-offsets-file)
 	$(call cmd,syscalls)
 
 # Check the manual modification of atomic headers
diff --git a/include/linux/preempt.h b/include/linux/preempt.h
index 1fad1c8a4c76..92237c319035 100644
--- a/include/linux/preempt.h
+++ b/include/linux/preempt.h
@@ -424,8 +424,6 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
  *       work-conserving schedulers.
  *
  */
-extern void migrate_disable(void);
-extern void migrate_enable(void);
 
 /**
  * preempt_disable_nested - Disable preemption inside a normally preempt disabled section
@@ -471,7 +469,6 @@ static __always_inline void preempt_enable_nested(void)
 
 DEFINE_LOCK_GUARD_0(preempt, preempt_disable(), preempt_enable())
 DEFINE_LOCK_GUARD_0(preempt_notrace, preempt_disable_notrace(), preempt_enable_notrace())
-DEFINE_LOCK_GUARD_0(migrate, migrate_disable(), migrate_enable())
 
 #ifdef CONFIG_PREEMPT_DYNAMIC
 
diff --git a/include/linux/sched.h b/include/linux/sched.h
index f8188b833350..ae73fb70cc05 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -49,6 +49,9 @@
 #include <linux/tracepoint-defs.h>
 #include <linux/unwind_deferred_types.h>
 #include <asm/kmap_size.h>
+#ifndef COMPILE_OFFSETS
+#include <generated/rq-offsets.h>
+#endif
 
 /* task_struct member predeclarations (sorted alphabetically): */
 struct audit_context;
@@ -2312,4 +2315,107 @@ static __always_inline void alloc_tag_restore(struct alloc_tag *tag, struct allo
 #define alloc_tag_restore(_tag, _old)		do {} while (0)
 #endif
 
+#ifndef MODULE
+#ifndef COMPILE_OFFSETS
+
+extern void ___migrate_enable(void);
+
+struct rq;
+DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
+
+/* The "struct rq" is not available here, so we can't access the
+ * "runqueues" with this_cpu_ptr(), as the compilation will fail in
+ * this_cpu_ptr() -> raw_cpu_ptr() -> __verify_pcpu_ptr():
+ *   typeof((ptr) + 0)
+ *
+ * So use arch_raw_cpu_ptr()/PERCPU_PTR() directly here.
+ */
+#ifdef CONFIG_SMP
+#define this_rq_raw() arch_raw_cpu_ptr(&runqueues)
+#else
+#define this_rq_raw() PERCPU_PTR(&runqueues)
+#endif
+#define this_rq_pinned() (*(unsigned int *)((void *)this_rq_raw() + RQ_nr_pinned))
+
+static inline void __migrate_enable(void)
+{
+	struct task_struct *p = current;
+
+#ifdef CONFIG_DEBUG_PREEMPT
+	/*
+	 * Check both overflow from migrate_disable() and superfluous
+	 * migrate_enable().
+	 */
+	if (WARN_ON_ONCE((s16)p->migration_disabled <= 0))
+		return;
+#endif
+
+	if (p->migration_disabled > 1) {
+		p->migration_disabled--;
+		return;
+	}
+
+	/*
+	 * Ensure stop_task runs either before or after this, and that
+	 * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule().
+	 */
+	guard(preempt)();
+	if (unlikely(p->cpus_ptr != &p->cpus_mask))
+		___migrate_enable();
+	/*
+	 * Mustn't clear migration_disabled() until cpus_ptr points back at the
+	 * regular cpus_mask, otherwise things that race (eg.
+	 * select_fallback_rq) get confused.
+	 */
+	barrier();
+	p->migration_disabled = 0;
+	this_rq_pinned()--;
+}
+
+static inline void __migrate_disable(void)
+{
+	struct task_struct *p = current;
+
+	if (p->migration_disabled) {
+#ifdef CONFIG_DEBUG_PREEMPT
+		/*
+		 *Warn about overflow half-way through the range.
+		 */
+		WARN_ON_ONCE((s16)p->migration_disabled < 0);
+#endif
+		p->migration_disabled++;
+		return;
+	}
+
+	guard(preempt)();
+	this_rq_pinned()++;
+	p->migration_disabled = 1;
+}
+#else /* !COMPILE_OFFSETS */
+static inline void __migrate_disable(void) { }
+static inline void __migrate_enable(void) { }
+#endif /* !COMPILE_OFFSETS */
+
+#ifndef CREATE_MIGRATE_DISABLE
+static inline void migrate_disable(void)
+{
+	__migrate_disable();
+}
+
+static inline void migrate_enable(void)
+{
+	__migrate_enable();
+}
+#else /* CREATE_MIGRATE_DISABLE */
+extern void migrate_disable(void);
+extern void migrate_enable(void);
+#endif /* CREATE_MIGRATE_DISABLE */
+
+#else /* MODULE */
+extern void migrate_disable(void);
+extern void migrate_enable(void);
+#endif /* MODULE */
+
+DEFINE_LOCK_GUARD_0(migrate, migrate_disable(), migrate_enable())
+
 #endif
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index c4f69a9e9af6..de9078a9df3a 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c
@@ -23855,6 +23855,7 @@ int bpf_check_attach_target(struct bpf_verifier_log *log,
 BTF_SET_START(btf_id_deny)
 BTF_ID_UNUSED
 #ifdef CONFIG_SMP
+BTF_ID(func, ___migrate_enable)
 BTF_ID(func, migrate_disable)
 BTF_ID(func, migrate_enable)
 #endif
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index be00629f0ba4..58164a69449d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7,6 +7,8 @@
  *  Copyright (C) 1991-2002  Linus Torvalds
  *  Copyright (C) 1998-2024  Ingo Molnar, Red Hat
  */
+#define CREATE_MIGRATE_DISABLE
+#include <linux/sched.h>
 #include <linux/highmem.h>
 #include <linux/hrtimer_api.h>
 #include <linux/ktime_api.h>
@@ -2381,28 +2383,7 @@ static void migrate_disable_switch(struct rq *rq, struct task_struct *p)
 	__do_set_cpus_allowed(p, &ac);
 }
 
-void migrate_disable(void)
-{
-	struct task_struct *p = current;
-
-	if (p->migration_disabled) {
-#ifdef CONFIG_DEBUG_PREEMPT
-		/*
-		 *Warn about overflow half-way through the range.
-		 */
-		WARN_ON_ONCE((s16)p->migration_disabled < 0);
-#endif
-		p->migration_disabled++;
-		return;
-	}
-
-	guard(preempt)();
-	this_rq()->nr_pinned++;
-	p->migration_disabled = 1;
-}
-EXPORT_SYMBOL_GPL(migrate_disable);
-
-void migrate_enable(void)
+void ___migrate_enable(void)
 {
 	struct task_struct *p = current;
 	struct affinity_context ac = {
@@ -2410,35 +2391,19 @@ void migrate_enable(void)
 		.flags     = SCA_MIGRATE_ENABLE,
 	};
 
-#ifdef CONFIG_DEBUG_PREEMPT
-	/*
-	 * Check both overflow from migrate_disable() and superfluous
-	 * migrate_enable().
-	 */
-	if (WARN_ON_ONCE((s16)p->migration_disabled <= 0))
-		return;
-#endif
+	__set_cpus_allowed_ptr(p, &ac);
+}
+EXPORT_SYMBOL_GPL(___migrate_enable);
 
-	if (p->migration_disabled > 1) {
-		p->migration_disabled--;
-		return;
-	}
+void migrate_disable(void)
+{
+	__migrate_disable();
+}
+EXPORT_SYMBOL_GPL(migrate_disable);
 
-	/*
-	 * Ensure stop_task runs either before or after this, and that
-	 * __set_cpus_allowed_ptr(SCA_MIGRATE_ENABLE) doesn't schedule().
-	 */
-	guard(preempt)();
-	if (p->cpus_ptr != &p->cpus_mask)
-		__set_cpus_allowed_ptr(p, &ac);
-	/*
-	 * Mustn't clear migration_disabled() until cpus_ptr points back at the
-	 * regular cpus_mask, otherwise things that race (eg.
-	 * select_fallback_rq) get confused.
-	 */
-	barrier();
-	p->migration_disabled = 0;
-	this_rq()->nr_pinned--;
+void migrate_enable(void)
+{
+	__migrate_enable();
 }
 EXPORT_SYMBOL_GPL(migrate_enable);
 
diff --git a/kernel/sched/rq-offsets.c b/kernel/sched/rq-offsets.c
new file mode 100644
index 000000000000..a23747bbe25b
--- /dev/null
+++ b/kernel/sched/rq-offsets.c
@@ -0,0 +1,12 @@
+// SPDX-License-Identifier: GPL-2.0
+#define COMPILE_OFFSETS
+#include <linux/kbuild.h>
+#include <linux/types.h>
+#include "sched.h"
+
+int main(void)
+{
+	DEFINE(RQ_nr_pinned, offsetof(struct rq, nr_pinned));
+
+	return 0;
+}
-- 
2.50.1

[PATCH v3 3/3] sched: fix some typos in include/linux/preempt.h

[Menglong Dong]

There are some typos in the comments of migrate in
include/linux/preempt.h:

  elegible -> eligible
  it's -> its
  migirate_disable -> migrate_disable
  abritrary -> arbitrary

Just fix them.

Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
---
 include/linux/preempt.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/include/linux/preempt.h b/include/linux/preempt.h
index 92237c319035..102202185d7a 100644
--- a/include/linux/preempt.h
+++ b/include/linux/preempt.h
@@ -372,7 +372,7 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
 /*
  * Migrate-Disable and why it is undesired.
  *
- * When a preempted task becomes elegible to run under the ideal model (IOW it
+ * When a preempted task becomes eligible to run under the ideal model (IOW it
  * becomes one of the M highest priority tasks), it might still have to wait
  * for the preemptee's migrate_disable() section to complete. Thereby suffering
  * a reduction in bandwidth in the exact duration of the migrate_disable()
@@ -387,7 +387,7 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
  * - a lower priority tasks; which under preempt_disable() could've instantly
  *   migrated away when another CPU becomes available, is now constrained
  *   by the ability to push the higher priority task away, which might itself be
- *   in a migrate_disable() section, reducing it's available bandwidth.
+ *   in a migrate_disable() section, reducing its available bandwidth.
  *
  * IOW it trades latency / moves the interference term, but it stays in the
  * system, and as long as it remains unbounded, the system is not fully
@@ -399,7 +399,7 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
  * PREEMPT_RT breaks a number of assumptions traditionally held. By forcing a
  * number of primitives into becoming preemptible, they would also allow
  * migration. This turns out to break a bunch of per-cpu usage. To this end,
- * all these primitives employ migirate_disable() to restore this implicit
+ * all these primitives employ migrate_disable() to restore this implicit
  * assumption.
  *
  * This is a 'temporary' work-around at best. The correct solution is getting
@@ -407,7 +407,7 @@ static inline void preempt_notifier_init(struct preempt_notifier *notifier,
  * per-cpu locking or short preempt-disable regions.
  *
  * The end goal must be to get rid of migrate_disable(), alternatively we need
- * a schedulability theory that does not depend on abritrary migration.
+ * a schedulability theory that does not depend on arbitrary migration.
  *
  *
  * Notes on the implementation.
-- 
2.50.1

Re: [PATCH v3 2/3] sched: make migrate_enable/migrate_disable inline

[Alexei Starovoitov]

On Thu, Aug 21, 2025 at 2:38 AM Menglong Dong <menglong8.dong@gmail.com> wrote:
>
> +
> +#ifndef CREATE_MIGRATE_DISABLE
> +static inline void migrate_disable(void)
> +{
> +       __migrate_disable();
> +}
> +
> +static inline void migrate_enable(void)
> +{
> +       __migrate_enable();
> +}
> +#else /* CREATE_MIGRATE_DISABLE */
> +extern void migrate_disable(void);
> +extern void migrate_enable(void);
> +#endif /* CREATE_MIGRATE_DISABLE */

I think the explanation from the commit log is better to be
copy pasted here as a comment, since the need for the macro
is quite hard to understand.

> +
> +#else /* MODULE */
> +extern void migrate_disable(void);
> +extern void migrate_enable(void);
> +#endif /* MODULE */
> +

...
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index be00629f0ba4..58164a69449d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -7,6 +7,8 @@
>   *  Copyright (C) 1991-2002  Linus Torvalds
>   *  Copyright (C) 1998-2024  Ingo Molnar, Red Hat
>   */
> +#define CREATE_MIGRATE_DISABLE
> +#include <linux/sched.h>

Also how about calling it
#define INSTANTIATE_EXPORTED_MIGRATE_DISABLE

When I asked AI what "instantiate exported migrate_disable"
means it guessed it nicely :)
while "create migrate_disable" had a vague answer.

Re: [PATCH v3 1/3] arch: add the macro COMPILE_OFFSETS to all the asm-offsets.c

[Alexei Starovoitov]

On Thu, Aug 21, 2025 at 2:38 AM Menglong Dong <menglong8.dong@gmail.com> wrote:
>
> The include/generated/asm-offsets.h is generated in Kbuild during
> compiling from arch/SRCARCH/kernel/asm-offsets.c. When we want to
> generate another similar offset header file, circular dependency can
> happen.

Is there a way to avoid all this churn?

> For example, we want to generate a offset file include/generated/test.h,
> which is included in include/sched/sched.h. If we generate asm-offsets.h
> first, it will fail, as include/sched/sched.h is included in asm-offsets.c

if so, may be don't add "static inline void migrate_disable()" to sched.h
and instead add it to preempt.h and it will avoid this issue?

> and include/generated/test.h doesn't exist; If we generate test.h first,
> it can't success neither, as include/generated/asm-offsets.h is included
> by it.
>
> In x86_64, the macro COMPILE_OFFSETS is used to avoid such circular
> dependency. We can generate asm-offsets.h first, and if the
> COMPILE_OFFSETS is defined, we don't include the "generated/test.h".
>
> And we define the macro COMPILE_OFFSETS for all the asm-offsets.c for this
> purpose.

Re: [PATCH v3 2/3] sched: make migrate_enable/migrate_disable inline

[Menglong Dong]

On Wed, Aug 27, 2025 at 10:58 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Thu, Aug 21, 2025 at 2:38 AM Menglong Dong <menglong8.dong@gmail.com> wrote:
> >
> > +
> > +#ifndef CREATE_MIGRATE_DISABLE
> > +static inline void migrate_disable(void)
> > +{
> > +       __migrate_disable();
> > +}
> > +
> > +static inline void migrate_enable(void)
> > +{
> > +       __migrate_enable();
> > +}
> > +#else /* CREATE_MIGRATE_DISABLE */
> > +extern void migrate_disable(void);
> > +extern void migrate_enable(void);
> > +#endif /* CREATE_MIGRATE_DISABLE */
>
> I think the explanation from the commit log is better to be
> copy pasted here as a comment, since the need for the macro
> is quite hard to understand.

Okay!

>
> > +
> > +#else /* MODULE */
> > +extern void migrate_disable(void);
> > +extern void migrate_enable(void);
> > +#endif /* MODULE */
> > +
>
> ...
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index be00629f0ba4..58164a69449d 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -7,6 +7,8 @@
> >   *  Copyright (C) 1991-2002  Linus Torvalds
> >   *  Copyright (C) 1998-2024  Ingo Molnar, Red Hat
> >   */
> > +#define CREATE_MIGRATE_DISABLE
> > +#include <linux/sched.h>
>
> Also how about calling it
> #define INSTANTIATE_EXPORTED_MIGRATE_DISABLE
>
> When I asked AI what "instantiate exported migrate_disable"
> means it guessed it nicely :)
> while "create migrate_disable" had a vague answer.

Okay, sounds nice!

Re: [PATCH v3 1/3] arch: add the macro COMPILE_OFFSETS to all the asm-offsets.c

[Menglong Dong]

On Wed, Aug 27, 2025 at 11:04 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Thu, Aug 21, 2025 at 2:38 AM Menglong Dong <menglong8.dong@gmail.com> wrote:
> >
> > The include/generated/asm-offsets.h is generated in Kbuild during
> > compiling from arch/SRCARCH/kernel/asm-offsets.c. When we want to
> > generate another similar offset header file, circular dependency can
> > happen.
>
> Is there a way to avoid all this churn?
>
> > For example, we want to generate a offset file include/generated/test.h,
> > which is included in include/sched/sched.h. If we generate asm-offsets.h
> > first, it will fail, as include/sched/sched.h is included in asm-offsets.c
>
> if so, may be don't add "static inline void migrate_disable()" to sched.h
> and instead add it to preempt.h and it will avoid this issue?

It's hard to avoid this churn. Take bounds.c for example, it defines
the macro __GENERATING_BOUNDS_H. For the header files that
it includes, it will exclude almost all the unnecessary code in
page-flags.h, mmzone.h if __GENERATING_BOUNDS_H is defined.
We can't use this approach, as it's hard to decide what to exclude
in sched.h.

We can't add migrate_disable or __migrate_disable to preempt.h,
as struct task is used in it, which is not available in preempt.h :/

I think this stuff can be reused in the feature if someone wants
to add such an offset.

Thanks!
Menglong Dong

>
> > and include/generated/test.h doesn't exist; If we generate test.h first,
> > it can't success neither, as include/generated/asm-offsets.h is included
> > by it.
> >
> > In x86_64, the macro COMPILE_OFFSETS is used to avoid such circular
> > dependency. We can generate asm-offsets.h first, and if the
> > COMPILE_OFFSETS is defined, we don't include the "generated/test.h".
> >
> > And we define the macro COMPILE_OFFSETS for all the asm-offsets.c for this
> > purpose.