Improve the performance of icache flushing by creating a new prctl flag
PR_RISCV_SET_ICACHE_FLUSH_CTX. The interface is left generic to allow
for future expansions such as with the proposed J extension [1].
Documentation is also provided to explain the use case.
Patch sent to add PR_RISCV_SET_ICACHE_FLUSH_CTX to man-pages [2].
[1] https://github.com/riscv/riscv-j-extension
[2] https://lore.kernel.org/linux-man/[email protected]
Signed-off-by: Charlie Jenkins <[email protected]>
---
Changes in v10:
- Fix fence.i condition to properly only flush on migration (Alex)
- Fix documentation wording (Alex)
- Link to v9: https://lore.kernel.org/r/[email protected]
Changes in v9:
- Remove prev_cpu from mm (Alex)
- Link to v8: https://lore.kernel.org/r/[email protected]
Changes in v8:
- Only flush icache if migrated to different cpu (Alex)
- Move flushing to switch_to to catch per-thread flushing properly
- Link to v7: https://lore.kernel.org/r/[email protected]
Changes in v7:
- Change "per_thread" parameter to "scope" and provide constants for the
parameter.
- Link to v6: https://lore.kernel.org/r/[email protected]
Changes in v6:
- Fixup documentation formatting
- Link to v5: https://lore.kernel.org/r/[email protected]
Changes in v5:
- Minor documentation changes (Randy)
- Link to v4: https://lore.kernel.org/r/[email protected]
Changes in v4:
- Add OFF flag to disallow fence.i in userspace (Atish)
- Fix documentation issues (Atish)
- Link to v3: https://lore.kernel.org/r/[email protected]
Changes in v3:
- Check if value force_icache_flush set on thread, rather than in mm
twice (Clément)
- Link to v2: https://lore.kernel.org/r/[email protected]
Changes in v2:
- Fix kernel-doc comment (Conor)
- Link to v1: https://lore.kernel.org/r/[email protected]
---
Charlie Jenkins (2):
riscv: Include riscv_set_icache_flush_ctx prctl
documentation: Document PR_RISCV_SET_ICACHE_FLUSH_CTX prctl
Documentation/arch/riscv/cmodx.rst | 96 ++++++++++++++++++++++++++++++++++++++
Documentation/arch/riscv/index.rst | 1 +
arch/riscv/include/asm/mmu.h | 2 +
arch/riscv/include/asm/processor.h | 7 +++
arch/riscv/include/asm/switch_to.h | 13 ++++++
arch/riscv/mm/cacheflush.c | 67 ++++++++++++++++++++++++++
arch/riscv/mm/context.c | 13 ++++--
include/uapi/linux/prctl.h | 6 +++
kernel/sys.c | 6 +++
9 files changed, 208 insertions(+), 3 deletions(-)
---
base-commit: 6613476e225e090cc9aad49be7fa504e290dd33d
change-id: 20231117-fencei-f9f60d784fa0
--
- Charlie
Provide documentation that explains how to properly do CMODX in riscv.
Signed-off-by: Charlie Jenkins <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
---
Documentation/arch/riscv/cmodx.rst | 96 ++++++++++++++++++++++++++++++++++++++
Documentation/arch/riscv/index.rst | 1 +
2 files changed, 97 insertions(+)
diff --git a/Documentation/arch/riscv/cmodx.rst b/Documentation/arch/riscv/cmodx.rst
new file mode 100644
index 000000000000..24aafa23a72b
--- /dev/null
+++ b/Documentation/arch/riscv/cmodx.rst
@@ -0,0 +1,96 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==============================================================================
+Concurrent Modification and Execution of Instructions (CMODX) for RISC-V Linux
+==============================================================================
+
+CMODX is a programming technique where a program executes instructions that were
+modified by the program itself. Instruction storage and the instruction cache
+(icache) are not guaranteed to be synchronized on RISC-V hardware. Therefore, the
+program must enforce its own synchronization with the unprivileged fence.i
+instruction.
+
+However, the default Linux ABI prohibits the use of fence.i in userspace
+applications. At any point the scheduler may migrate a task onto a new hart. If
+migration occurs after the userspace synchronized the icache and instruction
+storage with fence.i, the icache on the new hart will no longer be clean. This
+is due to the behavior of fence.i only affecting the hart that it is called on.
+Thus, the hart that the task has been migrated to may not have synchronized
+instruction storage and icache.
+
+There are two ways to solve this problem: use the riscv_flush_icache() syscall,
+or use the ``PR_RISCV_SET_ICACHE_FLUSH_CTX`` prctl() and emit fence.i in
+userspace. The syscall performs a one-off icache flushing operation. The prctl
+changes the Linux ABI to allow userspace to emit icache flushing operations.
+
+As an aside, "deferred" icache flushes can sometimes be triggered in the kernel.
+At the time of writing, this only occurs during the riscv_flush_icache() syscall
+and when the kernel uses copy_to_user_page(). These deferred flushes happen only
+when the memory map being used by a hart changes. If the prctl() context caused
+an icache flush, this deferred icache flush will be skipped as it is redundant.
+Therefore, there will be no additional flush when using the riscv_flush_icache()
+syscall inside of the prctl() context.
+
+prctl() Interface
+---------------------
+
+Call prctl() with ``PR_RISCV_SET_ICACHE_FLUSH_CTX`` as the first argument. The
+remaining arguments will be delegated to the riscv_set_icache_flush_ctx
+function detailed below.
+
+.. kernel-doc:: arch/riscv/mm/cacheflush.c
+ :identifiers: riscv_set_icache_flush_ctx
+
+Example usage:
+
+The following files are meant to be compiled and linked with each other. The
+modify_instruction() function replaces an add with 0 with an add with one,
+causing the instruction sequence in get_value() to change from returning a zero
+to returning a one.
+
+cmodx.c::
+
+ #include <stdio.h>
+ #include <sys/prctl.h>
+
+ extern int get_value();
+ extern void modify_instruction();
+
+ int main()
+ {
+ int value = get_value();
+ printf("Value before cmodx: %d\n", value);
+
+ // Call prctl before first fence.i is called inside modify_instruction
+ prctl(PR_RISCV_SET_ICACHE_FLUSH_CTX_ON, PR_RISCV_CTX_SW_FENCEI, PR_RISCV_SCOPE_PER_PROCESS);
+ modify_instruction();
+
+ value = get_value();
+ printf("Value after cmodx: %d\n", value);
+ return 0;
+ }
+
+cmodx.S::
+
+ .option norvc
+
+ .text
+ .global modify_instruction
+ modify_instruction:
+ lw a0, new_insn
+ lui a5,%hi(old_insn)
+ sw a0,%lo(old_insn)(a5)
+ fence.i
+ ret
+
+ .section modifiable, "awx"
+ .global get_value
+ get_value:
+ li a0, 0
+ old_insn:
+ addi a0, a0, 0
+ ret
+
+ .data
+ new_insn:
+ addi a0, a0, 1
diff --git a/Documentation/arch/riscv/index.rst b/Documentation/arch/riscv/index.rst
index 4dab0cb4b900..eecf347ce849 100644
--- a/Documentation/arch/riscv/index.rst
+++ b/Documentation/arch/riscv/index.rst
@@ -13,6 +13,7 @@ RISC-V architecture
patch-acceptance
uabi
vector
+ cmodx
features
--
2.34.1
Support new prctl with key PR_RISCV_SET_ICACHE_FLUSH_CTX to enable
optimization of cross modifying code. This prctl enables userspace code
to use icache flushing instructions such as fence.i with the guarantee
that the icache will continue to be clean after thread migration.
Signed-off-by: Charlie Jenkins <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
---
arch/riscv/include/asm/mmu.h | 2 ++
arch/riscv/include/asm/processor.h | 7 ++++
arch/riscv/include/asm/switch_to.h | 13 ++++++++
arch/riscv/mm/cacheflush.c | 67 ++++++++++++++++++++++++++++++++++++++
arch/riscv/mm/context.c | 13 ++++++--
include/uapi/linux/prctl.h | 6 ++++
kernel/sys.c | 6 ++++
7 files changed, 111 insertions(+), 3 deletions(-)
diff --git a/arch/riscv/include/asm/mmu.h b/arch/riscv/include/asm/mmu.h
index 355504b37f8e..60be458e94da 100644
--- a/arch/riscv/include/asm/mmu.h
+++ b/arch/riscv/include/asm/mmu.h
@@ -19,6 +19,8 @@ typedef struct {
#ifdef CONFIG_SMP
/* A local icache flush is needed before user execution can resume. */
cpumask_t icache_stale_mask;
+ /* Force local icache flush on all migrations. */
+ bool force_icache_flush;
#endif
#ifdef CONFIG_BINFMT_ELF_FDPIC
unsigned long exec_fdpic_loadmap;
diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h
index a8509cc31ab2..816cdc2395f4 100644
--- a/arch/riscv/include/asm/processor.h
+++ b/arch/riscv/include/asm/processor.h
@@ -123,6 +123,10 @@ struct thread_struct {
struct __riscv_v_ext_state vstate;
unsigned long align_ctl;
struct __riscv_v_ext_state kernel_vstate;
+#ifdef CONFIG_SMP
+ bool force_icache_flush;
+ unsigned int prev_cpu;
+#endif
};
/* Whitelist the fstate from the task_struct for hardened usercopy */
@@ -184,6 +188,9 @@ extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
#define GET_UNALIGN_CTL(tsk, addr) get_unalign_ctl((tsk), (addr))
#define SET_UNALIGN_CTL(tsk, val) set_unalign_ctl((tsk), (val))
+#define RISCV_SET_ICACHE_FLUSH_CTX(arg1, arg2) riscv_set_icache_flush_ctx(arg1, arg2)
+extern int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long per_thread);
+
#endif /* __ASSEMBLY__ */
#endif /* _ASM_RISCV_PROCESSOR_H */
diff --git a/arch/riscv/include/asm/switch_to.h b/arch/riscv/include/asm/switch_to.h
index 7efdb0584d47..ac99ab64499c 100644
--- a/arch/riscv/include/asm/switch_to.h
+++ b/arch/riscv/include/asm/switch_to.h
@@ -8,6 +8,7 @@
#include <linux/jump_label.h>
#include <linux/sched/task_stack.h>
+#include <linux/mm_types.h>
#include <asm/vector.h>
#include <asm/cpufeature.h>
#include <asm/processor.h>
@@ -72,14 +73,26 @@ static __always_inline bool has_fpu(void) { return false; }
extern struct task_struct *__switch_to(struct task_struct *,
struct task_struct *);
+static inline bool switch_to_should_flush_icache(struct task_struct *task)
+{
+ bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
+ bool stale_thread = task->thread.force_icache_flush;
+ bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
+
+ return thread_migrated && (stale_mm || stale_thread);
+}
+
#define switch_to(prev, next, last) \
do { \
struct task_struct *__prev = (prev); \
struct task_struct *__next = (next); \
+ __next->thread.prev_cpu = smp_processor_id(); \
if (has_fpu()) \
__switch_to_fpu(__prev, __next); \
if (has_vector()) \
__switch_to_vector(__prev, __next); \
+ if (switch_to_should_flush_icache(__next)) \
+ local_flush_icache_all(); \
((last) = __switch_to(__prev, __next)); \
} while (0)
diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
index 55a34f2020a8..2d5e1575f6c1 100644
--- a/arch/riscv/mm/cacheflush.c
+++ b/arch/riscv/mm/cacheflush.c
@@ -5,6 +5,7 @@
#include <linux/acpi.h>
#include <linux/of.h>
+#include <linux/prctl.h>
#include <asm/acpi.h>
#include <asm/cacheflush.h>
@@ -152,3 +153,69 @@ void __init riscv_init_cbo_blocksizes(void)
if (cboz_block_size)
riscv_cboz_block_size = cboz_block_size;
}
+
+/**
+ * riscv_set_icache_flush_ctx() - Enable/disable icache flushing instructions in
+ * userspace.
+ * @ctx: Set the type of icache flushing instructions permitted/prohibited in
+ * userspace. Supported values described below.
+ *
+ * Supported values for ctx:
+ *
+ * * %PR_RISCV_CTX_SW_FENCEI_ON: Allow fence.i in userspace.
+ *
+ * * %PR_RISCV_CTX_SW_FENCEI_OFF: Disallow fence.i in userspace. All threads in
+ * a process will be affected when ``scope == PR_RISCV_SCOPE_PER_PROCESS``.
+ * Therefore, caution must be taken -- only use this flag when you can
+ * guarantee that no thread in the process will emit fence.i from this point
+ * onward.
+ *
+ * @scope: Set scope of where icache flushing instructions are allowed to be
+ * emitted. Supported values described below.
+ *
+ * Supported values for scope:
+ *
+ * * %PR_RISCV_SCOPE_PER_PROCESS: Ensure the icache of any thread in this process
+ * is coherent with instruction storage upon
+ * migration.
+ *
+ * * %PR_RISCV_SCOPE_PER_THREAD: Ensure the icache of the current thread is
+ * coherent with instruction storage upon
+ * migration.
+ *
+ * When ``scope == PR_RISCV_SCOPE_PER_PROCESS``, all threads in the process are
+ * permitted to emit icache flushing instructions. Whenever any thread in the
+ * process is migrated, the corresponding hart's icache will be guaranteed to be
+ * consistent with instruction storage. Note this does not enforce any
+ * guarantees outside of migration. If a thread modifies an instruction that
+ * another thread may attempt to execute, the other thread must still emit an
+ * icache flushing instruction before attempting to execute the potentially
+ * modified instruction. This must be performed by the userspace program.
+ *
+ * In per-thread context (eg. ``scope == PR_RISCV_SCOPE_PER_THREAD``), only the
+ * thread calling this function is permitted to emit icache flushing
+ * instructions. When the thread is migrated, the corresponding hart's icache
+ * will be guaranteed to be consistent with instruction storage.
+ *
+ * On kernels configured without SMP, this function is a nop as migrations
+ * across harts will not occur.
+ */
+int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long scope)
+{
+#ifdef CONFIG_SMP
+ switch (ctx) {
+ case PR_RISCV_CTX_SW_FENCEI_ON:
+ switch (scope) {
+ case PR_RISCV_SCOPE_PER_PROCESS:
+ current->mm->context.force_icache_flush = true;
+ break;
+ case PR_RISCV_SCOPE_PER_THREAD:
+ current->thread.force_icache_flush = true;
+ break;
+ default:
+ return -EINVAL;
+ }
+ }
+#endif
+ return 0;
+}
diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
index 217fd4de6134..5cabbc449080 100644
--- a/arch/riscv/mm/context.c
+++ b/arch/riscv/mm/context.c
@@ -15,6 +15,7 @@
#include <asm/tlbflush.h>
#include <asm/cacheflush.h>
#include <asm/mmu_context.h>
+#include <asm/switch_to.h>
#ifdef CONFIG_MMU
@@ -297,7 +298,8 @@ static inline void set_mm(struct mm_struct *prev,
*
* The "cpu" argument must be the current local CPU number.
*/
-static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
+static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu,
+ struct task_struct *task)
{
#ifdef CONFIG_SMP
cpumask_t *mask = &mm->context.icache_stale_mask;
@@ -309,7 +311,12 @@ static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
* This pairs with a barrier in flush_icache_mm.
*/
smp_mb();
- local_flush_icache_all();
+
+ /*
+ * If cache will be flushed in switch_to, no need to flush here.
+ */
+ if (!(task && switch_to_should_flush_icache(task)))
+ local_flush_icache_all();
}
#endif
@@ -332,5 +339,5 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
set_mm(prev, next, cpu);
- flush_icache_deferred(next, cpu);
+ flush_icache_deferred(next, cpu, task);
}
diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
index 370ed14b1ae0..524d546d697b 100644
--- a/include/uapi/linux/prctl.h
+++ b/include/uapi/linux/prctl.h
@@ -306,4 +306,10 @@ struct prctl_mm_map {
# define PR_RISCV_V_VSTATE_CTRL_NEXT_MASK 0xc
# define PR_RISCV_V_VSTATE_CTRL_MASK 0x1f
+#define PR_RISCV_SET_ICACHE_FLUSH_CTX 71
+# define PR_RISCV_CTX_SW_FENCEI_ON 0
+# define PR_RISCV_CTX_SW_FENCEI_OFF 1
+# define PR_RISCV_SCOPE_PER_PROCESS 0
+# define PR_RISCV_SCOPE_PER_THREAD 1
+
#endif /* _LINUX_PRCTL_H */
diff --git a/kernel/sys.c b/kernel/sys.c
index e219fcfa112d..69afdd8b430f 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -146,6 +146,9 @@
#ifndef RISCV_V_GET_CONTROL
# define RISCV_V_GET_CONTROL() (-EINVAL)
#endif
+#ifndef RISCV_SET_ICACHE_FLUSH_CTX
+# define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
+#endif
/*
* this is where the system-wide overflow UID and GID are defined, for
@@ -2743,6 +2746,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
case PR_RISCV_V_GET_CONTROL:
error = RISCV_V_GET_CONTROL();
break;
+ case PR_RISCV_SET_ICACHE_FLUSH_CTX:
+ error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
+ break;
default:
error = -EINVAL;
break;
--
2.34.1
Hi Charlie,
On 25/01/2024 05:23, Charlie Jenkins wrote:
> Support new prctl with key PR_RISCV_SET_ICACHE_FLUSH_CTX to enable
> optimization of cross modifying code. This prctl enables userspace code
> to use icache flushing instructions such as fence.i with the guarantee
> that the icache will continue to be clean after thread migration.
>
> Signed-off-by: Charlie Jenkins <[email protected]>
> Reviewed-by: Atish Patra <[email protected]>
> ---
> arch/riscv/include/asm/mmu.h | 2 ++
> arch/riscv/include/asm/processor.h | 7 ++++
> arch/riscv/include/asm/switch_to.h | 13 ++++++++
> arch/riscv/mm/cacheflush.c | 67 ++++++++++++++++++++++++++++++++++++++
> arch/riscv/mm/context.c | 13 ++++++--
> include/uapi/linux/prctl.h | 6 ++++
> kernel/sys.c | 6 ++++
> 7 files changed, 111 insertions(+), 3 deletions(-)
>
> diff --git a/arch/riscv/include/asm/mmu.h b/arch/riscv/include/asm/mmu.h
> index 355504b37f8e..60be458e94da 100644
> --- a/arch/riscv/include/asm/mmu.h
> +++ b/arch/riscv/include/asm/mmu.h
> @@ -19,6 +19,8 @@ typedef struct {
> #ifdef CONFIG_SMP
> /* A local icache flush is needed before user execution can resume. */
> cpumask_t icache_stale_mask;
> + /* Force local icache flush on all migrations. */
> + bool force_icache_flush;
> #endif
> #ifdef CONFIG_BINFMT_ELF_FDPIC
> unsigned long exec_fdpic_loadmap;
> diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h
> index a8509cc31ab2..816cdc2395f4 100644
> --- a/arch/riscv/include/asm/processor.h
> +++ b/arch/riscv/include/asm/processor.h
> @@ -123,6 +123,10 @@ struct thread_struct {
> struct __riscv_v_ext_state vstate;
> unsigned long align_ctl;
> struct __riscv_v_ext_state kernel_vstate;
> +#ifdef CONFIG_SMP
> + bool force_icache_flush;
> + unsigned int prev_cpu;
> +#endif
> };
>
> /* Whitelist the fstate from the task_struct for hardened usercopy */
> @@ -184,6 +188,9 @@ extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
> #define GET_UNALIGN_CTL(tsk, addr) get_unalign_ctl((tsk), (addr))
> #define SET_UNALIGN_CTL(tsk, val) set_unalign_ctl((tsk), (val))
>
> +#define RISCV_SET_ICACHE_FLUSH_CTX(arg1, arg2) riscv_set_icache_flush_ctx(arg1, arg2)
> +extern int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long per_thread);
> +
> #endif /* __ASSEMBLY__ */
>
> #endif /* _ASM_RISCV_PROCESSOR_H */
> diff --git a/arch/riscv/include/asm/switch_to.h b/arch/riscv/include/asm/switch_to.h
> index 7efdb0584d47..ac99ab64499c 100644
> --- a/arch/riscv/include/asm/switch_to.h
> +++ b/arch/riscv/include/asm/switch_to.h
> @@ -8,6 +8,7 @@
>
> #include <linux/jump_label.h>
> #include <linux/sched/task_stack.h>
> +#include <linux/mm_types.h>
> #include <asm/vector.h>
> #include <asm/cpufeature.h>
> #include <asm/processor.h>
> @@ -72,14 +73,26 @@ static __always_inline bool has_fpu(void) { return false; }
> extern struct task_struct *__switch_to(struct task_struct *,
> struct task_struct *);
>
> +static inline bool switch_to_should_flush_icache(struct task_struct *task)
> +{
> + bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
> + bool stale_thread = task->thread.force_icache_flush;
> + bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
> +
> + return thread_migrated && (stale_mm || stale_thread);
> +}
> +
> #define switch_to(prev, next, last) \
> do { \
> struct task_struct *__prev = (prev); \
> struct task_struct *__next = (next); \
> + __next->thread.prev_cpu = smp_processor_id(); \
> if (has_fpu()) \
> __switch_to_fpu(__prev, __next); \
> if (has_vector()) \
> __switch_to_vector(__prev, __next); \
> + if (switch_to_should_flush_icache(__next)) \
> + local_flush_icache_all(); \
> ((last) = __switch_to(__prev, __next)); \
> } while (0)
>
> diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
> index 55a34f2020a8..2d5e1575f6c1 100644
> --- a/arch/riscv/mm/cacheflush.c
> +++ b/arch/riscv/mm/cacheflush.c
> @@ -5,6 +5,7 @@
>
> #include <linux/acpi.h>
> #include <linux/of.h>
> +#include <linux/prctl.h>
> #include <asm/acpi.h>
> #include <asm/cacheflush.h>
>
> @@ -152,3 +153,69 @@ void __init riscv_init_cbo_blocksizes(void)
> if (cboz_block_size)
> riscv_cboz_block_size = cboz_block_size;
> }
> +
> +/**
> + * riscv_set_icache_flush_ctx() - Enable/disable icache flushing instructions in
> + * userspace.
> + * @ctx: Set the type of icache flushing instructions permitted/prohibited in
> + * userspace. Supported values described below.
> + *
> + * Supported values for ctx:
> + *
> + * * %PR_RISCV_CTX_SW_FENCEI_ON: Allow fence.i in userspace.
> + *
> + * * %PR_RISCV_CTX_SW_FENCEI_OFF: Disallow fence.i in userspace. All threads in
> + * a process will be affected when ``scope == PR_RISCV_SCOPE_PER_PROCESS``.
> + * Therefore, caution must be taken -- only use this flag when you can
> + * guarantee that no thread in the process will emit fence.i from this point
> + * onward.
> + *
> + * @scope: Set scope of where icache flushing instructions are allowed to be
> + * emitted. Supported values described below.
> + *
> + * Supported values for scope:
> + *
> + * * %PR_RISCV_SCOPE_PER_PROCESS: Ensure the icache of any thread in this process
> + * is coherent with instruction storage upon
> + * migration.
> + *
> + * * %PR_RISCV_SCOPE_PER_THREAD: Ensure the icache of the current thread is
> + * coherent with instruction storage upon
> + * migration.
> + *
> + * When ``scope == PR_RISCV_SCOPE_PER_PROCESS``, all threads in the process are
> + * permitted to emit icache flushing instructions. Whenever any thread in the
> + * process is migrated, the corresponding hart's icache will be guaranteed to be
> + * consistent with instruction storage. Note this does not enforce any
> + * guarantees outside of migration. If a thread modifies an instruction that
> + * another thread may attempt to execute, the other thread must still emit an
> + * icache flushing instruction before attempting to execute the potentially
> + * modified instruction. This must be performed by the userspace program.
> + *
> + * In per-thread context (eg. ``scope == PR_RISCV_SCOPE_PER_THREAD``), only the
> + * thread calling this function is permitted to emit icache flushing
> + * instructions. When the thread is migrated, the corresponding hart's icache
> + * will be guaranteed to be consistent with instruction storage.
> + *
> + * On kernels configured without SMP, this function is a nop as migrations
> + * across harts will not occur.
> + */
> +int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long scope)
> +{
> +#ifdef CONFIG_SMP
> + switch (ctx) {
> + case PR_RISCV_CTX_SW_FENCEI_ON:
> + switch (scope) {
> + case PR_RISCV_SCOPE_PER_PROCESS:
> + current->mm->context.force_icache_flush = true;
> + break;
> + case PR_RISCV_SCOPE_PER_THREAD:
> + current->thread.force_icache_flush = true;
> + break;
> + default:
> + return -EINVAL;
> + }
> + }
> +#endif
> + return 0;
> +}
> diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
> index 217fd4de6134..5cabbc449080 100644
> --- a/arch/riscv/mm/context.c
> +++ b/arch/riscv/mm/context.c
> @@ -15,6 +15,7 @@
> #include <asm/tlbflush.h>
> #include <asm/cacheflush.h>
> #include <asm/mmu_context.h>
> +#include <asm/switch_to.h>
>
> #ifdef CONFIG_MMU
>
> @@ -297,7 +298,8 @@ static inline void set_mm(struct mm_struct *prev,
> *
> * The "cpu" argument must be the current local CPU number.
> */
> -static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
> +static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu,
> + struct task_struct *task)
> {
> #ifdef CONFIG_SMP
> cpumask_t *mask = &mm->context.icache_stale_mask;
> @@ -309,7 +311,12 @@ static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
> * This pairs with a barrier in flush_icache_mm.
> */
> smp_mb();
> - local_flush_icache_all();
> +
> + /*
> + * If cache will be flushed in switch_to, no need to flush here.
> + */
> + if (!(task && switch_to_should_flush_icache(task)))
> + local_flush_icache_all();
> }
>
> #endif
> @@ -332,5 +339,5 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
>
> set_mm(prev, next, cpu);
>
> - flush_icache_deferred(next, cpu);
> + flush_icache_deferred(next, cpu, task);
> }
> diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> index 370ed14b1ae0..524d546d697b 100644
> --- a/include/uapi/linux/prctl.h
> +++ b/include/uapi/linux/prctl.h
> @@ -306,4 +306,10 @@ struct prctl_mm_map {
> # define PR_RISCV_V_VSTATE_CTRL_NEXT_MASK 0xc
> # define PR_RISCV_V_VSTATE_CTRL_MASK 0x1f
>
> +#define PR_RISCV_SET_ICACHE_FLUSH_CTX 71
> +# define PR_RISCV_CTX_SW_FENCEI_ON 0
> +# define PR_RISCV_CTX_SW_FENCEI_OFF 1
> +# define PR_RISCV_SCOPE_PER_PROCESS 0
> +# define PR_RISCV_SCOPE_PER_THREAD 1
> +
> #endif /* _LINUX_PRCTL_H */
> diff --git a/kernel/sys.c b/kernel/sys.c
> index e219fcfa112d..69afdd8b430f 100644
> --- a/kernel/sys.c
> +++ b/kernel/sys.c
> @@ -146,6 +146,9 @@
> #ifndef RISCV_V_GET_CONTROL
> # define RISCV_V_GET_CONTROL() (-EINVAL)
> #endif
> +#ifndef RISCV_SET_ICACHE_FLUSH_CTX
> +# define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
> +#endif
>
> /*
> * this is where the system-wide overflow UID and GID are defined, for
> @@ -2743,6 +2746,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> case PR_RISCV_V_GET_CONTROL:
> error = RISCV_V_GET_CONTROL();
> break;
> + case PR_RISCV_SET_ICACHE_FLUSH_CTX:
> + error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
> + break;
> default:
> error = -EINVAL;
> break;
>
Great, thanks for the modifications, this looks good to me!
Reviewed-by: Alexandre Ghiti <[email protected]>
Thanks,
Alex
On 25/01/2024 05:23, Charlie Jenkins wrote:
> Provide documentation that explains how to properly do CMODX in riscv.
>
> Signed-off-by: Charlie Jenkins <[email protected]>
> Reviewed-by: Atish Patra <[email protected]>
> ---
> Documentation/arch/riscv/cmodx.rst | 96 ++++++++++++++++++++++++++++++++++++++
> Documentation/arch/riscv/index.rst | 1 +
> 2 files changed, 97 insertions(+)
>
> diff --git a/Documentation/arch/riscv/cmodx.rst b/Documentation/arch/riscv/cmodx.rst
> new file mode 100644
> index 000000000000..24aafa23a72b
> --- /dev/null
> +++ b/Documentation/arch/riscv/cmodx.rst
> @@ -0,0 +1,96 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +==============================================================================
> +Concurrent Modification and Execution of Instructions (CMODX) for RISC-V Linux
> +==============================================================================
> +
> +CMODX is a programming technique where a program executes instructions that were
> +modified by the program itself. Instruction storage and the instruction cache
> +(icache) are not guaranteed to be synchronized on RISC-V hardware. Therefore, the
> +program must enforce its own synchronization with the unprivileged fence.i
> +instruction.
> +
> +However, the default Linux ABI prohibits the use of fence.i in userspace
> +applications. At any point the scheduler may migrate a task onto a new hart. If
> +migration occurs after the userspace synchronized the icache and instruction
> +storage with fence.i, the icache on the new hart will no longer be clean. This
> +is due to the behavior of fence.i only affecting the hart that it is called on.
> +Thus, the hart that the task has been migrated to may not have synchronized
> +instruction storage and icache.
> +
> +There are two ways to solve this problem: use the riscv_flush_icache() syscall,
> +or use the ``PR_RISCV_SET_ICACHE_FLUSH_CTX`` prctl() and emit fence.i in
> +userspace. The syscall performs a one-off icache flushing operation. The prctl
> +changes the Linux ABI to allow userspace to emit icache flushing operations.
> +
> +As an aside, "deferred" icache flushes can sometimes be triggered in the kernel.
> +At the time of writing, this only occurs during the riscv_flush_icache() syscall
> +and when the kernel uses copy_to_user_page(). These deferred flushes happen only
> +when the memory map being used by a hart changes. If the prctl() context caused
> +an icache flush, this deferred icache flush will be skipped as it is redundant.
> +Therefore, there will be no additional flush when using the riscv_flush_icache()
> +syscall inside of the prctl() context.
> +
> +prctl() Interface
> +---------------------
> +
> +Call prctl() with ``PR_RISCV_SET_ICACHE_FLUSH_CTX`` as the first argument. The
> +remaining arguments will be delegated to the riscv_set_icache_flush_ctx
> +function detailed below.
> +
> +.. kernel-doc:: arch/riscv/mm/cacheflush.c
> + :identifiers: riscv_set_icache_flush_ctx
> +
> +Example usage:
> +
> +The following files are meant to be compiled and linked with each other. The
> +modify_instruction() function replaces an add with 0 with an add with one,
> +causing the instruction sequence in get_value() to change from returning a zero
> +to returning a one.
> +
> +cmodx.c::
> +
> + #include <stdio.h>
> + #include <sys/prctl.h>
> +
> + extern int get_value();
> + extern void modify_instruction();
> +
> + int main()
> + {
> + int value = get_value();
> + printf("Value before cmodx: %d\n", value);
> +
> + // Call prctl before first fence.i is called inside modify_instruction
> + prctl(PR_RISCV_SET_ICACHE_FLUSH_CTX_ON, PR_RISCV_CTX_SW_FENCEI, PR_RISCV_SCOPE_PER_PROCESS);
> + modify_instruction();
> +
> + value = get_value();
> + printf("Value after cmodx: %d\n", value);
> + return 0;
> + }
> +
> +cmodx.S::
> +
> + .option norvc
> +
> + .text
> + .global modify_instruction
> + modify_instruction:
> + lw a0, new_insn
> + lui a5,%hi(old_insn)
> + sw a0,%lo(old_insn)(a5)
> + fence.i
> + ret
> +
> + .section modifiable, "awx"
> + .global get_value
> + get_value:
> + li a0, 0
> + old_insn:
> + addi a0, a0, 0
> + ret
> +
> + .data
> + new_insn:
> + addi a0, a0, 1
> diff --git a/Documentation/arch/riscv/index.rst b/Documentation/arch/riscv/index.rst
> index 4dab0cb4b900..eecf347ce849 100644
> --- a/Documentation/arch/riscv/index.rst
> +++ b/Documentation/arch/riscv/index.rst
> @@ -13,6 +13,7 @@ RISC-V architecture
> patch-acceptance
> uabi
> vector
> + cmodx
>
> features
>
>
You can add:
Reviewed-by: Alexandre Ghiti <[email protected]>
Thanks,
Alex
On Wed, Feb 07, 2024 at 12:55:24PM -0600, Samuel Holland wrote:
> Hi Charlie,
>
> On 2024-01-24 10:23 PM, Charlie Jenkins wrote:
> > Support new prctl with key PR_RISCV_SET_ICACHE_FLUSH_CTX to enable
> > optimization of cross modifying code. This prctl enables userspace code
> > to use icache flushing instructions such as fence.i with the guarantee
> > that the icache will continue to be clean after thread migration.
> >
> > Signed-off-by: Charlie Jenkins <[email protected]>
> > Reviewed-by: Atish Patra <[email protected]>
> > ---
> > arch/riscv/include/asm/mmu.h | 2 ++
> > arch/riscv/include/asm/processor.h | 7 ++++
> > arch/riscv/include/asm/switch_to.h | 13 ++++++++
> > arch/riscv/mm/cacheflush.c | 67 ++++++++++++++++++++++++++++++++++++++
> > arch/riscv/mm/context.c | 13 ++++++--
> > include/uapi/linux/prctl.h | 6 ++++
> > kernel/sys.c | 6 ++++
> > 7 files changed, 111 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/riscv/include/asm/mmu.h b/arch/riscv/include/asm/mmu.h
> > index 355504b37f8e..60be458e94da 100644
> > --- a/arch/riscv/include/asm/mmu.h
> > +++ b/arch/riscv/include/asm/mmu.h
> > @@ -19,6 +19,8 @@ typedef struct {
> > #ifdef CONFIG_SMP
> > /* A local icache flush is needed before user execution can resume. */
> > cpumask_t icache_stale_mask;
> > + /* Force local icache flush on all migrations. */
> > + bool force_icache_flush;
> > #endif
> > #ifdef CONFIG_BINFMT_ELF_FDPIC
> > unsigned long exec_fdpic_loadmap;
> > diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h
> > index a8509cc31ab2..816cdc2395f4 100644
> > --- a/arch/riscv/include/asm/processor.h
> > +++ b/arch/riscv/include/asm/processor.h
> > @@ -123,6 +123,10 @@ struct thread_struct {
> > struct __riscv_v_ext_state vstate;
> > unsigned long align_ctl;
> > struct __riscv_v_ext_state kernel_vstate;
> > +#ifdef CONFIG_SMP
> > + bool force_icache_flush;
> > + unsigned int prev_cpu;
> > +#endif
> > };
> >
> > /* Whitelist the fstate from the task_struct for hardened usercopy */
> > @@ -184,6 +188,9 @@ extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
> > #define GET_UNALIGN_CTL(tsk, addr) get_unalign_ctl((tsk), (addr))
> > #define SET_UNALIGN_CTL(tsk, val) set_unalign_ctl((tsk), (val))
> >
> > +#define RISCV_SET_ICACHE_FLUSH_CTX(arg1, arg2) riscv_set_icache_flush_ctx(arg1, arg2)
> > +extern int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long per_thread);
> > +
> > #endif /* __ASSEMBLY__ */
> >
> > #endif /* _ASM_RISCV_PROCESSOR_H */
> > diff --git a/arch/riscv/include/asm/switch_to.h b/arch/riscv/include/asm/switch_to.h
> > index 7efdb0584d47..ac99ab64499c 100644
> > --- a/arch/riscv/include/asm/switch_to.h
> > +++ b/arch/riscv/include/asm/switch_to.h
> > @@ -8,6 +8,7 @@
> >
> > #include <linux/jump_label.h>
> > #include <linux/sched/task_stack.h>
> > +#include <linux/mm_types.h>
> > #include <asm/vector.h>
> > #include <asm/cpufeature.h>
> > #include <asm/processor.h>
> > @@ -72,14 +73,26 @@ static __always_inline bool has_fpu(void) { return false; }
> > extern struct task_struct *__switch_to(struct task_struct *,
> > struct task_struct *);
> >
> > +static inline bool switch_to_should_flush_icache(struct task_struct *task)
> > +{
> > + bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
>
> No parentheses needed here.
>
> > + bool stale_thread = task->thread.force_icache_flush;
> > + bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
>
> This fails to compile when SMP is disabled:
>
> In file included from arch/riscv/kernel/ptrace.c:14:
> ./arch/riscv/include/asm/switch_to.h:78:49: error: no member named
> 'force_icache_flush' in 'mm_context_t'
> bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
> ~~~~~~~~~~~~~~~~~ ^
> ./arch/riscv/include/asm/switch_to.h:79:35: error: no member named
> 'force_icache_flush' in 'struct thread_struct'
> bool stale_thread = task->thread.force_icache_flush;
> ~~~~~~~~~~~~ ^
> ./arch/riscv/include/asm/switch_to.h:80:60: error: no member named 'prev_cpu' in
> 'struct thread_struct'
> bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
> ~~~~~~~~~~~~ ^
> CC arch/riscv/kernel/sys_hwprobe.o
>
> > +
> > + return thread_migrated && (stale_mm || stale_thread);
> > +}
> > +
> > #define switch_to(prev, next, last) \
> > do { \
> > struct task_struct *__prev = (prev); \
> > struct task_struct *__next = (next); \
> > + __next->thread.prev_cpu = smp_processor_id(); \
>
> Shouldn't this instead update __prev (the task being scheduled out)?
Hmm, yes that makes sense. I shouldn't have moved this to above the
switch_to_should_flush_icache call. If this is after
switch_to_should_flush_icache, then setting __prev or __next will have
the same effect but __prev probably makes more sense.
>
> > if (has_fpu()) \
> > __switch_to_fpu(__prev, __next); \
> > if (has_vector()) \
> > __switch_to_vector(__prev, __next); \
> > + if (switch_to_should_flush_icache(__next)) \
> > + local_flush_icache_all(); \
> > ((last) = __switch_to(__prev, __next)); \
> > } while (0)
> >
> > diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
> > index 55a34f2020a8..2d5e1575f6c1 100644
> > --- a/arch/riscv/mm/cacheflush.c
> > +++ b/arch/riscv/mm/cacheflush.c
> > @@ -5,6 +5,7 @@
> >
> > #include <linux/acpi.h>
> > #include <linux/of.h>
> > +#include <linux/prctl.h>
> > #include <asm/acpi.h>
> > #include <asm/cacheflush.h>
> >
> > @@ -152,3 +153,69 @@ void __init riscv_init_cbo_blocksizes(void)
> > if (cboz_block_size)
> > riscv_cboz_block_size = cboz_block_size;
> > }
> > +
> > +/**
> > + * riscv_set_icache_flush_ctx() - Enable/disable icache flushing instructions in
> > + * userspace.
> > + * @ctx: Set the type of icache flushing instructions permitted/prohibited in
> > + * userspace. Supported values described below.
> > + *
> > + * Supported values for ctx:
> > + *
> > + * * %PR_RISCV_CTX_SW_FENCEI_ON: Allow fence.i in userspace.
> > + *
> > + * * %PR_RISCV_CTX_SW_FENCEI_OFF: Disallow fence.i in userspace. All threads in
> > + * a process will be affected when ``scope == PR_RISCV_SCOPE_PER_PROCESS``.
> > + * Therefore, caution must be taken -- only use this flag when you can
> > + * guarantee that no thread in the process will emit fence.i from this point
> > + * onward.
> > + *
> > + * @scope: Set scope of where icache flushing instructions are allowed to be
> > + * emitted. Supported values described below.
> > + *
> > + * Supported values for scope:
> > + *
> > + * * %PR_RISCV_SCOPE_PER_PROCESS: Ensure the icache of any thread in this process
> > + * is coherent with instruction storage upon
> > + * migration.
> > + *
> > + * * %PR_RISCV_SCOPE_PER_THREAD: Ensure the icache of the current thread is
> > + * coherent with instruction storage upon
> > + * migration.
> > + *
> > + * When ``scope == PR_RISCV_SCOPE_PER_PROCESS``, all threads in the process are
> > + * permitted to emit icache flushing instructions. Whenever any thread in the
> > + * process is migrated, the corresponding hart's icache will be guaranteed to be
> > + * consistent with instruction storage. Note this does not enforce any
> > + * guarantees outside of migration. If a thread modifies an instruction that
> > + * another thread may attempt to execute, the other thread must still emit an
> > + * icache flushing instruction before attempting to execute the potentially
> > + * modified instruction. This must be performed by the userspace program.
> > + *
> > + * In per-thread context (eg. ``scope == PR_RISCV_SCOPE_PER_THREAD``), only the
> > + * thread calling this function is permitted to emit icache flushing
> > + * instructions. When the thread is migrated, the corresponding hart's icache
> > + * will be guaranteed to be consistent with instruction storage.
> > + *
> > + * On kernels configured without SMP, this function is a nop as migrations
> > + * across harts will not occur.
> > + */
> > +int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long scope)
> > +{
> > +#ifdef CONFIG_SMP
> > + switch (ctx) {
> > + case PR_RISCV_CTX_SW_FENCEI_ON:
> > + switch (scope) {
> > + case PR_RISCV_SCOPE_PER_PROCESS:
> > + current->mm->context.force_icache_flush = true;
> > + break;
> > + case PR_RISCV_SCOPE_PER_THREAD:
> > + current->thread.force_icache_flush = true;
> > + break;
> > + default:
> > + return -EINVAL;
> > + }
>
> At some point you dropped the implementation for PR_RISCV_CTX_SW_FENCEI_OFF.
> It's still documented, so I don't think that was intentional.
>
Yikes!
> Also, PR_RISCV_CTX_SW_FENCEI_OFF needs to mark the icache on all other CPUs as
> stale, as in flush_icache_mm(). Consider the following sequence:
> 1. Migrate from CPU A to CPU B
> 2. CMODX on CPU B
> 3. fence.i in userspace on CPU B
> 4. prctl(PR_RISCV_CTX_SW_FENCEI_OFF) on CPU B
> 5. Migrate back to CPU A -- force_icache_flush is false, no flush happens
> 6. Execute the modified code -- fails, because CPU A's I$ never got flushed
>
Oh good catch, thank you.
> > + }
> > +#endif
> > + return 0;
>
> This makes the prtcl() call silently succeed for invalid ctx values, whereas it
> should fail for those instead.
>
> Regards,
> Samuel
>
> > +}
> > diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
> > index 217fd4de6134..5cabbc449080 100644
> > --- a/arch/riscv/mm/context.c
> > +++ b/arch/riscv/mm/context.c
> > @@ -15,6 +15,7 @@
> > #include <asm/tlbflush.h>
> > #include <asm/cacheflush.h>
> > #include <asm/mmu_context.h>
> > +#include <asm/switch_to.h>
> >
> > #ifdef CONFIG_MMU
> >
> > @@ -297,7 +298,8 @@ static inline void set_mm(struct mm_struct *prev,
> > *
> > * The "cpu" argument must be the current local CPU number.
> > */
> > -static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
> > +static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu,
> > + struct task_struct *task)
> > {
> > #ifdef CONFIG_SMP
> > cpumask_t *mask = &mm->context.icache_stale_mask;
> > @@ -309,7 +311,12 @@ static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
> > * This pairs with a barrier in flush_icache_mm.
> > */
> > smp_mb();
> > - local_flush_icache_all();
> > +
> > + /*
> > + * If cache will be flushed in switch_to, no need to flush here.
> > + */
> > + if (!(task && switch_to_should_flush_icache(task)))
> > + local_flush_icache_all();
> > }
> >
> > #endif
> > @@ -332,5 +339,5 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
> >
> > set_mm(prev, next, cpu);
> >
> > - flush_icache_deferred(next, cpu);
> > + flush_icache_deferred(next, cpu, task);
> > }
> > diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> > index 370ed14b1ae0..524d546d697b 100644
> > --- a/include/uapi/linux/prctl.h
> > +++ b/include/uapi/linux/prctl.h
> > @@ -306,4 +306,10 @@ struct prctl_mm_map {
> > # define PR_RISCV_V_VSTATE_CTRL_NEXT_MASK 0xc
> > # define PR_RISCV_V_VSTATE_CTRL_MASK 0x1f
> >
> > +#define PR_RISCV_SET_ICACHE_FLUSH_CTX 71
> > +# define PR_RISCV_CTX_SW_FENCEI_ON 0
> > +# define PR_RISCV_CTX_SW_FENCEI_OFF 1
> > +# define PR_RISCV_SCOPE_PER_PROCESS 0
> > +# define PR_RISCV_SCOPE_PER_THREAD 1
> > +
> > #endif /* _LINUX_PRCTL_H */
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index e219fcfa112d..69afdd8b430f 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -146,6 +146,9 @@
> > #ifndef RISCV_V_GET_CONTROL
> > # define RISCV_V_GET_CONTROL() (-EINVAL)
> > #endif
> > +#ifndef RISCV_SET_ICACHE_FLUSH_CTX
> > +# define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
> > +#endif
> >
> > /*
> > * this is where the system-wide overflow UID and GID are defined, for
> > @@ -2743,6 +2746,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> > case PR_RISCV_V_GET_CONTROL:
> > error = RISCV_V_GET_CONTROL();
> > break;
> > + case PR_RISCV_SET_ICACHE_FLUSH_CTX:
> > + error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
> > + break;
> > default:
> > error = -EINVAL;
> > break;
> >
>
I will send out a new verson.
- Charlie
On 07/02/2024 19:55, Samuel Holland wrote:
> Hi Charlie,
>
> On 2024-01-24 10:23 PM, Charlie Jenkins wrote:
>> Support new prctl with key PR_RISCV_SET_ICACHE_FLUSH_CTX to enable
>> optimization of cross modifying code. This prctl enables userspace code
>> to use icache flushing instructions such as fence.i with the guarantee
>> that the icache will continue to be clean after thread migration.
>>
>> Signed-off-by: Charlie Jenkins <[email protected]>
>> Reviewed-by: Atish Patra <[email protected]>
>> ---
>> arch/riscv/include/asm/mmu.h | 2 ++
>> arch/riscv/include/asm/processor.h | 7 ++++
>> arch/riscv/include/asm/switch_to.h | 13 ++++++++
>> arch/riscv/mm/cacheflush.c | 67 ++++++++++++++++++++++++++++++++++++++
>> arch/riscv/mm/context.c | 13 ++++++--
>> include/uapi/linux/prctl.h | 6 ++++
>> kernel/sys.c | 6 ++++
>> 7 files changed, 111 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/riscv/include/asm/mmu.h b/arch/riscv/include/asm/mmu.h
>> index 355504b37f8e..60be458e94da 100644
>> --- a/arch/riscv/include/asm/mmu.h
>> +++ b/arch/riscv/include/asm/mmu.h
>> @@ -19,6 +19,8 @@ typedef struct {
>> #ifdef CONFIG_SMP
>> /* A local icache flush is needed before user execution can resume. */
>> cpumask_t icache_stale_mask;
>> + /* Force local icache flush on all migrations. */
>> + bool force_icache_flush;
>> #endif
>> #ifdef CONFIG_BINFMT_ELF_FDPIC
>> unsigned long exec_fdpic_loadmap;
>> diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h
>> index a8509cc31ab2..816cdc2395f4 100644
>> --- a/arch/riscv/include/asm/processor.h
>> +++ b/arch/riscv/include/asm/processor.h
>> @@ -123,6 +123,10 @@ struct thread_struct {
>> struct __riscv_v_ext_state vstate;
>> unsigned long align_ctl;
>> struct __riscv_v_ext_state kernel_vstate;
>> +#ifdef CONFIG_SMP
>> + bool force_icache_flush;
>> + unsigned int prev_cpu;
>> +#endif
>> };
>>
>> /* Whitelist the fstate from the task_struct for hardened usercopy */
>> @@ -184,6 +188,9 @@ extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
>> #define GET_UNALIGN_CTL(tsk, addr) get_unalign_ctl((tsk), (addr))
>> #define SET_UNALIGN_CTL(tsk, val) set_unalign_ctl((tsk), (val))
>>
>> +#define RISCV_SET_ICACHE_FLUSH_CTX(arg1, arg2) riscv_set_icache_flush_ctx(arg1, arg2)
>> +extern int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long per_thread);
>> +
>> #endif /* __ASSEMBLY__ */
>>
>> #endif /* _ASM_RISCV_PROCESSOR_H */
>> diff --git a/arch/riscv/include/asm/switch_to.h b/arch/riscv/include/asm/switch_to.h
>> index 7efdb0584d47..ac99ab64499c 100644
>> --- a/arch/riscv/include/asm/switch_to.h
>> +++ b/arch/riscv/include/asm/switch_to.h
>> @@ -8,6 +8,7 @@
>>
>> #include <linux/jump_label.h>
>> #include <linux/sched/task_stack.h>
>> +#include <linux/mm_types.h>
>> #include <asm/vector.h>
>> #include <asm/cpufeature.h>
>> #include <asm/processor.h>
>> @@ -72,14 +73,26 @@ static __always_inline bool has_fpu(void) { return false; }
>> extern struct task_struct *__switch_to(struct task_struct *,
>> struct task_struct *);
>>
>> +static inline bool switch_to_should_flush_icache(struct task_struct *task)
>> +{
>> + bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
> No parentheses needed here.
>
>> + bool stale_thread = task->thread.force_icache_flush;
>> + bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
> This fails to compile when SMP is disabled:
>
> In file included from arch/riscv/kernel/ptrace.c:14:
> ./arch/riscv/include/asm/switch_to.h:78:49: error: no member named
> 'force_icache_flush' in 'mm_context_t'
> bool stale_mm = task->mm && (task->mm->context.force_icache_flush);
> ~~~~~~~~~~~~~~~~~ ^
> ./arch/riscv/include/asm/switch_to.h:79:35: error: no member named
> 'force_icache_flush' in 'struct thread_struct'
> bool stale_thread = task->thread.force_icache_flush;
> ~~~~~~~~~~~~ ^
> ./arch/riscv/include/asm/switch_to.h:80:60: error: no member named 'prev_cpu' in
> 'struct thread_struct'
> bool thread_migrated = smp_processor_id() != task->thread.prev_cpu;
> ~~~~~~~~~~~~ ^
> CC arch/riscv/kernel/sys_hwprobe.o
>
>> +
>> + return thread_migrated && (stale_mm || stale_thread);
>> +}
>> +
>> #define switch_to(prev, next, last) \
>> do { \
>> struct task_struct *__prev = (prev); \
>> struct task_struct *__next = (next); \
>> + __next->thread.prev_cpu = smp_processor_id(); \
> Shouldn't this instead update __prev (the task being scheduled out)?
We need to know what was the last cpu a task was executed on, so either
we store this info before it gets scheduled (what we do here) or when it
is about to get scheduled out, to me that makes no difference.
>
>> if (has_fpu()) \
>> __switch_to_fpu(__prev, __next); \
>> if (has_vector()) \
>> __switch_to_vector(__prev, __next); \
>> + if (switch_to_should_flush_icache(__next)) \
>> + local_flush_icache_all(); \
>> ((last) = __switch_to(__prev, __next)); \
>> } while (0)
>>
>> diff --git a/arch/riscv/mm/cacheflush.c b/arch/riscv/mm/cacheflush.c
>> index 55a34f2020a8..2d5e1575f6c1 100644
>> --- a/arch/riscv/mm/cacheflush.c
>> +++ b/arch/riscv/mm/cacheflush.c
>> @@ -5,6 +5,7 @@
>>
>> #include <linux/acpi.h>
>> #include <linux/of.h>
>> +#include <linux/prctl.h>
>> #include <asm/acpi.h>
>> #include <asm/cacheflush.h>
>>
>> @@ -152,3 +153,69 @@ void __init riscv_init_cbo_blocksizes(void)
>> if (cboz_block_size)
>> riscv_cboz_block_size = cboz_block_size;
>> }
>> +
>> +/**
>> + * riscv_set_icache_flush_ctx() - Enable/disable icache flushing instructions in
>> + * userspace.
>> + * @ctx: Set the type of icache flushing instructions permitted/prohibited in
>> + * userspace. Supported values described below.
>> + *
>> + * Supported values for ctx:
>> + *
>> + * * %PR_RISCV_CTX_SW_FENCEI_ON: Allow fence.i in userspace.
>> + *
>> + * * %PR_RISCV_CTX_SW_FENCEI_OFF: Disallow fence.i in userspace. All threads in
>> + * a process will be affected when ``scope == PR_RISCV_SCOPE_PER_PROCESS``.
>> + * Therefore, caution must be taken -- only use this flag when you can
>> + * guarantee that no thread in the process will emit fence.i from this point
>> + * onward.
>> + *
>> + * @scope: Set scope of where icache flushing instructions are allowed to be
>> + * emitted. Supported values described below.
>> + *
>> + * Supported values for scope:
>> + *
>> + * * %PR_RISCV_SCOPE_PER_PROCESS: Ensure the icache of any thread in this process
>> + * is coherent with instruction storage upon
>> + * migration.
>> + *
>> + * * %PR_RISCV_SCOPE_PER_THREAD: Ensure the icache of the current thread is
>> + * coherent with instruction storage upon
>> + * migration.
>> + *
>> + * When ``scope == PR_RISCV_SCOPE_PER_PROCESS``, all threads in the process are
>> + * permitted to emit icache flushing instructions. Whenever any thread in the
>> + * process is migrated, the corresponding hart's icache will be guaranteed to be
>> + * consistent with instruction storage. Note this does not enforce any
>> + * guarantees outside of migration. If a thread modifies an instruction that
>> + * another thread may attempt to execute, the other thread must still emit an
>> + * icache flushing instruction before attempting to execute the potentially
>> + * modified instruction. This must be performed by the userspace program.
>> + *
>> + * In per-thread context (eg. ``scope == PR_RISCV_SCOPE_PER_THREAD``), only the
>> + * thread calling this function is permitted to emit icache flushing
>> + * instructions. When the thread is migrated, the corresponding hart's icache
>> + * will be guaranteed to be consistent with instruction storage.
>> + *
>> + * On kernels configured without SMP, this function is a nop as migrations
>> + * across harts will not occur.
>> + */
>> +int riscv_set_icache_flush_ctx(unsigned long ctx, unsigned long scope)
>> +{
>> +#ifdef CONFIG_SMP
>> + switch (ctx) {
>> + case PR_RISCV_CTX_SW_FENCEI_ON:
>> + switch (scope) {
>> + case PR_RISCV_SCOPE_PER_PROCESS:
>> + current->mm->context.force_icache_flush = true;
>> + break;
>> + case PR_RISCV_SCOPE_PER_THREAD:
>> + current->thread.force_icache_flush = true;
>> + break;
>> + default:
>> + return -EINVAL;
>> + }
> At some point you dropped the implementation for PR_RISCV_CTX_SW_FENCEI_OFF.
> It's still documented, so I don't think that was intentional.
>
> Also, PR_RISCV_CTX_SW_FENCEI_OFF needs to mark the icache on all other CPUs as
> stale, as in flush_icache_mm(). Consider the following sequence:
> 1. Migrate from CPU A to CPU B
> 2. CMODX on CPU B
> 3. fence.i in userspace on CPU B
> 4. prctl(PR_RISCV_CTX_SW_FENCEI_OFF) on CPU B
> 5. Migrate back to CPU A -- force_icache_flush is false, no flush happens
> 6. Execute the modified code -- fails, because CPU A's I$ never got flushed
>
>> + }
>> +#endif
>> + return 0;
> This makes the prtcl() call silently succeed for invalid ctx values, whereas it
> should fail for those instead.
>
> Regards,
> Samuel
>
>> +}
>> diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
>> index 217fd4de6134..5cabbc449080 100644
>> --- a/arch/riscv/mm/context.c
>> +++ b/arch/riscv/mm/context.c
>> @@ -15,6 +15,7 @@
>> #include <asm/tlbflush.h>
>> #include <asm/cacheflush.h>
>> #include <asm/mmu_context.h>
>> +#include <asm/switch_to.h>
>>
>> #ifdef CONFIG_MMU
>>
>> @@ -297,7 +298,8 @@ static inline void set_mm(struct mm_struct *prev,
>> *
>> * The "cpu" argument must be the current local CPU number.
>> */
>> -static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
>> +static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu,
>> + struct task_struct *task)
>> {
>> #ifdef CONFIG_SMP
>> cpumask_t *mask = &mm->context.icache_stale_mask;
>> @@ -309,7 +311,12 @@ static inline void flush_icache_deferred(struct mm_struct *mm, unsigned int cpu)
>> * This pairs with a barrier in flush_icache_mm.
>> */
>> smp_mb();
>> - local_flush_icache_all();
>> +
>> + /*
>> + * If cache will be flushed in switch_to, no need to flush here.
>> + */
>> + if (!(task && switch_to_should_flush_icache(task)))
>> + local_flush_icache_all();
>> }
>>
>> #endif
>> @@ -332,5 +339,5 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
>>
>> set_mm(prev, next, cpu);
>>
>> - flush_icache_deferred(next, cpu);
>> + flush_icache_deferred(next, cpu, task);
>> }
>> diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
>> index 370ed14b1ae0..524d546d697b 100644
>> --- a/include/uapi/linux/prctl.h
>> +++ b/include/uapi/linux/prctl.h
>> @@ -306,4 +306,10 @@ struct prctl_mm_map {
>> # define PR_RISCV_V_VSTATE_CTRL_NEXT_MASK 0xc
>> # define PR_RISCV_V_VSTATE_CTRL_MASK 0x1f
>>
>> +#define PR_RISCV_SET_ICACHE_FLUSH_CTX 71
>> +# define PR_RISCV_CTX_SW_FENCEI_ON 0
>> +# define PR_RISCV_CTX_SW_FENCEI_OFF 1
>> +# define PR_RISCV_SCOPE_PER_PROCESS 0
>> +# define PR_RISCV_SCOPE_PER_THREAD 1
>> +
>> #endif /* _LINUX_PRCTL_H */
>> diff --git a/kernel/sys.c b/kernel/sys.c
>> index e219fcfa112d..69afdd8b430f 100644
>> --- a/kernel/sys.c
>> +++ b/kernel/sys.c
>> @@ -146,6 +146,9 @@
>> #ifndef RISCV_V_GET_CONTROL
>> # define RISCV_V_GET_CONTROL() (-EINVAL)
>> #endif
>> +#ifndef RISCV_SET_ICACHE_FLUSH_CTX
>> +# define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
>> +#endif
>>
>> /*
>> * this is where the system-wide overflow UID and GID are defined, for
>> @@ -2743,6 +2746,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
>> case PR_RISCV_V_GET_CONTROL:
>> error = RISCV_V_GET_CONTROL();
>> break;
>> + case PR_RISCV_SET_ICACHE_FLUSH_CTX:
>> + error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
>> + break;
>> default:
>> error = -EINVAL;
>> break;
>>
>
> _______________________________________________
> linux-riscv mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-riscv