* [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe()
@ 2026-09-18 14:32 Adriano Cordova
2026-09-18 14:32 ` [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths Adriano Cordova
2026-09-20 14:35 ` [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Oleg Nesterov
0 siblings, 2 replies; 4+ messages in thread
From: Adriano Cordova @ 2026-09-18 14:32 UTC (permalink / raw)
To: Masami Hiramatsu, Oleg Nesterov, Peter Zijlstra
Cc: linux-trace-kernel, linux-kernel, Ingo Molnar,
Arnaldo Carvalho de Melo, Namhyung Kim, Mark Rutland,
Alexander Shishkin, Jiri Olsa, Ian Rogers, Adrian Hunter,
James Clark, Adriano Cordova
prepare_uprobe() serializes its callers by taking consumer_rwsem for
writing. That rwsem protects the consumer list, not the instruction
copy.
Give struct uprobe a dedicated prepare_mutex and use it instead.
No functional change.
Signed-off-by: Adriano Cordova <adrianox@gmail.com>
---
kernel/events/uprobes.c | 8 +++++---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 7709ea882477..686585539ebf 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -64,6 +64,7 @@ struct uprobe {
refcount_t ref;
struct rw_semaphore register_rwsem;
struct rw_semaphore consumer_rwsem;
+ struct mutex prepare_mutex; /* serializes prepare_uprobe() */
struct list_head pending_list;
struct list_head consumers;
struct inode *inode; /* Also hold a ref to inode */
@@ -1007,6 +1008,7 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
INIT_LIST_HEAD(&uprobe->consumers);
init_rwsem(&uprobe->register_rwsem);
init_rwsem(&uprobe->consumer_rwsem);
+ mutex_init(&uprobe->prepare_mutex);
RB_CLEAR_NODE(&uprobe->rb_node);
refcount_set(&uprobe->ref, 1);
@@ -1104,8 +1106,8 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
return ret;
- /* TODO: move this into _register, until then we abuse this sem. */
- down_write(&uprobe->consumer_rwsem);
+ /* Serialize concurrent prepare_uprobe() calls from uprobe_mmap(). */
+ mutex_lock(&uprobe->prepare_mutex);
if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
goto out;
@@ -1125,7 +1127,7 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
set_bit(UPROBE_COPY_INSN, &uprobe->flags);
out:
- up_write(&uprobe->consumer_rwsem);
+ mutex_unlock(&uprobe->prepare_mutex);
return ret;
}
--
2.51.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths
2026-09-18 14:32 [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Adriano Cordova
@ 2026-09-18 14:32 ` Adriano Cordova
2026-09-20 14:38 ` Oleg Nesterov
2026-09-20 14:35 ` [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Oleg Nesterov
1 sibling, 1 reply; 4+ messages in thread
From: Adriano Cordova @ 2026-09-18 14:32 UTC (permalink / raw)
To: Masami Hiramatsu, Oleg Nesterov, Peter Zijlstra
Cc: linux-trace-kernel, linux-kernel, Ingo Molnar,
Arnaldo Carvalho de Melo, Namhyung Kim, Mark Rutland,
Alexander Shishkin, Jiri Olsa, Ian Rogers, Adrian Hunter,
James Clark, Adriano Cordova
find_uprobe_rcu() backs handle_swbp() and handle_syscall_uprobe(). Make
it skip a uprobe whose instruction has not been analyzed yet, so neither
path can use an uninitialized ->arch. An unprepared uprobe has no
breakpoint installed (install_breakpoint() prepares before set_swbp()),
so the trap path treats it as not registered yet. This also gives
handle_syscall_uprobe() the check it was missing.
Signed-off-by: Adriano Cordova <adrianox@gmail.com>
---
kernel/events/uprobes.c | 35 +++++++++++++++--------------------
1 file changed, 15 insertions(+), 20 deletions(-)
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 686585539ebf..bdc05fdce0a0 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -924,8 +924,20 @@ static struct uprobe *find_uprobe_rcu(struct inode *inode, loff_t offset)
* try again as we might have missed the element (false
* negative). If seqcount is unchanged, search truly failed.
*/
- if (node)
- return __node_2_uprobe(node);
+ if (node) {
+ struct uprobe *uprobe = __node_2_uprobe(node);
+
+ /*
+ * A uprobe is inserted before it is prepared; do
+ * not hand it out until ->arch is usable.
+ */
+ if (!test_bit(UPROBE_COPY_INSN, &uprobe->flags))
+ return NULL;
+
+ /* Pairs with the smp_wmb() in prepare_uprobe(). */
+ smp_rmb();
+ return uprobe;
+ }
} while (read_seqcount_retry(&uprobes_seqcount, seq));
return NULL;
@@ -1123,7 +1135,7 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
if (ret)
goto out;
- smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
+ smp_wmb(); /* pairs with the smp_rmb() in find_uprobe_rcu() */
set_bit(UPROBE_COPY_INSN, &uprobe->flags);
out:
@@ -2743,23 +2755,6 @@ static void handle_swbp(struct pt_regs *regs)
/* change it in advance for ->handler() and restart */
instruction_pointer_set(regs, bp_vaddr);
- /*
- * TODO: move copy_insn/etc into _register and remove this hack.
- * After we hit the bp, _unregister + _register can install the
- * new and not-yet-analyzed uprobe at the same address, restart.
- */
- if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
- goto out;
-
- /*
- * Pairs with the smp_wmb() in prepare_uprobe().
- *
- * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
- * we must also see the stores to &uprobe->arch performed by the
- * prepare_uprobe() call.
- */
- smp_rmb();
-
/* Tracing handlers use ->utask to communicate with fetch methods */
if (!get_utask())
goto out;
--
2.51.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe()
2026-09-18 14:32 [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Adriano Cordova
2026-09-18 14:32 ` [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths Adriano Cordova
@ 2026-09-20 14:35 ` Oleg Nesterov
1 sibling, 0 replies; 4+ messages in thread
From: Oleg Nesterov @ 2026-09-20 14:35 UTC (permalink / raw)
To: Adriano Cordova
Cc: Masami Hiramatsu, Peter Zijlstra, linux-trace-kernel,
linux-kernel, Ingo Molnar, Arnaldo Carvalho de Melo,
Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
Ian Rogers, Adrian Hunter, James Clark
On 09/18, Adriano Cordova wrote:
>
> prepare_uprobe() serializes its callers by taking consumer_rwsem for
> writing. That rwsem protects the consumer list, not the instruction
> copy.
>
> Give struct uprobe a dedicated prepare_mutex and use it instead.
>
> No functional change.
So why? Is it preparation for 2/2 ? At first glance it doesn't need
this change... Please add more details.
Oleg.
> Signed-off-by: Adriano Cordova <adrianox@gmail.com>
> ---
> kernel/events/uprobes.c | 8 +++++---
> 1 file changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index 7709ea882477..686585539ebf 100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -64,6 +64,7 @@ struct uprobe {
> refcount_t ref;
> struct rw_semaphore register_rwsem;
> struct rw_semaphore consumer_rwsem;
> + struct mutex prepare_mutex; /* serializes prepare_uprobe() */
> struct list_head pending_list;
> struct list_head consumers;
> struct inode *inode; /* Also hold a ref to inode */
> @@ -1007,6 +1008,7 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
> INIT_LIST_HEAD(&uprobe->consumers);
> init_rwsem(&uprobe->register_rwsem);
> init_rwsem(&uprobe->consumer_rwsem);
> + mutex_init(&uprobe->prepare_mutex);
> RB_CLEAR_NODE(&uprobe->rb_node);
> refcount_set(&uprobe->ref, 1);
>
> @@ -1104,8 +1106,8 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
> if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
> return ret;
>
> - /* TODO: move this into _register, until then we abuse this sem. */
> - down_write(&uprobe->consumer_rwsem);
> + /* Serialize concurrent prepare_uprobe() calls from uprobe_mmap(). */
> + mutex_lock(&uprobe->prepare_mutex);
> if (test_bit(UPROBE_COPY_INSN, &uprobe->flags))
> goto out;
>
> @@ -1125,7 +1127,7 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
> set_bit(UPROBE_COPY_INSN, &uprobe->flags);
>
> out:
> - up_write(&uprobe->consumer_rwsem);
> + mutex_unlock(&uprobe->prepare_mutex);
>
> return ret;
> }
> --
> 2.51.0
>
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths
2026-09-18 14:32 ` [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths Adriano Cordova
@ 2026-09-20 14:38 ` Oleg Nesterov
0 siblings, 0 replies; 4+ messages in thread
From: Oleg Nesterov @ 2026-09-20 14:38 UTC (permalink / raw)
To: Adriano Cordova
Cc: Masami Hiramatsu, Peter Zijlstra, linux-trace-kernel,
linux-kernel, Ingo Molnar, Arnaldo Carvalho de Melo,
Namhyung Kim, Mark Rutland, Alexander Shishkin, Jiri Olsa,
Ian Rogers, Adrian Hunter, James Clark, Andrii Nakryiko
On 09/18, Adriano Cordova wrote:
>
> find_uprobe_rcu() backs handle_swbp() and handle_syscall_uprobe(). Make
> it skip a uprobe whose instruction has not been analyzed yet, so neither
> path can use an uninitialized ->arch. An unprepared uprobe has no
> breakpoint installed (install_breakpoint() prepares before set_swbp()),
> so the trap path treats it as not registered yet. This also gives
> handle_syscall_uprobe() the check it was missing.
Sorry, I am confused.
Can you explain in more details what problem(s) this series tries to fix?
handle_swbp() checks UPROBE_COPY_INSN.
handle_syscall_uprobe() is only called by sys_uprobe() which checks
in_uprobe_trampoline(). So it can only be called after the probed insn
has been analyzed and optimized.
Oleg.
> Signed-off-by: Adriano Cordova <adrianox@gmail.com>
> ---
> kernel/events/uprobes.c | 35 +++++++++++++++--------------------
> 1 file changed, 15 insertions(+), 20 deletions(-)
>
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index 686585539ebf..bdc05fdce0a0 100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -924,8 +924,20 @@ static struct uprobe *find_uprobe_rcu(struct inode *inode, loff_t offset)
> * try again as we might have missed the element (false
> * negative). If seqcount is unchanged, search truly failed.
> */
> - if (node)
> - return __node_2_uprobe(node);
> + if (node) {
> + struct uprobe *uprobe = __node_2_uprobe(node);
> +
> + /*
> + * A uprobe is inserted before it is prepared; do
> + * not hand it out until ->arch is usable.
> + */
> + if (!test_bit(UPROBE_COPY_INSN, &uprobe->flags))
> + return NULL;
> +
> + /* Pairs with the smp_wmb() in prepare_uprobe(). */
> + smp_rmb();
> + return uprobe;
> + }
> } while (read_seqcount_retry(&uprobes_seqcount, seq));
>
> return NULL;
> @@ -1123,7 +1135,7 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
> if (ret)
> goto out;
>
> - smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
> + smp_wmb(); /* pairs with the smp_rmb() in find_uprobe_rcu() */
> set_bit(UPROBE_COPY_INSN, &uprobe->flags);
>
> out:
> @@ -2743,23 +2755,6 @@ static void handle_swbp(struct pt_regs *regs)
> /* change it in advance for ->handler() and restart */
> instruction_pointer_set(regs, bp_vaddr);
>
> - /*
> - * TODO: move copy_insn/etc into _register and remove this hack.
> - * After we hit the bp, _unregister + _register can install the
> - * new and not-yet-analyzed uprobe at the same address, restart.
> - */
> - if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
> - goto out;
> -
> - /*
> - * Pairs with the smp_wmb() in prepare_uprobe().
> - *
> - * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
> - * we must also see the stores to &uprobe->arch performed by the
> - * prepare_uprobe() call.
> - */
> - smp_rmb();
> -
> /* Tracing handlers use ->utask to communicate with fetch methods */
> if (!get_utask())
> goto out;
> --
> 2.51.0
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-09-20 14:38 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-09-18 14:32 [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Adriano Cordova
2026-09-18 14:32 ` [PATCH 2/2] uprobes: Hide unprepared uprobes from the breakpoint paths Adriano Cordova
2026-09-20 14:38 ` Oleg Nesterov
2026-09-20 14:35 ` [PATCH 1/2] uprobes: Use a dedicated mutex to serialize prepare_uprobe() Oleg Nesterov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®