Skip to contentSkip to conversation

uprobes: Add uprobe_task->dup_work/dup_addr

PATCH Superseded Oleg Nesterov 3 participants 13 messages Nov 8, 2013 – Nov 25, 2013

Subsystems (2)
lore
Share & tools
lore

Oleg Nesterov

author maintainer
#0 lore raw

PATCH 1/1 uprobes: Add uprobe_task->dup_work/dup_addr

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

---
 include/linux/uprobes.h |   21 ++++++++++++++++-----
 kernel/events/uprobes.c |   16 ++++------------
 2 files changed, 20 insertions(+), 17 deletions(-)
include/linux/uprobes.h +16 -5
index 319eae7..366b8b2 100644
@@ -26,6 +26,7 @@
 
 #include <linux/errno.h>
 #include <linux/rbtree.h>
+#include <linux/types.h>
 
 struct vm_area_struct;
 struct mm_struct;
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
+		struct {
+			struct callback_head	dup_work;
+			unsigned long		dup_addr;
+		};
+	};
 
+	struct uprobe			*active_uprobe;
 	unsigned long			xol_vaddr;
-	unsigned long			vaddr;
+
+	struct return_instance		*return_instances;
+	unsigned int			depth;
 };
 
 /*
kernel/events/uprobes.c +4 -12
index 24b7d6c..2546a7b 100644
@@ -1403,12 +1403,10 @@ static void uprobe_warn(struct task_struct *t, const char *msg)
 
 static void dup_xol_work(struct callback_head *work)
 {
-	kfree(work);
-
 	if (current->flags & PF_EXITING)
 		return;
 
-	if (!__create_xol_area(current->utask->vaddr))
+	if (!__create_xol_area(current->utask->dup_addr))
 		uprobe_warn(current, "dup xol area");
 }
 
@@ -1419,7 +1417,6 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 {
 	struct uprobe_task *utask = current->utask;
 	struct mm_struct *mm = current->mm;
-	struct callback_head *work;
 	struct xol_area *area;
 
 	t->utask = NULL;
@@ -1441,14 +1438,9 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 	if (mm == t->mm)
 		return;
 
-	/* TODO: move it into the union in uprobe_task */
-	work = kmalloc(sizeof(*work), GFP_KERNEL);
-	if (!work)
-		return uprobe_warn(t, "dup xol area");
-
-	t->utask->vaddr = area->vaddr;
-	init_task_work(work, dup_xol_work);
-	task_work_add(t, work, true);
+	t->utask->dup_addr = area->vaddr;
+	init_task_work(&t->utask->dup_work, dup_xol_work);
+	task_work_add(t, &t->utask->dup_work, true);
 }
 
 /*
Signature · 1 line
1.5.5.1
#1 lore raw
Quoted 15 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

Quoted 8 lines

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Hmm, I'm not so sure about uprobes implementation so deeply. Is there no possibility to run xol preparing code (e.g. adding new uprobe?) between the task_work_add() and task_work_run()?

Quoted 4 lines

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

If those data structures are small, I think we don't need to use such union...

Thank you,

Signature · 4 lines
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: [email protected]
author maintainer
#2 lore raw
Quoted 22 lines · Masami Hiramatsu
Quoted 20 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

Quoted 7 lines
Quoted 3 lines

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Hmm, I'm not so sure about uprobes implementation so deeply. Is there no possibility to run xol preparing code (e.g. adding new uprobe?) between the task_work_add() and task_work_run()?

No, task_work_run() must be called before the new child returns to user-mode.

And it obviously can't hit the breakpoint until it returns to user mode. "adding new uprobe" doesn't matter at all, the task itself runs xol preparing code when it hits the bp first time.

Quoted 7 lines
Quoted 4 lines

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

If those data structures are small, I think we don't need to use such union...

Well, I disagree. First of all, to me this patch cleanups the code but this is subjective.

Why should we blow the size of task_struct->utask if there is no reason?

For example, should we instead add utask->dup_addr outiside of this union? Or create dup_xol_struct which holds this argument along with callback_head ? I don't think so. The scope of autask/vaddr and dup_work/addr is not interactable.

The same for the new ->private (or whatever) we are going to add for FETCH_MTD_relative. It will only have a meaning inside the ->handler() paths, to me it would be strange to not reuse the "free" memory we already have.

Oleg.

#3 lore raw
Quoted 31 lines · Oleg Nesterov
Quoted 22 lines · Masami Hiramatsu
Quoted 20 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

Quoted 7 lines
Quoted 3 lines

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Hmm, I'm not so sure about uprobes implementation so deeply. Is there no possibility to run xol preparing code (e.g. adding new uprobe?) between the task_work_add() and task_work_run()?

Quoted 6 lines

No, task_work_run() must be called before the new child returns to user-mode.

And it obviously can't hit the breakpoint until it returns to user mode. "adding new uprobe" doesn't matter at all, the task itself runs xol preparing code when it hits the bp first time.

Ah, I misunderstood. XOL area should be placed in each process address space, thus until it hits the probe, uprobe can't create XOL code, I got it.

Quoted 11 lines
Quoted 7 lines
Quoted 4 lines

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

If those data structures are small, I think we don't need to use such union...

Well, I disagree. First of all, to me this patch cleanups the code but this is subjective.

Quoted 7 lines

Why should we blow the size of task_struct->utask if there is no reason?

For example, should we instead add utask->dup_addr outiside of this union? Or create dup_xol_struct which holds this argument along with callback_head ? I don't think so. The scope of autask/vaddr and dup_work/addr is not interactable.

I see your point.

Quoted 4 lines

The same for the new ->private (or whatever) we are going to add for FETCH_MTD_relative. It will only have a meaning inside the ->handler() paths, to me it would be strange to not reuse the "free" memory we already have.

Looks nice ;)

Thank you,

Signature · 4 lines
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: [email protected]
Reviewed-by on v1 1/1
#4 lore raw
In reply to Oleg Nesterov
Quoted 20 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

Quoted 7 lines

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

OK, those are used in the different phase as a scratchpad.

Quoted 1 line

Thanks!

Quoted 92 lines
---
 include/linux/uprobes.h |   21 ++++++++++++++++-----
 kernel/events/uprobes.c |   16 ++++------------
 2 files changed, 20 insertions(+), 17 deletions(-)
include/linux/uprobes.h +16 -5
index 319eae7..366b8b2 100644
@@ -26,6 +26,7 @@
 
 #include <linux/errno.h>
 #include <linux/rbtree.h>
+#include <linux/types.h>
 
 struct vm_area_struct;
 struct mm_struct;
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
+		struct {
+			struct callback_head	dup_work;
+			unsigned long		dup_addr;
+		};
+	};
 
+	struct uprobe			*active_uprobe;
 	unsigned long			xol_vaddr;
-	unsigned long			vaddr;
+
+	struct return_instance		*return_instances;
+	unsigned int			depth;
 };
 
 /*
kernel/events/uprobes.c +4 -12
index 24b7d6c..2546a7b 100644
@@ -1403,12 +1403,10 @@ static void uprobe_warn(struct task_struct *t, const char *msg)
 
 static void dup_xol_work(struct callback_head *work)
 {
-	kfree(work);
-
 	if (current->flags & PF_EXITING)
 		return;
 
-	if (!__create_xol_area(current->utask->vaddr))
+	if (!__create_xol_area(current->utask->dup_addr))
 		uprobe_warn(current, "dup xol area");
 }
 
@@ -1419,7 +1417,6 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 {
 	struct uprobe_task *utask = current->utask;
 	struct mm_struct *mm = current->mm;
-	struct callback_head *work;
 	struct xol_area *area;
 
 	t->utask = NULL;
@@ -1441,14 +1438,9 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 	if (mm == t->mm)
 		return;
 
-	/* TODO: move it into the union in uprobe_task */
-	work = kmalloc(sizeof(*work), GFP_KERNEL);
-	if (!work)
-		return uprobe_warn(t, "dup xol area");
-
-	t->utask->vaddr = area->vaddr;
-	init_task_work(work, dup_xol_work);
-	task_work_add(t, work, true);
+	t->utask->dup_addr = area->vaddr;
+	init_task_work(&t->utask->dup_work, dup_xol_work);
+	task_work_add(t, &t->utask->dup_work, true);
 }
 
 /*
Signature · 4 lines
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: [email protected]
Acked-by on v1 1/1
#5 lore raw
In reply to Oleg Nesterov
Quoted 6 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

While the code changes look good, I would disagree with the above statement. uprobe_task->vaddr is a bit strange only in uprobe_copy_process() and not in arch_uprobe_pre_xol() context. uprobe_task->vaddr was used to refer to the actual instruction pointer and do the necessary fixups after single stepping out of line.

The casual reading of this commit message, one can get an impression that vaddr is never needed. Your change still retains it. So can we modify the commit message accordingly?

Quoted 14 lines

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line. This means it is safe to reuse both in UTASK_RUNNING state.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

Quoted 7 lines

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_addr which can be used instead ->vaddr, and ->dup_work which can be used to avoid kmalloc() and simplify the code.

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

One nit below + request for change in the above commit message. Otherwise

Quoted 27 lines
---
 include/linux/uprobes.h |   21 ++++++++++++++++-----
 kernel/events/uprobes.c |   16 ++++------------
 2 files changed, 20 insertions(+), 17 deletions(-)
include/linux/uprobes.h +1 -3
index 319eae7..366b8b2 100644
@@ -26,6 +26,7 @@
 
 #include <linux/errno.h>
 #include <linux/rbtree.h>
+#include <linux/types.h>
 
 struct vm_area_struct;
 struct mm_struct;
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;

include/linux/uprobes.h enum uprobe_task_state { · line 85

Quoted 10 lines
include/linux/uprobes.h +9 -1
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
+		struct {
+			struct callback_head	dup_work;
+			unsigned long		dup_addr;

Nit:
Can we rename dup_addr to mean that it refers to the xol; something like dup_xol_addr or even xol_addr. So that its more clear what address it refers to.

Signature · 2 lines
Thanks and Regards
Srikar Dronamraju
author maintainer
#6 lore raw
Quoted 10 lines · Srikar Dronamraju
Quoted 6 lines · Oleg Nesterov

uprobe_task->vaddr is a bit strange. First of all it is not really needed, we can move it into arch_uprobe_task. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

Quoted 4 lines

While the code changes look good, I would disagree with the above statement. uprobe_task->vaddr is a bit strange only in uprobe_copy_process() and not in arch_uprobe_pre_xol() context. uprobe_task->vaddr was used to refer to the actual instruction pointer

Yes, and it is always equal to regs->ip when pre_ssout() is called,

Quoted 1 line

and do the necessary fixups after single stepping out of line.

Exactly. So it is write-only (and meaningless) to the generic uprobe code. We can (and perhaps should) move it into autask->saved_vaddr, arch_uprobe_pre_xol() can initialize it.

Quoted 2 lines

The casual reading of this commit message, one can get an impression that vaddr is never needed.

See above. The changelog doesn't say we can simply remove it, it says "move it".

Quoted 1 line

Your change still retains it.

Of course we can't kill utas->vaddr until we change x86/powerpc. And just in case, of course I understand that this is minor and even subjective.

But at least this patch logically "joins" autask and vaddr, and I believe this is good because they are always used together, because, well, logically vaddr belongs to autask ;)

Quoted 2 lines

So can we
modify the commit message accordingly?

Well, I'll try... but perhaps you can help? I mean, I am not sure about how I can improve it. Could you suggest a better wording?

Quoted 2 lines

Otherwise

Thanks!

Quoted 15 lines
Quoted 15 lines
include/linux/uprobes.h +6 -4
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
Quoted 8 lines

include/linux/uprobes.h enum uprobe_task_state { · line 85

Quoted 3 lines
include/linux/uprobes.h +3 -0
+		struct {
+			struct callback_head	dup_work;
+			unsigned long		dup_addr;

Nit:
Can we rename dup_addr to mean that it refers to the xol; something like dup_xol_addr or even xol_addr. So that its more clear what address it refers to.

OK. How about dup_xol_work/dup_xol_vaddr ?

Oleg.

author maintainer
#7 lore raw
Quoted 9 lines · Oleg Nesterov
Quoted 5 lines · Srikar Dronamraju

Nit:
Can we rename dup_addr to mean that it refers to the xol; something like dup_xol_addr or even xol_addr. So that its more clear what address it refers to.

OK. How about dup_xol_work/dup_xol_vaddr ?

I didn't notice you suggested the same naming ;) OK, will rename to dup_xol_work/addr.

Oleg.

#8 lore raw
In reply to Oleg Nesterov
Quoted 8 lines

Yes, and it is always equal to regs->ip when pre_ssout() is called,

Quoted 1 line

and do the necessary fixups after single stepping out of line.

Exactly. So it is write-only (and meaningless) to the generic uprobe code. We can (and perhaps should) move it into autask->saved_vaddr, arch_uprobe_pre_xol() can initialize it.

Quoted 5 lines
Quoted 2 lines

The casual reading of this commit message, one can get an impression that vaddr is never needed.

See above. The changelog doesn't say we can simply remove it, it says "move it".

Okay, moving to arch_uprobe_task is fine. I probably got confused by "First of all it is not really needed,"

Quoted 6 lines
Quoted 2 lines

Your change still retains it.

OK. How about dup_xol_work/dup_xol_vaddr ?

Yes fine with me.

Signature · 2 lines
Thanks and Regards
Srikar Dronamraju
author maintainer
#9 lore raw

PATCH v2 1/1 uprobes: Add uprobe_task->dup_work/dup_addr

Quoted 3 lines · Srikar Dronamraju

Okay, moving to arch_uprobe_task is fine. I probably got confused by "First of all it is not really needed,"

OK, this doesn't look good, I agree.

Please see v2 below, I tried to improve the changelog.

Quoted 3 lines
Quoted 1 line

OK. How about dup_xol_work/dup_xol_vaddr ?

Yes fine with me.

Done.

I added your ack optimistically, please let me know if you think I should change something else.

Oleg.
---

Subject: [PATCH] uprobes: Add uprobe_task->dup_xol_work/dup_xol_addr

uprobe_task->vaddr is a bit strange. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line, so it is safe to reuse both in UTASK_RUNNING state.

This all means that logically ->vaddr belongs to arch_uprobe_task and we should probably move it there, arch_uprobe_pre_xol() can record instruction_pointer() itself.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_xol_addr which can be used instead ->vaddr, and ->dup_xol_work which can be used to avoid kmalloc() and simplify the code.

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

---
 include/linux/uprobes.h |   21 ++++++++++++++++-----
 kernel/events/uprobes.c |   16 ++++------------
 2 files changed, 20 insertions(+), 17 deletions(-)
include/linux/uprobes.h +16 -5
index 319eae7..2225542 100644
@@ -26,6 +26,7 @@
 
 #include <linux/errno.h>
 #include <linux/rbtree.h>
+#include <linux/types.h>
 
 struct vm_area_struct;
 struct mm_struct;
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
+		struct {
+			struct callback_head	dup_xol_work;
+			unsigned long		dup_xol_addr;
+		};
+	};
 
+	struct uprobe			*active_uprobe;
 	unsigned long			xol_vaddr;
-	unsigned long			vaddr;
+
+	struct return_instance		*return_instances;
+	unsigned int			depth;
 };
 
 /*
kernel/events/uprobes.c +4 -12
index 24b7d6c..df4ef09 100644
@@ -1403,12 +1403,10 @@ static void uprobe_warn(struct task_struct *t, const char *msg)
 
 static void dup_xol_work(struct callback_head *work)
 {
-	kfree(work);
-
 	if (current->flags & PF_EXITING)
 		return;
 
-	if (!__create_xol_area(current->utask->vaddr))
+	if (!__create_xol_area(current->utask->dup_xol_addr))
 		uprobe_warn(current, "dup xol area");
 }
 
@@ -1419,7 +1417,6 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 {
 	struct uprobe_task *utask = current->utask;
 	struct mm_struct *mm = current->mm;
-	struct callback_head *work;
 	struct xol_area *area;
 
 	t->utask = NULL;
@@ -1441,14 +1438,9 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 	if (mm == t->mm)
 		return;
 
-	/* TODO: move it into the union in uprobe_task */
-	work = kmalloc(sizeof(*work), GFP_KERNEL);
-	if (!work)
-		return uprobe_warn(t, "dup xol area");
-
-	t->utask->vaddr = area->vaddr;
-	init_task_work(work, dup_xol_work);
-	task_work_add(t, work, true);
+	t->utask->dup_xol_addr = area->vaddr;
+	init_task_work(&t->utask->dup_xol_work, dup_xol_work);
+	task_work_add(t, &t->utask->dup_xol_work, true);
 }
 
 /*
Signature · 1 line
1.5.5.1
#10 lore raw
Quoted 9 lines · Oleg Nesterov
Quoted 3 lines · Srikar Dronamraju

Okay, moving to arch_uprobe_task is fine. I probably got confused by "First of all it is not really needed,"

OK, this doesn't look good, I agree.

Please see v2 below, I tried to improve the changelog.

Quoted 8 lines
Quoted 3 lines
Quoted 1 line

OK. How about dup_xol_work/dup_xol_vaddr ?

Yes fine with me.

Done.

I added your ack optimistically, please let me know if you think I should change something else.

Looks nice to me.

Signature · 2 lines
Thanks and Regards
Srikar Dronamraju
#11 lore raw
In reply to Oleg Nesterov

Ping?

Is this already pulled?
I think it is enough discussed and reviewed.

Thank you,

Quoted 148 lines · Oleg Nesterov
Quoted 3 lines · Srikar Dronamraju

Okay, moving to arch_uprobe_task is fine. I probably got confused by "First of all it is not really needed,"

OK, this doesn't look good, I agree.

Please see v2 below, I tried to improve the changelog.

Quoted 3 lines
Quoted 1 line

OK. How about dup_xol_work/dup_xol_vaddr ?

Yes fine with me.

Done.

I added your ack optimistically, please let me know if you think I should change something else.

Oleg.
---

Subject: [PATCH] uprobes: Add uprobe_task->dup_xol_work/dup_xol_addr

uprobe_task->vaddr is a bit strange. The generic code uses it only to pass the additional argument to arch_uprobe_pre_xol(), and since it is always equal to instruction_pointer() this looks even more strange.

And both utask->vaddr and and utask->autask have the same scope, they only have the meaning when the task executes the probed insn out-of-line, so it is safe to reuse both in UTASK_RUNNING state.

This all means that logically ->vaddr belongs to arch_uprobe_task and we should probably move it there, arch_uprobe_pre_xol() can record instruction_pointer() itself.

OTOH, it is also used by uprobe_copy_process() and dup_xol_work() for another purpose, this doesn't look clean and doesn't allow to move this member into arch_uprobe_task.

This patch adds the union with 2 anonymous structs into uprobe_task.

The first struct is autask + vaddr, this way we "almost" move vaddr into autask.

The second struct has 2 new members for uprobe_copy_process() paths:
->dup_xol_addr which can be used instead ->vaddr, and ->dup_xol_work which can be used to avoid kmalloc() and simplify the code.

Note that this union will likely have another member(s), we need something like "private_data_for_handlers" so that the tracing handlers could use it to communicate with call_fetch() methods.

---
 include/linux/uprobes.h |   21 ++++++++++++++++-----
 kernel/events/uprobes.c |   16 ++++------------
 2 files changed, 20 insertions(+), 17 deletions(-)
include/linux/uprobes.h +16 -5
index 319eae7..2225542 100644
@@ -26,6 +26,7 @@
 
 #include <linux/errno.h>
 #include <linux/rbtree.h>
+#include <linux/types.h>
 
 struct vm_area_struct;
 struct mm_struct;
@@ -72,14 +73,24 @@ enum uprobe_task_state {
  */
 struct uprobe_task {
 	enum uprobe_task_state		state;
-	struct arch_uprobe_task		autask;
 
-	struct return_instance		*return_instances;
-	unsigned int			depth;
-	struct uprobe			*active_uprobe;
+	union {
+		struct {
+			struct arch_uprobe_task	autask;
+			unsigned long		vaddr;
+		};
+
+		struct {
+			struct callback_head	dup_xol_work;
+			unsigned long		dup_xol_addr;
+		};
+	};
 
+	struct uprobe			*active_uprobe;
 	unsigned long			xol_vaddr;
-	unsigned long			vaddr;
+
+	struct return_instance		*return_instances;
+	unsigned int			depth;
 };
 
 /*
kernel/events/uprobes.c +4 -12
index 24b7d6c..df4ef09 100644
@@ -1403,12 +1403,10 @@ static void uprobe_warn(struct task_struct *t, const char *msg)
 
 static void dup_xol_work(struct callback_head *work)
 {
-	kfree(work);
-
 	if (current->flags & PF_EXITING)
 		return;
 
-	if (!__create_xol_area(current->utask->vaddr))
+	if (!__create_xol_area(current->utask->dup_xol_addr))
 		uprobe_warn(current, "dup xol area");
 }
 
@@ -1419,7 +1417,6 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 {
 	struct uprobe_task *utask = current->utask;
 	struct mm_struct *mm = current->mm;
-	struct callback_head *work;
 	struct xol_area *area;
 
 	t->utask = NULL;
@@ -1441,14 +1438,9 @@ void uprobe_copy_process(struct task_struct *t, unsigned long flags)
 	if (mm == t->mm)
 		return;
 
-	/* TODO: move it into the union in uprobe_task */
-	work = kmalloc(sizeof(*work), GFP_KERNEL);
-	if (!work)
-		return uprobe_warn(t, "dup xol area");
-
-	t->utask->vaddr = area->vaddr;
-	init_task_work(work, dup_xol_work);
-	task_work_add(t, work, true);
+	t->utask->dup_xol_addr = area->vaddr;
+	init_task_work(&t->utask->dup_xol_work, dup_xol_work);
+	task_work_add(t, &t->utask->dup_xol_work, true);
 }
 
 /*
Signature · 4 lines
Masami HIRAMATSU
IT Management Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: [email protected]
author maintainer
#12 lore raw
Quoted 5 lines · Masami Hiramatsu

Ping?

Is this already pulled?
I think it is enough discussed and reviewed.

Yes, thanks, this is already in tip/perf/core.

Oleg.