Skip to contentSkip to conversation

smp_wmb() in prepare_uprobe()

Question Andrea Parri 2 participants 5 messages Nov 21, 2018 – Nov 22, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

The comment for the smp_wmb() in prepare_uprobe() says:

"pairs with rmb() in find_active_uprobe()"

but I see no (smp_)rmb() in find_active_uprobe(); I see the smp_rmb() in handle_swbp(): is this the intended pairing barrier?

Which memory accesses do you want to "order" with this pairing?

Thanks,
Andrea

maintainer
#1 lore raw

Hi,

Quoted 4 lines · Andrea Parri

The comment for the smp_wmb() in prepare_uprobe() says:

"pairs with rmb() in find_active_uprobe()"

it seems that this comment was wrong from the very beginning,

Quoted 2 lines

but I see no (smp_)rmb() in find_active_uprobe(); I see the smp_rmb() in handle_swbp(): is this the intended pairing barrier?

Yes, and the comment near this rmb() says "pairs with wmb() in install_breakpoint()", today this is not right too.

Quoted 1 line

Which memory accesses do you want to "order" with this pairing?

See 142b18ddc81439acda4bc4231b291e99fe67d507 ("uprobes: Fix handle_swbp() vs unregister() + register() race") and the comment above this rmb().

Oleg.

#2 lore raw
Quoted 8 lines · Oleg Nesterov

Hi,

Quoted 4 lines · Andrea Parri

The comment for the smp_wmb() in prepare_uprobe() says:

"pairs with rmb() in find_active_uprobe()"

Quoted 8 lines

it seems that this comment was wrong from the very beginning,

Quoted 2 lines

but I see no (smp_)rmb() in find_active_uprobe(); I see the smp_rmb() in handle_swbp(): is this the intended pairing barrier?

Yes, and the comment near this rmb() says "pairs with wmb() in install_breakpoint()", today this is not right too.

Thanks for the confirmation. So, this is the easy part ;-), maybe something like:

kernel/events/uprobes.c +2 -2
index 96d4bee83489b..2d29977522017 100644
@@ -829,7 +829,7 @@ static int prepare_uprobe(struct uprobe *uprobe, struct file *file,
 	BUG_ON((uprobe->offset & ~PAGE_MASK) +
 			UPROBE_SWBP_INSN_SIZE > PAGE_SIZE);
 
-	smp_wmb(); /* pairs with rmb() in find_active_uprobe() */
+	smp_wmb(); /* pairs with the smp_rmb() in handle_swbp() */
 	set_bit(UPROBE_COPY_INSN, &uprobe->flags);
 
  out:
@@ -2178,7 +2178,7 @@ static void handle_swbp(struct pt_regs *regs)
 	 * After we hit the bp, _unregister + _register can install the
 	 * new and not-yet-analyzed uprobe at the same address, restart.
 	 */
-	smp_rmb(); /* pairs with wmb() in install_breakpoint() */
+	smp_rmb(); /* pairs with the smp_wmb() in prepare_uprobe() */
 	if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
 		goto out;
 
Quoted 5 lines
Quoted 2 lines

Which memory accesses do you want to "order" with this pairing?

See 142b18ddc81439acda4bc4231b291e99fe67d507 ("uprobes: Fix handle_swbp() vs unregister() + register() race") and the comment above this rmb().

Mmh..., at first glance, this suggests me that the above set_bit() and test_bit() to/from uprobe->flags are among these memory accesses. But this doesn't make sense to me: these accesses do not "alternate" (i.e., they both appear after the corresponding barrier..); instead I'd expect something like (on top of the above diff):

kernel/events/uprobes.c +9 -1
index 2d29977522017..a75b9a08dee54 100644
@@ -2178,10 +2178,18 @@ static void handle_swbp(struct pt_regs *regs)
 	 * After we hit the bp, _unregister + _register can install the
 	 * new and not-yet-analyzed uprobe at the same address, restart.
 	 */
-	smp_rmb(); /* pairs with the smp_wmb() in prepare_uprobe() */
 	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 (can) also see the stores to &uprobe->arch performed
+	 * by prepare_uprobe() (say).
+	 */
+	smp_rmb();
+
 	/* Tracing handlers use ->utask to communicate with fetch methods */
 	if (!get_utask())
 		goto out;

Thoughts?

Andrea

Quoted 3 lines

Oleg.

maintainer
#3 lore raw
Quoted 20 lines · Andrea Parri
Quoted 3 lines

See 142b18ddc81439acda4bc4231b291e99fe67d507 ("uprobes: Fix handle_swbp() vs unregister() + register() race") and the comment above this rmb().

Mmh..., at first glance, this suggests me that the above set_bit() and test_bit() to/from uprobe->flags are among these memory accesses. But this doesn't make sense to me: these accesses do not "alternate" (i.e., they both appear after the corresponding barrier..); instead I'd expect something like (on top of the above diff):

kernel/events/uprobes.c +0 -1
index 2d29977522017..a75b9a08dee54 100644
@@ -2178,10 +2178,18 @@ static void handle_swbp(struct pt_regs *regs)
 	 * After we hit the bp, _unregister + _register can install the
 	 * new and not-yet-analyzed uprobe at the same address, restart.
 	 */
-	smp_rmb(); /* pairs with the smp_wmb() in prepare_uprobe() */
 	if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))

kernel/events/uprobes.c static void handle_swbp(struct pt_regs *regs) · line 2191

Quoted 10 lines
kernel/events/uprobes.c +8 -0
 		goto out;
 
+	/*
+	 * Pairs with the smp_wmb() in prepare_uprobe().
+	 *
+	 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
+	 * we must (can) also see the stores to &uprobe->arch performed
+	 * by prepare_uprobe() (say).
+	 */
+	smp_rmb();

OOPS, you are right! Thanks.

Oleg.

#4 lore raw
Quoted 23 lines · Oleg Nesterov
Quoted 22 lines · Andrea Parri
Quoted 3 lines

See 142b18ddc81439acda4bc4231b291e99fe67d507 ("uprobes: Fix handle_swbp() vs unregister() + register() race") and the comment above this rmb().

Mmh..., at first glance, this suggests me that the above set_bit() and test_bit() to/from uprobe->flags are among these memory accesses. But this doesn't make sense to me: these accesses do not "alternate" (i.e., they both appear after the corresponding barrier..); instead I'd expect something like (on top of the above diff):

kernel/events/uprobes.c +0 -1
index 2d29977522017..a75b9a08dee54 100644
@@ -2178,10 +2178,18 @@ static void handle_swbp(struct pt_regs *regs)
 	 * After we hit the bp, _unregister + _register can install the
 	 * new and not-yet-analyzed uprobe at the same address, restart.
 	 */
-	smp_rmb(); /* pairs with the smp_wmb() in prepare_uprobe() */
 	if (unlikely(!test_bit(UPROBE_COPY_INSN, &uprobe->flags)))
 		goto out;
 
Quoted 10 lines

kernel/events/uprobes.c static void handle_swbp(struct pt_regs *regs) · line 2191

Quoted 8 lines
kernel/events/uprobes.c +8 -0
+	/*
+	 * Pairs with the smp_wmb() in prepare_uprobe().
+	 *
+	 * Guarantees that if we see the UPROBE_COPY_INSN bit set, then
+	 * we must (can) also see the stores to &uprobe->arch performed
+	 * by prepare_uprobe() (say).
+	 */
+	smp_rmb();

OOPS, you are right! Thanks.

Thank you for the clarification; I'll send a patch with the fix shortly.

Andrea

Quoted 3 lines

Oleg.