Skip to contentSkip to conversation

jump_label: Robustify jump label patching

PATCH Applied Josh Poimboeuf 7 participants 17 messages Feb 14, 2018 – Feb 16, 2018

Subsystems (5)

v1 of 4 Applied Latest is v4 →

lore
Share & tools
lore
#0 lore raw

cover jump_label: Robustify jump label patching

Warn if a jump_label fails to patch, and fix up a couple of issues found by the warning.

Josh Poimboeuf (3):
jump_label: Warn on failed jump_label patch
x86/xen: Remove inert tracepoint in __init code
percpu: Remove inert tracepoint in __init code

 arch/x86/xen/enlighten_pv.c | 2 --
 kernel/jump_label.c         | 8 ++++++--
 mm/percpu.c                 | 1 -
 3 files changed, 6 insertions(+), 5 deletions(-)
Signature · 1 line
2.14.3
#1 lore raw

PATCH 1/3 jump_label: Warn on failed jump_label patch

When the jump label code encounters an address which isn't recognized by kernel_text_address(), it just silently fails.

This can be dangerous because jump labels are used in a variety of places, and are generally expected to work. Convert the silent failure to a warning.

---
 kernel/jump_label.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
kernel/jump_label.c +6 -2
index b4517095db6a..c71fb7cdfc41 100644
@@ -370,8 +370,12 @@ static void __jump_label_update(struct static_key *key,
 		 * kernel_text_address() verifies we are not in core kernel
 		 * init code, see jump_label_invalidate_module_init().
 		 */
-		if (entry->code && kernel_text_address(entry->code))
-			arch_jump_label_transform(entry, jump_label_type(entry));
+		if (entry->code) {
+			if (kernel_text_address(entry->code))
+				arch_jump_label_transform(entry, jump_label_type(entry));
+			else
+				WARN_ONCE(1, "can't patch jump_label at %pS", (void *)entry->code);
+		}
 	}
 }
 
Signature · 1 line
2.14.3
#2 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 6 lines · On Wed, 14 Feb 2018 10:40:41 -0600 Josh Poimboeuf <[email protected]> wrote:

When the jump label code encounters an address which isn't recognized by kernel_text_address(), it just silently fails.

This can be dangerous because jump labels are used in a variety of places, and are generally expected to work. Convert the silent failure to a warning.

I made ftrace function tracing work on init code, can we do the same with tracepoints (aka jump labels)?

But I have to say that this goes with my argument that there exists tracepoints in the kernel that nobody cares about ;-)

-- Steve

Quoted 25 lines
---
 kernel/jump_label.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
kernel/jump_label.c +6 -2
index b4517095db6a..c71fb7cdfc41 100644
@@ -370,8 +370,12 @@ static void __jump_label_update(struct static_key *key,
 		 * kernel_text_address() verifies we are not in core kernel
 		 * init code, see jump_label_invalidate_module_init().
 		 */
-		if (entry->code && kernel_text_address(entry->code))
-			arch_jump_label_transform(entry, jump_label_type(entry));
+		if (entry->code) {
+			if (kernel_text_address(entry->code))
+				arch_jump_label_transform(entry, jump_label_type(entry));
+			else
+				WARN_ONCE(1, "can't patch jump_label at %pS", (void *)entry->code);
+		}
 	}
 }
 
maintainer
#3 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 6 lines · Steven Rostedt
Quoted 3 lines · On Wed, 14 Feb 2018 10:40:41 -0600 Josh Poimboeuf <[email protected]> wrote:

When the jump label code encounters an address which isn't recognized by kernel_text_address(), it just silently fails.

Quoted 7 lines
Quoted 3 lines

This can be dangerous because jump labels are used in a variety of places, and are generally expected to work. Convert the silent failure to a warning.

I made ftrace function tracing work on init code, can we do the same with tracepoints (aka jump labels)?

jump labels do work on init code, except they don't work on it after it has been freed :)

It uses 'kernel_text_address()', which will return true for init code if system_state < SYSTEM_RUNNING. See: core_kernel_text().

So I'm guessing that the warnings here are coming from init code that has already been freed. Are we sure that these warnings are coming from init code that hasn't already been freed?

Thanks,

-Jason

Quoted 31 lines

But I have to say that this goes with my argument that there exists tracepoints in the kernel that nobody cares about ;-)

-- Steve

Quoted 25 lines
---
 kernel/jump_label.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
kernel/jump_label.c +6 -2
index b4517095db6a..c71fb7cdfc41 100644
@@ -370,8 +370,12 @@ static void __jump_label_update(struct static_key *key,
 		 * kernel_text_address() verifies we are not in core kernel
 		 * init code, see jump_label_invalidate_module_init().
 		 */
-		if (entry->code && kernel_text_address(entry->code))
-			arch_jump_label_transform(entry, jump_label_type(entry));
+		if (entry->code) {
+			if (kernel_text_address(entry->code))
+				arch_jump_label_transform(entry, jump_label_type(entry));
+			else
+				WARN_ONCE(1, "can't patch jump_label at %pS", (void *)entry->code);
+		}
 	}
 }
 
#4 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 20 lines · Jason Baron
Quoted 6 lines · Steven Rostedt
Quoted 3 lines · On Wed, 14 Feb 2018 10:40:41 -0600 Josh Poimboeuf <[email protected]> wrote:

When the jump label code encounters an address which isn't recognized by kernel_text_address(), it just silently fails.

Quoted 7 lines
Quoted 3 lines

This can be dangerous because jump labels are used in a variety of places, and are generally expected to work. Convert the silent failure to a warning.

I made ftrace function tracing work on init code, can we do the same with tracepoints (aka jump labels)?

jump labels do work on init code, except they don't work on it after it has been freed :)

Quoted 6 lines

It uses 'kernel_text_address()', which will return true for init code if system_state < SYSTEM_RUNNING. See: core_kernel_text().

So I'm guessing that the warnings here are coming from init code that has already been freed. Are we sure that these warnings are coming from init code that hasn't already been freed?

Ah, I see. I missed that 'system_state < SYSTEM_RUNNING' check. It was nicely hidden. :-)

So I guess patches 2 & 3 can be dropped, as those tracepoints can still be used during boot.

And in patch 1 the warning conditions need to be tweaked a bit to exclude the __init case.

Signature · 1 line
Josh
#5 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 3 lines · On Wed, 14 Feb 2018 11:34:07 -0600 Josh Poimboeuf <[email protected]> wrote:

And in patch 1 the warning conditions need to be tweaked a bit to exclude the __init case.

Hmm, I wonder if we could simply remove the references from the jump label table when init is freed.

-- Steve

#6 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 10 lines · Steven Rostedt
Quoted 3 lines · On Wed, 14 Feb 2018 11:34:07 -0600 Josh Poimboeuf <[email protected]> wrote:

And in patch 1 the warning conditions need to be tweaked a bit to exclude the __init case.

Hmm, I wonder if we could simply remove the references from the jump label table when init is freed.

Interesting idea. The static key's entries (i.e. tracepoint sites) are in a static array. Maybe we could rewrite the affected arrays when init memory is freed. I'm not sure whether that would be overkill though. I can tinker with it and see how it compares to the current solution.

Signature · 1 line
Josh
#7 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 9 lines · Josh Poimboeuf
Quoted 8 lines · Steven Rostedt
Quoted 3 lines · On Wed, 14 Feb 2018 11:34:07 -0600 Josh Poimboeuf <[email protected]> wrote:

And in patch 1 the warning conditions need to be tweaked a bit to exclude the __init case.

Quoted 7 lines
Quoted 2 lines

Hmm, I wonder if we could simply remove the references from the jump label table when init is freed.

Interesting idea. The static key's entries (i.e. tracepoint sites) are in a static array. Maybe we could rewrite the affected arrays when init memory is freed. I'm not sure whether that would be overkill though. I can tinker with it and see how it compares to the current solution.

It turns out that something like this is already done for module init code. It doesn't *remove* the entry, but it does disable it by setting entry->code to zero. I'll just do the same thing right before initmem is freed.

Signature · 1 line
Josh
#8 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

Quoted 9 lines · On Thu, 15 Feb 2018 16:35:58 -0600 Josh Poimboeuf <[email protected]> wrote:
Quoted 4 lines

Interesting idea. The static key's entries (i.e. tracepoint sites) are in a static array. Maybe we could rewrite the affected arrays when init memory is freed. I'm not sure whether that would be overkill though. I can tinker with it and see how it compares to the current solution.

It turns out that something like this is already done for module init code. It doesn't *remove* the entry, but it does disable it by setting entry->code to zero. I'll just do the same thing right before initmem is freed.

Yes, I was actually going to mention to do it like the module code does.

Thanks!

-- Steve

Reviewed-by on v1 1/3
#9 lore raw

Re: [PATCH 1/3] jump_label: Warn on failed jump_label patch

In reply to Josh Poimboeuf
Quoted 33 lines · Josh Poimboeuf

When the jump label code encounters an address which isn't recognized by kernel_text_address(), it just silently fails.

This can be dangerous because jump labels are used in a variety of places, and are generally expected to work. Convert the silent failure to a warning.

---
 kernel/jump_label.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
kernel/jump_label.c +6 -2
index b4517095db6a..c71fb7cdfc41 100644
@@ -370,8 +370,12 @@ static void __jump_label_update(struct static_key *key,
 		 * kernel_text_address() verifies we are not in core kernel
 		 * init code, see jump_label_invalidate_module_init().
 		 */
-		if (entry->code && kernel_text_address(entry->code))
-			arch_jump_label_transform(entry, jump_label_type(entry));
+		if (entry->code) {
+			if (kernel_text_address(entry->code))
+				arch_jump_label_transform(entry, jump_label_type(entry));
+			else
+				WARN_ONCE(1, "can't patch jump_label at %pS", (void *)entry->code);
+		}
 	}
 }
 
Signature · 1 line
2.14.3

Good idea.

Signature · 5 lines
Regards/Gruss,
    Boris.

SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg)
-- 
#10 lore raw

PATCH 2/3 x86/xen: Remove inert tracepoint in __init code

In reply to Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

---
 arch/x86/xen/enlighten_pv.c | 2 --
 1 file changed, 2 deletions(-)
arch/x86/xen/enlighten_pv.c +0 -2
index c047f42552e1..0dbfad92f442 100644
@@ -803,8 +803,6 @@ static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
 static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
 					    const void *desc, int type)
 {
-	trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
-
 	switch (type) {
 	case DESC_LDT:
 	case DESC_TSS:
Signature · 1 line
2.14.3
#12 lore raw

PATCH 3/3 percpu: Remove inert tracepoint in __init code

In reply to Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

---
 mm/percpu.c | 1 -
 1 file changed, 1 deletion(-)
mm/percpu.c +0 -1
index 50e7fdf84055..3216364ad91b 100644
@@ -2156,7 +2156,6 @@ int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
 	pcpu_chunk_relocate(pcpu_first_chunk, -1);
 
 	pcpu_stats_chunk_alloc();
-	trace_percpu_create_chunk(base_addr);
 
 	/* we're done */
 	pcpu_base_addr = base_addr;
Signature · 1 line
2.14.3
maintainer Applied
#13 lore raw

Re: [PATCH 3/3] percpu: Remove inert tracepoint in __init code

Quoted 5 lines · Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

Applied to percpu/for-4.17.

Thanks.

Signature · 1 line
tejun
#14 lore raw

Re: [PATCH 3/3] percpu: Remove inert tracepoint in __init code

Quoted 8 lines · Tejun Heo
Quoted 5 lines · Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

Applied to percpu/for-4.17.

Actually, I think this patch should be dropped. As Jason pointed out, tracepoints *can* be enabled at boot time for __init code, so it's possible for this tracepoint to be used.

Signature · 1 line
Josh
#15 lore raw

Re: [PATCH 3/3] percpu: Remove inert tracepoint in __init code

Quoted 5 lines · Josh Poimboeuf
Quoted 4 lines · Tejun Heo
Quoted 3 lines · Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

Applied to percpu/for-4.17.

Actually, I think this patch should be dropped. As Jason pointed out, tracepoints *can* be enabled at boot time for __init code, so it's possible for this tracepoint to be used.

Would it still be useful to have the warning for failed patching?

Thanks,

Ingo

#16 lore raw

Re: [PATCH 3/3] percpu: Remove inert tracepoint in __init code

Quoted 11 lines · Ingo Molnar
Quoted 8 lines · Josh Poimboeuf
Quoted 7 lines · Tejun Heo
Quoted 5 lines · Josh Poimboeuf

The jump_label code doesn't patch init code, so this tracepoint can never be enabled. Remove it.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

Applied to percpu/for-4.17.

Actually, I think this patch should be dropped. As Jason pointed out, tracepoints *can* be enabled at boot time for __init code, so it's possible for this tracepoint to be used.

Would it still be useful to have the warning for failed patching?

Yes, I'm working on a v2.

Signature · 1 line
Josh