Skip to contentSkip to conversation

ARM: mm: harden branch predictor before opening interrupts during fault

PATCH In review Lecopzer Chen 3 participants 7 messages Dec 3, 2020 – Jan 26, 2021 Arm (32-bit)

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: mm: harden branch predictor before opening interrupts during fault

This patch was send by "Sebastian Andrzej Siewior <[email protected]>" in [1] and rebase in v5.10-rc6.

The original commit message:

Quoted 10 lines

On non-LPAE systems a write to 0xbffffff0 (modules area) from userland results in:

| BUG: using smp_processor_id() in preemptible [00000000] code: mem-tc/521
| caller is __do_user_fault.constprop.2+0x4c/0x74
| CPU: 1 PID: 521 Comm: mem-tc Not tainted 5.1.0-rc1 #4
| [<c04614e4>] (debug_smp_processor_id) from [<c0116378>] (__do_user_fault.constprop.2+0x4c/0x74)
| [<c0116378>] (__do_user_fault.constprop.2) from [<c011668c>] (do_page_fault+0x278/0x37c)
| [<c011668c>] (do_page_fault) from [<c0116904>] (do_DataAbort+0x3c/0xa8)
| [<c0116904>] (do_DataAbort) from [<c0101e1c>] (__dabt_usr+0x3c/0x40)
Quoted 4 lines

Move harden_branch_predictor() from __do_user_fault() to its both callers (do_bad_area() and do_page_fault()). The invocation in do_page_fault() is added before interrupst are enabled. The invocation in do_bad_area() is added just before __do_user_fault() is invoked.

The BUG still exists in v5.10-rc, and the previous disscussion was [2]. This issue can be easily reproduced in ARM with CONFIG_DEBUG_PREEMPT and CONFIG_HARDEN_BRANCH_PREDICTOR by the following userspace program:

  int *p = 0xffff3ff4;
  *p = 123;

[1]:
https://patchwork.kernel.org/project/linux-arm-kernel/patch/[email protected]/
[2]:
https://lkml.org/lkml/2019/6/3/426

Fixes f5fe12b1eaee2 ("ARM: spectre-v2: harden user aborts in kernel space")
Cc YJ Chiang <[email protected]>
---
 arch/arm/mm/fault.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
arch/arm/mm/fault.c +7 -4
index efa402025031..f1b57b7d5a0c 100644
@@ -138,9 +138,6 @@ __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
 {
 	struct task_struct *tsk = current;
 
-	if (addr > TASK_SIZE)
-		harden_branch_predictor();
-
 #ifdef CONFIG_DEBUG_USER
 	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
 	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
@@ -173,8 +170,11 @@ void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	 * If we are in kernel mode at this point, we
 	 * have no context to handle this fault with.
 	 */
-	if (user_mode(regs))
+	if (user_mode(regs)) {
+		if (addr > TASK_SIZE)
+			harden_branch_predictor();
 		__do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
+	}
 	else
 		__do_kernel_fault(mm, addr, fsr, regs);
 }
@@ -251,6 +251,9 @@ do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 	tsk = current;
 	mm  = tsk->mm;
 
+	if (addr > TASK_SIZE && user_mode(regs))
+		harden_branch_predictor();
+
 	/* Enable interrupts if they were enabled in the parent context. */
 	if (interrupts_enabled(regs))
 		local_irq_enable();
Signature · 1 line
2.18.0
#2 lore raw
Quoted 5 lines · Lecopzer Chen

Hi all,

I don't see any fix for this issue now(maybe I missed it..?), could we fix this if there is better solution?
This issue exists almost two years.

I don't think anyone provided an acceptable patch.

The first patch moved the hardening out of the translation/section fault handling. Since the kernel is mapped with sections, these are above TASK_SIZE, and the whole point of the branch prediction hardening is to prevent the prediction in the kernel being exploited, missing the hardening effectively makes the mitigation useless.

The discussion in February 2019 never concluded from what I can see.

Signature · 7 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#3 lore raw
Quoted 9 lines · Russell King - ARM Linux admin
Quoted 5 lines · Lecopzer Chen

Hi all,

I don't see any fix for this issue now(maybe I missed it..?), could we fix this if there is better solution?
This issue exists almost two years.

I don't think anyone provided an acceptable patch.

Quoted 7 lines

The first patch moved the hardening out of the translation/section fault handling. Since the kernel is mapped with sections, these are above TASK_SIZE, and the whole point of the branch prediction hardening is to prevent the prediction in the kernel being exploited, missing the hardening effectively makes the mitigation useless.

The discussion in February 2019 never concluded from what I can see.

My memory is that I never got a reply which I understood. Let me try again this week with the information above.

Sebastian

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#4 lore raw
Quoted 16 lines · Russell King - ARM Linux admin
Quoted 15 lines · Lecopzer Chen
Quoted 5 lines

Hi all,

I don't see any fix for this issue now(maybe I missed it..?), could we fix this if there is better solution?
This issue exists almost two years.

I don't think anyone provided an acceptable patch.

The first patch moved the hardening out of the translation/section fault handling. Since the kernel is mapped with sections, these are above TASK_SIZE, and the whole point of the branch prediction hardening is to prevent the prediction in the kernel being exploited, missing the hardening effectively makes the mitigation useless.

Quoted 4 lines
Quoted 1 line

The discussion in February 2019 never concluded from what I can see.

My memory is that I never got a reply which I understood. Let me try again this week with the information above.

NOTE:
Before sending this mail, I had searched the relative threads and there are two solutions in general:
1. Add get_pcpu()/put_cpu() https://lkml.org/lkml/2019/6/3/426
Reject by Marc:
> The right fix would be to move the call to a point where we haven't > enabled preemption yet.

2. Move out like the patch from Sebastian:
This seems follow the concept of 1.
(move the call to a point where we haven't enabled preemption yet). But I can't find any reply in the thread.

Now the CONFIG_HARDEN_BRANCH_PREDICTOR has already backported to LTS, and after upgrading ARM CONFIG_CPU_V7 products to latest LTS, the CONFIG_HARDEN_BRANCH_PREDICTOR will be default y and this issue makes our devices panic and we have to either disable HARDEN_BRANCH_PREDICTOR or hack in-house to avoid the kernel panic.

I'll wait for anyone have acceptable idea and patching(maybe Sebastian could? :)), and I'll also take some time to find other solution too.

Thanks,
Lecopzer
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

#5 lore raw
Quoted 35 lines · Lecopzer Chen
Quoted 16 lines · Russell King - ARM Linux admin
Quoted 15 lines · Lecopzer Chen
Quoted 5 lines

Hi all,

I don't see any fix for this issue now(maybe I missed it..?), could we fix this if there is better solution?
This issue exists almost two years.

I don't think anyone provided an acceptable patch.

The first patch moved the hardening out of the translation/section fault handling. Since the kernel is mapped with sections, these are above TASK_SIZE, and the whole point of the branch prediction hardening is to prevent the prediction in the kernel being exploited, missing the hardening effectively makes the mitigation useless.

Quoted 4 lines
Quoted 1 line

The discussion in February 2019 never concluded from what I can see.

My memory is that I never got a reply which I understood. Let me try again this week with the information above.

NOTE:
Before sending this mail, I had searched the relative threads and there are two solutions in general:
1. Add get_pcpu()/put_cpu() https://lkml.org/lkml/2019/6/3/426
Reject by Marc:
> The right fix would be to move the call to a point where we haven't > enabled preemption yet.

2. Move out like the patch from Sebastian:
This seems follow the concept of 1.
(move the call to a point where we haven't enabled preemption yet). But I can't find any reply in the thread.

Quoted 5 lines

Now the CONFIG_HARDEN_BRANCH_PREDICTOR has already backported to LTS, and after upgrading ARM CONFIG_CPU_V7 products to latest LTS, the CONFIG_HARDEN_BRANCH_PREDICTOR will be default y and this issue makes our devices panic and we have to either disable HARDEN_BRANCH_PREDICTOR or hack in-house to avoid the kernel panic.

It does _not_ cause the kernel to panic, ever. A kernel panic takes out the system. This is not the case here.

It merely causes a noisy message to be emitted in the kernel log, and the system survives. That is way more preferable than breaking the effect of branch predictor hardening.

If it is taking out your kernel with a real panic, then there is something wrong elsewhere - and this is _not_ something that should be happening during normal system operation.

Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
#6 lore raw
Quoted 45 lines · Lecopzer Chen
Quoted 35 lines
Quoted 16 lines · Russell King - ARM Linux admin
Quoted 15 lines · Lecopzer Chen
Quoted 5 lines

Hi all,

I don't see any fix for this issue now(maybe I missed it..?), could we fix this if there is better solution?
This issue exists almost two years.

I don't think anyone provided an acceptable patch.

The first patch moved the hardening out of the translation/section fault handling. Since the kernel is mapped with sections, these are above TASK_SIZE, and the whole point of the branch prediction hardening is to prevent the prediction in the kernel being exploited, missing the hardening effectively makes the mitigation useless.

Quoted 4 lines
Quoted 1 line

The discussion in February 2019 never concluded from what I can see.

My memory is that I never got a reply which I understood. Let me try again this week with the information above.

NOTE:
Before sending this mail, I had searched the relative threads and there are two solutions in general:
1. Add get_pcpu()/put_cpu() https://lkml.org/lkml/2019/6/3/426
Reject by Marc:
> The right fix would be to move the call to a point where we haven't > enabled preemption yet.

2. Move out like the patch from Sebastian:
This seems follow the concept of 1.
(move the call to a point where we haven't enabled preemption yet). But I can't find any reply in the thread.

Quoted 5 lines

Now the CONFIG_HARDEN_BRANCH_PREDICTOR has already backported to LTS, and after upgrading ARM CONFIG_CPU_V7 products to latest LTS, the CONFIG_HARDEN_BRANCH_PREDICTOR will be default y and this issue makes our devices panic and we have to either disable HARDEN_BRANCH_PREDICTOR or hack in-house to avoid the kernel panic.

It does _not_ cause the kernel to panic, ever. A kernel panic takes out the system. This is not the case here.

Quoted 7 lines

It merely causes a noisy message to be emitted in the kernel log, and the system survives. That is way more preferable than breaking the effect of branch predictor hardening.

If it is taking out your kernel with a real panic, then there is something wrong elsewhere - and this is _not_ something that should be happening during normal system operation.

Oh, yes, you're right;

After reread the panic log, our panic happened because -> invalid userspace memory access
-> debug_preempt log
-> the program seg fault
-> main service need the program but it crash -> panic

Sorry for wrong information and thanks a lot for the correctness. I think I have to see why the in-house hacking is working...

Thanks!!

BRs,
Lecopzer
_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel