Skip to contentSkip to conversation

ARM: zynq: Use memcpy_toio instead of memcpy on smp bring-up

RFC PATCH In review Luis Araneda 3 participants 5 messages Jul 30, 2019 – Aug 6, 2019

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 ARM: zynq: Use memcpy_toio instead of memcpy on smp bring-up

This fixes a kernel panic (read overflow) on memcpy when FORTIFY_SOURCE is enabled.

The computed size of memcpy args are:
- p_size (dst): 4294967295 = (size_t) -1
- q_size (src): 1
- size (len): 8

Additionally, the memory is marked as __iomem, so one of the memcpy_* functions should be used for read/write

---

For anyone trying to reproduce / debug this, it panics before the console has any output.
I used JTAG to find the panic, but I had to comment-out the call to "zynq_slcr_cpu_stop" as it stops the JTAG interface and the connection is dropped, at least with OpenOCD.

I run-tested this on a Digilent Zybo Z7 board

---
 arch/arm/mach-zynq/platsmp.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-zynq/platsmp.c +1 -1
index a7cfe07156f4..407abade7336 100644
@@ -57,7 +57,7 @@ int zynq_cpun_start(u32 address, int cpu)
 			* 0x4: Jump by mov instruction
 			* 0x8: Jumping address
 			*/
-			memcpy((__force void *)zero, &zynq_secondary_trampoline,
+			memcpy_toio(zero, &zynq_secondary_trampoline,
 							trampoline_size);
 			writel(address, zero + trampoline_size);
 
Signature · 1 line
2.22.0
#1 lore raw
Quoted 30 lines · Luis Araneda

This fixes a kernel panic (read overflow) on memcpy when FORTIFY_SOURCE is enabled.

The computed size of memcpy args are:
- p_size (dst): 4294967295 = (size_t) -1
- q_size (src): 1
- size (len): 8

Additionally, the memory is marked as __iomem, so one of the memcpy_* functions should be used for read/write

---

For anyone trying to reproduce / debug this, it panics before the console has any output.
I used JTAG to find the panic, but I had to comment-out the call to "zynq_slcr_cpu_stop" as it stops the JTAG interface and the connection is dropped, at least with OpenOCD.

I run-tested this on a Digilent Zybo Z7 board

---
 arch/arm/mach-zynq/platsmp.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-zynq/platsmp.c +0 -0
index a7cfe07156f4..407abade7336 100644

arch/arm/mach-zynq/platsmp.c int zynq_cpun_start(u32 address, int cpu) · line 62

Quoted 7 lines
arch/arm/mach-zynq/platsmp.c +1 -1
@@ -57,7 +57,7 @@ int zynq_cpun_start(u32 address, int cpu)
 			* 0x4: Jump by mov instruction
 			* 0x8: Jumping address
 			*/
-			memcpy((__force void *)zero, &zynq_secondary_trampoline,
+			memcpy_toio(zero, &zynq_secondary_trampoline,
 							trampoline_size);
 			writel(address, zero + trampoline_size);

I'm not convinced that this is correct. It looks like zynq_secondary_trampoline could be either ARM or Thumb code - there is no .arm directive before it. If it's ARM code, then this is fine. If Thumb code, then zynq_secondary_trampoline will be offset by one, and we will miss copying the first byte of code.

Signature · 3 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up
#2 lore raw

Hi Russell,

Thanks for reviewing.

Quoted 4 lines · Russell King - ARM Linux admin
Quoted 2 lines · Luis Araneda

This fixes a kernel panic (read overflow) on memcpy when FORTIFY_SOURCE is enabled.

[...]

Quoted 6 lines

I'm not convinced that this is correct. It looks like zynq_secondary_trampoline could be either ARM or Thumb code - there is no .arm directive before it. If it's ARM code, then this is fine. If Thumb code, then zynq_secondary_trampoline will be offset by one, and we will miss copying the first byte of code.

You're right, I tested what happens if the zynq_secondary_trampoline is ARM or Thumb by editing the file where it's defined, headsmp.S

When the .arm directive is used, the CPU is brought-up correctly, but if I use .thumb, I get the following message (no panic):

Quoted 1 line

CPU1: failed to come online

This seems unrelated to solving the panic, as the message even appears with memcpy and FORTIFY_SOURCE disabled.

I could add the .arm directive to headsmp.S
Is that your expected solution?
Should that change be on a separate commit?

I'd like to know Michal's opinion, as he wrote the code.

#3 lore raw
Quoted 25 lines · Luis Araneda

Hi Russell,

Thanks for reviewing.

Quoted 4 lines · Russell King - ARM Linux admin
Quoted 2 lines · Luis Araneda

This fixes a kernel panic (read overflow) on memcpy when FORTIFY_SOURCE is enabled.

[...]

Quoted 6 lines

I'm not convinced that this is correct. It looks like zynq_secondary_trampoline could be either ARM or Thumb code - there is no .arm directive before it. If it's ARM code, then this is fine. If Thumb code, then zynq_secondary_trampoline will be offset by one, and we will miss copying the first byte of code.

You're right, I tested what happens if the zynq_secondary_trampoline is ARM or Thumb by editing the file where it's defined, headsmp.S

When the .arm directive is used, the CPU is brought-up correctly, but if I use .thumb, I get the following message (no panic):

Quoted 1 line

CPU1: failed to come online

Quoted 9 lines

This seems unrelated to solving the panic, as the message even appears with memcpy and FORTIFY_SOURCE disabled.

I could add the .arm directive to headsmp.S
Is that your expected solution?
Should that change be on a separate commit?

I'd like to know Michal's opinion, as he wrote the code.

There are two things together. Thanks Russel to pointing to it.
1. How to support SMP in thumb2 mode?
Adding .arm mode to headsmp.S which ensure that cpu starts in proper mode and correct code size is copied.
And also point to secondary_startup_arm in zynq_boot_secondary to switch cpu from arm to thumb mode.

2. And the second is this patch to fix FORTIFY_SOURCE.

Feel free to create the first patch too or I will do it myself.

Thanks,
Michal

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

#4 lore raw

Hi Michal,

Thanks for the review.

Quoted 44 lines · Michal Simek
Quoted 25 lines · Luis Araneda

Hi Russell,

Thanks for reviewing.

Quoted 4 lines · Russell King - ARM Linux admin
Quoted 2 lines · Luis Araneda

This fixes a kernel panic (read overflow) on memcpy when FORTIFY_SOURCE is enabled.

[...]

Quoted 6 lines

I'm not convinced that this is correct. It looks like zynq_secondary_trampoline could be either ARM or Thumb code - there is no .arm directive before it. If it's ARM code, then this is fine. If Thumb code, then zynq_secondary_trampoline will be offset by one, and we will miss copying the first byte of code.

You're right, I tested what happens if the zynq_secondary_trampoline is ARM or Thumb by editing the file where it's defined, headsmp.S

When the .arm directive is used, the CPU is brought-up correctly, but if I use .thumb, I get the following message (no panic):

Quoted 1 line

CPU1: failed to come online

Quoted 9 lines

This seems unrelated to solving the panic, as the message even appears with memcpy and FORTIFY_SOURCE disabled.

I could add the .arm directive to headsmp.S
Is that your expected solution?
Should that change be on a separate commit?

I'd like to know Michal's opinion, as he wrote the code.

There are two things together. Thanks Russel to pointing to it.
1. How to support SMP in thumb2 mode?
Adding .arm mode to headsmp.S which ensure that cpu starts in proper mode and correct code size is copied.
And also point to secondary_startup_arm in zynq_boot_secondary to switch cpu from arm to thumb mode.

Quoted 3 lines

2. And the second is this patch to fix FORTIFY_SOURCE.

Feel free to create the first patch too or I will do it myself.

I'll be sending the two patches as a series (I already tested that they work), so they can be picked by the stable trees.

Thanks,
Luis Araneda.