Skip to contentSkip to conversation

Enable kprobes on AST2500 and later

RFC PATCH In review Andrew Jeffery 3 participants 11 messages May 14, 2020 – May 18, 2020

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

cover Enable kprobes on AST2500 and later

Hello,

kprobes are pretty handy, so turn them on. I'm expecting opinions, especially with the first and last and middle patches, so the series is RFC.

Please opine!

Andrew

Andrew Jeffery (4):
ARM: configs: aspeed-g5: Disable netfilter
ARM: configs: aspeed-g5: Enable modules
ARM: configs: aspeed-g5: Enable kprobes
ARM: configs: aspeed-g5: Set CONFIG_FORTIFY_SOURCE=n to appease
kprobes

 arch/arm/configs/aspeed_g5_defconfig | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
Signature · 1 line
2.25.1
#1 lore raw

RFC PATCH 1/4 ARM: configs: aspeed-g5: Disable netfilter

Enabling modules with CONFIG_NETFILTER=y and CONFIG_NETFILTER_ADVANCED=n results in a diverse set of netfilter functions being built as modules. These functions were not previously enabled in any way because modules were disabled, therefore we can't have been using them. Given that we aren't using them, disable netfilter as well to prevent them from being built.

CONFIG_NETFILTER=y was introduced in 20c90af9ea6a ("ARM: config: aspeed:
Update defconfig") with little fanfare in the commit message, which makes me wonder whether it was enabled because we needed it or whether its addition slipped through the cracks.

---
 arch/arm/configs/aspeed_g5_defconfig | 2 --
 1 file changed, 2 deletions(-)
arch/arm/configs/aspeed_g5_defconfig +0 -2
index cdf30fbe613e..e124b4ddd442 100644
@@ -69,8 +69,6 @@ CONFIG_IPV6_ROUTER_PREF=y
 CONFIG_IPV6_ROUTE_INFO=y
 CONFIG_IPV6_OPTIMISTIC_DAD=y
 CONFIG_IPV6_MULTIPLE_TABLES=y
-CONFIG_NETFILTER=y
-# CONFIG_NETFILTER_ADVANCED is not set
 CONFIG_VLAN_8021Q=y
 CONFIG_NET_NCSI=y
 # CONFIG_WIRELESS is not set
Signature · 1 line
2.25.1
#2 lore raw

RFC PATCH 2/4 ARM: configs: aspeed-g5: Enable modules

In reply to Andrew Jeffery

This is a step on the path to enabling kprobes, which depend on CONFIG_MODULES=y.

Setting CONFIG_MODULES=y results in the following change in kernel size:

Before:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8140500 2440582  178904 10759986         a42f32 vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3445200

After:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8240349 2388862  172304 10801515         a4d16b vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3458368

This gives a 41,529 byte increase to vmlinux and a 13,168 byte increase to the size of the compressed kernel image. In the most constrained BMC flash layout (arch/arm/boot/dts/openbmc-flash-layout.dtsi) this change decreases the free space in the kernel partition from 1,011,248 bytes to 998,080 bytes.

CONFIG_HW_RANDOM=y seems to come along for the ride with the ./scripts/config -e modules && make olddefconfig && make savedefconfig. The measurements above include it, I haven't investigated why it gets enabled or what effect disabling it might have on the numbers.

---
 arch/arm/configs/aspeed_g5_defconfig | 2 ++
 1 file changed, 2 insertions(+)
arch/arm/configs/aspeed_g5_defconfig +2 -0
index e124b4ddd442..e30f27a413f4 100644
@@ -46,6 +46,7 @@ CONFIG_NEON=y
 CONFIG_KERNEL_MODE_NEON=y
 CONFIG_FIRMWARE_MEMMAP=y
 CONFIG_JUMP_LABEL=y
+CONFIG_MODULES=y
 # CONFIG_BLK_DEV_BSG is not set
 # CONFIG_BLK_DEBUG_FS is not set
 # CONFIG_MQ_IOSCHED_DEADLINE is not set
@@ -152,6 +153,7 @@ CONFIG_SERIAL_8250_DW=y
 CONFIG_SERIAL_OF_PLATFORM=y
 CONFIG_ASPEED_KCS_IPMI_BMC=y
 CONFIG_ASPEED_BT_IPMI_BMC=y
+CONFIG_HW_RANDOM=y
 CONFIG_HW_RANDOM_TIMERIOMEM=y
 # CONFIG_I2C_COMPAT is not set
 CONFIG_I2C_CHARDEV=y
Signature · 1 line
2.25.1
Reviewed-by on v1 2/4
#3 lore raw

Re: [RFC PATCH linux dev-5.4 2/4] ARM: configs: aspeed-g5: Enable modules

Quoted 59 lines · Andrew Jeffery

This is a step on the path to enabling kprobes, which depend on CONFIG_MODULES=y.

Setting CONFIG_MODULES=y results in the following change in kernel size:

Before:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8140500 2440582  178904 10759986         a42f32 vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3445200

After:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8240349 2388862  172304 10801515         a4d16b vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3458368

This gives a 41,529 byte increase to vmlinux and a 13,168 byte increase to the size of the compressed kernel image. In the most constrained BMC flash layout (arch/arm/boot/dts/openbmc-flash-layout.dtsi) this change decreases the free space in the kernel partition from 1,011,248 bytes to 998,080 bytes.

CONFIG_HW_RANDOM=y seems to come along for the ride with the ./scripts/config -e modules && make olddefconfig && make savedefconfig. The measurements above include it, I haven't investigated why it gets enabled or what effect disabling it might have on the numbers.

---
 arch/arm/configs/aspeed_g5_defconfig | 2 ++
 1 file changed, 2 insertions(+)
arch/arm/configs/aspeed_g5_defconfig +2 -0
index e124b4ddd442..e30f27a413f4 100644
@@ -46,6 +46,7 @@ CONFIG_NEON=y
 CONFIG_KERNEL_MODE_NEON=y
 CONFIG_FIRMWARE_MEMMAP=y
 CONFIG_JUMP_LABEL=y
+CONFIG_MODULES=y
 # CONFIG_BLK_DEV_BSG is not set
 # CONFIG_BLK_DEBUG_FS is not set
 # CONFIG_MQ_IOSCHED_DEADLINE is not set
@@ -152,6 +153,7 @@ CONFIG_SERIAL_8250_DW=y
 CONFIG_SERIAL_OF_PLATFORM=y
 CONFIG_ASPEED_KCS_IPMI_BMC=y
 CONFIG_ASPEED_BT_IPMI_BMC=y
+CONFIG_HW_RANDOM=y
 CONFIG_HW_RANDOM_TIMERIOMEM=y
 # CONFIG_I2C_COMPAT is not set
 CONFIG_I2C_CHARDEV=y
Signature · 1 line
2.25.1

We're already using modules on the facebook/openbmc kernels and it hasn't seemed to cause any issues. Is there any argument why we shouldn't begin moving some of the drivers to be modules? Any easy change would be to make all the hwmon content as modules and load them based on the device tree.

I think there is a DISTRO_FEATURE we should enable for userspace support of modules. Let me know if you want me to investigate that.

Signature · 1 line
Patrick Williams
#4 lore raw

RE: [RFC PATCH linux dev-5.4 2/4] ARM: configs: aspeed-g5: Enable modules

Quoted 74 lines · Patrick Williams

On Thu, May 14, 2020 at 10:57:01PM +0930, Andrew Jeffery wrote:>> This is a step on the path to enabling kprobes, which depend on

Quoted 3 lines

CONFIG_MODULES=y.

Setting CONFIG_MODULES=y results in the following change in kernel

size:

Quoted 8 lines

Before:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8140500 2440582  178904 10759986         a42f32 vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3445200
Quoted 8 lines

After:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8240349 2388862  172304 10801515         a4d16b vmlinux
         $ stat --printf '%s\n' arch/arm/boot/zImage
         3458368

This gives a 41,529 byte increase to vmlinux and a 13,168 byte

increase

Quoted 1 line

to the size of the compressed kernel image. In the most constrained

BMC

Quoted 1 line

flash layout (arch/arm/boot/dts/openbmc-flash-layout.dtsi) this

change

Quoted 1 line

decreases the free space in the kernel partition from 1,011,248

bytes to

Quoted 6 lines

998,080 bytes.

CONFIG_HW_RANDOM=y seems to come along for the ride with the ./scripts/config -e modules && make olddefconfig && make savedefconfig. The measurements above include it, I haven't investigated why it gets enabled or what effect disabling it might

have

Review

Quoted 8 lines

on the numbers.

---
 arch/arm/configs/aspeed_g5_defconfig | 2 ++
 1 file changed, 2 insertions(+)
(quoted diff) +0 -0

b/arch/arm/configs/aspeed_g5_defconfig

Quoted 22 lines

index e124b4ddd442..e30f27a413f4 100644

arch/arm/configs/aspeed_g5_defconfig +2 -0
@@ -46,6 +46,7 @@ CONFIG_NEON=y
 CONFIG_KERNEL_MODE_NEON=y
 CONFIG_FIRMWARE_MEMMAP=y
 CONFIG_JUMP_LABEL=y
+CONFIG_MODULES=y
 # CONFIG_BLK_DEV_BSG is not set
 # CONFIG_BLK_DEBUG_FS is not set
 # CONFIG_MQ_IOSCHED_DEADLINE is not set
@@ -152,6 +153,7 @@ CONFIG_SERIAL_8250_DW=y
 CONFIG_SERIAL_OF_PLATFORM=y
 CONFIG_ASPEED_KCS_IPMI_BMC=y
 CONFIG_ASPEED_BT_IPMI_BMC=y
+CONFIG_HW_RANDOM=y
 CONFIG_HW_RANDOM_TIMERIOMEM=y
 # CONFIG_I2C_COMPAT is not set
 CONFIG_I2C_CHARDEV=y
Signature · 1 line
2.25.1

We're already using modules on the facebook/openbmc kernels and it hasn't seemed to cause any issues. Is there any argument why we shouldn't begin moving some of the drivers to be modules? Any easy change would be to make all the hwmon content as modules and load them
based on the device tree.

Quoted 3 lines

I think there is a DISTRO_FEATURE we should enable for userspace support
of modules. Let me know if you want me to investigate that.

This gerrit is still open. It also adds content to the rootfs b ecause it tries to pull in vmlinux into the root fs as part of the base kernel recipie:

https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/26092

actually this one that has failed since September:

https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/25536/2

milton

#5 lore raw

RFC PATCH 3/4 ARM: configs: aspeed-g5: Enable kprobes

In reply to Andrew Jeffery

Enable kprobes to improve the debugging experience on the AST2500 and later.

Setting CONFIG_KPROBES=y results in the following change in kernel size:

Before:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8240349 2388862  172304 10801515         a4d16b vmlinux
         $ stat --printf '%s\n' *.zImage
         3458368

After:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8279214 2399530  176912 10855656         a5a4e8 vmlinux
         $ stat --printf '%s\n' *.zImage
         3471880

This gives a 54,141 byte increase to vmlinux and a 13,512 byte increase to the size of the compressed kernel image. In the most constrained BMC flash layout (arch/arm/boot/dts/openbmc-flash-layout.dtsi) this change decreases the free space in the kernel partition from 998,080 bytes to 984,568 bytes.

---
 arch/arm/configs/aspeed_g5_defconfig | 1 +
 1 file changed, 1 insertion(+)
arch/arm/configs/aspeed_g5_defconfig +1 -0
index e30f27a413f4..254fb7562738 100644
@@ -45,6 +45,7 @@ CONFIG_VFP=y
 CONFIG_NEON=y
 CONFIG_KERNEL_MODE_NEON=y
 CONFIG_FIRMWARE_MEMMAP=y
+CONFIG_KPROBES=y
 CONFIG_JUMP_LABEL=y
 CONFIG_MODULES=y
 # CONFIG_BLK_DEV_BSG is not set
Signature · 1 line
2.25.1
#6 lore raw

Re: [RFC PATCH linux dev-5.4 3/4] ARM: configs: aspeed-g5: Enable kprobes

Quoted 19 lines · Andrew Jeffery

Enable kprobes to improve the debugging experience on the AST2500 and later.

Setting CONFIG_KPROBES=y results in the following change in kernel size:

Before:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8240349 2388862  172304 10801515         a4d16b vmlinux
         $ stat --printf '%s\n' *.zImage
         3458368

After:

         $ size vmlinux
            text    data     bss     dec     hex filename
         8279214 2399530  176912 10855656         a5a4e8 vmlinux
         $ stat --printf '%s\n' *.zImage
         3471880
Quoted 5 lines

This gives a 54,141 byte increase to vmlinux and a 13,512 byte increase to the size of the compressed kernel image. In the most constrained BMC flash layout (arch/arm/boot/dts/openbmc-flash-layout.dtsi) this change decreases the free space in the kernel partition from 998,080 bytes to 984,568 bytes.

Just realised I should be measuring the FIT not the zImage so ignore the free-space numbers for the moment.

Andrew

#7 lore raw

RFC PATCH 4/4 ARM: configs: aspeed-g5: Set CONFIG_FORTIFY_SOURCE=n to appease kprobes

In reply to Andrew Jeffery

Setting CONFIG_FORTIFY_SOURCE=y and CONFIG_KPROBES=y on arm gives a panic when trying to insert a new probe:

$ echo r:myretprobe do_sys_open '$retval' >> /sys/kernel/debug/tracing/kprobe_events
[   51.688589] detected buffer overflow in memcpy
[   51.689430] ------------[ cut here ]------------
[   51.689627] kernel BUG at lib/string.c:1096!
[   51.689800] Internal error: Oops - BUG: 0 [#1] SMP ARM
[   51.690107] Modules linked in:
[   51.690442] CPU: 0 PID: 107 Comm: sh Not tainted 5.4.39-00272-ga1ee7c973659 #6
[   51.690674] Hardware name: Generic DT based system
[   51.691184] PC is at fortify_panic+0x18/0x20
[   51.691371] LR is at __irq_work_queue_local+0x40/0x7c
[   51.691575] pc : [<8083eaa4>]    lr : [<8020e0ec>]    psr: 60000013
[   51.691812] sp : bd237cb0  ip : bd237bc0  fp : bd237cbc
[   51.692129] r10: 00000007  r9 : 00000000  r8 : 00000060
[   51.692395] r7 : 8011f2f0  r6 : b5092480  r5 : 7f000000  r4 : b4c53b4c
[   51.692723] r3 : 80c04c48  r2 : 00000000  r1 : bd7c5448  r0 : 00000022
[   51.693088] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   51.693449] Control: 10c5387d  Table: b50f406a  DAC: 00000051
[   51.693769] Process sh (pid: 107, stack limit = 0x4e9b7225)
[   51.694126] Stack: (0xbd237cb0 to 0xbd238000)

YOLO it and disable fortified source.

---
 arch/arm/configs/aspeed_g5_defconfig | 1 -
 1 file changed, 1 deletion(-)
arch/arm/configs/aspeed_g5_defconfig +0 -1
index 254fb7562738..b7f8ccc99000 100644
@@ -269,7 +269,6 @@ CONFIG_SQUASHFS_XZ=y
 CONFIG_SQUASHFS_ZSTD=y
 # CONFIG_NETWORK_FILESYSTEMS is not set
 CONFIG_HARDENED_USERCOPY=y
-CONFIG_FORTIFY_SOURCE=y
 CONFIG_CRYPTO_HMAC=y
 CONFIG_CRYPTO_SHA256=y
 CONFIG_CRYPTO_USER_API_HASH=y
Signature · 1 line
2.25.1
#8 lore raw

Re: [RFC PATCH linux dev-5.4 4/4] ARM: configs: aspeed-g5: Set CONFIG_FORTIFY_SOURCE=n to appease kprobes

Quoted 23 lines · Andrew Jeffery

Setting CONFIG_FORTIFY_SOURCE=y and CONFIG_KPROBES=y on arm gives a panic when trying to insert a new probe:

$ echo r:myretprobe do_sys_open '$retval' >> /sys/kernel/debug/tracing/kprobe_events
[   51.688589] detected buffer overflow in memcpy
[   51.689430] ------------[ cut here ]------------
[   51.689627] kernel BUG at lib/string.c:1096!
[   51.689800] Internal error: Oops - BUG: 0 [#1] SMP ARM
[   51.690107] Modules linked in:
[   51.690442] CPU: 0 PID: 107 Comm: sh Not tainted 5.4.39-00272-ga1ee7c973659 #6
[   51.690674] Hardware name: Generic DT based system
[   51.691184] PC is at fortify_panic+0x18/0x20
[   51.691371] LR is at __irq_work_queue_local+0x40/0x7c
[   51.691575] pc : [<8083eaa4>]    lr : [<8020e0ec>]    psr: 60000013
[   51.691812] sp : bd237cb0  ip : bd237bc0  fp : bd237cbc
[   51.692129] r10: 00000007  r9 : 00000000  r8 : 00000060
[   51.692395] r7 : 8011f2f0  r6 : b5092480  r5 : 7f000000  r4 : b4c53b4c
[   51.692723] r3 : 80c04c48  r2 : 00000000  r1 : bd7c5448  r0 : 00000022
[   51.693088] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   51.693449] Control: 10c5387d  Table: b50f406a  DAC: 00000051
[   51.693769] Process sh (pid: 107, stack limit = 0x4e9b7225)
[   51.694126] Stack: (0xbd237cb0 to 0xbd238000)
Quoted 6 lines

YOLO it and disable fortified source.

---
 arch/arm/configs/aspeed_g5_defconfig | 1 -
 1 file changed, 1 deletion(-)

Doesn't this mean we're just hiding a bug with KPROBE code and potentially hiding more bugs with other drivers?

Signature · 1 line
Patrick Williams
#9 lore raw

Re: [RFC PATCH linux dev-5.4 4/4] ARM: configs: aspeed-g5: Set CONFIG_FORTIFY_SOURCE=n to appease kprobes

Quoted 24 lines · Patrick Williams
Quoted 23 lines · Andrew Jeffery

Setting CONFIG_FORTIFY_SOURCE=y and CONFIG_KPROBES=y on arm gives a panic when trying to insert a new probe:

$ echo r:myretprobe do_sys_open '$retval' >> /sys/kernel/debug/tracing/kprobe_events
[   51.688589] detected buffer overflow in memcpy
[   51.689430] ------------[ cut here ]------------
[   51.689627] kernel BUG at lib/string.c:1096!
[   51.689800] Internal error: Oops - BUG: 0 [#1] SMP ARM
[   51.690107] Modules linked in:
[   51.690442] CPU: 0 PID: 107 Comm: sh Not tainted 5.4.39-00272-ga1ee7c973659 #6
[   51.690674] Hardware name: Generic DT based system
[   51.691184] PC is at fortify_panic+0x18/0x20
[   51.691371] LR is at __irq_work_queue_local+0x40/0x7c
[   51.691575] pc : [<8083eaa4>]    lr : [<8020e0ec>]    psr: 60000013
[   51.691812] sp : bd237cb0  ip : bd237bc0  fp : bd237cbc
[   51.692129] r10: 00000007  r9 : 00000000  r8 : 00000060
[   51.692395] r7 : 8011f2f0  r6 : b5092480  r5 : 7f000000  r4 : b4c53b4c
[   51.692723] r3 : 80c04c48  r2 : 00000000  r1 : bd7c5448  r0 : 00000022
[   51.693088] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   51.693449] Control: 10c5387d  Table: b50f406a  DAC: 00000051
[   51.693769] Process sh (pid: 107, stack limit = 0x4e9b7225)
[   51.694126] Stack: (0xbd237cb0 to 0xbd238000)
Quoted 8 lines
Quoted 6 lines

YOLO it and disable fortified source.

---
 arch/arm/configs/aspeed_g5_defconfig | 1 -
 1 file changed, 1 deletion(-)

Doesn't this mean we're just hiding a bug with KPROBE code

Possibly.

Quoted 2 lines

and
potentially hiding more bugs with other drivers?

Yes.

You should view this series as "things I did to get kprobes 'working'" and not "patches we should apply to the kernel tree". This one in particular was a low-key troll patch to flush out any information on why CONFIG_FORTIFY_SOURCE=y breaks kprobes in particular. I mean, it even features "YOLO" in the commit mesage, so don't take it too seriously :)

Andrew

#10 lore raw

Re: [RFC PATCH linux dev-5.4 4/4] ARM: configs: aspeed-g5: Set CONFIG_FORTIFY_SOURCE=n to appease kprobes

Quoted 41 lines · Andrew Jeffery
Quoted 24 lines · Patrick Williams
Quoted 23 lines · Andrew Jeffery

Setting CONFIG_FORTIFY_SOURCE=y and CONFIG_KPROBES=y on arm gives a panic when trying to insert a new probe:

$ echo r:myretprobe do_sys_open '$retval' >> /sys/kernel/debug/tracing/kprobe_events
[   51.688589] detected buffer overflow in memcpy
[   51.689430] ------------[ cut here ]------------
[   51.689627] kernel BUG at lib/string.c:1096!
[   51.689800] Internal error: Oops - BUG: 0 [#1] SMP ARM
[   51.690107] Modules linked in:
[   51.690442] CPU: 0 PID: 107 Comm: sh Not tainted 5.4.39-00272-ga1ee7c973659 #6
[   51.690674] Hardware name: Generic DT based system
[   51.691184] PC is at fortify_panic+0x18/0x20
[   51.691371] LR is at __irq_work_queue_local+0x40/0x7c
[   51.691575] pc : [<8083eaa4>]    lr : [<8020e0ec>]    psr: 60000013
[   51.691812] sp : bd237cb0  ip : bd237bc0  fp : bd237cbc
[   51.692129] r10: 00000007  r9 : 00000000  r8 : 00000060
[   51.692395] r7 : 8011f2f0  r6 : b5092480  r5 : 7f000000  r4 : b4c53b4c
[   51.692723] r3 : 80c04c48  r2 : 00000000  r1 : bd7c5448  r0 : 00000022
[   51.693088] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   51.693449] Control: 10c5387d  Table: b50f406a  DAC: 00000051
[   51.693769] Process sh (pid: 107, stack limit = 0x4e9b7225)
[   51.694126] Stack: (0xbd237cb0 to 0xbd238000)
Quoted 8 lines
Quoted 6 lines

YOLO it and disable fortified source.

---
 arch/arm/configs/aspeed_g5_defconfig | 1 -
 1 file changed, 1 deletion(-)

Doesn't this mean we're just hiding a bug with KPROBE code

Possibly.

Quoted 2 lines

and
potentially hiding more bugs with other drivers?

Quoted 6 lines

Yes.

You should view this series as "things I did to get kprobes 'working'" and not "patches we should apply to the kernel tree". This one in particular was a low-key troll patch to flush out any information on why CONFIG_FORTIFY_SOURCE=y breaks kprobes in particular.

In [1] Kees Cook suggests it's a false-positive and provides a potential fix. This was submitted at [2]. [3] then fixes the fix. Turns out the issue needs fixing again due to compiler advances, so I sent [4] but it looks like we might instead resolve it by [5].

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1639397
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e46daee53bb5
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=0ac569bf6a798
[4] https://lore.kernel.org/linux-arm-kernel/[email protected]/
[5] https://lore.kernel.org/linux-arm-kernel/202005171447.00CFE0C@keescook/

Andrew