[PATCH 0/4] microblaze: KUnit support

[Thomas Weißschuh]

Add the necessary prerequisites and configuration to run KUnit on
microblaze.

Please note that various KUnit tests will fail with the default
configuration as the memmove() implementation from
arch/microblaze/lib/memmove.c seems to be broken.

Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
---
Thomas Weißschuh (4):
microblaze: uaccess: Zero out destination on failed get_user()
microblaze: reset: Call POWER_OFF handlers
microblaze: reset: Provide a power off handler through an unaligned PC
kunit: qemu_configs: Add microblaze configuration

arch/microblaze/Kconfig | 9 +++++++++
arch/microblaze/include/asm/uaccess.h | 3 ++-
arch/microblaze/kernel/reset.c | 22 ++++++++++++++++++++++
tools/testing/kunit/qemu_configs/microblaze.py | 17 +++++++++++++++++
4 files changed, 50 insertions(+), 1 deletion(-)
---
base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
change-id: 20260802-kunit-microblaze-7a3f6ac88b4d

Best regards,
--
Thomas Weißschuh <li...@weissschuh.net>

[Thomas Weißschuh]

Add a basic configuration to run kunit tests on microblaze.

Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
---

tools/testing/kunit/qemu_configs/microblaze.py | 17 +++++++++++++++++
1 file changed, 17 insertions(+)

diff --git a/tools/testing/kunit/qemu_configs/microblaze.py b/tools/testing/kunit/qemu_configs/microblaze.py
new file mode 100644
index 000000000000..ff012095e77d
--- /dev/null
+++ b/tools/testing/kunit/qemu_configs/microblaze.py
@@ -0,0 +1,17 @@
+# SPDX-License-Identifier: GPL-2.0-only
+from ..qemu_config import QemuArchParams
+
+QEMU_ARCH = QemuArchParams(linux_arch='microblaze',
+ kconfig='''
+CONFIG_CPU_BIG_ENDIAN=y
+CONFIG_SERIAL_UARTLITE=y
+CONFIG_SERIAL_UARTLITE_CONSOLE=y
+CONFIG_MB_POWER_OFF_THROUGH_UNALIGNED_PC=y
+''',
+ qemu_arch='microblaze',
+ kernel_path='arch/microblaze/boot/linux.bin',
+ kernel_command_line='kunit_shutdown=poweroff',
+ extra_qemu_params=[
+ '-M', 'petalogix-s3adsp1800',
+ ],
+)

--
2.55.0

[Thomas Weißschuh]

On failure get_user() is supposed to zero out the destination variable.
This is documented in the kdoc of the microblaze get_user()
implementation and validated in lib/tests/usercopy_kunit.c.

Currently that zeroing is missing.

Add it.

Fixes: 0d6de9532663 ("microblaze_mmu_v2: uaccess MMU update")

Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
---

arch/microblaze/include/asm/uaccess.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/microblaze/include/asm/uaccess.h b/arch/microblaze/include/asm/uaccess.h
index afa0dd8d013f..77203af255e5 100644
--- a/arch/microblaze/include/asm/uaccess.h
+++ b/arch/microblaze/include/asm/uaccess.h
@@ -95,7 +95,8 @@ extern long __user_bad(void);
#define get_user(x, ptr) ({ \
const typeof(*(ptr)) __user *__gu_ptr = (ptr); \
access_ok(__gu_ptr, sizeof(*__gu_ptr)) ? \
- __get_user(x, __gu_ptr) : -EFAULT; \
+ __get_user(x, __gu_ptr) : \
+ ((x) = 0, -EFAULT); \
})

#define __get_user(x, ptr) \

--
2.55.0

[Thomas Weißschuh]

microblaze is missing a generic architecture-wide power off mechanism.

To enable KUnit for microblaze it is necessary for KUnit to shut down
the machine in a way that QEMU will recognize. The machines emulated by
QEMU do not provide machine-specific power off functionality which could
be used.
However at least the petalogix-s3adsp1800 machine will abort if an
unaligned instruction is executed. An ugly message will be printed but
that is not an issue for KUnit.

Make use of this to provide a power off handler.

Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>

---

This is quite hacky. But I didn't find a better solution.
---
arch/microblaze/Kconfig | 9 +++++++++
arch/microblaze/kernel/reset.c | 21 +++++++++++++++++++++
2 files changed, 30 insertions(+)

diff --git a/arch/microblaze/Kconfig b/arch/microblaze/Kconfig
index 484ebb3baedf..0ca8999dd770 100644
--- a/arch/microblaze/Kconfig
+++ b/arch/microblaze/Kconfig
@@ -216,3 +216,12 @@ config MB_MANAGER
Say N here unless you know what you are doing.

endmenu
+
+config MB_POWER_OFF_THROUGH_UNALIGNED_PC
+ bool "Power off through unaligned PC"
+ help
+ This options adds a power off handler which executes an unaligned PC
+ so the machine resets in a generic way. This works for the
+ petalogix-s3adsp1800 QEMU machine.
+
+ Say N here unless you know what you are doing.
diff --git a/arch/microblaze/kernel/reset.c b/arch/microblaze/kernel/reset.c
index 3612a20ca16d..4af660deed3b 100644
--- a/arch/microblaze/kernel/reset.c
+++ b/arch/microblaze/kernel/reset.c
@@ -41,3 +41,24 @@ void machine_restart(char *cmd)
pr_emerg("Reboot failed -- System halted\n");
while (1);
}
+
+#ifdef CONFIG_MB_POWER_OFF_THROUGH_UNALIGNED_PC
+static int unaligned_pc_sys_off(struct sys_off_data *data)
+{
+ __asm__(
+ "bri 1\n"
+ );
+
+ return NOTIFY_DONE;
+}
+
+static int __init register_unaligned_pc_sys_off(void)
+{
+ struct sys_off_handler *sys_off;
+
+ sys_off = register_sys_off_handler(SYS_OFF_MODE_POWER_OFF, SYS_OFF_PRIO_LOW,
+ unaligned_pc_sys_off, NULL);
+ return PTR_ERR_OR_ZERO(sys_off);
+}
+device_initcall(register_unaligned_pc_sys_off);
+#endif /* CONFIG_MB_POWER_OFF_THROUGH_UNALIGNED_PC */

--
2.55.0

[David Gow]

Le 04/08/2026 à 13:32, Thomas Weißschuh a écrit :
> On failure get_user() is supposed to zero out the destination variable.
> This is documented in the kdoc of the microblaze get_user()
> implementation and validated in lib/tests/usercopy_kunit.c.
>
> Currently that zeroing is missing.
>
> Add it.
>
> Fixes: 0d6de9532663 ("microblaze_mmu_v2: uaccess MMU update")
> Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
> ---

This test seems to be the gift which keeps on giving. Every architecture
seems to have had a buggy get_user() at some point.

Reviewed-by: David Gow <da...@davidgow.net>

Cheers,
-- David

[David Gow]

Le 04/08/2026 à 13:32, Thomas Weißschuh a écrit :

> microblaze is missing a generic architecture-wide power off mechanism.
>
> To enable KUnit for microblaze it is necessary for KUnit to shut down
> the machine in a way that QEMU will recognize. The machines emulated by
> QEMU do not provide machine-specific power off functionality which could
> be used.
> However at least the petalogix-s3adsp1800 machine will abort if an
> unaligned instruction is executed. An ugly message will be printed but
> that is not an issue for KUnit.
>
> Make use of this to provide a power off handler.
>
> Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
>
> ---
> This is quite hacky. But I didn't find a better solution.

This is hacky. It does work fine here, though, and I also don't have a
better solution. But I'd feel better about it if a Microblaze person has
seen it and at least not complained too loudly.

Tested-by: David Gow <da...@davidgow.net>

[David Gow]

Le 04/08/2026 à 13:32, Thomas Weißschuh a écrit :

> Add a basic configuration to run kunit tests on microblaze.
>
> Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
> ---

Works well here, though there are still a couple of test failures (and a
longer timeout is needed to get through them all on my machine):

> Testing complete. Ran 965 tests: passed: 866, failed: 3, skipped: 96
> Failures: blake2s.test_hash_alignment_consistency,
memcpy.memmove_overlap_test, printf.dentry


I'm okay with this going in via either the microblaze or KUnit trees.
Any preferences?

Reviewed-by: David Gow <da...@davidgow.net>

Cheers,
-- David

[Thomas Weißschuh]

On 2026-08-05 21:59:16+0800, David Gow wrote:
> Le 04/08/2026 à 13:32, Thomas Weißschuh a écrit :
> > Add a basic configuration to run kunit tests on microblaze.
> >
> > Signed-off-by: Thomas Weißschuh <li...@weissschuh.net>
> > ---
>
> Works well here, though there are still a couple of test failures (and a
> longer timeout is needed to get through them all on my machine):
>
> > Testing complete. Ran 965 tests: passed: 866, failed: 3, skipped: 96
> > Failures: blake2s.test_hash_alignment_consistency,
> memcpy.memmove_overlap_test, printf.dentry

These are all due to the broken memmove() implementation.

It should be fixed by *either*:
- Disabling CONFIG_OPT_LIB_FUNCTION
- Enabling CONFIG_OPT_LIB_ASM

> I'm okay with this going in via either the microblaze or KUnit trees. Any
> preferences?

None from me. I would have expected it to go through microblaze, though.

> Reviewed-by: David Gow <da...@davidgow.net>

Thanks!

(...)