[PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[Neal Frager]

C functions are invoked with arguments without allocating stack space for
register params (for potential register spills). As a result kernel built
with gcc v15 boot hangs at the point of starting init -.

This patch adds necessary stack space for few C function calls from entry.S
(not all). After this patch, the kernel boots all the way to login prompt and
root login works. Simple commands like ps work and show normal output.

Signed-off-by: Gopi Kumar Bulusu <gopi@sankhya.com>
Tested-by: Neal Frager <neal.frager@amd.com>
---
 arch/microblaze/kernel/entry.S | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
index 582d7256d815..2f9604640486 100644
--- a/arch/microblaze/kernel/entry.S
+++ b/arch/microblaze/kernel/entry.S
@@ -339,6 +339,9 @@ C_ENTRY(mb_invalidate_icache):
  * Trap entered via brki instruction, so BIP bit is set, and interrupts
  * are masked. This is nice, means we don't have to CLI before state save
  */
+
+#define SYSCALL_ARGS_SIZE 36
+
 C_ENTRY(_user_exception):
 	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
 	addi	r14, r14, 4	/* return address is 4 byte after call */
@@ -422,16 +425,19 @@ C_ENTRY(_user_exception):
 	lwi	r12, r12, sys_call_table
 	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
 	addi	r15, r0, ret_from_trap-8
+	addik 	r1, r1, -SYSCALL_ARGS_SIZE
 	bra	r12
 
 	/* The syscall number is invalid, return an error.  */
 5:
-	braid	ret_from_trap
+	braid	ret_from_trap2
 	addi	r3, r0, -ENOSYS;
 
 /* Entry point used to return from a syscall/trap */
 /* We re-enable BIP bit before state restore */
 C_ENTRY(ret_from_trap):
+	addik	r1, r1, SYSCALL_ARGS_SIZE
+C_ENTRY(ret_from_trap2):
 	swi	r3, r1, PT_R3
 	swi	r4, r1, PT_R4
 
@@ -504,7 +510,7 @@ C_ENTRY(ret_from_fork):
 	bralid	r15, schedule_tail; /* ...which is schedule_tail's arg */
 	add	r5, r3, r0;	/* switch_thread returns the prev task */
 				/* ( in the delay slot ) */
-	brid	ret_from_trap;	/* Do normal trap return */
+	brid	ret_from_trap2; /* Do normal trap return */
 	add	r3, r0, r0;	/* Child's fork call should return 0. */
 
 C_ENTRY(ret_from_kernel_thread):
@@ -513,13 +519,13 @@ C_ENTRY(ret_from_kernel_thread):
 				/* ( in the delay slot ) */
 	brald	r15, r20	/* fn was left in r20 */
 	addk	r5, r0, r19	/* ... and argument - in r19 */
-	brid	ret_from_trap
+	brid	ret_from_trap2
 	add	r3, r0, r0
 
 C_ENTRY(sys_rt_sigreturn_wrapper):
 	addik	r30, r0, 0		/* no restarts */
 	brid	sys_rt_sigreturn	/* Do real work */
-	addik	r5, r1, 0;		/* add user context as 1st arg */
+	addik	r5, r1, SYSCALL_ARGS_SIZE;	/* add user context as 1st arg */
 
 /*
  * HW EXCEPTION rutine start
@@ -729,11 +735,13 @@ C_ENTRY(_interrupt):
 	lwi	CURRENT_TASK, r0, TOPHYS(PER_CPU(CURRENT_SAVE));
 	tovirt(r1,r1)
 	addik	r15, r0, irq_call;
+	addik 	r1, r1, -8
 irq_call:rtbd	r0, do_IRQ;
-	addik	r5, r1, 0;
+	addik	r5, r1, 8;
 
 /* MS: we are in virtual mode */
 ret_from_irq:
+	addik 	r1, r1, 8
 	lwi	r11, r1, PT_MODE;
 	bnei	r11, 2f;
 
@@ -1302,7 +1310,7 @@ type_SYSCALL_PREEMPT:
 .global microblaze_trap_handlers
 microblaze_trap_handlers:
 	/* Exact matches come first */
-	.word ret_from_trap; .word ret_from_trap   ; .word type_SYSCALL
+	.word ret_from_trap; .word ret_from_trap2  ; .word type_SYSCALL
 	.word ret_from_irq ; .word ret_from_irq    ; .word type_IRQ
 	/* Fuzzy matches go here */
 	.word ret_from_irq ; .word no_intr_resched ; .word type_IRQ_PREEMPT
-- 
2.25.1

Re: [PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[Michal Simek]

On 3/27/26 09:54, Neal Frager wrote:
> C functions are invoked with arguments without allocating stack space for
> register params (for potential register spills). As a result kernel built
> with gcc v15 boot hangs at the point of starting init -.

Remove -.

> 
> This patch adds necessary stack space for few C function calls from entry.S

Add necessary.

> (not all). After this patch, the kernel boots all the way to login prompt and
> root login works. Simple commands like ps work and show normal output.
> 
> Signed-off-by: Gopi Kumar Bulusu <gopi@sankhya.com>
> Tested-by: Neal Frager <neal.frager@amd.com>

If you are sending it you should have sob line here.

> ---
>   arch/microblaze/kernel/entry.S | 20 ++++++++++++++------
>   1 file changed, 14 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
> index 582d7256d815..2f9604640486 100644
> --- a/arch/microblaze/kernel/entry.S
> +++ b/arch/microblaze/kernel/entry.S
> @@ -339,6 +339,9 @@ C_ENTRY(mb_invalidate_icache):
>    * Trap entered via brki instruction, so BIP bit is set, and interrupts
>    * are masked. This is nice, means we don't have to CLI before state save
>    */
> +
> +#define SYSCALL_ARGS_SIZE 36
> +
>   C_ENTRY(_user_exception):
>   	swi	r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
>   	addi	r14, r14, 4	/* return address is 4 byte after call */
> @@ -422,16 +425,19 @@ C_ENTRY(_user_exception):
>   	lwi	r12, r12, sys_call_table
>   	/* where the trap should return need -8 to adjust for rtsd r15, 8 */
>   	addi	r15, r0, ret_from_trap-8
> +	addik 	r1, r1, -SYSCALL_ARGS_SIZE

Please check coding style here.

I will test it on real HW and will talk to you directly about results.

Thanks,
Michal

RE: [PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[Frager, Neal]

[AMD Official Use Only - AMD Internal Distribution Only]

Hi Michal,

Thank you for your thorough review.

> C functions are invoked with arguments without allocating stack space for
> register params (for potential register spills). As a result kernel built
> with gcc v15 boot hangs at the point of starting init -.

> Remove -.

Fixed for v2.

>
> This patch adds necessary stack space for few C function calls from entry.S

> Add necessary.

Fixed for v2.

> (not all). After this patch, the kernel boots all the way to login prompt and
> root login works. Simple commands like ps work and show normal output.
>
> Signed-off-by: Gopi Kumar Bulusu <gopi@sankhya.com>
> Tested-by: Neal Frager <neal.frager@amd.com>

> If you are sending it you should have sob line here.

Fixed for v2.

> ---
>   arch/microblaze/kernel/entry.S | 20 ++++++++++++++------
>   1 file changed, 14 insertions(+), 6 deletions(-)
>
> diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
> index 582d7256d815..2f9604640486 100644
> --- a/arch/microblaze/kernel/entry.S
> +++ b/arch/microblaze/kernel/entry.S
> @@ -339,6 +339,9 @@ C_ENTRY(mb_invalidate_icache):
>    * Trap entered via brki instruction, so BIP bit is set, and interrupts
>    * are masked. This is nice, means we don't have to CLI before state save
>    */
> +
> +#define SYSCALL_ARGS_SIZE 36

I am testing if 36 is really needed and am running a test using 24
for the 6 registers that are used for arguments.

> +
>   C_ENTRY(_user_exception):
>       swi     r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
>       addi    r14, r14, 4     /* return address is 4 byte after call */
> @@ -422,16 +425,19 @@ C_ENTRY(_user_exception):
>       lwi     r12, r12, sys_call_table
>       /* where the trap should return need -8 to adjust for rtsd r15, 8 */
>       addi    r15, r0, ret_from_trap-8
> +     addik   r1, r1, -SYSCALL_ARGS_SIZE

> Please check coding style here.

Fixed for v2.

> I will test it on real HW and will talk to you directly about results.

Yes please.  Thank you!

Best regards,
Neal Frager
AMD

RE: [PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[Frager, Neal]

[AMD Official Use Only - AMD Internal Distribution Only]

Hi Gopi

> There should be 4 bytes for return address (place for link register) as well.

I tested with 24 bytes and I boot normally with your patch and gcc v15.
#define SYSCALL_ARGS_SIZE 24

If you are sure 28 bytes are needed, I will make the patch 28 with a comment
explaining the 6 argument registers plus the link register.

> Hi Michal,

> Thank you for your thorough review.

> C functions are invoked with arguments without allocating stack space for
> register params (for potential register spills). As a result kernel built
> with gcc v15 boot hangs at the point of starting init -.

> Remove -.

> Fixed for v2.

>
> This patch adds necessary stack space for few C function calls from entry.S

> Add necessary.

> Fixed for v2.

> (not all). After this patch, the kernel boots all the way to login prompt and
> root login works. Simple commands like ps work and show normal output.
>
> Signed-off-by: Gopi Kumar Bulusu <mailto:gopi@sankhya.com>
> Tested-by: Neal Frager <mailto:neal.frager@amd.com>

> If you are sending it you should have sob line here.

> Fixed for v2.

> ---
>   arch/microblaze/kernel/entry.S | 20 ++++++++++++++------
>   1 file changed, 14 insertions(+), 6 deletions(-)
>
> diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
> index 582d7256d815..2f9604640486 100644
> --- a/arch/microblaze/kernel/entry.S
> +++ b/arch/microblaze/kernel/entry.S
> @@ -339,6 +339,9 @@ C_ENTRY(mb_invalidate_icache):
>    * Trap entered via brki instruction, so BIP bit is set, and interrupts
>    * are masked. This is nice, means we don't have to CLI before state save
>    */
> +
> +#define SYSCALL_ARGS_SIZE 36

> I am testing if 36 is really needed and am running a test using 24
> for the 6 registers that are used for arguments.

> +
>   C_ENTRY(_user_exception):
>       swi     r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
>       addi    r14, r14, 4     /* return address is 4 byte after call */
> @@ -422,16 +425,19 @@ C_ENTRY(_user_exception):
>       lwi     r12, r12, sys_call_table
>       /* where the trap should return need -8 to adjust for rtsd r15, 8 */
>       addi    r15, r0, ret_from_trap-8
> +     addik   r1, r1, -SYSCALL_ARGS_SIZE

> Please check coding style here.

> Fixed for v2.

> I will test it on real HW and will talk to you directly about results.

> Yes please.  Thank you!

Best regards,
Neal Frager
AMD

Re: [PATCH v1 1/1] arch: microblaze: allocate stack space for register params

[Gopi Kumar Bulusu]

namaskaaram

Yes, please refer to Microblaze Processor Reference Guide, ABI section

https://docs.amd.com/r/en-US/ug984-vivado-microblaze-ref/Stack-Convention

dhanyavaadaaha
gopi

Sankhya Technologies
www.sankhya.com

On Tue, Apr 21, 2026 at 4:16 PM Frager, Neal <neal.frager@amd.com> wrote:
>
> [AMD Official Use Only - AMD Internal Distribution Only]
>
> Hi Gopi
>
> > There should be 4 bytes for return address (place for link register) as well.
>
> I tested with 24 bytes and I boot normally with your patch and gcc v15.
> #define SYSCALL_ARGS_SIZE 24
>
> If you are sure 28 bytes are needed, I will make the patch 28 with a comment
> explaining the 6 argument registers plus the link register.
>
> > Hi Michal,
>
> > Thank you for your thorough review.
>
> > C functions are invoked with arguments without allocating stack space for
> > register params (for potential register spills). As a result kernel built
> > with gcc v15 boot hangs at the point of starting init -.
>
> > Remove -.
>
> > Fixed for v2.
>
> >
> > This patch adds necessary stack space for few C function calls from entry.S
>
> > Add necessary.
>
> > Fixed for v2.
>
> > (not all). After this patch, the kernel boots all the way to login prompt and
> > root login works. Simple commands like ps work and show normal output.
> >
> > Signed-off-by: Gopi Kumar Bulusu <mailto:gopi@sankhya.com>
> > Tested-by: Neal Frager <mailto:neal.frager@amd.com>
>
> > If you are sending it you should have sob line here.
>
> > Fixed for v2.
>
> > ---
> >   arch/microblaze/kernel/entry.S | 20 ++++++++++++++------
> >   1 file changed, 14 insertions(+), 6 deletions(-)
> >
> > diff --git a/arch/microblaze/kernel/entry.S b/arch/microblaze/kernel/entry.S
> > index 582d7256d815..2f9604640486 100644
> > --- a/arch/microblaze/kernel/entry.S
> > +++ b/arch/microblaze/kernel/entry.S
> > @@ -339,6 +339,9 @@ C_ENTRY(mb_invalidate_icache):
> >    * Trap entered via brki instruction, so BIP bit is set, and interrupts
> >    * are masked. This is nice, means we don't have to CLI before state save
> >    */
> > +
> > +#define SYSCALL_ARGS_SIZE 36
>
> > I am testing if 36 is really needed and am running a test using 24
> > for the 6 registers that are used for arguments.
>
> > +
> >   C_ENTRY(_user_exception):
> >       swi     r1, r0, TOPHYS(PER_CPU(ENTRY_SP)) /* save stack */
> >       addi    r14, r14, 4     /* return address is 4 byte after call */
> > @@ -422,16 +425,19 @@ C_ENTRY(_user_exception):
> >       lwi     r12, r12, sys_call_table
> >       /* where the trap should return need -8 to adjust for rtsd r15, 8 */
> >       addi    r15, r0, ret_from_trap-8
> > +     addik   r1, r1, -SYSCALL_ARGS_SIZE
>
> > Please check coding style here.
>
> > Fixed for v2.
>
> > I will test it on real HW and will talk to you directly about results.
>
> > Yes please.  Thank you!
>
> Best regards,
> Neal Frager
> AMD