Skip to contentSkip to conversation

riscv: smp: Align secondary_start_sbi to 4 bytes

PATCH Superseded [email protected] 3 participants 3 messages Feb 5 RISC-V

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 riscv: smp: Align secondary_start_sbi to 4 bytes

From: Chen Pei <[email protected]>

During SMP boot, the secondary_start_sbi address is passed to the slave core via sbi_hsm_hart_start. In OpenSBI, this address is written to STVEC in sbi_hart_switch_mode.

According to the privileged specification, the BASE field of STVEC must always be aligned on a 4-byte boundary. However, System.map reveals that secondary_start_sbi is not a 4-byte aligned address.

For example, the address of secondary_start_sbi is 0xffffffff80001066, and the disassembly is as follows:

Dump of assembler code from 0xffffffff80001052 to 0xffffffff8000107a:

   0xffffffff80001052 <_start+4178>:    c.nop   -11
   0xffffffff80001054 <_start+4180>:    auipc   gp,0x1a1f
   0xffffffff80001058 <_start+4184>:    addi    gp,gp,84
   0xffffffff8000105c <_start+4188>:    csrw    satp,a2
   0xffffffff80001060 <_start+4192>:    sfence.vma
   0xffffffff80001064 <_start+4196>:    ret
   0xffffffff80001066 <_start+4198>:    csrw    sie,zero
   0xffffffff8000106a <_start+4202>:    csrw    sip,zero
   0xffffffff8000106e <_start+4206>:    li      t0,2
   0xffffffff80001070 <_start+4208>:    csrw    scounteren,t0
   0xffffffff80001074 <_start+4212>:    auipc   gp,0x1a1f
   0xffffffff80001078 <_start+4216>:    addi    gp,gp,52

When writing to STVEC at address 0xffffffff80001066, the actual write address is 0xffffffff80001064, corresponding to the address of the previous ret instruction. This is unexpected, and if an interrupt occurs at this point, it will cause unpredictable results.

However, secondary_start_sbi immediately masks all interrupts and updates STVEC, so no problems occurred.

In summary, it is more reasonable to make secondary_start_sbi satisfy 4-byte alignment.

---
 arch/riscv/kernel/head.S | 1 +
 1 file changed, 1 insertion(+)
arch/riscv/kernel/head.S +1 -0
index bdf3352acf4c..f35dbfa79505 100644
@@ -124,6 +124,7 @@ relocate_enable_mmu:
 
 	ret
 #endif /* CONFIG_MMU */
+.align 2
 #ifdef CONFIG_SMP
 	.global secondary_start_sbi
 secondary_start_sbi:
Signature · 1 line
2.50.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 57 lines · On Thu, Feb 5, 2026 at 8:16 PM <[email protected]> wrote:

From: Chen Pei <[email protected]>

During SMP boot, the secondary_start_sbi address is passed to the slave core via sbi_hsm_hart_start. In OpenSBI, this address is written to STVEC in sbi_hart_switch_mode.

According to the privileged specification, the BASE field of STVEC must always be aligned on a 4-byte boundary. However, System.map reveals that secondary_start_sbi is not a 4-byte aligned address.

For example, the address of secondary_start_sbi is 0xffffffff80001066, and the disassembly is as follows:

Dump of assembler code from 0xffffffff80001052 to 0xffffffff8000107a:

   0xffffffff80001052 <_start+4178>:    c.nop   -11
   0xffffffff80001054 <_start+4180>:    auipc   gp,0x1a1f
   0xffffffff80001058 <_start+4184>:    addi    gp,gp,84
   0xffffffff8000105c <_start+4188>:    csrw    satp,a2
   0xffffffff80001060 <_start+4192>:    sfence.vma
   0xffffffff80001064 <_start+4196>:    ret
   0xffffffff80001066 <_start+4198>:    csrw    sie,zero
   0xffffffff8000106a <_start+4202>:    csrw    sip,zero
   0xffffffff8000106e <_start+4206>:    li      t0,2
   0xffffffff80001070 <_start+4208>:    csrw    scounteren,t0
   0xffffffff80001074 <_start+4212>:    auipc   gp,0x1a1f
   0xffffffff80001078 <_start+4216>:    addi    gp,gp,52

When writing to STVEC at address 0xffffffff80001066, the actual write address is 0xffffffff80001064, corresponding to the address of the previous ret instruction. This is unexpected, and if an interrupt occurs at this point, it will cause unpredictable results.

However, secondary_start_sbi immediately masks all interrupts and updates STVEC, so no problems occurred.

In summary, it is more reasonable to make secondary_start_sbi satisfy 4-byte alignment.

---
 arch/riscv/kernel/head.S | 1 +
 1 file changed, 1 insertion(+)
arch/riscv/kernel/head.S +1 -1
index bdf3352acf4c..f35dbfa79505 100644
@@ -124,6 +124,7 @@ relocate_enable_mmu:
 
        ret
 #endif /* CONFIG_MMU */
+.align 2
 #ifdef CONFIG_SMP
        .global secondary_start_sbi
 secondary_start_sbi:
--
Quoted 2 lines

2.50.1

In Linux/arch/riscv/kernel/cpu_ops_sbi.c:
unsigned long boot_addr = __pa_symbol(secondary_start_sbi); return sbi_hsm_hart_start(hartid, boot_addr, hsm_data);

OpenSBI sets boot_addr to both sepc and stval, and stval requires 4-byte alignment.

Signature · 2 lines
Best Regards
 Guo Ren
Reviewed-by on v1 1/1
#2 lore raw
In reply to [email protected]
Quoted 58 lines · On Thu, Feb 05, 2026 at 08:16:25PM +0800, [email protected] wrote:

From: Chen Pei <[email protected]>

During SMP boot, the secondary_start_sbi address is passed to the slave core via sbi_hsm_hart_start. In OpenSBI, this address is written to STVEC in sbi_hart_switch_mode.

According to the privileged specification, the BASE field of STVEC must always be aligned on a 4-byte boundary. However, System.map reveals that secondary_start_sbi is not a 4-byte aligned address.

For example, the address of secondary_start_sbi is 0xffffffff80001066, and the disassembly is as follows:

Dump of assembler code from 0xffffffff80001052 to 0xffffffff8000107a:

   0xffffffff80001052 <_start+4178>:    c.nop   -11
   0xffffffff80001054 <_start+4180>:    auipc   gp,0x1a1f
   0xffffffff80001058 <_start+4184>:    addi    gp,gp,84
   0xffffffff8000105c <_start+4188>:    csrw    satp,a2
   0xffffffff80001060 <_start+4192>:    sfence.vma
   0xffffffff80001064 <_start+4196>:    ret
   0xffffffff80001066 <_start+4198>:    csrw    sie,zero
   0xffffffff8000106a <_start+4202>:    csrw    sip,zero
   0xffffffff8000106e <_start+4206>:    li      t0,2
   0xffffffff80001070 <_start+4208>:    csrw    scounteren,t0
   0xffffffff80001074 <_start+4212>:    auipc   gp,0x1a1f
   0xffffffff80001078 <_start+4216>:    addi    gp,gp,52

When writing to STVEC at address 0xffffffff80001066, the actual write address is 0xffffffff80001064, corresponding to the address of the previous ret instruction. This is unexpected, and if an interrupt occurs at this point, it will cause unpredictable results.

However, secondary_start_sbi immediately masks all interrupts and updates STVEC, so no problems occurred.

In summary, it is more reasonable to make secondary_start_sbi satisfy 4-byte alignment.

---
 arch/riscv/kernel/head.S | 1 +
 1 file changed, 1 insertion(+)
arch/riscv/kernel/head.S +1 -0
index bdf3352acf4c..f35dbfa79505 100644
@@ -124,6 +124,7 @@ relocate_enable_mmu:
 
 	ret
 #endif /* CONFIG_MMU */
+.align 2
 #ifdef CONFIG_SMP
 	.global secondary_start_sbi
 secondary_start_sbi:
Signature · 1 line
2.50.1

Looks good, but the .align 2 would ideally go inside the #ifdef CONFIG_SMP, just above the label.

Thanks,
drew