Since commit 6b9f29b81b15 ("riscv: Enable pcpu page first chunk allocator"), if NUMA is enabled, the page percpu allocator may be used on very sparse configurations, or when requested on boot with percpu_alloc=page.
In that case, percpu data gets put in the vmalloc area. However, sbi_hsm_hart_start() needs the physical address of a sbi_hart_boot_data, and simply assumes that __pa() would work. This causes the just started hart to immediately access an invalid address and hang.
Fortunately, struct sbi_hart_boot_data is not too large, so we can simply allocate an array for boot_data statically, putting it in the kernel image.
This fixes NUMA=y SMP boot on Sophgo SG2042.
To reproduce on QEMU: Set CONFIG_NUMA=y and CONFIG_DEBUG_VIRTUAL=y, then run with:
qemu-system-riscv64 -M virt -smp 2 -nographic \ -kernel arch/riscv/boot/Image \ -append "percpu_alloc=page"
Kernel output:
[ 0.000000] Booting Linux on hartid 0 [ 0.000000] Linux version 6.16.0-rc1 (dram@sakuya) (riscv64-unknown-linux-gnu-gcc (GCC) 14.2.1 20250322, GNU ld (GNU Binutils) 2.44) #11 SMP Tue Jun 24 14:56:22 CST 2025 ... [ 0.000000] percpu: 28 4K pages/cpu s85784 r8192 d20712 ... [ 0.083192] smp: Bringing up secondary CPUs ... [ 0.086722] ------------[ cut here ]------------ [ 0.086849] virt_to_phys used for non-linear address: (____ptrval____) (0xff2000000001d080) [ 0.088001] WARNING: CPU: 0 PID: 1 at arch/riscv/mm/physaddr.c:14 __virt_to_phys+0xae/0xe8 [ 0.088376] Modules linked in: [ 0.088656] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.16.0-rc1 #11 NONE [ 0.088833] Hardware name: riscv-virtio,qemu (DT) [ 0.088948] epc : __virt_to_phys+0xae/0xe8 [ 0.089001] ra : __virt_to_phys+0xae/0xe8 [ 0.089037] epc : ffffffff80021eaa ra : ffffffff80021eaa sp : ff2000000004bbc0 [ 0.089057] gp : ffffffff817f49c0 tp : ff60000001d60000 t0 : 5f6f745f74726976 [ 0.089076] t1 : 0000000000000076 t2 : 705f6f745f747269 s0 : ff2000000004bbe0 [ 0.089095] s1 : ff2000000001d080 a0 : 0000000000000000 a1 : 0000000000000000 [ 0.089113] a2 : 0000000000000000 a3 : 0000000000000000 a4 : 0000000000000000 [ 0.089131] a5 : 0000000000000000 a6 : 0000000000000000 a7 : 0000000000000000 [ 0.089155] s2 : ffffffff8130dc00 s3 : 0000000000000001 s4 : 0000000000000001 [ 0.089174] s5 : ffffffff8185eff8 s6 : ff2000007f1eb000 s7 : ffffffff8002a2ec [ 0.089193] s8 : 0000000000000001 s9 : 0000000000000001 s10: 0000000000000000 [ 0.089211] s11: 0000000000000000 t3 : ffffffff8180a9f7 t4 : ffffffff8180a9f7 [ 0.089960] t5 : ffffffff8180a9f8 t6 : ff2000000004b9d8 [ 0.089984] status: 0000000200000120 badaddr: ffffffff80021eaa cause: 0000000000000003 [ 0.090101] [<ffffffff80021eaa>] __virt_to_phys+0xae/0xe8 [ 0.090228] [<ffffffff8001d796>] sbi_cpu_start+0x6e/0xe8 [ 0.090247] [<ffffffff8001a5da>] __cpu_up+0x1e/0x8c [ 0.090260] [<ffffffff8002a32e>] bringup_cpu+0x42/0x258 [ 0.090277] [<ffffffff8002914c>] cpuhp_invoke_callback+0xe0/0x40c [ 0.090292] [<ffffffff800294e0>] __cpuhp_invoke_callback_range+0x68/0xfc [ 0.090320] [<ffffffff8002a96a>] _cpu_up+0x11a/0x244 [ 0.090334] [<ffffffff8002aae6>] cpu_up+0x52/0x90 [ 0.090384] [<ffffffff80c09350>] bringup_nonboot_cpus+0x78/0x118 [ 0.090411] [<ffffffff80c11060>] smp_init+0x34/0xb8 [ 0.090425] [<ffffffff80c01220>] kernel_init_freeable+0x148/0x2e4 [ 0.090442] [<ffffffff80b83802>] kernel_init+0x1e/0x14c [ 0.090455] [<ffffffff800124ca>] ret_from_fork_kernel+0xe/0xf0 [ 0.090471] [<ffffffff80b8d9c2>] ret_from_fork_kernel_asm+0x16/0x18 [ 0.090560] ---[ end trace 0000000000000000 ]--- [ 1.179875] CPU1: failed to come online [ 1.190324] smp: Brought up 1 node, 1 CPU
Cc: stable@vger.kernel.org Reported-by: Han Gao rabenda.cn@gmail.com Fixes: 9a2451f18663 ("RISC-V: Avoid using per cpu array for ordered booting") Signed-off-by: Vivian Wang wangruikang@iscas.ac.cn --- arch/riscv/kernel/cpu_ops_sbi.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/riscv/kernel/cpu_ops_sbi.c b/arch/riscv/kernel/cpu_ops_sbi.c index e6fbaaf549562d6e9ca63a66371441fe8b230cb3..87d6559448039cdd2e8a604a19bd832ab5a98fc2 100644 --- a/arch/riscv/kernel/cpu_ops_sbi.c +++ b/arch/riscv/kernel/cpu_ops_sbi.c @@ -18,10 +18,10 @@ const struct cpu_operations cpu_ops_sbi;
/* * Ordered booting via HSM brings one cpu at a time. However, cpu hotplug can - * be invoked from multiple threads in parallel. Define a per cpu data + * be invoked from multiple threads in parallel. Define an array of boot data * to handle that. */ -static DEFINE_PER_CPU(struct sbi_hart_boot_data, boot_data); +static struct sbi_hart_boot_data boot_data[NR_CPUS];
static int sbi_hsm_hart_start(unsigned long hartid, unsigned long saddr, unsigned long priv) @@ -67,7 +67,7 @@ static int sbi_cpu_start(unsigned int cpuid, struct task_struct *tidle) unsigned long boot_addr = __pa_symbol(secondary_start_sbi); unsigned long hartid = cpuid_to_hartid_map(cpuid); unsigned long hsm_data; - struct sbi_hart_boot_data *bdata = &per_cpu(boot_data, cpuid); + struct sbi_hart_boot_data *bdata = &boot_data[cpuid];
/* Make sure tidle is updated */ smp_mb();
--- base-commit: 19272b37aa4f83ca52bdf9c16d5d81bdd1354494 change-id: 20250624-riscv-hsm-boot-data-array-1d392098bd22
Best regards,
Hi Vivian,
On Tue, Jun 24, 2025 at 10:05 AM Vivian Wang wangruikang@iscas.ac.cn wrote:
Since commit 6b9f29b81b15 ("riscv: Enable pcpu page first chunk allocator"), if NUMA is enabled, the page percpu allocator may be used on very sparse configurations, or when requested on boot with percpu_alloc=page.
In that case, percpu data gets put in the vmalloc area. However, sbi_hsm_hart_start() needs the physical address of a sbi_hart_boot_data, and simply assumes that __pa() would work. This causes the just started hart to immediately access an invalid address and hang.
Fortunately, struct sbi_hart_boot_data is not too large, so we can simply allocate an array for boot_data statically, putting it in the kernel image.
This fixes NUMA=y SMP boot on Sophgo SG2042.
To reproduce on QEMU: Set CONFIG_NUMA=y and CONFIG_DEBUG_VIRTUAL=y, then run with:
qemu-system-riscv64 -M virt -smp 2 -nographic \ -kernel arch/riscv/boot/Image \ -append "percpu_alloc=page"
Kernel output:
[ 0.000000] Booting Linux on hartid 0 [ 0.000000] Linux version 6.16.0-rc1 (dram@sakuya) (riscv64-unknown-linux-gnu-gcc (GCC) 14.2.1 20250322, GNU ld (GNU Binutils) 2.44) #11 SMP Tue Jun 24 14:56:22 CST 2025 ... [ 0.000000] percpu: 28 4K pages/cpu s85784 r8192 d20712 ... [ 0.083192] smp: Bringing up secondary CPUs ... [ 0.086722] ------------[ cut here ]------------ [ 0.086849] virt_to_phys used for non-linear address: (____ptrval____) (0xff2000000001d080) [ 0.088001] WARNING: CPU: 0 PID: 1 at arch/riscv/mm/physaddr.c:14 __virt_to_phys+0xae/0xe8 [ 0.088376] Modules linked in: [ 0.088656] CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.16.0-rc1 #11 NONE [ 0.088833] Hardware name: riscv-virtio,qemu (DT) [ 0.088948] epc : __virt_to_phys+0xae/0xe8 [ 0.089001] ra : __virt_to_phys+0xae/0xe8 [ 0.089037] epc : ffffffff80021eaa ra : ffffffff80021eaa sp : ff2000000004bbc0 [ 0.089057] gp : ffffffff817f49c0 tp : ff60000001d60000 t0 : 5f6f745f74726976 [ 0.089076] t1 : 0000000000000076 t2 : 705f6f745f747269 s0 : ff2000000004bbe0 [ 0.089095] s1 : ff2000000001d080 a0 : 0000000000000000 a1 : 0000000000000000 [ 0.089113] a2 : 0000000000000000 a3 : 0000000000000000 a4 : 0000000000000000 [ 0.089131] a5 : 0000000000000000 a6 : 0000000000000000 a7 : 0000000000000000 [ 0.089155] s2 : ffffffff8130dc00 s3 : 0000000000000001 s4 : 0000000000000001 [ 0.089174] s5 : ffffffff8185eff8 s6 : ff2000007f1eb000 s7 : ffffffff8002a2ec [ 0.089193] s8 : 0000000000000001 s9 : 0000000000000001 s10: 0000000000000000 [ 0.089211] s11: 0000000000000000 t3 : ffffffff8180a9f7 t4 : ffffffff8180a9f7 [ 0.089960] t5 : ffffffff8180a9f8 t6 : ff2000000004b9d8 [ 0.089984] status: 0000000200000120 badaddr: ffffffff80021eaa cause: 0000000000000003 [ 0.090101] [<ffffffff80021eaa>] __virt_to_phys+0xae/0xe8 [ 0.090228] [<ffffffff8001d796>] sbi_cpu_start+0x6e/0xe8 [ 0.090247] [<ffffffff8001a5da>] __cpu_up+0x1e/0x8c [ 0.090260] [<ffffffff8002a32e>] bringup_cpu+0x42/0x258 [ 0.090277] [<ffffffff8002914c>] cpuhp_invoke_callback+0xe0/0x40c [ 0.090292] [<ffffffff800294e0>] __cpuhp_invoke_callback_range+0x68/0xfc [ 0.090320] [<ffffffff8002a96a>] _cpu_up+0x11a/0x244 [ 0.090334] [<ffffffff8002aae6>] cpu_up+0x52/0x90 [ 0.090384] [<ffffffff80c09350>] bringup_nonboot_cpus+0x78/0x118 [ 0.090411] [<ffffffff80c11060>] smp_init+0x34/0xb8 [ 0.090425] [<ffffffff80c01220>] kernel_init_freeable+0x148/0x2e4 [ 0.090442] [<ffffffff80b83802>] kernel_init+0x1e/0x14c [ 0.090455] [<ffffffff800124ca>] ret_from_fork_kernel+0xe/0xf0 [ 0.090471] [<ffffffff80b8d9c2>] ret_from_fork_kernel_asm+0x16/0x18 [ 0.090560] ---[ end trace 0000000000000000 ]--- [ 1.179875] CPU1: failed to come online [ 1.190324] smp: Brought up 1 node, 1 CPU
Cc: stable@vger.kernel.org Reported-by: Han Gao rabenda.cn@gmail.com Fixes: 9a2451f18663 ("RISC-V: Avoid using per cpu array for ordered booting")
The Fixes should point to commit 6b9f29b81b15 ("riscv: Enable pcpu page first chunk allocator"), but no need to send it back, I fixed it when I merged it in my fixes branch.
Reviewed-by: Alexandre Ghiti alexghiti@rivosinc.com Tested-by: Alexandre Ghiti alexghiti@rivosinc.com
That's a good catch, thanks,
Alex
Signed-off-by: Vivian Wang wangruikang@iscas.ac.cn
arch/riscv/kernel/cpu_ops_sbi.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/riscv/kernel/cpu_ops_sbi.c b/arch/riscv/kernel/cpu_ops_sbi.c index e6fbaaf549562d6e9ca63a66371441fe8b230cb3..87d6559448039cdd2e8a604a19bd832ab5a98fc2 100644 --- a/arch/riscv/kernel/cpu_ops_sbi.c +++ b/arch/riscv/kernel/cpu_ops_sbi.c @@ -18,10 +18,10 @@ const struct cpu_operations cpu_ops_sbi;
/*
- Ordered booting via HSM brings one cpu at a time. However, cpu hotplug can
- be invoked from multiple threads in parallel. Define a per cpu data
*/
- be invoked from multiple threads in parallel. Define an array of boot data
- to handle that.
-static DEFINE_PER_CPU(struct sbi_hart_boot_data, boot_data); +static struct sbi_hart_boot_data boot_data[NR_CPUS];
static int sbi_hsm_hart_start(unsigned long hartid, unsigned long saddr, unsigned long priv) @@ -67,7 +67,7 @@ static int sbi_cpu_start(unsigned int cpuid, struct task_struct *tidle) unsigned long boot_addr = __pa_symbol(secondary_start_sbi); unsigned long hartid = cpuid_to_hartid_map(cpuid); unsigned long hsm_data;
struct sbi_hart_boot_data *bdata = &per_cpu(boot_data, cpuid);
struct sbi_hart_boot_data *bdata = &boot_data[cpuid]; /* Make sure tidle is updated */ smp_mb();
base-commit: 19272b37aa4f83ca52bdf9c16d5d81bdd1354494 change-id: 20250624-riscv-hsm-boot-data-array-1d392098bd22
Best regards,
Vivian "dramforever" Wang
Hello:
This patch was applied to riscv/linux.git (fixes) by Alexandre Ghiti alexghiti@rivosinc.com:
On Tue, 24 Jun 2025 16:04:46 +0800 you wrote:
Since commit 6b9f29b81b15 ("riscv: Enable pcpu page first chunk allocator"), if NUMA is enabled, the page percpu allocator may be used on very sparse configurations, or when requested on boot with percpu_alloc=page.
In that case, percpu data gets put in the vmalloc area. However, sbi_hsm_hart_start() needs the physical address of a sbi_hart_boot_data, and simply assumes that __pa() would work. This causes the just started hart to immediately access an invalid address and hang.
[...]
Here is the summary with links: - riscv: cpu_ops_sbi: Use static array for boot_data https://git.kernel.org/riscv/c/2b29be967ae4
You are awesome, thank you!
linux-stable-mirror@lists.linaro.org
-
Alexandre Ghiti -
patchwork-bot+linux-riscv＠kernel.org -
Vivian Wang