Re: [PATCH v10 05/14] riscv: vector: Use vlenb from DT for thead

[Yangyu Chen <cyy@cyyself.name>]

Hi Charlie,

I have tested this patchset with ghostwrite rebased to linux commit da4373fbcf ("Merge tag 'thermal-6.12-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm") [1] on my D1 Nezha board, with defconfig + CONFIG_ERRATA_THEAD_GHOSTWRITE=n, I got this message during boot:

[    0.027584] Kernel panic - not syncing: __kmem_cache_create_args: Failed to create slab 'riscv_vector_ctx'. Error -22
[    0.038057] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Not tainted 6.12.0-rc6-00310-gb276cf69df24-dirty #11
[    0.047240] Hardware name: Allwinner D1 Nezha (DT)
[    0.052007] Call Trace:
[    0.054434] [<ffffffff80007172>] dump_backtrace+0x1c/0x24
[    0.059806] [<ffffffff809f6834>] show_stack+0x2c/0x38
[    0.064833] [<ffffffff80a040f0>] dump_stack_lvl+0x52/0x74
[    0.070206] [<ffffffff80a04126>] dump_stack+0x14/0x1c
[    0.075233] [<ffffffff809f6db6>] panic+0x10c/0x300
[    0.080000] [<ffffffff8017b5a0>] __kmem_cache_create_args+0x24a/0x2b6
[    0.086413] [<ffffffff80c04c68>] riscv_v_setup_ctx_cache+0x56/0x84
[    0.092566] [<ffffffff80c04288>] arch_task_cache_init+0x10/0x1c
[    0.098460] [<ffffffff80c07d02>] fork_init+0x68/0x1a8
[    0.103486] [<ffffffff80c00ed2>] start_kernel+0x77e/0x822
[    0.108870] ---[ end Kernel panic - not syncing: __kmem_cache_create_args: Failed to create slab 'riscv_vector_ctx'. Error -22 ]---

[1] https://github.com/cyyself/linux/tree/xtheadvector_20241110

On 9/12/24 13:55, Charlie Jenkins wrote:
>  diff --git a/arch/riscv/kernel/vector.c b/arch/riscv/kernel/vector.c
> index 682b3feee451..9775d6a9c8ee 100644
> --- a/arch/riscv/kernel/vector.c
> +++ b/arch/riscv/kernel/vector.c
> @@ -33,7 +33,17 @@ int riscv_v_setup_vsize(void)
>  {
>  	unsigned long this_vsize;
>  -	/* There are 32 vector registers with vlenb length. */
> +	/*
> +	 * There are 32 vector registers with vlenb length.
> +	 *
> +	 * If the thead,vlenb property was provided by the firmware, use that
> +	 * instead of probing the CSRs.
> +	 */
> +	if (thead_vlenb_of) {
> +		this_vsize = thead_vlenb_of * 32;

Then, I patched here which replaces "this_vsize" with "riscv_v_vsize". The kernel boots normally and I can see “xtheadvector" in /proc/cpuinfo.

However, when I try to run the "v_exec_initval_nolibc" test, the kernel panics with these outputs:

[  978.788878] Oops - illegal instruction [#1]
[  978.788897] Modules linked in:
[  978.788908] CPU: 0 UID: 1000 PID: 461 Comm: v_exec_initval_ Not tainted 6.12.0-rc6-00310-gb276cf69df24-dirty #12
[  978.788924] Hardware name: Allwinner D1 Nezha (DT)
[  978.788929] epc : do_trap_ecall_u+0x56/0x20a
[  978.788956]  ra : _new_vmalloc_restore_context_a0+0xc2/0xce
[  978.788974] epc : ffffffff80a04afe ra : ffffffff80a0e742 sp : ffffffc6003fbeb0
[  978.788983]  gp : ffffffff81717080 tp : ffffffd60723b300 t0 : ffffffff81001268
[  978.788991]  t1 : ffffffff80a04aa8 t2 : ffffffff810012a8 s0 : ffffffc6003fbee0
[  978.789000]  s1 : ffffffc6003fbee0 a0 : ffffffc6003fbee0 a1 : 000000000000005d
[  978.789007]  a2 : 0000000000000000 a3 : ffffffffffffffda a4 : 0000000000000003
[  978.789015]  a5 : 0000000000000000 a6 : 0000000002adb5fe a7 : 000000000000005d
[  978.789022]  s2 : 00000000000108a8 s3 : 0000000000000000 s4 : 0000000000000008
[  978.789030]  s5 : 0000003fb42ab780 s6 : 0000002adb5fe420 s7 : 0000002adb5fb9e0
[  978.789038]  s8 : 0000002adb5fe440 s9 : 0000002adb5fe420 s10: 0000002adb572ad4
[  978.789046]  s11: 0000002adb572ad0 t3 : 0000003fb43c5e3c t4 : 622f7273752f3d5f
[  978.789053]  t5 : 0000002adb5fd5a1 t6 : 0000000002adb5ff
[  978.789060] status: 8000000201800100 badaddr: 000000005e0fb057 cause: 0000000000000002
[  978.789069] [<ffffffff80a04afe>] do_trap_ecall_u+0x56/0x20a
[  978.789086] [<ffffffff80a0e742>] _new_vmalloc_restore_context_a0+0xc2/0xce
[  978.789113] Code: a073 1007 006f 1a60 7057 0c30 57fd 17fe 77d7 0c30 (b057) 5e0f
[  978.789123] ---[ end trace 0000000000000000 ]---
[  978.789131] Kernel panic - not syncing: Fatal exception in interrupt
[  978.937158] ---[ end Kernel panic - not syncing: Fatal exception in interrupt ]---

Is something wrong with my setup?

Thanks,
Yangyu Chen

> +		return 0;
> +	}
> +
>  	riscv_v_enable();
>  	this_vsize = csr_read(CSR_VLENB) * 32;
>  	riscv_v_disable();
> diff --git a/arch/riscv/kernel/vendor_extensions/thead.c b/arch/riscv/kernel/vendor_extensions/thead.c
> index 0f27baf8d245..519dbf70710a 100644
> --- a/arch/riscv/kernel/vendor_extensions/thead.c
> +++ b/arch/riscv/kernel/vendor_extensions/thead.c
> @@ -5,6 +5,7 @@
>  #include <asm/vendor_extensions/thead.h>
>    #include <linux/array_size.h>
> +#include <linux/cpumask.h>
>  #include <linux/types.h>
>    /* All T-Head vendor extensions supported in Linux */
> @@ -16,3 +17,13 @@ struct riscv_isa_vendor_ext_data_list riscv_isa_vendor_ext_list_thead = {
>  	.ext_data_count = ARRAY_SIZE(riscv_isa_vendor_ext_thead),
>  	.ext_data = riscv_isa_vendor_ext_thead,
>  };
> +
> +void disable_xtheadvector(void)
> +{
> +	int cpu;
> +
> +	for_each_possible_cpu(cpu)
> +		clear_bit(RISCV_ISA_VENDOR_EXT_XTHEADVECTOR, riscv_isa_vendor_ext_list_thead.per_hart_isa_bitmap[cpu].isa);
> +
> +	clear_bit(RISCV_ISA_VENDOR_EXT_XTHEADVECTOR, riscv_isa_vendor_ext_list_thead.all_harts_isa_bitmap.isa);
> +}