[PATCH v4 06/10] arm64: cpufeature: Detect HCR_EL2.NV1 being RES0

[Marek Szyprowski]

On 12.02.2024 15:00, Marc Zyngier wrote:
> On Mon, 12 Feb 2024 12:48:35 +0000,
> Marek Szyprowski <m.szyprowski at samsung.com> wrote:
>> On 22.01.2024 19:13, Marc Zyngier wrote:
>>> A variant of FEAT_E2H0 not being implemented exists in the form of
>>> HCR_EL2.E2H being RES1 *and* HCR_EL2.NV1 being RES0 (indicating that
>>> only VHE is supported on the host and nested guests).
>>>
>>> Add the necessary infrastructure for this new CPU capability.
>>>
>>> Reviewed-by: Suzuki K Poulose <suzuki.poulose at arm.com>
>>> Signed-off-by: Marc Zyngier <maz at kernel.org>
>>> ---
>> This patch landed recently in linux-next as commit da9af5071b25 ("arm64:
>> cpufeature: Detect HCR_EL2.NV1 being RES0"). I found that it causes a
>> following regression in the CPU hot-plug operation:
>>
>> # for i in /sys/devices/system/cpu/cpu[1-9]; do echo 1 >$i/online; done
>> ------------[ cut here ]------------
>> kernel BUG at arch/arm64/kernel/cpufeature.c:1468!
>> Internal error: Oops - BUG: 00000000f2000800 [#1] PREEMPT SMP
>> Modules linked in: dw_hdmi_cec dw_hdmi_i2s_audio crct10dif_ce
>> snd_soc_simple_card rockchip_saradc industrialio_triggered_buffer
>> hantro_vpu kfifo_buf snd_soc_simple_card_utils rockchip_thermal
>> phy_rockchip_naneng_combphy display_connector gpio_ir_recv v4l2_vp9
>> v4l2_h264 v4l2_mem2mem dwmac_rk videobuf2_dma_contig stmmac_platform
>> videobuf2_memops videobuf2_v4l2 stmmac rockchipdrm videodev
>> snd_soc_rockchip_i2s_tdm snd_soc_rk817 pcs_xpcs panfrost rtc_rk808
>> rk805_pwrkey analogix_dp rk817_charger spi_rockchip_sfc dw_mipi_dsi
>> videobuf2_common dw_hdmi drm_shmem_helper mc gpu_sched
>> drm_display_helper ahci_dwc ip_tables x_tables ipv6
>> CPU: 1 PID: 0 Comm: swapper/1 Not tainted 6.8.0-rc1+ #14563
>> Hardware name: Hardkernel ODROID-M1 (DT)
>> pstate: 800001c9 (Nzcv dAIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>> pc : __read_sysreg_by_encoding+0x38c/0x390
>> lr : read_scoped_sysreg+0x4c/0x70
>> ...
>> Call trace:
>>    __read_sysreg_by_encoding+0x38c/0x390
>>    read_scoped_sysreg+0x4c/0x70
>>    has_nv1+0x18/0x48
>>    verify_local_cpu_caps+0x54/0x124
>>    check_local_cpu_capabilities+0x28/0x208
>>    secondary_start_kernel+0xb0/0x154
>>    __secondary_switched+0xb8/0xbc
>> Code: d53802f3 17ffff3b d5380253 17ffff39 (d4210000)
>> ---[ end trace 0000000000000000 ]---
>> Kernel panic - not syncing: Attempted to kill the idle task!
>> SMP: stopping secondary CPUs
>> Kernel Offset: disabled
>> CPU features: 0x1,80000000,80050295,2100721b
>> Memory Limit: none
>> ---[ end Kernel panic - not syncing: Attempted to kill the idle task! ]---
>>
>> It looks that some additional checks are needed for the CPU hot-plug case.
> Nah, that's much simpler.
>
> I simply missed a spot where ID_AA64MMFR4_EL1 must be handled (it is
> getting annoying that we have more than a single place where these
> "read all the ID registers" are handled).
>
> Can you please give the following a go? If that works for you (it does
> for me here), I'll post an actual fix (plus another fix for another
> buglet I just noticed).

Works fine now.

Reported-by: Marek Szyprowski <m.szyprowski at samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski at samsung.com>
> Thanks,
>
> 	M.
>
> diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> index 0f29ac43c7a2..2f8958f27e9e 100644
> --- a/arch/arm64/kernel/cpufeature.c
> +++ b/arch/arm64/kernel/cpufeature.c
> @@ -1456,6 +1456,7 @@ u64 __read_sysreg_by_encoding(u32 sys_id)
>   	read_sysreg_case(SYS_ID_AA64MMFR1_EL1);
>   	read_sysreg_case(SYS_ID_AA64MMFR2_EL1);
>   	read_sysreg_case(SYS_ID_AA64MMFR3_EL1);
> +	read_sysreg_case(SYS_ID_AA64MMFR4_EL1);
>   	read_sysreg_case(SYS_ID_AA64ISAR0_EL1);
>   	read_sysreg_case(SYS_ID_AA64ISAR1_EL1);
>   	read_sysreg_case(SYS_ID_AA64ISAR2_EL1);
>
Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland