[U-Boot] [PATCH v4 1/2] armv8: Support loading 32-bit OS in AArch32 execution state
york sun
york.sun at nxp.com
Wed Jul 6 18:55:28 CEST 2016
On 07/03/2016 08:39 PM, Huan Wang wrote:
>> On 30.06.16 09:16, Huan Wang wrote:
>>> Hi, Alex,
>>>
>>>>> Am 15.06.2016 um 05:04 schrieb Huan Wang <alison.wang at nxp.com>:
>>>>>
>>>>> Hi, Alex,
>>>>>
>>>>>>> On 06/08/2016 07:14 AM, Alison Wang wrote:
>>>>>>> To support loading a 32-bit OS, the execution state will change
>>>>>>> from
>>>>>>> AArch64 to AArch32 when jumping to kernel.
>>>>>>>
>>>>>>> The architecture information will be got through checking FIT
>>>>>>> image, then U-Boot will load 32-bit OS or 64-bit OS automatically.
>>>>>>>
>>>>>>> Signed-off-by: Ebony Zhu <ebony.zhu at nxp.com>
>>>>>>> Signed-off-by: Alison Wang <alison.wang at nxp.com>
>>>>>>> Signed-off-by: Chenhui Zhao <chenhui.zhao at nxp.com>
>>>>>>> ---
>>>>>>> Changes in v4:
>>>>>>> - Correct config ARM64_SUPPORT_AARCH32.
>>>>>>> - Omit arch and ftaddr arguments.
>>>>>>> - Rename "xreg5" to "tmp".
>>>>>>> - Use xxx_RES1 to combine all RES1 fields in xxx register.
>>>>>>> - Use an immediate cmp directly.
>>>>>>> - Use #ifdef for CONFIG_ARM64_SUPPORT_AARCH32.
>>>>>>>
>>>>>>> Changes in v3:
>>>>>>> - Comments the functions and the arguments.
>>>>>>> - Rename the real parameters.
>>>>>>> - Use the macros instead of the magic values.
>>>>>>> - Remove the redundant codes.
>>>>>>> - Clean up all of the mess in boot_jump_linux().
>>>>>>> - Add CONFIG_ARM64_SUPPORT_AARCH32 to detect for some ARM64 system
>>>>>> doesn't support AArch32 state.
>>>>>>>
>>>>>>> Changes in v2:
>>>>>>> - armv8_switch_to_el2_aarch32() is removed. armv8_switch_to_el2_m
>>>>>>> is
>>>>>> used
>>>>>>> to switch to AArch64 EL2 or AArch32 Hyp.
>>>>>>> - armv8_switch_to_el1_aarch32() is removed. armv8_switch_to_el1_m
>>>>>>> is
>>>>>> used
>>>>>>> to switch to AArch64 EL1 or AArch32 SVC.
>>>>>>>
>>>>>>> arch/arm/Kconfig | 6 ++
>>>>>>> arch/arm/cpu/armv8/start.S | 1 +
>>>>>>> arch/arm/cpu/armv8/transition.S | 8 +-
>>>>>>> arch/arm/include/asm/macro.h | 172
>>>>>> ++++++++++++++++++++++++++++++----------
>>>>>>> arch/arm/include/asm/system.h | 111 +++++++++++++++++++++++++-
>>>>>>> arch/arm/lib/bootm.c | 19 ++++-
>>>>>>> common/image-fit.c | 19 ++++-
>>>>>>> 7 files changed, 284 insertions(+), 52 deletions(-)
>>>>>>>
>>>>>>> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index
>>>>>>> 77eab66..9cf4acd 100644
>>>>>>> --- a/arch/arm/Kconfig
>>>>>>> +++ b/arch/arm/Kconfig
>>>>>>> @@ -91,6 +91,12 @@ config SYS_L2CACHE_OFF
>>>>>>> If SoC does not support L2CACHE or one do not want to
>> enable
>>>>>>> L2CACHE, choose this option.
>>>>>>>
>>>>>>> +config ARM64_SUPPORT_AARCH32
>>>>>>> + bool "ARM64 system support AArch32 execution state"
>>>>>>> + default y if ARM64 && !TARGET_THUNDERX_88XX
>>>>>>> + help
>>>>>>> + This ARM64 system supports AArch32 execution state.
>>>>>>> +
>>>>>>> choice
>>>>>>> prompt "Target select"
>>>>>>> default TARGET_HIKEY
>>>>>>> diff --git a/arch/arm/cpu/armv8/start.S
>>>>>>> b/arch/arm/cpu/armv8/start.S index e933021..dd69501 100644
>>>>>>> --- a/arch/arm/cpu/armv8/start.S
>>>>>>> +++ b/arch/arm/cpu/armv8/start.S
>>>>>>> @@ -234,6 +234,7 @@ WEAK(lowlevel_init)
>>>>>>> /*
>>>>>>> * All slaves will enter EL2 and optionally EL1.
>>>>>>> */
>>>>>>> + ldr x3, =ES_TO_AARCH64
>>>>>>> bl armv8_switch_to_el2
>>>>>>> #ifdef CONFIG_ARMV8_SWITCH_TO_EL1
>>>>>>> bl armv8_switch_to_el1
>>>>>>> diff --git a/arch/arm/cpu/armv8/transition.S
>>>>>> b/arch/arm/cpu/armv8/transition.S
>>>>>>> index 253a39b..e61b6ae 100644
>>>>>>> --- a/arch/arm/cpu/armv8/transition.S
>>>>>>> +++ b/arch/arm/cpu/armv8/transition.S
>>>>>>> @@ -11,13 +11,13 @@
>>>>>>> #include <asm/macro.h>
>>>>>>>
>>>>>>> ENTRY(armv8_switch_to_el2)
>>>>>>> - switch_el x0, 1f, 0f, 0f
>>>>>>> + switch_el x4, 1f, 0f, 0f
>>>>>>> 0: ret
>>>>>>> -1: armv8_switch_to_el2_m x0
>>>>>>> +1: armv8_switch_to_el2_m x0, x3, x4
>>>>>>> ENDPROC(armv8_switch_to_el2)
>>>>>>>
>>>>>>> ENTRY(armv8_switch_to_el1)
>>>>>>> - switch_el x0, 0f, 1f, 0f
>>>>>>> + switch_el x4, 0f, 1f, 0f
>>>>>>> 0: ret
>>>>>>> -1: armv8_switch_to_el1_m x0, x1
>>>>>>> +1: armv8_switch_to_el1_m x0, x3, x4
>>>>>>> ENDPROC(armv8_switch_to_el1)
>>>>>>> diff --git a/arch/arm/include/asm/macro.h
>>>>>> b/arch/arm/include/asm/macro.h
>>>>>>> index 9bb0efa..109724f 100644
>>>>>>> --- a/arch/arm/include/asm/macro.h
>>>>>>> +++ b/arch/arm/include/asm/macro.h
>>>>>>> @@ -8,6 +8,9 @@
>>>>>>>
>>>>>>> #ifndef __ASM_ARM_MACRO_H__
>>>>>>> #define __ASM_ARM_MACRO_H__
>>>>>>> +
>>>>>>> +#include <asm/system.h>
>>>>>>> +
>>>>>>> #ifdef __ASSEMBLY__
>>>>>>>
>>>>>>> /*
>>>>>>> @@ -135,13 +138,20 @@ lr .req x30
>>>>>>> #endif
>>>>>>> .endm
>>>>>>>
>>>>>>> -.macro armv8_switch_to_el2_m, xreg1
>>>>>>> - /* 64bit EL2 | HCE | SMD | RES1 (Bits[5:4]) | Non-secure
>>>> EL0/EL1
>>>>>> */
>>>>>>> - mov \xreg1, #0x5b1
>>>>>>> - msr scr_el3, \xreg1
>>>>>>> +/*
>>>>>>> + * Switch from EL3 to EL2 for ARMv8
>>>>>>> + * @ep: kernel entry point
>>>>>>> + * @flag: The execution state flag for lower exception
>>>>>>> + * level, ES_TO_AARCH64 or ES_TO_AARCH32
>>>>>>> + * @tmp: temporary register
>>>>>>> + *
>>>>>>> + * x1 is machine nr and x2 is ftaddr, they will be passed
>>>>>>> + * to the guest.
>>>>>>> + */
>>>>>>> +.macro armv8_switch_to_el2_m, ep, flag, tmp
>>>>>>> msr cptr_el3, xzr /* Disable coprocessor traps to
>> EL3
>>>>>> */
>>>>>>> - mov \xreg1, #0x33ff
>>>>>>> - msr cptr_el2, \xreg1 /* Disable coprocessor traps to
>> EL2
>>>> */
>>>>>>> + mov \tmp, #CPTR_EL2_RES1
>>>>>>> + msr cptr_el2, \tmp /* Disable coprocessor traps to
>>>> EL2
>>>>>> */
>>>>>>>
>>>>>>> /* Initialize Generic Timers */
>>>>>>> msr cntvoff_el2, xzr
>>>>>>> @@ -152,45 +162,91 @@ lr .req x30
>>>>>>> * and RES0 bits (31,30,27,26,24,21,20,17,15-13,10-6) +
>>>>>>> * EE,WXN,I,SA,C,A,M to 0
>>>>>>> */
>>>>>>> - mov \xreg1, #0x0830
>>>>>>> - movk \xreg1, #0x30C5, lsl #16
>>>>>>> - msr sctlr_el2, \xreg1
>>>>>>> + ldr \tmp, =(SCTLR_EL2_RES1 | SCTLR_EL2_EE_LE |\
>>>>>>> + SCTLR_EL2_WXN_DIS | SCTLR_EL2_ICACHE_DIS |\
>>>>>>> + SCTLR_EL2_SA_DIS | SCTLR_EL2_DCACHE_DIS |\
>>>>>>> + SCTLR_EL2_ALIGN_DIS | SCTLR_EL2_MMU_DIS)
>>>>>>> + msr sctlr_el2, \tmp
>>>>>>> +
>>>>>>> + mov \tmp, sp
>>>>>>> + msr sp_el2, \tmp /* Migrate SP */
>>>>>>> + mrs \tmp, vbar_el3
>>>>>>> + msr vbar_el2, \tmp /* Migrate VBAR */
>>>>>>> +
>>>>>>> + /* Check switch to AArch64 EL2 or AArch32 Hypervisor mode */
>>>>>>> + cmp \flag, #ES_TO_AARCH32
>>>>>>> + b.eq 1f
>>>>>>> +
>>>>>>> + /*
>>>>>>> + * The next lower exception level is AArch64, 64bit EL2 | HCE
>> |
>>>>>>> + * SMD | RES1 (Bits[5:4]) | Non-secure EL0/EL1.
>>>>>>> + */
>>>>>>> + ldr \tmp, =(SCR_EL3_RW_AARCH64 | SCR_EL3_HCE_EN |\
>>>>>>> + SCR_EL3_SMD_DIS | SCR_EL3_RES1 |\
>>>>>>> + SCR_EL3_NS_EN)
>>>>>>> + msr scr_el3, \tmp
>>>>>>>
>>>>>>> /* Return to the EL2_SP2 mode from EL3 */
>>>>>>> - mov \xreg1, sp
>>>>>>> - msr sp_el2, \xreg1 /* Migrate SP */
>>>>>>> - mrs \xreg1, vbar_el3
>>>>>>> - msr vbar_el2, \xreg1 /* Migrate VBAR */
>>>>>>> - mov \xreg1, #0x3c9
>>>>>>> - msr spsr_el3, \xreg1 /* EL2_SP2 | D | A | I | F */
>>>>>>> + ldr \tmp, =(SPSR_EL_DEBUG_MASK | SPSR_EL_SERR_MASK |\
>>>>>>> + SPSR_EL_IRQ_MASK | SPSR_EL_FIQ_MASK |\
>>>>>>> + SPSR_EL_M_AARCH64 | SPSR_EL_M_EL2H)
>>>>>>> + msr spsr_el3, \tmp
>>>>>>> msr elr_el3, lr
>>>>>>
>>>>>> So if we switch into AArch64 mode, we return ...
>>>>>>
>>>>>>> eret
>>>>>>> +
>>>>>>> +1:
>>>>>>> + /*
>>>>>>> + * The next lower exception level is AArch32, 32bit EL2 | HCE
>> |
>>>>>>> + * SMD | RES1 (Bits[5:4]) | Non-secure EL0/EL1.
>>>>>>> + */
>>>>>>> + ldr \tmp, =(SCR_EL3_RW_AARCH32 | SCR_EL3_HCE_EN |\
>>>>>>> + SCR_EL3_SMD_DIS | SCR_EL3_RES1 |\
>>>>>>> + SCR_EL3_NS_EN)
>>>>>>> + msr scr_el3, \tmp
>>>>>>> +
>>>>>>> + /* Return to AArch32 Hypervisor mode */
>>>>>>> + ldr \tmp, =(SPSR_EL_END_LE | SPSR_EL_ASYN_MASK |\
>>>>>>> + SPSR_EL_IRQ_MASK | SPSR_EL_FIQ_MASK |\
>>>>>>> + SPSR_EL_T_A32 | SPSR_EL_M_AARCH32 |\
>>>>>>> + SPSR_EL_M_HYP)
>>>>>>> + msr spsr_el3, \tmp
>>>>>>> + msr elr_el3, \ep
>>>>>>> +
>>>>>>> + mov \ep, #0
>>>>>>
>>>>>> ... while if we switch to AArch32 mode we jump to ep.
>>>>>>
>>>>>> I think it would make a lot of sense if we could *always* jump to
>> ep.
>>>>>> Just swizzle the argument order so that you get
>>>>>>
>>>>>> if ((IH_ARCH_DEFAULT == IH_ARCH_ARM64) &&
>>>>>> (images->os.arch == IH_ARCH_ARM))
>>>>>> armv8_switch_to_el2(0,
>>>>>> gd->bd->bi_arch_number, (uintptr_t)images->ft_addr, (u64)images->ep,
>> ES_TO_AARCH32);
>>>>>> else
>>>>>> armv8_switch_to_el2((uintptr_t)images->ft_addr, 0 0, images->ep,
>>>>>> ES_TO_AARCH64);
>>>>> [Alison Wang] I don't agree it would make a lot of sense if we could
>>>>> *always* jump to ep.
>>>>> If we switch to EL2 AArch32 mode, it will happen at the last minute
>>>>> from U-Boot to kernel. The ep is the entry pointer for kernel.
>>>>> If we switch to EL2 AArch64 mode, it will happen earlier. For
>>>>> primary core, it will call kernel entry after the switch. For
>>>>> secondary cores, it will switch from EL3 AArch64 to EL2 AArch64
>>>>> first, then it will wait until the spin-table is written by the
>>>>> kernel, later it will jump
>>>> to kernel.
>>>>
>>>> Then jump to a separate function that continues the execution stream.
>>>> I really think we should keep the interface between both flags as
>>>> identical as possible.
>>> [Alison Wang] I think you are right. I will change the code to
>>> *always* jump to ep in v5.
>>>>
>>>>>>
>>>>>> If you *really really really* are attached to booting into EL1,
>>>>>> just add a small function that calls armv8_switch_to_el1() for you
>>>>>> and pass the pointer to that as ep to the el2 switch. If I were you
>>>>>> I'd just remove the EL1 mess.
>>>>> [Alison Wang] I agreed with you to remove the EL1 mess.
>>>>
>>>> Awesome, thanks.
>>> [Alison Wang] When I tried to remove the EL1 mess, I found
>>> CONFIG_ARMV8_SWITCH_TO_EL1 is enabled for Versatile Express Aemv8a. If
>>> that platform uses it, I have a concern about removing it.
>>
>> Do you have access to the respective simulator? It's probably only
>> enabled because in the very very very early days of Linux bringup Linux
>> didn't know how to run in EL2 yet. So we can probably just drop it.
>>
>> But let's CC people who might know :).
>>
> [Alison Wang] Alex, thanks for the explanation.
>
> Does anyone disagree to remove CONFIG_ARMV8_SWITCH_TO_EL1 and other corresponding functions?
> If there is no disagreement, I will remove them in the v5 patches.
>
If nobody uses EL1, go ahead to drop it. It may be a good idea to check
this macro and throw an error if found.
York
More information about the U-Boot
mailing list