1、ATF里都有什么?
最初的功能很简单:
- cpu_context的保存和恢复,即: 双系统的切换
- 电源管理、PSCI等
但是随着技术的发展,功能也越来越多,越来越复杂,以下列举了当前的部分功能:
- 安全世界的初始化,例如异常向量表、一些控制寄存器和中断寄
- CPU reset和power down的时序。包括Arm DynamIQ cpu的支持。
- 标准的system IP的驱动,例如Generic Interrupt Controller (GIC), Cache Coherent Interconnect (CCI), Cache Coherent Network (CCN), Network Interconnect (NIC) and TrustZone Controller (TZC).
- 一种通用的SCMI驱动程序, 适用于电源控制接口,例如ARM SYSTEM Control Processor(SCP)
- smc处理,using an EL3 runtime services framework
- PSCI库的支持,用于CPU/Cluster/system的电源管理,这个库集成到了aarch64 el3的runtime中,也适用于aarch32 el3
- secure monitor代码,用于world切换、中断routing
- SPDs for the OP-TEE Secure OS, NVIDIA Trusted Little Kernel and Trusty Secure OS
- SecureBoot实现
- 预集成TBB与Arm CryptoCell产品,利用其硬件Root的信任和加密加速服务。
如需更详细,请参考《ATF里面都有什么?》 一文
2、ATF的编译
不同平台之间的设计肯定都是不一样的,但大多数类似如下,请注意 RESET_TO_BL31=1
,表示该ATF从BL31启动。
make -C $DIRPATH RESET_TO_BL31=1 PLAT=xxx clean
make -C $DIRPATH RESET_TO_BL31=1 PLAT=xxx HIGHADDR_DEVICE=1 all
3、ATF的启动
废话不多说,直接上图,请自行理解:BL1 BL2 BL31 BL32 BL33的概念、EL3 S-EL1 NS-EL1的概念。
4、进入ATF的和退出ATF方式
透过事务看本质, 进入和退出ATF,就是就是EL等级切换的过程,那么EL等级都是怎么切换的呢?通过下面一张图就可以说明这一切:
(事实上除了以上同步异常指令,如果是触发异步异常,也会trapped到ATF)
(1)、进入ATF的方式
进入ATF的方式触发异常:同步异常SMC、异步异常(irq,fiq)
➨ 如果是同步异常,那么一定是在linux或tee中发生了smc调用,此时进入跳转ATF中异常向量表中的同步异常程序smc_handler64或smc_handler32
在该程序中,解析smc id,来选择跳转到具体哪一个rt-svc(runtime service)
➨如果是异步异常,那么一定是触发了irq或fiq或serror中断等,此时进入跳转ATF中异常向量表中的异步异常程序,进而跳转到响应的中断处理函数.
在ATF中仅实现irq_aarch64、fiq_aarch64、irq_aarch32、fiq_aarch32 四个异常中断处理函数
vector_entry irq_aarch64
check_and_unmask_ea
handle_interrupt_exception irq_aarch64
end_vector_entry irq_aarch64
vector_entry fiq_aarch64
check_and_unmask_ea
handle_interrupt_exception fiq_aarch64
end_vector_entry fiq_aarch64
vector_entry irq_aarch32
check_and_unmask_ea
handle_interrupt_exception irq_aarch32
end_vector_entry irq_aarch32
vector_entry fiq_aarch32
check_and_unmask_ea
handle_interrupt_exception fiq_aarch32
end_vector_entry fiq_aarch32
在中断函数中,先调用plat_ic_get_pending_interrupt_type获取interrupt_type,其实就是通过读取寄存器read_icc_hppir0_el1() ,判断中断是从那里来的,然后返回下面三种interrupt_type:
define INTR_TYPE_S_EL1 U(0)
define INTR_TYPE_EL3 U(1)
define INTR_TYPE_NS U(2)
有了type,再get_interrupt_type_handler获取handler程序,进而跳转到相应的handler程序。
ATF中中断的注册(这三种类型的handler程序的注册),以INTR_TYPE_S_EL1为例:
在开机bl32_main调用opteed_setup()时,将opteed_sel1_interrupt_handler()函数注册成了INTR_TYPE_S_EL1类型中断,同时也会将REE(Linux)使用的SCR_EL3.FIQ配置成1,也意味着当CPU运行在TEE时,来了一个secure group1的中断,此中断在REE中被标记FIQ后将被target到EL3,进入EL3(ATF)的中断处理函数,也就是刚才注册的opteed_sel1_interrupt_handler()函数,在该函数中,会将cpu切换到TEE中,去处理这个中断;
(2)、退出ATF的方式
进入EL3(ATF)的方式触发异常:ERET指令、或是主动修改PSTATE寄存器
在ATF中执行smc_handler或中断handler结束后,会调用el3_exit,el3_exit会调用ERET指令,恢复Secure或non-secure的PC指针和PSTATE,回到secure EL1或non-secure EL1.
下图是一个smc进入ATF,处理完任务后再返回EL1的过程:
5、ATF中向量表的介绍
请参考《Linux Kernel/optee/ATF等操作系统的异常向量表的速查》 一文
6、ATF中栈的设置
请参考《思想解读:TF-A(ATF)中栈指针和栈内存的设计思想解读》 一文
7、ATF中寄存器的保存和恢复
请参考《TF-A代码阅读: 双系统切换时是如何保存寄存器的(cpu_context介绍)》 一文
8、ATF的rt_svc介绍(runtime service)
(1)、SMC Calling convention文档
我们重点看下这张表,对应smc id的定义
- bit31决定是fast call,还是std call(yield对应的就是std call)
- bit30表示是以32位传参,还是以64位传参, 注意我们看了optee在linux的driver,都是以32位方式
- bit29:24 决定服务的类型
- bit23:16 reserved
- bit15:0 每种call类型下,表示range
bit31、bit30、bit23:16、bit15:0 都是很好理解,我们来讲一下bit29:24
在ATF中定义rt_svc(runtime service)时,需按照该文档的描述来定义
例如在opteed_main.c中,定义了一个service,它的call类型是OEN_TOS_START–OEN_TOS_END,对应的恰好是bit29:24 = 50–63
DECLARE_RT_SVC(
opteed_fast,
OEN_TOS_START,
OEN_TOS_END,
SMC_TYPE_FAST,
opteed_setup,
opteed_smc_handler
);
那么我们在linux kernel中,调用smc时的smc id的bit29:24需要等于50,那么此次的smc调用才会调用到这个runtime service的handler程序
例如在arm_arch_svc_setup.c中,定义了一个service,它的call类型是OEN_ARM_START–OEN_ARM_END,对应的恰好是bit29:24 = 0–0
/* Register Standard Service Calls as runtime service */
DECLARE_RT_SVC(
arm_arch_svc,
OEN_ARM_START,
OEN_ARM_END,
SMC_TYPE_FAST,
NULL,
arm_arch_svc_smc_handler
);
那么我们在linux kernel中,调用smc时的smc id的bit29:24需要等于0,那么此次的smc调用才会调用到这个runtime service的handler程序
例如mtk代码中,mtk_sip_svc.c中,定义了一个service,它的call类型是OEN_SIP_START–OEN_SIP_END,对应的恰好是bit29:24 = 2–2
/* Define a runtime service descriptor for fast SMC calls */
DECLARE_RT_SVC(
mediatek_sip_svc,
OEN_SIP_START,
OEN_SIP_END,
SMC_TYPE_FAST,
NULL,
sip_smc_handler
);
那么我们在linux kernel中,调用smc时的smc id的bit29:24需要等于2,那么此次的smc调用才会调用到这个runtime service的handler程序
(2)、DECLARE_RT_SVC的使用
DECLARE_RT_SVC是一个宏,用于定义一组service,例如在opteed_main.c中的使用
/* Define an OPTEED runtime service descriptor for fast SMC calls */
DECLARE_RT_SVC(
opteed_fast,
OEN_TOS_START,
OEN_TOS_END,
SMC_TYPE_FAST,
opteed_setup,
opteed_smc_handler
);
/* Define an OPTEED runtime service descriptor for yielding SMC calls */
DECLARE_RT_SVC(
opteed_std,
OEN_TOS_START,
OEN_TOS_END,
SMC_TYPE_YIELD,
NULL,
opteed_smc_handler
);
其中OEN_TOS_START和OEN_TOS_END、SMC_TYPE_FAST和SMC_TYPE_YIELD都是按照SMC Calling convention文档来定义的
#define OEN_ARM_START U(0)
#define OEN_ARM_END U(0)
#define OEN_CPU_START U(1)
#define OEN_CPU_END U(1)
#define OEN_SIP_START U(2)
#define OEN_SIP_END U(2)
#define OEN_OEM_START U(3)
#define OEN_OEM_END U(3)
#define OEN_STD_START U(4) /* Standard Service Calls */
#define OEN_STD_END U(4)
#define OEN_STD_HYP_START U(5) /* Standard Hypervisor Service calls */
#define OEN_STD_HYP_END U(5)
#define OEN_VEN_HYP_START U(6) /* Vendor Hypervisor Service calls */
#define OEN_VEN_HYP_END U(6)
#define OEN_TAP_START U(48) /* Trusted Applications */
#define OEN_TAP_END U(49)
#define OEN_TOS_START U(50) /* Trusted OS */
#define OEN_TOS_END U(63)
#define OEN_LIMIT U(64)
#define SMC_TYPE_FAST ULL(1)
#define SMC_TYPE_YIELD ULL(0)
SMC_TYPE_FAST和SMC_TYPE_YIELD也是根据SMC Calling convention文档定义
(3)、DECLARE_RT_SVC的定义
在runtime_svc.h中,其实就是在section(“rt_svc_descs”)段中定义了一个全局变量.
/*
* Convenience macros to declare a service descriptor
*/
#define DECLARE_RT_SVC(_name, _start, _end, _type, _setup, _smch) \
static const rt_svc_desc_t __svc_desc_ ## _name \
__section("rt_svc_descs") __used = {
.start_oen = (_start), \
.end_oen = (_end),
.call_type = (_type),
.name = #_name, \
.init = (_setup),
.handle = (_smch)
}
section “rt_svc_descs”在RT_SVC_DESCS宏中
#define RT_SVC_DESCS
. = ALIGN(STRUCT_ALIGN);
__RT_SVC_DESCS_START__ = .;
KEEP(*(rt_svc_descs))
__RT_SVC_DESCS_END__ = .;
而在rodata_common的宏中,定义了RT_SVC_DESCS
#define RODATA_COMMON
RT_SVC_DESCS
FCONF_POPULATOR
PMF_SVC_DESCS
PARSER_LIB_DESCS
CPU_OPS
GOT
BASE_XLAT_TABLE_RO
在bl31.ld.S中,将RODATA_COMMON放入了rodata段
.rodata . : {
__RODATA_START__ = .;
*(SORT_BY_ALIGNMENT(.rodata*))
RODATA_COMMON
/* Place pubsub sections for events */
. = ALIGN(8);
#include <lib/el3_runtime/pubsub_events.h>
. = ALIGN(PAGE_SIZE);
__RODATA_END__ = .;
} >RAM
(4)、在同步异常中smc_handler64,跳转到响应的rt_svc
附上完整代码和注释
smc_handler64:
/* NOTE: The code below must preserve x0-x4 */
/*
* Save general purpose and ARMv8.3-PAuth registers (if enabled).
* If Secure Cycle Counter is not disabled in MDCR_EL3 when
* ARMv8.5-PMU is implemented, save PMCR_EL0 and disable Cycle Counter.
*/
bl save_gp_pmcr_pauth_regs
#if ENABLE_PAUTH
/* Load and program APIAKey firmware key */
bl pauth_load_bl31_apiakey
#endif
/*
* Populate the parameters for the SMC handler.
* We already have x0-x4 in place. x5 will point to a cookie (not used
* now). x6 will point to the context structure (SP_EL3) and x7 will
* contain flags we need to pass to the handler.
*/
mov x5, xzr
mov x6, sp
/*
* Restore the saved C runtime stack value which will become the new
* SP_EL0 i.e. EL3 runtime stack. It was saved in the 'cpu_context'
* structure prior to the last ERET from EL3.
*/
ldr x12, [x6, #CTX_EL3STATE_OFFSET + CTX_RUNTIME_SP]
/* Switch to SP_EL0 */
msr spsel, #MODE_SP_EL0
/*
* Save the SPSR_EL3, ELR_EL3, & SCR_EL3 in case there is a world
* switch during SMC handling.
* TODO: Revisit if all system registers can be saved later.
*/
mrs x16, spsr_el3
mrs x17, elr_el3
mrs x18, scr_el3
stp x16, x17, [x6, #CTX_EL3STATE_OFFSET + CTX_SPSR_EL3]
str x18, [x6, #CTX_EL3STATE_OFFSET + CTX_SCR_EL3]
/* Copy SCR_EL3.NS bit to the flag to indicate caller's security */
bfi x7, x18, #0, #1
mov sp, x12
/* Get the unique owning entity number */
ubfx x16, x0, #FUNCID_OEN_SHIFT, #FUNCID_OEN_WIDTH ---------------- 获取FUNCID_OEN_SHIFT,对应第一节中的OEN_TOS_START
ubfx x15, x0, #FUNCID_TYPE_SHIFT, #FUNCID_TYPE_WIDTH ---------------- 获取FUNCID_TYPE_SHIFT,对应第一节中的SMC_TYPE_FAST(fast还是yield,yield其实就是standard)
orr x16, x16, x15, lsl #FUNCID_OEN_WIDTH
/* Load descriptor index from array of indices */
adrp x14, rt_svc_descs_indices ----在runtime_svc_init()中会将所有的section rt_svc_descs段放入rt_svc_descs_indices数组,这里获取该数组地址
add x14, x14, :lo12:rt_svc_descs_indices
ldrb w15, [x14, x16] ---找到rt_svc在rt_svc_descs_indices数组中的index
/* Any index greater than 127 is invalid. Check bit 7. */
tbnz w15, 7, smc_unknown
/*
* Get the descriptor using the index
* x11 = (base + off), w15 = index -------------------------重要的注释
*
* handler = (base + off) + (index << log2(size)) ------ 这句注释特别重要,整段汇编看不懂没关系,这句注释看懂就行
*/
adr x11, (__RT_SVC_DESCS_START__ + RT_SVC_DESC_HANDLE)
lsl w10, w15, #RT_SVC_SIZE_LOG2
ldr x15, [x11, w10, uxtw] ------------------------------这句话对应的就是上述注释:handler = (base + off) + (index << log2(size))
/*
* Call the Secure Monitor Call handler and then drop directly into
* el3_exit() which will program any remaining architectural state
* prior to issuing the ERET to the desired lower EL.
*/
#if DEBUG
cbz x15, rt_svc_fw_critical_error
#endif
blr x15 -------------------------------------跳转到handler
b el3_exit
(5)、smc在驱动中的调用
在optee_smc.h中,我们可以查看linux kernel中给driver定义的smc的类型有:
首先是两个宏,一个用于定义fast call,一个用于定义std call
#define OPTEE_SMC_STD_CALL_VAL(func_num) \
ARM_SMCCC_CALL_VAL(ARM_SMCCC_STD_CALL, ARM_SMCCC_SMC_32, \
ARM_SMCCC_OWNER_TRUSTED_OS, (func_num))
#define OPTEE_SMC_FAST_CALL_VAL(func_num) \
ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_32, \
ARM_SMCCC_OWNER_TRUSTED_OS, (func_num))
std call只有两个cmd,一个用于正向调用,一个用于rpc调用
#define OPTEE_SMC_FUNCID_RETURN_FROM_RPC 3
#define OPTEE_SMC_CALL_RETURN_FROM_RPC \
OPTEE_SMC_STD_CALL_VAL(OPTEE_SMC_FUNCID_RETURN_FROM_RPC)
#define OPTEE_SMC_FUNCID_CALL_WITH_ARG OPTEE_MSG_FUNCID_CALL_WITH_ARG
#define OPTEE_SMC_CALL_WITH_ARG \
OPTEE_SMC_STD_CALL_VAL(OPTEE_SMC_FUNCID_CALL_WITH_ARG)
fast call有5个分别用于: get_os_uuid、get_shm_config、exchange_capabilities、disable_shm_cache、enable_shm_cache
#define OPTEE_SMC_FUNCID_GET_OS_UUID OPTEE_MSG_FUNCID_GET_OS_UUID
#define OPTEE_SMC_CALL_GET_OS_UUID \
OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_GET_OS_UUID)
#define OPTEE_SMC_FUNCID_GET_SHM_CONFIG 7
#define OPTEE_SMC_GET_SHM_CONFIG \
OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_GET_SHM_CONFIG)
#define OPTEE_SMC_FUNCID_EXCHANGE_CAPABILITIES 9
#define OPTEE_SMC_EXCHANGE_CAPABILITIES \
OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_EXCHANGE_CAPABILITIES)
#define OPTEE_SMC_FUNCID_DISABLE_SHM_CACHE 10
#define OPTEE_SMC_DISABLE_SHM_CACHE \
OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_DISABLE_SHM_CACHE)
#define OPTEE_SMC_FUNCID_ENABLE_SHM_CACHE 11
#define OPTEE_SMC_ENABLE_SHM_CACHE \
OPTEE_SMC_FAST_CALL_VAL(OPTEE_SMC_FUNCID_ENABLE_SHM_CACHE)
(6)、smc流程下的代码分析
关注"Arm精选"公众号,备注进ARM交流讨论区。