文章总结： 本文深度解析iOSSPTM机制，详述启动、GL模式切换及状态机调度。核心发现是SPTM接管页表映射，利用SPRR索引实现内存retype分类与隔离以防止非法访问。文章还分析异常处理及TXM运行机制，揭示了苹果内核安全的深层防护逻辑，为研究提供参考。
综合评分： 90
文章分类： 逆向分析,二进制安全,移动安全,漏洞分析

IOS SPTM深度分析

原创

wzt

kernsec

2026年1月2日 17:57
浙江

1 SPTM启动

2 GL Entry分析

3 Exception处理分析

4 Sptm页表保护

5 Sptm retype分类保护

6 TXM分析

1. sptm启动

MRS X0,&nbsp;#0, c4, c2, #2 ; CurrentELCMP X0,&nbsp;#8B.NE loc_FFFFFFF00706B074 ; != EL2MRS X0,&nbsp;#4, c1, c1, #0 ; HCR_EL2ORR X0, X0,&nbsp;#0x8000000 ; TGE, bit [27] All exceptions that would be routed to EL1 are routed to EL2.ORR X0, X0,&nbsp;#0x400000000 ; E2H, bit [34], The facilities to support a Host Operating System at EL2 are enabled.MSR&nbsp;#4, c1, c1, #0, X0 ;

首先判断如果在el2状态，则设置hcr_el2的TGE和E2H状态。 sptm由iboot加载，iboot跳转到sptm时处于el2状态。

ADRL&nbsp;X30, sptm_init_subsystemMOV&nbsp;X21, X20ADRL&nbsp;X0, unk_FFFFFFF0070B0000ADD&nbsp;X0, X0, X22SUB&nbsp;X0, X0, X23MSR&nbsp;#5, #0MOV&nbsp;SP, X0MOV&nbsp;X24, #1B&nbsp;init_hardware_registers ;

执行init_hardware_registers，后在跳转到sptm_init_subsystem继续执行。

init_hardware_registers的执行流程：

– 设置SPRR_PPERM_EL1寄存器为0x2020a52a302afaf5。

– 设置TCR_EL1为0x0800236511a511

– 设置TTBR0_EL1为0xFFFFFFF007010000

– 设置TTBR1_EL1为0xFFFFFFF0070A0000

– 重新设置APCTL_EL1。

– 用0xFEEDFACEFEEDFACF依次设置APIBKey_EL1、APDBKey_EL1、KERNKE_EL1（苹果新增）、 APIAKey_EL1、APDAKey_EL1、APGAKey_EL1。

– 设置SCTLR_EL1为0x01000fc14713d。

– 设置若干HID*寄存器（苹果新增）。

sptm_init_subsystem的执行流程：

– 调用copy_and_clear_random_seed生成若干随机种子。

– 调用init_xnu_ro_data初始化xnu_ro_pagetables_begin、xnu_ro_pagetables_end、

、xnu_exc_return_handler、xnu_ctrr_dispatch_table等全局变量。

– 调用validate_region_order检查txm的地址范围。

– 调用register_papt_range依次注册SPTM-rx、SPTM-rw、SPTM-le、TXM-ro、TXM-rx、TXM-bx、TXM-rw、BootKC-ro、BootKC-rs、BootKC-rx、BootKC-bx、BootKC-rw、BootKC-le、TrustCache、CL4-ro、CL4-rx、CL4-rw、CL4-le、RAMDisk、RTBuddySeg、SEPFW、SEPPatches、uStuff、preoslog、BootArgs、ExclaveOSIntegrityCatalog、ExclaveOSTrustCache、AVAILABLE、PHYS_SLIDE等地址范围信息。

register_papt_range(char&nbsp;*name,&nbsp;uint64_t&nbsp;type,&nbsp;uint64_t&nbsp;flag){&nbsp; &nbsp;init_get_image_region(name)；&nbsp; &nbsp;bootstrap_register_papt_range();}

bootstrap_register_papt_range向papt_range_array数组注册一个地址范围信息：

struct&nbsp;papt_range_array&nbsp;{&nbsp; &nbsp;char&nbsp;*name;+0x8:&nbsp; &nbsp;uint64_t&nbsp;type;+0x10:&nbsp; &nbsp;uint64_t&nbsp;start_address;+0x18:&nbsp; &nbsp;uint64_t&nbsp;base;+0x20:&nbsp; &nbsp;uint64_t&nbsp;end_address;+0x24:&nbsp; &nbsp;char&nbsp;flag;+0x28:};

bootstrap_register_papt_range(char&nbsp;*name,&nbsp;uint64_t&nbsp;type,&nbsp;uint64_t&nbsp;start_addr,&nbsp;uint64_t&nbsp;end_addr,&nbsp;uint64_t&nbsp;flag){&nbsp; papt_range_array[papt_range_array_length].name = name;&nbsp; papt_range_array[papt_range_array_length].type = type;&nbsp; papt_range_array[papt_range_array_length].start_address = start_addr;&nbsp; papt_range_array[papt_range_array_length]end_address = end_addr;&nbsp; papt_range_array[papt_range_array_length].flag = flag;&nbsp; papt_range_array_length++

| |
| — |
| |

| | | |
| — | — | — |
| Name | Type | Flag |
| DeviceTree | 0xb | 0x20 |
| SPTM-ro | 0x2 | 0x28 |
| SPTM-rx | 0x4 | 0x21 |
| SPTM-rw | 0x2 | 0x21 |
| SPTM-le | 0x2 | 0x21 |
| TXM-ro | 0x20 | 0x48 |
| TXM-rx | 0x20 | 0x41 |
| TXM-bx | 0x5 | 0x41 |
| TXM-rw | 0x20 | 0x41 |
| TXM-le | 0x20 | 0x41 |
| BootKC-ro | 0xb | 0x48 |
| BootKC-rs | 0xb | 0x41 |
| BootKC-rx | 0xb | 0x41 |
| BootKC-bx | x | 0x41 |
| BootKC-rw | 0xb | 0x51 |
| BootKC-le | 0xb | 0x51 |
| TrustCache | 0x0 | 0x40 |
| CL4-ro | 0x32 | 0x6 |
| CL4-rx | 0x32 | 0x6 |
| CL4-rw | 0x32 | 0x6 |
| CL4-le | 0x32 | 0x6 |
| RAMDisk | 0xb | 0x2 |
| RTBuddySeg | 0xb | 0x2 |
| SEPFW | 0xb | 0x2 |
| SEPPatches | 0xb | 0x2 |
| uStuff | 0xb | 0x2 |
| Preoslog | 0xb | 0x2 |
| BootArgs | 0xb | 0x0 |
| ExclaveOSIntegrityCatalog | 0x32 | 0x6 |
| ExclaveOSTrustCache | 0x32 | 0x6 |
| AVAILABLE | 0x0 | 0x0 |
| PHYS_SLIDE | 0xb | 0x0 |

– 调用sptm_bootstrap_early

– 调用初始化sprr相关寄存器：

设置SPRR_PPERM_EL1 – sprr kernel permission寄存器值为0x2020a52a302afaf5。

设置SPRR_UPERM_EL0 – sprr user permission寄存器值为0x2010002030100000。

– 调用init_gxf_mode_prev

设置GXF_CONFIG_EL1寄存器为1。

设置GXF_PABENTRY_EL1。

设置GXF_ENTRY_EL1为sub_FFFFFFF00706C808。

执行0x201420指令进入Guard Level模式， GL的入口地址设置为sub_FFFFFFF00706C808，

__text:FFFFFFF00706C808 sub_FFFFFFF00706C808 ; DATA XREF:MSR&nbsp;#6, c15, c11, #1, X0 ; TPIDR_GL2 - Software Thread ID Register (GL2)MOVK&nbsp;X0, #0x2020,LSL#48MOVK&nbsp;X0, #0xA52A,LSL#32MOVK&nbsp;X0, #0x302A,LSL#16MOVK&nbsp;X0, #0xFAE6 ;&nbsp;0x2020a52a302afae6MSR&nbsp;#6, c15, c1, #6, X0 ; SPRR_PPERM_EL1 - SPRR Kernel Permission Configuration Register (EL1)MOVK&nbsp;X0, #0,LSL#48MOVK&nbsp;X0, #0,LSL#32MOVK&nbsp;X0, #0,LSL#16MOVK&nbsp;X0, #0xFFMSR&nbsp;#6, c15, c1, #0, X0 ; SPRR_CONFIG_EL1 - SPRR Configuration Register (EL1)RET

首先设置TPIDR_GL2寄存器为init_gl2_thread。

gl2_thread {+0x0:&nbsp; &nbsp;char&nbsp;domain_type;+0x8:&nbsp; &nbsp;uint64_t&nbsp;tls_index;+0x18:&nbsp; &nbsp;uint64_t&nbsp;gl2_sp;+0x20:&nbsp; &nbsp;uint64_t&nbsp;unknown;+0x28:&nbsp; &nbsp;char&nbsp;sptm_state;+0x30:&nbsp; &nbsp;uint64_t&nbsp;hop_flag;+0x38:&nbsp;struct&nbsp;save_caller_reg&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;x0,x1,x2,x3,x4,x5,x6,x7;};+0x78:&nbsp;struct&nbsp;save_exception_reg&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;0x0;&nbsp; &nbsp;...&nbsp; &nbsp;uint64_t&nbsp;0xb0;&nbsp;};+0xf8:&nbsp; &nbsp;uint64_t&nbsp;gl1_sp;+0x130:&nbsp;struct&nbsp;save_exception_reg&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;0x0;&nbsp; &nbsp;...&nbsp; &nbsp;uint64_t&nbsp;0xb0;&nbsp;};+0x1f0:&nbsp;uint64_t&nbsp;gl0_sp;+0x520:&nbsp;union&nbsp;{&nbsp;struct&nbsp;config_reg&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;save_caller_arg_address;&nbsp; &nbsp;uint64_t&nbsp;sprr_user_permission;&nbsp; &nbsp;uint64_t&nbsp;id_isar2_el1;&nbsp; &nbsp;uint64_t&nbsp;unknown;&nbsp; &nbsp;uint64_t&nbsp;apctl_el1;&nbsp; &nbsp;uint64_t&nbsp;sctlr_el1;&nbsp;};&nbsp;struct&nbsp;config_reg1&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;tpidr_el0;&nbsp; &nbsp;uint64_t&nbsp;tpidrro_el0;&nbsp; &nbsp;uint64_t&nbsp;sprr_pperm_el12;&nbsp; &nbsp;uint64_t&nbsp;sprr_uperm_el02&nbsp; &nbsp;uint64_t&nbsp;gxf_config_el12;&nbsp; &nbsp;uint64_t&nbsp;sprr_config_el12;&nbsp; &nbsp;...&nbsp; &nbsp;uint64_t&nbsp;tpidr_gl1;&nbsp; &nbsp;uint64_t&nbsp;unknown;&nbsp; &nbsp;uint64_t&nbsp;vbar_gl12;&nbsp; &nbsp;uint64_t&nbsp;unknown;+0x88:&nbsp; &nbsp;uint64_t&nbsp;hcr_el2;&nbsp;};+0x5b0:&nbsp;strucct config_reg2 {&nbsp; &nbsp;uint64_t;&nbsp;+0x88:&nbsp; &nbsp;uint64_t;&nbsp;};+0x640:&nbsp;struct&nbsp;unknown_reg_store&nbsp;{&nbsp; &nbsp;uint64_t&nbsp;x0,x1,x2,x3,x4,x5,x6,x7;&nbsp;};+0x668:&nbsp;char&nbsp;el_level;+0x680:&nbsp;uint64_t&nbsp;current_cpu;+0x7f0:&nbsp;uint64_t&nbsp;value;+0x7f8:&nbsp;uint64_t&nbsp;unknown;+0x870:&nbsp;struct&nbsp;unknown_ctrr_struct&nbsp;*ctrr;};struct&nbsp;unknown_ctrr_struct&nbsp;{&nbsp;uint64_t&nbsp;value1;+0x800:&nbsp;uint64_t&nbsp;value2;};

重新设置SPRR_PPERM_EL1 – sprr kernel permission寄存器值为0x2020a52a302afae6。

重新设置SPRR_CONFIG_EL1为0xff。

注意随后使用的是ret指令，而不是0x201400，说明并没有推出GL模式，从而让sptm从普通的el2模式进入到GL2模式继续运行。

– 调用init_sptm_mode继续初始化sptm。

锁定VMSA_LOCK_EL1寄存器。

用CTRR_A_UPR/LOW_EL1寄存器锁定el1地址范围。

用CTRR_A_UPR/LOW_EL2寄存器锁定el2地址范围。

调用init_gxf_mode_final，它将GXF_ENTRY_EL1设置为gxf_entry_el1函数地址，将VBAR_GL1设置为syn_handler_sp0_g12地址，将GXF_CONFIG_EL1重新设置为0x2d。

– 分别设置elr_gl1_cl4、elr_gl1_txm、elr_gl1_kernelcache三个全局变量。

– 调用sptm_register_dispatch_table注册当前domain的三个dispatch_table函数。

| |
| — |
| |

| | | |
| — | — | — |
| table_id | Dispatch function | Table_permissions |
| 0 | sptm_dispatch | 2 |
| 1 | sptm_dispatch | 4 |
| 2 | sptm_dispatch | 8 |

sptm_register_dispatch_table(uint64_t table_id, uint64_t dispatch_entry_point, uint64_t table_permissions) 当gl2_thread->domain_type < 2时选择xnu_ctrr_dispatch_table，否则选择domain_id_array。

domain_id_struct {&nbsp;table_id_struct {&nbsp;uint64_t&nbsp;entry_point;&nbsp;uint64_t&nbsp;table_permissions;&nbsp;uint64_t&nbsp;value2;&nbsp;}[16];}[16];

domain_id_array[gl2_thread->domain_type][table_id].entry_point = dispatch_entry_point;domain_id_array[gl2_thread->domain_type][table_id].table_permissions = table_permissions;

– 至此sptm初始化完毕。

2 GL Entry分析

2.1 gxf_entry_el1分析

Kernelcache和sptm启动阶段的代码会调用0x201420指令进入Guard Level模式，在sptm的启动阶段将gxf_entry_el1写入GXF_ENTRY_EL1寄存器，通过x16对其传递参数。

除了0x201420指令，在kernelcache代码中，还看到了其他三种指令也能进入GL模式：

com.apple.kernel:__text:FFFFFFF0284F63F0 sptm_txm_call ; CODE XREF: com.apple.kernel:__text:FFFFFFF028266D8C↑pPACIBSPMOV&nbsp;W16, W0MOVK&nbsp;X16, #2,LSL#48MOVK&nbsp;X16, #0,LSL#32&nbsp;;&nbsp;0x20000w0000000MOV&nbsp;X10, X1LDP&nbsp;X0, X1,&nbsp;[X10]LDP&nbsp;X2, X3,&nbsp;[X10,#0x10]LDP&nbsp;X4, X5,&nbsp;[X10,#0x20]LDP&nbsp;X6, X7,&nbsp;[X10,#0x30]DCD&nbsp;0x201420RETABDCD&nbsp;0x201421DCD&nbsp;0x201422DCD&nbsp;0x201423

在对gxf_entry_el1代码的追踪中，发现只有0x201420可以向sptm传递参数，其他三个只能调用特定的几个dispatch函数。

gxf_entry_el1(uint64_t&nbsp;event_metadata){&nbsp; &nbsp;uint64_t&nbsp;esr_gl1 =&nbsp;msr(ESR_GL1) &&nbsp;0x1f; &nbsp; &nbsp; &nbsp; &nbsp;// genter arg&nbsp; &nbsp;if&nbsp;(!(word_FFFFFFF007094018 &&nbsp;0xf)) {&nbsp; &nbsp; &nbsp;if&nbsp;(word_FFFFFFF007094018 != gl2_thread->0x680) {&nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(0xb,&nbsp;0); &nbsp; &nbsp;&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(esr_gl1 ==&nbsp;0) {&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;uint64_t&nbsp;domain_id = (metadata >>&nbsp;48) &&nbsp;0xffff;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(domain_id ==&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(2, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(domain_id ==&nbsp;3)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(4, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(domain_id ==&nbsp;2)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(3, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;else&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;genter_dispatch_entry_error();&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(esr_gl1 ==&nbsp;1) {&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(0xa,&nbsp;0);&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp;genter_panic();&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp;}&nbsp;}&nbsp;else&nbsp;{&nbsp; &nbsp;if&nbsp;(esr_gl1 ==&nbsp;0) {&nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp;&nbsp; &nbsp; &nbsp; &nbsp;uint64_t&nbsp;domain_id = (metadata >>&nbsp;48) &&nbsp;0xffff;&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(domain_id ==&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(2, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(domain_id ==&nbsp;3)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(4, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp;&nbsp;else&nbsp;if&nbsp;(domain_id ==&nbsp;2)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(3, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp;else&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;genter_dispatch_entry_error();&nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(esr_gl1 ==&nbsp;1) {&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(8,&nbsp;0);&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(esr_gl1 ==&nbsp;2) {&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(3,&nbsp;0x200000000000);&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(esr_gl1 ==&nbsp;3) {&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->unkown_flag !=&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(4,&nbsp;0x3000000000000);&nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp;}}

0x201420可以触发如下调用路径，可以传递参数：

dispatch_state_machine(2, event_metadata);dispatch_state_machine(4, event_metadata);dispatch_state_machine(3, event_metadata);

0x201421可以触发如下调用路径，不能传递参数：

dispatch_state_machine(0xa,&nbsp;0);dispatch_state_machine(8,&nbsp;0);

0x201422可以触发如下调用路径，不能传递参数：

dispatch_state_machine(3, 0x200000000000);

0x201423可以触发如下调用路径，不能传递参数：

dispatch_state_machine(4, 0x3000000000000);

2.2 dispatch_state_machine分析

dispatch_state_machine根据sptm的状态数据库sptm_state_array进行不同的操作。

sptm_state_array是一个二维数组：

sptm_state_struct {&nbsp; dispatch_table_struct {&nbsp; &nbsp; char next_state;&nbsp; &nbsp; void *func_addr;&nbsp; &nbsp; char domain_type;&nbsp; &nbsp; +0x18:&nbsp; &nbsp; &nbsp; uint64_t unknown_value;&nbsp; &nbsp; +0x20:&nbsp; }[12];&nbsp; +0x180:}[16];

通过gl2_thread->sptm_state定位一维数组索引，通过第一个参数event_type定位二维数组索引。

Func_addr为不同sptm state状态的驱动函数。

在对sptm_state_array这个庞大的状态数组进行分析后，得到了所有的状态驱动函数：

(gdb) p/x 0x598E0+0x180$3&nbsp;= 0x59a60(gdb) p/x 0x5D548+0x4000$4&nbsp;= 0x61548(gdb) p/x 0x5C370+0x4000$5&nbsp;= 0x60370(gdb) p/x 0x5D04C+0x4000$6&nbsp;= 0x6104c(gdb) p/x 0x5D098+0x4000$7&nbsp;= 0x61098(gdb) p/x 0x5CCFC+0x4000$8&nbsp;= 0x60cfc(gdb) p/x 0x5CD8C+0x4000$9&nbsp;= 0x60d8c(gdb) p/x 0x5C0E4+0x4000$10&nbsp;= 0x600e4(gdb) p/x 0x5D57C+0x4000$11&nbsp;= 0x6157c(gdb) p/x 0x5C47C+0x4000$12&nbsp;= 0x6047c(gdb) p/x 0x5C294+0x4000$13&nbsp;= 0x60294(gdb) p/x 0x5D44C+0x4000$14&nbsp;= 0x6144c(gdb) p/x 0x85DA0+0x4000$15&nbsp;= 0x89da0(gdb) p/x 0x5CEE0+0x4000$16&nbsp;= 0x60ee0(gdb) p/x 0x5C8E0+0x4000$17&nbsp;= 0x608e0(gdb) p/x 0x5D32C+0x4000$18&nbsp;= 0x6132c(gdb) p/x 0x5D364+0x4000$19&nbsp;= 0x61364

对其重新命名：

接着对参数event_metadata进行提取domain_type和table_id，event_metadata的48-53位保存domain_type， 32-39保存table_id。

然后测试提取出的domain_type有没有这个table_id的对应权限。

最终提取的是table_id_struct里的entry_point函数，接着调用sptm state的驱动函数，将这个entry_point作为回调参数：

Entry_point在sptm启动阶段通过sptm_register_dispatch_table进行了注册：

IOMMU子系统也进行了对应的注册：

在iommu_register_dispatch_table函数内部也是将sptm_dispatch作为entry_point进行注册：

在分析Sptm_dispatch之前，还需要弄清sptm state驱动函数在何时调用sptm_dispatch。

Sptm在不同的状态将会调用不同的驱动函数，sptm_state_func1函数只会将elr_gl1设置为elr_gl1_cl4，返回gl0运行，不会调用sptm_dispatch。sptm_state_func2函数同上，在gl0运行elr_gl1_txm。而sptm_state_func3函数才会在gl2调用sptm_dispatch:

X1就是驱动函数的第2个参数，即sptm_dispatch。

而sptm_state_func8比较有意思，它将sptm_dispatch放到了gl0运行：

完整的dispatch_state_machine函数如下：

void&nbsp;dispatch_state_machine(char&nbsp;event_type,&nbsp;uint64_t&nbsp;event_metadata){&nbsp; &nbsp;check_GXF_STATUS_EL1();&nbsp; &nbsp;if&nbsp;(gl2_thread->stpm_state >&nbsp;0x11)&nbsp; &nbsp; &nbsp;invalid_state_branch();&nbsp; &nbsp;if&nbsp;(event_type >&nbsp;0xc)&nbsp; &nbsp; &nbsp;invalid_event_type_branch();&nbsp;&nbsp; &nbsp; &nbsp;struct&nbsp;sptm_state_struct&nbsp;*sptm_state = sptm_state_array[gl2_thread->sptm_state *&nbsp;0x180];&nbsp; &nbsp; &nbsp;struct&nbsp;dispatch_table_struct&nbsp;*target_elem = sptm_state_array[gl2_thread->sptm_state *&nbsp;0x180][event_type *&nbsp;32];&nbsp; &nbsp; &nbsp;if&nbsp;(target_elem->next_state >&nbsp;0x11)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;invalid_next_state_branch();&nbsp; &nbsp; &nbsp;if&nbsp;(!target_elem->func_addr)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;invalid_state_transition_branch();&nbsp; &nbsp; &nbsp;gl2_thread->domain_type = target_elem->domain_type;&nbsp; &nbsp; &nbsp;if&nbsp;(!(target_elem->unknown_value &&nbsp;0x1)) {&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(!(target_elem->unknown_value &&nbsp;0x2)) {&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;gl2_thread->sptm_state = target_elem->next_state;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;target_elem->func_addr(event_metadata,&nbsp;NULL);&nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->hop_flag !=&nbsp;1)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;invalid_hop_detected_branch();&nbsp;&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_err_log();&nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp;}&nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp;if&nbsp;(event_metadata &&nbsp;0xFC000000000000)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;domain_id_branch();&nbsp; &nbsp; &nbsp;if&nbsp;(event_metadata &&nbsp;0xF000000000)&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;table_id_branch(); &nbsp;&nbsp;&nbsp; &nbsp; &nbsp;&nbsp;uint64_t&nbsp;domain_id_index = (event_metadata >>&nbsp;48) &&nbsp;0xff&nbsp;-&nbsp;1;&nbsp; &nbsp; &nbsp;&nbsp;uint64_t&nbsp;table_id_index = (event_metadata >>&nbsp;32) &&nbsp;0xff&nbsp; &nbsp; &nbsp;if&nbsp;(domain_id >&nbsp;1) {&nbsp; &nbsp; &nbsp; &nbsp;struct&nbsp;domain_id_struct *domain_id = domain_id_array[domain_id_index];&nbsp; &nbsp; &nbsp; &nbsp;struct&nbsp;table_id_struct&nbsp;*table_id = domain_id_array[domain_id_index][table_id_index *&nbsp;0x18];&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp;struct&nbsp;domain_id_struct *domain_id = domain_id_array[domain_id_index];&nbsp; &nbsp; &nbsp; &nbsp;struct&nbsp;table_id_struct&nbsp;*table_id = domain_id_array[domain_id_index][table_id_index *&nbsp;0x18];&nbsp; &nbsp;}&nbsp; &nbsp;if&nbsp;(!(table_id->table_permissions >> gl2_thread->caller_domain) &&nbsp;0x1))&nbsp; &nbsp; &nbsp;table_permissions_branch();&nbsp; &nbsp; &nbsp;if&nbsp;(table_id->entry_point ==&nbsp;0)&nbsp; &nbsp; &nbsp; &nbsp;illegal_dispatch_branch();&nbsp; &nbsp; &nbsp;if&nbsp;(!(target_elem->unknown_value &&nbsp;0x2)) {&nbsp; &nbsp; &nbsp; &nbsp;gl2_thread->sptm_state = target_elem->next_state;&nbsp; &nbsp; &nbsp; &nbsp;target_elem->func_addr(event_metadata, table_id->entry_point);&nbsp; &nbsp;}&nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp;if&nbsp;(gl2_thread->hop_flag !=&nbsp;1)&nbsp; &nbsp; &nbsp; &nbsp;invalid_hop_detected_branch();&nbsp;&nbsp; &nbsp; &nbsp; &nbsp;dispatch_err_log();&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp;}}

2.3 sptm_dispatch

sptm注册了7个子系统，每个子系统都有一个dispatch table。

| |
| — |
| |

| | | |
| — | — | — |
| 编号 | 子系统名称 | Dispatch table函数表数目 |
| 0 | Xnu | 22 |
| 1 | txm | 5 |
| 2 | unknown | 3 |
| 3 | dart | 17 |
| 4 | unknown | 2 |
| 5 | sart | 3 |
| 6 | uat | 13 |

函数原型：sptm_dispatch(uint64_t event_metadata)

event_metadata的32-39位保存的是dispatch table表中的索引。

使用以下脚本将table表中的值转为对应的函数地址：

awk&nbsp;'{print $3}' c.txt|sed -e 's/8[0-9][0-9]/0/' |awk '{a=strtonum($1)+0x4000+0x7000000; printf("%d 0xfffffff0%08x\n", NR, a)}'

| |
| — |
| |

| |
| — |
| |

| | | |
| — | — | — |
| txm | 函数地址 | 函数名称 |
| 1 | 0xfffffff0070751cc | |
| 2 | 0xfffffff007089e60 | sptm_register_dispatch_table |
| 3 | 0xfffffff00708a2ec | sptm_retype |
| 4 | 0xfffffff00708ee28 | |

| |
| — |
| |

| | | |
| — | — | — |
| Unknown | 函数地址 | 函数名称 |
| 1 | 0xfffffff007089e60 | sptm_register_dispatch_table |
| 2 | 0xfffffff00708a2ec | sptm_retype |
| 3 | 0xfffffff0070878f8 | |

| |
| — |
| |

| | | |
| — | — | — |
| dart | 函数地址 | 函数名称 |
| 1 | 0xfffffff00707dee0 | |
| 2 | 0xfffffff00707d498 | |
| 3 | 0xfffffff00707cdec | |
| 4 | 0xfffffff00707c1bc | |
| 5 | 0xfffffff00707bdf4 | |
| 6 | 0xfffffff00707aeb4 | |
| 7 | 0xfffffff007079e0c | |
| 8 | 0xfffffff007079b54 | |
| 9 | 0xfffffff0070798d8 | |
| 10 | 0xfffffff00707962c | |
| 11 | 0xfffffff0070794f4 | |
| 12 | 0xfffffff007079358 | |
| 13 | 0xfffffff007079068 | |
| 14 | 0xfffffff007078e8c | |
| 15 | 0xfffffff007078ca8 | |
| 16 | 0xfffffff007078a60 | |
| 17 | 0xfffffff007078044 | |

| |
| — |
| |

| | | |
| — | — | — |
| Unknown | 函数地址 | 函数名称 |
| 1 | 0xfffffff00707e928 | |
| 2 | 0xfffffff00707e800 | |

| |
| — |
| |

| | | |
| — | — | — |
| sart | 函数地址 | 函数名称 |
| 1 | 0xfffffff00707791c | |
| 2 | 0xfffffff0070774dc | |
| 3 | 0xfffffff007077000 | |

| |
| — |
| |

| | | |
| — | — | — |
| nvme | 函数地址 | 函数名称 |
| 1 | 0xfffffff007076078 | |
| 2 | 0xfffffff0070759e4 | |
| 3 | 0xfffffff0070756fc | |
| 4 | 0xfffffff0070754c0 | |
| 5 | 0xfffffff007075000 | |

| |
| — |
| |

| | | |
| — | — | — |
| uat | 函数地址 | 函数名称 |
| 1 | 0xfffffff0070718ec | |
| 2 | 0xfffffff0070716e8 | |
| 3 | 0xfffffff00707135c | |
| 4 | 0xfffffff0070710c8 | |
| 5 | 0xfffffff007070fdc | |
| 6 | 0xfffffff007070978 | |
| 7 | 0xfffffff0070707c0 | |
| 8 | 0xfffffff0070704d4 | |
| 9 | 0xfffffff007070190 | |
| 10 | 0xfffffff00706f628 | |
| 11 | 0xfffffff00706f340 | |
| 12 | 0xfffffff00706eb14 | |
| 13 | 0xfffffff00706e97c | |

不同的switch case最后统一调用sptm_dispatch_entry_point函数：

函数原型：sptm_dispatch_entry_point(uint64_t table_entry_point)

首先从gl2_thread中恢复sptm_dispatch需要的各种参数，这些参数在gxf_entry_el1开始处进行了保存。

然后就可以调用具体的entry_point了。

3 Exception处理分析

sptm在启动阶段，将syn_handler_sp0_g12地址设置为vbar_gl1的异常处理地址。

以syn_handler_sp0_g12+0x400的异常处理函数为例，追踪sptm如何处理来自gl0的异常，逻辑还原如下：

syn_handler_low_gl2(uint64_t&nbsp;event_metadata){&nbsp; &nbsp;uint64_t&nbsp;ec_value = (msr(ESR_GL1) >>&nbsp;26) &&nbsp;3f;&nbsp; &nbsp;uint64_t&nbsp;svc_number =&nbsp;msr(ESR_GL1) &&nbsp;0xffff;&nbsp; &nbsp;if&nbsp;(ec_value ==&nbsp;0x15) {&nbsp; &nbsp; &nbsp;if&nbsp;(svc_number ==&nbsp;0) {&nbsp; &nbsp; &nbsp; &nbsp;uint64_t&nbsp;value = (event_metadata >>&nbsp;32) &&nbsp;0xffff;&nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(value !=&nbsp;0xfd) {&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(value !=&nbsp;0xfe) {&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;if&nbsp;(value !=&nbsp;0xff) {&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(2, event_metadata);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(7,&nbsp;1);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(0xb,&nbsp;0);&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp; &nbsp; }&nbsp; &nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp; &nbsp;dispatch_state_machine(5,&nbsp;0);&nbsp; &nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(svc_number ==&nbsp;37) {&nbsp; &nbsp; &nbsp; &nbsp;&nbsp;uint64_t&nbsp;value =&nbsp;mrs(SPSR_GL1);&nbsp; &nbsp; &nbsp; &nbsp;msr(value & ~0x1C0);&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp; &nbsp;else&nbsp;if&nbsp;(svc_number ==&nbsp;38) {&nbsp; &nbsp; &nbsp; &nbsp;&nbsp;uint64_t&nbsp;value =&nbsp;mrs(SPSR_GL1);&nbsp; &nbsp; &nbsp; &nbsp;&nbsp;msr(value |&nbsp;0x1C0);&nbsp; &nbsp; &nbsp; }&nbsp; &nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp; &nbsp;dead_loop();&nbsp; &nbsp; &nbsp;}&nbsp; &nbsp;}&nbsp; &nbsp;else&nbsp;if&nbsp;(ec_value ==&nbsp;0x16) {&nbsp; &nbsp;}&nbsp; &nbsp;else&nbsp;{&nbsp; &nbsp; &nbsp;handle_unkdown_exception_gl12();&nbsp; &nbsp;}}

syn_handler_low_gl2提供了3个系统调用：

Svc&nbsp;0Svc&nbsp;37Svc&nbsp;38

后两个系统调用用于关闭中断。

Svc 0有如下触发路径：

dispatch_state_machine(2, event_metadata);dispatch_state_machine(7,&nbsp;1);dispatch_state_machine(0xb,&nbsp;0);dispatch_state_machine(5,&nbsp;0);

同时可以看到，异常处理中并没有对关键寄存器进行保护。

4 sptm页表保护

sptm是secure page table monitor的缩写，ios将原本xnu内核中的页表映射代码转移到了sptm，也就是gl2模式进行处理。

页表映射的函数原型：sptm_map_page(uint64_t paddr, uint64_t vaddr, uint64_t new_pte)

首先调用acquire_root_pt获取root页目录表的地址。

接着调用check_vaddr_range检查虚拟地址vaddr是否在root页目标表所能翻译的地址范围内，注意每层目录表都有一个能翻译的地址范围。

接着检查new_pte合法性。

然后通过root_pt对应的attr_idx，在attr_struct_array中找到对应的结构体，判断物理地址paddr是否合法。

接着获取中间目录表和最后的页表结构体。

重点来了，sptm_type_params_database数组中隐藏着巨大秘密。

首先通过页表page_fte->type在sptm_type_params_database数组中找到对应的struct current_type_params结构体，它保存着此种类型页表对应的sprr_index，然后在new_pte中提取出新的页表对应的sprr_index，将两者对比，如果不相等则中止此次页表映射。也就是说sptm维护了一组sprr_index数据库，只有类型匹配才能有权限进行映射。

5 sptm retype分类保护

Sptm将不同的数据段或代码段进行分类保护，每种不同的类型都有对应的地址范围，不同的类型不能互相映射，用分类概念完成了地址隔离保护。

在sptm启动阶段，调用sptm_bootstrap_early对若干地址区域进行了分类注册：

对papt_range_array中的所有地址范围进行注册， retype类型为0x9。这些地址范围包括：SPTM-rx、SPTM-rw、SPTM-le、TXM-ro、TXM-rx、TXM-bx、TXM-rw、BootKC-ro、BootKC-rs、BootKC-rx、BootKC-bx、BootKC-rw、BootKC-le、TrustCache、CL4-ro、CL4-rx、CL4-rw、CL4-le、RAMDisk、RTBuddySeg、SEPFW、SEPPatches、uStuff、preoslog、BootArgs、ExclaveOSIntegrityCatalog、ExclaveOSTrustCache、AVAILABLE、PHYS_SLIDE等。

还有其他的地址范围注册了以下retype类型：0xb、0x13、0x36。

__text:FFFFFFF0070866A8 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; X0, #0xFFFFFFF000000000__text:FFFFFFF0070866AC MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W1, #1__text:FFFFFFF0070866B0 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W2, #0__text:FFFFFFF0070866B4 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W3, #0x13__text:FFFFFFF0070866B8 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W4, #3__text:FFFFFFF0070866BC MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W5, #1__text:FFFFFFF0070866C0 BL &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;retype_handlerretype_handler(0xFFFFFFF000000000,1,0,0x13,3,1)__text:FFFFFFF0070866D8 ADD &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; X22, X22, #4,LSL#12__text:FFFFFFF0070866DC MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; X0, X22__text:FFFFFFF0070866E0 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W1, #2__text:FFFFFFF0070866E4 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W2, #0__text:FFFFFFF0070866E8 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W3, #0x13__text:FFFFFFF0070866EC MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W4, #3__text:FFFFFFF0070866F0 MOV &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; W5, #1__text:FFFFFFF0070866F4 BL &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;retype_handler

retype_handler(x22,2,0,0×13,3,1)

retype_handler原型：

retype_handler(uint64_t&nbsp;base,&nbsp;uint64_t&nbsp;length, x2,&nbsp;uint64_t&nbsp;frame_type,&nbsp;uint64_t&nbsp;max_num,&nbsp;bool&nbsp;flag)；

首先根据参数base在retype_range_array数组中找到其对应的retype_array数组地址。

接着调用bootstrap_alloc_frames分配一个retype帧结构体。

然后调用bootstrap_retype_frames将刚分配到的retype帧结构进行下一步操作。

bootstrap_retype_frames函数原型：

bootstrap_retype_frames(uint64_t&nbsp;frame_start_address,&nbsp;uint64_t&nbsp;frame_num,&nbsp;uint64_t&nbsp;old_type,&nbsp;uint64_t&nbsp;new_type)；

bootstrap_retype_frames调用set_retype_by_sprr完成类型编码的存储操作。

set_retype_by_sprr原型：

set_retype_by_sprr(uint64_t&nbsp;*retype_address,&nbsp;uint64_t&nbsp;sprr_index)

最终看到是将sprr_index编码进了retype_array数组。

6 txm

txm为trust execution monitor的缩写，运行在gl0，通过svc请求sptm服务：

__bootcode:FFFFFFF017054084 loc_FFFFFFF017054084 ; CODE XREF: sub_FFFFFFF017021C44+291BC↑j__bootcode:FFFFFFF017054084 ; sub_FFFFFFF01704AE04+10↑j ...__bootcode:FFFFFFF017054084 PACIBSP__bootcode:FFFFFFF017054088 SVC &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 0__bootcode:FFFFFFF01705408C RETAB__bootcode:FFFFFFF01705408C ; END OF FUNCTION CHUNK FOR sub_FFFFFFF017021C44__bootcode:FFFFFFF017054090 sub_FFFFFFF017054090 ; CODE XREF: start+48↑p__bootcode:FFFFFFF017054090__bootcode:FFFFFFF017054090 ; FUNCTION CHUNK AT __bootcode:FFFFFFF017054084 SIZE 0000000C BYTES__bootcode:FFFFFFF017054090__bootcode:FFFFFFF017054090 MOVK &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;X16, #0,LSL#48__bootcode:FFFFFFF017054094 MOVK &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;X16, #1,LSL#32__bootcode:FFFFFFF017054098 MOVK &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;X16, #0,LSL#16__bootcode:FFFFFFF01705409C MOVK &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;X16, #1__bootcode:FFFFFFF0170540A0 B &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; loc_FFFFFFF017054084

免责声明：

本文所载程序、技术方法仅面向合法合规的安全研究与教学场景，旨在提升网络安全防护能力，具有明确的技术研究属性。

任何单位或个人未经授权，将本文内容用于攻击、破坏等非法用途的，由此引发的全部法律责任、民事赔偿及连带责任，均由行为人独立承担，本站不承担任何连带责任。

本站内容均为技术交流与知识分享目的发布，若存在版权侵权或其他异议，请通过邮件联系处理，具体联系方式可点击页面上方的联系我。

本文转载自：kernsec wzt《IOS SPTM深度分析》

Tags: 二进制安全漏洞分析移动安全逆向分析

你可能错过

10月17日专题会议：数据安全防护与治理前沿学术会议

【漏洞通告】Cisco ISE远程代码执行漏洞（CVE-2026-20211)

2026年人工智能大模型安全众测结果发布

【漏洞通告】Cisco ISE远程代码执行漏洞（CVE-2026-20211)