Hi Yanyan, According to your advice,I modified the avail_p_regs[] in hardware.h file as tk1. 1 MiB starting from 0x37f00000 is reserved for monitor mode.
{ /* .start = */ 0x10000000, /* .end = */ 0x37f00000} And then #define MON_PA_START (0x10000000 + 0x27f00000) #define MON_PA_SIZE (1 << 20) #define MON_PA_END (MON_PA_START + MON_PA_SIZE) #define MON_PA_STACK (MON_PA_END - 0x10) #define MON_VECTOR_START (MON_PA_START) set sp: asm volatile ("cps %0\n\t" "isb\n" "mov sp, %1\n\t" ::"I"(MONITOR_MODE),"r"(MON_PA_STACK)); copy monitor mode vector and write into MVBAR: uint32_t size = arm_monitor_vector_end - arm_monitor_vector; printf("Copy monitor mode vector from %x to %x size %x \n", (arm_monitor_vector), MON_VECTOR_START, size); memcpy((void *)MON_VECTOR_START, (void *)(arm_monitor_vector), size); asm volatile ("dmb\n isb\n"); asm volatile ("mcr p15, 0, %0, c12, c0, 1"::"r"(MON_VECTOR_START)); I used the arm_monitor_vector provided by the source code in monitor.S and just didn't perform the operation 'blx r0'. I think the it can jump to smc_handler and return successfully. #define VECTOR_BASE 0x37f00000 #define STACK_TOP (VECTOR_BASE + (1 << 20) - 0x10) arm_monitor_vector: ldr pc, [pc, #28] ldr pc, [pc, #24] ldr pc, [pc, #16] ldr pc, [pc, #16] ldr pc, [pc, #12] ldr pc, [pc, #8] ldr pc, [pc, #4] ldr pc, [pc, #0] smc_handler_addr: .word VECTOR_BASE + (smc_handler - arm_monitor_vector)//0x1000002c smc_halt_addr: .word VECTOR_BASE + (smc_halt - arm_monitor_vector) smc_stack: .word STACK_TOP smc_handler: /* always have a valid stack */ ldr sp, [pc, #-12] push {lr} //blx r0 isb mrc p15, 0, r5, c1, c1, 0 /* set the NS bit */ orr r5, r5, #1 mcr p15, 0, r5, c1, c1, 0 pop {lr} isb movs pc, lr However, the problem still exists. And I have another question that why set the sp in monitor mode to sp in SVC mode. asm volatile ("mov r8, sp\n\t" "cps %0\n\t" "isb\n" "mov sp, r8\n\t" ::"I"(MONITOR_MODE)); Thank you, Dongxu Ji <yanyan.s...@data61.csiro.au> 于2018年8月28日周二 下午10:10写道: > Hi Dongxu, > > > As you can see, the arm_monitor_vector uses PC-relative addressing so that > the code can be moved around in memory. I think ldr pc, =smc_handler > breaks this. Also, please set the NS bit in SCR to 1 before returning. > > > To reserve a memory region for the monitor-mode code and data, I suggest > you modify the avail_p_regs[] in > kernel/include/plat/imx6/plat/machine/hardware.h > file. See the kernel/include/plat/tk1/plat/machine/hardware.h as an example. > > > > > Regards, > > Yanyan > > > ------------------------------ > *From:* Dongxu Ji <jidongxu1...@gmail.com> > *Sent:* Wednesday, August 29, 2018 12:02 AM > *To:* devel@sel4.systems; Shen, Yanyan (Data61, Kensington NSW) > *Subject:* Fwd: [seL4] SMC in seL4 > > Hi Yanyan, > 1. It doesn't set the NS bit to 1 in SCR(I just want it to return without > do anything). The arm_monitor_vector and the smc_handler(): > > arm_monitor_vector: > ldr pc, [pc, #28] > ldr pc, [pc, #24] > ldr pc, =smc_handler > ldr pc, [pc, #16] > ldr pc, [pc, #12] > ldr pc, [pc, #8] > ldr pc, [pc, #4] > ldr pc, [pc, #0] > > smc_handler: > movs pc, lr > > 2. I didn't do any extra work other than the boot log: > > .......... > ELF-loader started on CPU: ARM Ltd. Cortex-A9 r2p10 > > paddr=[20000000..203fbfff] > > ELF-loading image 'kernel' > > paddr=[10000000..10026fff] > > vaddr=[e0000000..e0026fff] > > virt_entry=e0000000 > > ELF-loading image 'sel4test-driver' > > paddr=[10027000..10500fff] > > vaddr=[8000..4e1fff] > > virt_entry=25a6c > > Enabling MMU and paging > > Jumping to kernel-image entry point... > > 3. The initialization operations in platform_init.c: > set sp: > #define MONITOR_MODE (0x16) > #define MON_VECTOR_START (0x11000000) > #define VECTOR_BASE 0x11000000 > #define STACK_TOP (VECTOR_BASE + (1 << 12) - 0x10) > > asm volatile ( "mrs r1, cpsr\n\t" > "cps %0\n\t" > "isb\n" > "mov sp, %1\n\t" > "msr cpsr, r1\n\t" > ::"I"(MONITOR_MODE),"r"(STACK_TOP)); > > copy monitor mode vector to MON_VECTOR_START and write into MVBAR: > uint32_t size = arm_monitor_vector_end - arm_monitor_vector; > printf("Copy monitor mode vector from %x to %x size %x\n", > (arm_monitor_vector), MON_VECTOR_START, size); > memcpy((void *)MON_VECTOR_START, (void *)(arm_monitor_vector), size); > asm volatile ("dmb\n isb\n"); > asm volatile ("mcr p15, 0, %0, c12, c0, 1"::"r"(MON_VECTOR_START)); > > I enter into SVC mode by software interrupt and call the function smc(): > asm (".arch_extension sec\n"); > asm volatile ("stmfd sp!, {r3-r11, lr}\n\t" > "dsb\n" > "smc #0\n" > "ldmfd sp!, {r3-r11, pc}"); > > and then the problem arises. > > Thank you, > Dongxu Ji > > > <yanyan.s...@data61.csiro.au> 于2018年8月28日周二 下午8:30写道: > >> Hi, >> >> The smc_handle() in monitor.S, it does nothing but "movs pc, lr". >> >> Does it set the NS bit to 1 in SCR? >> >> Also, what did you do to ensure that 0x11000000 is not used by the kernel? >> >> Could you share the code (if possible) so that I could better understand >> the problem. >> >> Regards, >> Yanyan >> >> >> ------------------------------ >> *From:* Devel <devel-bounces@sel4.systems> on behalf of 冀东旭 < >> jidongxu1...@gmail.com> >> *Sent:* Tuesday, August 28, 2018 1:02 PM >> *To:* devel@sel4.systems >> *Subject:* [seL4] SMC in seL4 >> >> Hello, >> >> I'm porting sel4 to imx6q sabrelite as the trusted OS in trustzone. I >> initialize the monitor mode by setting the sp to STACK_TOP and copying >> arm_monitor_vector to MON_VECTOR_START according to the functions >> "install_monitor_hook()" and "switch_to_mon_mode()" in "platform_init.c". >> >> #define VECTOR_BASE 0x11000000(addr is not used by the seL4 kernel) >> >> #define STACK_TOP (VECTOR_BASE + (1 << 12) - 0x10) >> >> #define MON_VECTOR_START 0x11000000(The VECTOR_BASE is the same as >> MON_VECTOR_START) >> >> The smc_handle() in monitor.S, it does nothing but "movs pc, lr". After >> calling smc in SVC mode, it hangs without any log. If I comment the "smc >> #0", it can return the caller successfully in usr mode. >> >> stmfd sp!, {r3-r11, lr} >> dsb >> smc #0 >> ldmfd sp!, {r3-r11, pc} >> >> Is the sp in monitor mode appropriate? Or I need to do something else in >> initialization operations? What's wrong with it? Do you have any ideas? >> >> Thank you! >> >> Dongxu Ji >> >>
_______________________________________________ Devel mailing list Devel@sel4.systems https://sel4.systems/lists/listinfo/devel