[PATCH 1/1] Add RISC-V TEE support

liushiwei liushiwei at eswincomputing.com
Sat Jan 28 00:33:34 PST 2023


Hi, Atish 
	Thank you for your reply.
	I submitted my ideas to https://github.com/riscv-non-isa/riscv-sbi-doc/pull/106 as requested by other members of the email. 
	You can preview them at https://github.com/liushiwei007/riscv-sbi-doc/blob/master/riscv-sbi.adoc .
	I'm not sure if I need to send an email to tech-prs at lists.riscv.org or sig-trusted-computing at lists.riscv.org after submitting to riscv-sbi-doc? 
	If so, Is it to send the patch of opensbi I submitted before?

Regards,
Liushiwei


-----邮件原件-----
发件人: Atish Patra [mailto:atishp at atishpatra.org] 
发送时间: 2023年1月25日 3:12
收件人: liushiwei <liushiwei at eswincomputing.com>
抄送: Himanshu Chauhan <hchauhan at ventanamicro.com>; opensbi at lists.infradead.org; chenchaokai at eswincomputing.com
主题: Re: 答复: [PATCH 1/1] Add RISC-V TEE support

On Wed, Jan 11, 2023 at 11:08 PM liushiwei <liushiwei at eswincomputing.com> wrote:
>
> Hi, Himanshu ,  These are my description:
>
> In my design, The entire linux space is called REE (Rich Execution 
> Environment), and TEE OS includes its user state called TEE (Trusted execution environment).
> adding TEE functionality to opensbi requires two configuration items.
> For example:
>    add CONFIG_SBI_ECALL_TEE=y in platform/generic/configs/defconfig file.
>    add CONFIG_TEE_LOAD_ADDR=0x27C000000 platform/generic/objects.mk.
>    The value of CONFIG_TEE_LOAD_ADDR depends on the actual memory layout,
>    It's a physical address.
>
> When TEE is configured, opensbi adds the following functionality:
> 1. Boot TEE OS.
>    If TEE is enabled, tee_os_init() is called before entering
>    sbi_hart_switch_mode(). tee_os_init saves the current context,
>    sets a new trap stack address, and runs to CONFIG_TEE_LOAD_ADDR as configured
>    to complete TEE OS initialization. TEE OS returns via ecall, Go to opensbi
>    sbi_ecall_tee_handler and use the characteristic value RETURN_ENTRY_DONE
>    to indicate the return after the TEE OS completes booting. Check whether the
>    TEE OS boot successfully according to the parameter. If fails,
>    the system enters wfi and terminates the startup process of opensbi. If successful,
>    it returns REE to switch into TEE's vector. then trap stack memory is restored,
>    the context is restored, the tee_os_init call is returned,
>    and the rest of the process is performed.
> 2. REE switches to TEE.
>    When the TEE OS boot successfully, it returns an entry vector for REE into the TEE.
>    It represents various entry points into TEE OS and is stored in opensbi's global variables.
>    It is the optee_vectors_t structure, which contains nine entry cases:
>       1. yield_smc_entry;
>       2. fast_smc_entry;
>       3. cpu_on_entry;
>       4. cpu_off_entry;
>       5. cpu_resume_entry;
>       6. cpu_suspend_entry;
>       7. fiq_entry;
>       8. system_off_entry;
>       9. system_reset_entry;
>    yield_smc_entry means that this function entry TEE will start the thread function
>    and enter the user state of TEE. It may also switch back to REE with RPC function,
>    and then return to TEE after REE completes the corresponding function.
>    For the REE process that sent you this call, it may cause sleep.
>
>    fast_smc_entry indicates that this is a quick function that returns after
>    the TEE OS does something, and that it does not cause the caller to sleep.
>
>    yield_smc_entry and fast_smc_entry return opensbi use eigenvalue RETURN_CALL_DONE.
>    for yield_smc_entry, whether the call returns or the RPC returns is decided by linux.
>
>    cpu_on_entry/cpu_off_entry/cpu_resume_entry/cpu_suspend_entry and
>    fiq_entry/system_off_entry/system_reset_entry they are not implemented currently.
>
> 3. TEE switches to REE.
>    opensbi needs to save the context when REE enters the TEE,
>    and restore the context when it returns from the TEE.
>
> 4. TEE/REE Request a special function.
>    We have some specific functions, like get hartid from TEE,
>    it need save and restore the TEE context.
>
> Data structure.
> opensbi adds the sbi_save_context declaration
>    struct sbi_save_context {
>       struct sbi_trap_regs regs;
>       unsigned long sepc;
>       unsigned long satp;
>       unsigned long sstatus;
>       unsigned long sie;
>       unsigned long stvec;
>       unsigned long sscratch;
>       unsigned long scounteren;
>       unsigned long scause;
>       unsigned long stval;
>       unsigned long sip;
>    };
>    sbi_save_context include sbi_trap_regs and S mode csr. Used to hold 
> the context of TEE or REE
>
>    typedef struct optee_vectors {
>       optee_vector_isn_t yield_smc_entry;
>       optee_vector_isn_t fast_smc_entry;
>       optee_vector_isn_t cpu_on_entry;
>       optee_vector_isn_t cpu_off_entry;
>       optee_vector_isn_t cpu_resume_entry;
>       optee_vector_isn_t cpu_suspend_entry;
>       optee_vector_isn_t fiq_entry;
>       optee_vector_isn_t system_off_entry;
>       optee_vector_isn_t system_reset_entry;
>    } optee_vectors_t;
>    optee_vectors represent the various entry points into TEE OS.
>
> Variables defined by opensbi
>    optee_vectors_t *optee_vector_table;
>    struct sbi_save_context nsec_cpu_context[OPTEED_CORE_COUNT];
>    struct sbi_save_context sec_cpu_context[OPTEED_CORE_COUNT];
>    struct sbi_trap_regs cpu_start_context[OPTEED_CORE_COUNT];
>    typedef ulong tee_tmp_trap_stack[1024];
>    static tee_tmp_trap_stack tmp_stack[OPTEED_CORE_COUNT];
>
>    optee_vector_table value is assigned after the TEE OS boot succeeds.
>    nsec_cpu_context and sec_cpu_context is to save the context of TEE and REE,
>    cpu_start_context save the context of opensbi before tee_os_init enter TEE OS,
>    tee_tmp_stack is the trap stack when TEE OS return tee_os_init.
>
> Under the current design, REE does not enable interrupts when entering 
> TEE, and the entire TEE, including opensbi, is the process context of linux.
> TEE processing must be brief and quick.
>

Great. Thanks. Your PR should include more details like this with proper FID allocation and description.
In order to make it a standard RISC-V SBI extension, you need to present this in tech-prs mailing list
(https://lists.riscv.org/g/tech-prs)
and get it approved within the group. Here is the policy document[1]

https://docs.google.com/document/d/1bQGHU-wD4uN4mU07oH9adpXbH51NqUbFxcL8qTTIqhA/edit

I think it would be good if you can present your work in trusted computing SIG (https://lists.riscv.org/g/sig-trusted-computing) as well.

>
> Regards,
> liushiwei
> -----邮件原件-----
> 发件人: Himanshu Chauhan [mailto:hchauhan at ventanamicro.com]
> 发送时间: 2023年1月11日 23:39
> 收件人: liushiwei <liushiwei at eswincomputing.com>
> 抄送: opensbi at lists.infradead.org; chenchaokai at eswincomputing.com
> 主题: Re: 答复: [PATCH 1/1] Add RISC-V TEE support
>
> On Wed, Jan 11, 2023 at 08:27:59PM +0800, liushiwei wrote:
> > Do you mean hardware? Our hardware design referred to arm's 
> > trustzone technology. optee os is a software solution using arm 
> > trustzone hardware, which mainly includes REE(linux), TEE(optee os), 
> > ATF(ARM Trusted firmware), and then our software also developed 
> > these three parts. opensbi is similar to ATF. whether if this is what you want?
> > The current committed code is not hardware-dependent, but just 
> > continues the idea of this workaround, and we may commit hardware-dependent code later.
> >
> No, I meant the software specification.
>
> > -----邮件原件-----
> > 发件人: hchauhan at ventanamicro.com [mailto:hchauhan at ventanamicro.com]
> > 发送时间: 2023年1月11日 20:03
> > 收件人: 'liushiwei' <liushiwei at eswincomputing.com>; opensbi at lists.infradead.
> > org
> > 抄送: chenchaokai at eswincomputing.com
> > 主题: RE: [PATCH 1/1] Add RISC-V TEE support
> >
> > -----Original Message-----
> > > From: opensbi <opensbi-bounces at lists.infradead.org> On Behalf Of 
> > > liushiwei
> > > Sent: 11 January 2023 07:32
> > > To: opensbi at lists.infradead.org
> > > Cc: chenchaokai at eswincomputing.com; liushiwei
> > <liushiwei at eswincomputing.com>
> > > Subject: [PATCH 1/1] Add RISC-V TEE support
> >
> > >RISC-V Trusted Executable Environment security software includes 
> > >linux,
> > opensbi, and OP-TEE OS. linux is the non-secure domain, and OP-TEE 
> > OS is the secure domain. At boot time, opensbi boots OP->TEE OS and then starts linux.
> > At runtime, opensbi acts as a secure monitor, responsible for 
> > context saving and restoring when switching between linux and OP-TEE OS.
> > >TEE function is off by default, when using configuration is added 
> > >in the
> > config and objects file, such as platform/generic/configs/defconfig
> > add CONFIG_SBI_ECALL_TEE = y, In the >platform/generic/objects.mk 
> > add CONFIG_TEE_LOAD_ADDR = 0x27c000000, CONFIG_TEE_LOAD_ADDR is the 
> > starting address of the OP-TEE OS.
> >
> > Hi Liushiwei,
> >
> > Was there any formal specification or draft for this? Could you 
> > please point me to the draft or specification?
> >
> > Regards
> > Himanshu
> >
> > --
> > opensbi mailing list
> > opensbi at lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/opensbi
> >
>
>
> --
> opensbi mailing list
> opensbi at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/opensbi



--
Regards,
Atish




More information about the opensbi mailing list