Re: [PATCH v6 10/11] virt: arm-cca-guest: TSM_REPORT support for realms

[Gavin Shan <gshan@redhat.com>]

On 10/15/24 12:46 AM, Suzuki K Poulose wrote:
> On 14/10/2024 15:41, Steven Price wrote:
>> On 14/10/2024 09:56, Suzuki K Poulose wrote:
>>> On 12/10/2024 07:06, Gavin Shan wrote:
>>>> On 10/12/24 2:22 AM, Suzuki K Poulose wrote:
>>>>> On 11/10/2024 15:14, Steven Price wrote:
>>>>>> On 08/10/2024 05:12, Gavin Shan wrote:
>>>>>>> On 10/5/24 12:43 AM, Steven Price wrote:
>>>>>>>> From: Sami Mujawar <sami.mujawar@arm.com>
>>>>>>>>
>>>>>>>> Introduce an arm-cca-guest driver that registers with
>>>>>>>> the configfs-tsm module to provide user interfaces for
>>>>>>>> retrieving an attestation token.
>>>>>>>>
>>>>>>>> When a new report is requested the arm-cca-guest driver
>>>>>>>> invokes the appropriate RSI interfaces to query an
>>>>>>>> attestation token.
>>>>>>>>
>>>>>>>> The steps to retrieve an attestation token are as follows:
>>>>>>>>      1. Mount the configfs filesystem if not already mounted
>>>>>>>>         mount -t configfs none /sys/kernel/config
>>>>>>>>      2. Generate an attestation token
>>>>>>>>         report=/sys/kernel/config/tsm/report/report0
>>>>>>>>         mkdir $report
>>>>>>>>         dd if=/dev/urandom bs=64 count=1 > $report/inblob
>>>>>>>>         hexdump -C $report/outblob
>>>>>>>>         rmdir $report
>>>>>>>>
>>>>>>>> Signed-off-by: Sami Mujawar <sami.mujawar@arm.com>
>>>>>>>> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
>>>>>>>> Signed-off-by: Steven Price <steven.price@arm.com>
>>>>>>>> ---
>>>>>>>> v3: Minor improvements to comments and adapt to the renaming of
>>>>>>>> GRANULE_SIZE to RSI_GRANULE_SIZE.
>>>>>>>> ---
>>>>>>>>     drivers/virt/coco/Kconfig                     |   2 +
>>>>>>>>     drivers/virt/coco/Makefile                    |   1 +
>>>>>>>>     drivers/virt/coco/arm-cca-guest/Kconfig       |  11 +
>>>>>>>>     drivers/virt/coco/arm-cca-guest/Makefile      |   2 +
>>>>>>>>     .../virt/coco/arm-cca-guest/arm-cca-guest.c   | 211
>>>>>>>> ++++++++++++ ++++++
>>>>>>>>     5 files changed, 227 insertions(+)
>>>>>>>>     create mode 100644 drivers/virt/coco/arm-cca-guest/Kconfig
>>>>>>>>     create mode 100644 drivers/virt/coco/arm-cca-guest/Makefile
>>>>>>>>     create mode 100644 drivers/virt/coco/arm-cca-guest/arm-cca-guest.c
>>>>
>>>> [...]
>>>>
>>>>>>>> +/**
>>>>>>>> + * arm_cca_report_new - Generate a new attestation token.
>>>>>>>> + *
>>>>>>>> + * @report: pointer to the TSM report context information.
>>>>>>>> + * @data:  pointer to the context specific data for this module.
>>>>>>>> + *
>>>>>>>> + * Initialise the attestation token generation using the
>>>>>>>> challenge data
>>>>>>>> + * passed in the TSM descriptor. Allocate memory for the attestation
>>>>>>>> token
>>>>>>>> + * and schedule calls to retrieve the attestation token on the
>>>>>>>> same CPU
>>>>>>>> + * on which the attestation token generation was initialised.
>>>>>>>> + *
>>>>>>>> + * The challenge data must be at least 32 bytes and no more than 64
>>>>>>>> bytes. If
>>>>>>>> + * less than 64 bytes are provided it will be zero padded to 64
>>>>>>>> bytes.
>>>>>>>> + *
>>>>>>>> + * Return:
>>>>>>>> + * * %0        - Attestation token generated successfully.
>>>>>>>> + * * %-EINVAL  - A parameter was not valid.
>>>>>>>> + * * %-ENOMEM  - Out of memory.
>>>>>>>> + * * %-EFAULT  - Failed to get IPA for memory page(s).
>>>>>>>> + * * A negative status code as returned by
>>>>>>>> smp_call_function_single().
>>>>>>>> + */
>>>>>>>> +static int arm_cca_report_new(struct tsm_report *report, void *data)
>>>>>>>> +{
>>>>>>>> +    int ret;
>>>>>>>> +    int cpu;
>>>>>>>> +    long max_size;
>>>>>>>> +    unsigned long token_size;
>>>>>>>> +    struct arm_cca_token_info info;
>>>>>>>> +    void *buf;
>>>>>>>> +    u8 *token __free(kvfree) = NULL;
>>>>>>>> +    struct tsm_desc *desc = &report->desc;
>>>>>>>> +
>>>>>>>> +    if (!report)
>>>>>>>> +        return -EINVAL;
>>>>>>>> +
>>>>>>>
>>>>>>> This check seems unnecessary and can be dropped.
>>>>>>
>>>>>> Ack
>>>>>>
>>>>>>>> +    if (desc->inblob_len < 32 || desc->inblob_len > 64)
>>>>>>>> +        return -EINVAL;
>>>>>>>> +
>>>>>>>> +    /*
>>>>>>>> +     * Get a CPU on which the attestation token generation will be
>>>>>>>> +     * scheduled and initialise the attestation token generation.
>>>>>>>> +     */
>>>>>>>> +    cpu = get_cpu();
>>>>>>>> +    max_size = rsi_attestation_token_init(desc->inblob,
>>>>>>>> desc->inblob_len);
>>>>>>>> +    put_cpu();
>>>>>>>> +
>>>>>>>
>>>>>>> It seems that put_cpu() is called early, meaning the CPU can go
>>>>>>> away before
>>>>>>> the subsequent call to arm_cca_attestation_continue() ?
>>>>>>
>>>>>> Indeed, good spot. I'll move it to the end of the function and update
>>>>>> the error paths below.
>>>>>
>>>>> Actually this was on purpose, not to block the CPU hotplug. The
>>>>> attestation must be completed on the same CPU.
>>>>>
>>>>> We can detect the failure from "smp_call" further down and make sure
>>>>> we can safely complete the operation or restart it.
>>>>>
>>>>
>>>> Yes, It's fine to call put_cpu() early since we're tolerant to error
>>>> introduced
>>>> by CPU unplug. It's a bit confused that rsi_attestation_token_init()
>>>> is called
>>>> on the local CPU while arm_cca_attestation_continue() is called on
>>>> same CPU
>>>> with help of smp_call_function_single(). Does it make sense to unify
>>>> so that
>>>> both will be invoked with the help of smp_call_function_single() ?
>>>>
>>>>       int cpu = smp_processor_id();
>>>>
>>>>       /*
>>>>        * The calling and target CPU can be different after the calling
>>>> process
>>>>        * is migrated to another different CPU. It's guaranteed the
>>>> attestatation
>>>>        * always happen on the target CPU with smp_call_function_single().
>>>>        */
>>>>       ret = smp_call_function_single(cpu,
>>>> rsi_attestation_token_init_wrapper,
>>>>                                      (void *)&info, true);
>>>
>>> Well, we want to allocate sufficient size buffer (size returned from
>>> token_init())  outside an atomic context (thus not in smp_call_function()).
>>>
>>> May be we could make this "allocation" restriction in a comment to
>>> make it clear, why we do it this way.
>>
>> So if I've followed this correctly the get_cpu() route doesn't work
>> because of the need to allocate outblob. So using
>> smp_call_function_single() for all calls seems to be the best approach,
>> along with a comment explaining what's going on. So how about:
>>
>>     /*
>>      * The attestation token 'init' and 'continue' calls must be
>>      * performed on the same CPU. smp_call_function_single() is used
>>      * instead of simply calling get_cpu() because of the need to
>>      * allocate outblob based on the returned value from the 'init'
>>      * call and that cannot be done in an atomic context.
>>      */
>>     cpu = smp_processor_id();
>>
>>     info.challenge = desc->inblob;
>>     info.challenge_size = desc->inblob_len;
>>
>>     ret = smp_call_function_single(cpu, arm_cca_attestation_init,
>>                        &info, true);
>>     if (ret)
>>         return ret;
>>     max_size = info.result;
>>
>> (with appropriate updates to the 'info' struct and a new
>> arm_cca_attestation_init() wrapper for rsi_attestation_token_init()).
> 
> That sounds good to me.
> 

+1, it looks good to me as well.

Thanks,
Gavin