Re: [PACTH v3 3/5] tpm_spapr: Support suspend and resume

[Stefan Berger <stefanb@linux.ibm.com>]

On 12/12/19 6:00 AM, Marc-André Lureau wrote:
> Hi
>
> On Wed, Dec 11, 2019 at 8:27 PM Stefan Berger
> <stefanb@linux.vnet.ibm.com> wrote:
>> Extend the tpm_spapr frontend with VM suspend and resume support.
>>
>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
>> ---
>>   hw/tpm/tpm_spapr.c | 42 +++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 41 insertions(+), 1 deletion(-)
>>
>> diff --git a/hw/tpm/tpm_spapr.c b/hw/tpm/tpm_spapr.c
>> index c4a67e2403..d9153bd95c 100644
>> --- a/hw/tpm/tpm_spapr.c
>> +++ b/hw/tpm/tpm_spapr.c
>> @@ -87,6 +87,8 @@ typedef struct {
>>       TPMVersion be_tpm_version;
>>
>>       size_t be_buffer_size;
>> +
>> +    bool deliver_response; /* whether to deliver response after VM resume */
>>   } SPAPRvTPMState;
>>
>>   static void tpm_spapr_show_buffer(const unsigned char *buffer,
>> @@ -339,9 +341,47 @@ static enum TPMVersion tpm_spapr_get_version(TPMIf *ti)
>>       return tpm_backend_get_tpm_version(s->be_driver);
>>   }
>>
>> +/* persistent state handling */
>> +
>> +static int tpm_spapr_pre_save(void *opaque)
>> +{
>> +    SPAPRvTPMState *s = opaque;
>> +
>> +    s->deliver_response = tpm_backend_finish_sync(s->be_driver);
>> +    /*
>> +     * we cannot deliver the results to the VM since DMA would touch VM memory
>> +     */
>> +
>> +    return 0;
>> +}
>> +
>> +static int tpm_spapr_post_load(void *opaque, int version_id)
>> +{
>> +    SPAPRvTPMState *s = opaque;
>> +
>> +    if (s->deliver_response) {
>> +        /* deliver the results to the VM via DMA */
>> +        tpm_spapr_request_completed(TPM_IF(s), 0);
> Why isn't it enough to rely on tpm_spapr_request_completed callback
> being called during pre-save when tpm_backend_finish_sync() is called?
> (like tis & crb)


When .pre_save is called the VM memory has been fully replicated and 
only the devices need to save their state, right? So TIS and CRB save 
the response in memory of the device for the OS driver to pick up after 
resume. The SPAPR device model is expected to write the response into VM 
memory using DMA but memory won't be marked dirty anymore and replicated 
(afaik). So we may have the mechanism of having 
tpm_spapr_request_completed() invoked but in addition we need to 
re-deliver a response after resume so that the OS driver reads the 
proper response then. I'll investigate, though...