Re: [PATCH v1] misc: fastrpc: Pass proper scm arguments for static process init

[Ekansh Gupta <quic_ekangupt@quicinc.com>]

On 6/12/2023 4:40 PM, Srinivas Kandagatla wrote:
> Thanks Ekansh for the patch.
> 
> 
> On 05/06/2023 14:48, Ekansh Gupta wrote:
>> Memory is allocated for dynamic loading when audio daemon is trying
>> to attach to audioPD on DSP side. This memory is allocated from
>> reserved CMA memory region and needs ownership assignment to
>> new VMID in order to use it from audioPD.
>>
>> In the current implementation, arguments are not correctly passed
>> to the scm call which might result in failure of dynamic loading
>> on audioPD. Added changes to pass correct arguments during daemon
>> attach request.
>>
>> Fixes:     0871561055e6 ("misc: fastrpc: Add support for audiopd")
>> Cc: stable <stable@kernel.org>
>> Tested-by: Ekansh Gupta <quic_ekangupt@quicinc.com>
>> Signed-off-by: Ekansh Gupta <quic_ekangupt@quicinc.com>
>> ---
>>   drivers/misc/fastrpc.c | 31 +++++++++++++++++++++++++------
>>   1 file changed, 25 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
>> index 30d4d04..b7335dd 100644
>> --- a/drivers/misc/fastrpc.c
>> +++ b/drivers/misc/fastrpc.c
>> @@ -1278,10 +1278,23 @@ static int 
>> fastrpc_init_create_static_process(struct fastrpc_user *fl,
>>           /* Map if we have any heap VMIDs associated with this ADSP 
>> Static Process. */
>>           if (fl->cctx->vmcount) {
>> +            u64 src_perms = BIT(QCOM_SCM_VMID_HLOS);
> 
> This is redundant too, we already have cctx->perms initialized to 
> exactly same bit map.
> 
>> +            struct qcom_scm_vmperm *dst_perms;
>> +            u32 i;
>> +
>> +            dst_perms = kcalloc(fl->cctx->vmcount,
>> +                            sizeof(struct qcom_scm_vmperm), GFP_KERNEL);
>> +            if (!dst_perms)
>> +                return -ENOMEM;
>> +            for (i = 0; i < fl->cctx->vmcount; i++) {
>> +                dst_perms[i].vmid = fl->cctx->vmperms[i].vmid;
>> +                dst_perms[i].perm = fl->cctx->vmperms[i].perm;
> 
> why do we need to copy this to another struct when we already have this 
> information in fl->cctx->vmperms ?
> 
>> +            }
>> +
>>               err = qcom_scm_assign_mem(fl->cctx->remote_heap->phys,
>>                               (u64)fl->cctx->remote_heap->size,
>> -                            &fl->cctx->perms,
>> -                            fl->cctx->vmperms, fl->cctx->vmcount);
>> +                            &src_perms, dst_perms, fl->cctx->vmcount);
> 
> 
>> +            kfree(dst_perms);
>>               if (err) {
>>                   dev_err(fl->sctx->dev, "Failed to assign memory with 
>> phys 0x%llx size 0x%llx err %d",
>>                       fl->cctx->remote_heap->phys, 
>> fl->cctx->remote_heap->size, err);
>> @@ -1322,13 +1335,19 @@ static int 
>> fastrpc_init_create_static_process(struct fastrpc_user *fl,
>>       return 0;
>>   err_invoke:
>>       if (fl->cctx->vmcount) {
>> -        struct qcom_scm_vmperm perm;
>> +        u64 src_perms = 0;
>> +        struct qcom_scm_vmperm dst_perms;
>> +        u32 i;
>> -        perm.vmid = QCOM_SCM_VMID_HLOS;
>> -        perm.perm = QCOM_SCM_PERM_RWX;
>> +        for (i = 0; i < fl->cctx->vmcount; i++) {
>> +            src_perms |= BIT(fl->cctx->vmperms[i].vmid);
>> +        }
> no need for brackets here.
Thanks Srini for your comments. I'll address them in next patch.

-ekansh
>> +
>> +        dst_perms.vmid = QCOM_SCM_VMID_HLOS;
>> +        dst_perms.perm = QCOM_SCM_PERM_RWX;
>>           err = qcom_scm_assign_mem(fl->cctx->remote_heap->phys,
>>                           (u64)fl->cctx->remote_heap->size,
>> -                        &fl->cctx->perms, &perm, 1);
>> +                        &src_perms, &dst_perms, 1);
>>           if (err)
>>               dev_err(fl->sctx->dev, "Failed to assign memory phys 
>> 0x%llx size 0x%llx err %d",
>>                   fl->cctx->remote_heap->phys, 
>> fl->cctx->remote_heap->size, err);