On 8/8/19 11:52 AM, Cezary Rojewski wrote:
On 2019-08-08 16:50, Pierre-Louis Bossart wrote:
Thanks Cezary, the split makes it much easier to review. I have a couple of minor comments below, looks good otherwise.
On 8/8/19 5:24 AM, Cezary Rojewski wrote:
LARGE_CONFIG_GET is mainly used to retrieve requested module parameters but it may also carry TX payload with them. Update its implementation to account for both TX and RX data. First reply.header carries total payload size within data_off_sizefield. Make use of reply.header to realloc returned buffer with correct size.
Failure of IPC request is permissive - error-payload may be returned, an informative data why GET for given param failed - and thus function should not collapse before entire processing is finished. Caller is responsible for checking returned payload and bytes parameters.
but that is the same as before, yes? this patch does not change the behavior on errors?
To evaluate this statement you have to take a look at old code (before addition of reply saving):
ret = sst_ipc_tx_message_wait(ipc, *ipc_header, NULL, 0, ((char *)param) + data_offset, msg->param_data_size); if (ret < 0) { dev_err(ipc->dev, "ipc: get large config fail, err: %d\n", ret); return ret; }
Third and four parameter for sst_ipc_tx_message_wait denoted tx_data and tx_size in older version. As you can see, these are empty - no payload is carried, _get_ only awaits reply and copied data retrieved (breaks vendor case as described earlier).
Immediately after failure, function was collapsing. Some data might have been appended to "param" but size is unknown. However, due to lack of retrieval of said size, indeed one may say the behavior on errors did not really change. It's vital to highlight that changes added here (2/2 patch) do not make function return right after receiving error though - buffer and size are updated for the caller before leaving the scope.
ok, got it. The change actually happened in patch1, I couldn't see the change in that specific patch. Thanks for the precisions.