On 3/9/22 10:59, Cezary Rojewski wrote:
On 2022-03-08 8:25 PM, Pierre-Louis Bossart wrote:
For devices designed for Windows, the SSP information should be listed in the NHLT, and when present can be used to set quirks automatically in the machine driver.
The NHLT information exposes BT and analog audio connections separately, for now we are only interested in the analog audio parts.
The use of dev_info() for the SSP mask is intentional so that we can immediately flag devices with an ES8336 codec. Since NHLT is not used for recent Chromebooks these messages should be rare.
...
+static int check_nhlt_ssp_mask(struct snd_sof_dev *sdev) +{ + struct nhlt_acpi_table *nhlt; + int ssp_mask = 0;
+ nhlt = intel_nhlt_init(sdev->dev); + if (!nhlt) + return ssp_mask;
+ if (intel_nhlt_has_endpoint_type(nhlt, NHLT_LINK_SSP)) { + ssp_mask = intel_nhlt_ssp_endpoint_mask(nhlt, NHLT_DEVICE_I2S); + if (ssp_mask) + dev_info(sdev->dev, "NHLT_DEVICE_I2S detected, ssp_mask %#x\n", ssp_mask); + } + intel_nhlt_free(nhlt);
NHLT "toggling" found in this function looks weird. Why not cache NHLT pointer i.e.: get it once and put when driver is no longer required? Initializing and freeing NHLT (AKA get/put ACPI table) every time a request is made does not look like an optimal solution.
I agree with your remark, but this is an optimization that we plan on doing later. There are other changes coming wrt to NHTL to extract DMIC blobs, and it's better to change all the functions using the same programming flow when we are in 'stable' state.
The concern isn't really optimization at this point but just to get audio to work. Keep in mind all this patchset was generated with tests crowdsourced to the community, and the empirical detection of the SSP-codec link could be broken on some platforms - the NHLT does not give *any* information on where the codec is actually connected.