On 1/7/2021 2:51 AM, Peter Geis wrote:
External email: Use caution opening links or attachments
On Tue, Jan 5, 2021 at 1:30 AM Sameer Pujar <spujar@xxxxxxxxxx> wrote:
On 12/25/2020 6:50 AM, Peter Geis wrote:
External email: Use caution opening links or attachments
Currently hda on tegra30 fails to open a stream with an input/output error.
This is similar to the issue referenced in [1].
For example:
speaker-test -Dhw:0,3 -c 2
speaker-test 1.2.2
Playback device is hw:0,3
Stream parameters are 48000Hz, S16_LE, 2 channels
Using 16 octaves of pink noise
Rate set to 48000Hz (requested 48000Hz)
Buffer size range from 64 to 16384
Period size range from 32 to 8192
Using max buffer size 16384
Periods = 4
was set period_size = 4096
was set buffer_size = 16384
0 - Front Left
Write error: -5,Input/output error
xrun_recovery failed: -5,Input/output error
Transfer failed: Input/output error
[1] states "Due to a legacy HW design problem", implying the issue applies to all previous tegra-hda devices.
The tegra-hda device was introduced in tegra30 but only utilized in tegra124 until now.
For this reason it is unknown when this issue first manifested.
Applying the fix in [1] universally resolves this issue on tegra30.
Tested on the Ouya game console and the tf201 tablet.
[1] 60019d8c650d ("ALSA: hda/tegra: workaround playback failure on Tegra194")
This issue was never seen on Tegra210/Tegra186 and hence at that time it
was thought to be specific to Tegra194. I never tested this on Tegra30
since I don't have this device. I will clarify this with HW folks if
workaround is safer for all chips.
So this is confirmed to not affect Tegra210 and Tegra186, but it does
affect Tegra194 and Tegra30.
Is it possible for the hardware team to pitch on on where the fix was
implemented?
I discussed this with HW team. The issue is applicable to all Tegra chips.
Below is the striping formula referenced from HD audio spec.
{ ((num_channels * bits_per_sample) / number of SDOs) >= 8 }
The current issue is seen because Tegra HW has a problem with boundary
condition (= 8) for striping. The reason why it is not seen on
Tegra210/Tegra186 is because it uses max 2SDO lines. Max SDO lines is
read from GCAP register.
For the given stream (channels = 2, bps = 16);
ratio = (channels * bps) / NSDO = 32 / NSDO;
On Tegra30, ratio = 32/4 = 8 (FAIL)
On Tegra210/186, ratio = 32/2 = 16 (PASS)
On Tegra194, ratio = 32/4 = 8 (FAIL) ==> Earlier workaround was
applied for it
If Tegra210/186 is forced to use 4SDO, it fails there as well. So the
behavior is consistent across all these chips.
Signed-off-by: Peter Geis <pgwipeout@xxxxxxxxx>
Tested-by: Ion Agorria <ion@xxxxxxxxxxx>
---
sound/pci/hda/hda_tegra.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/sound/pci/hda/hda_tegra.c b/sound/pci/hda/hda_tegra.c
index 70164d1428d4..f8d61e677a09 100644
--- a/sound/pci/hda/hda_tegra.c
+++ b/sound/pci/hda/hda_tegra.c
@@ -388,8 +388,7 @@ static int hda_tegra_first_init(struct azx *chip, struct platform_device *pdev)
* in powers of 2, next available ratio is 16 which can be
* used as a limiting factor here.
*/
- if (of_device_is_compatible(np, "nvidia,tegra194-hda"))
- chip->bus.core.sdo_limit = 16;
+ chip->bus.core.sdo_limit = 16;
Future Tegra chips address this problem and hence cannot be enforced by
default. May be we can have like below:
if (of_device_is_compatible(np, "nvidia,tegra30-hda"))
chip->bus.core.sdo_limit = 16;
It will need to be a bit more complicated than that, since the
tegra186 and tegra210 device trees have "nvidia,tegra30-hda" as a
fallback.
Looking at the generation map, tegra30-hda can be the fallback for the
broken implementation and tegra210-hda can be the fallback for the
working implementation.
Does that work for you?
As per above explanation, it is fine to apply the workaround for
Tegra210/186 as well. So it simplifies things for all existing chips.
/* codec detection */
if (!bus->codec_mask) {
--
2.25.1
