[Bluez,v1] device - If HFP is supported, ignore HSP

Commit Message

Yu Liu June 30, 2020, 10:49 p.m. UTC

From: Hsin-Yu Chao <[e-mail]>

For a BT headset that supports both HSP and HFP, BlueZ creates
service instances for these two profiles and connects them.
It's uncertain that which of HSP and HFP eventually get connected
and being used for SCO audio. And we start observing some problem
because of this uncertainty:

- For headset that supports WBS, we need HFP connect for codec
negotiation. If HSP connects but not HFP, WBS cannot be used.
- For WH-1000XM3, if BlueZ ever initiated HFP connection but failed,
headset won't have working SCO audio even HSP is connected.

Fix this at when device probes services, if HFP is in the uuid list,
don't bother create one for HSP.

Reviewed-by: Sonny Sasaka <[e-mail]>
---

Changes in v1:
- Initial change

 src/device.c | 5 +++++
 1 file changed, 5 insertions(+)

Comments

Luiz Augusto von Dentz July 1, 2020, 12:30 a.m. UTC | #1

Hi Yu Liu,

On Tue, Jun 30, 2020 at 3:55 PM Yu Liu <[e-mail]> wrote:
>
> From: Hsin-Yu Chao <[e-mail]>
>
> For a BT headset that supports both HSP and HFP, BlueZ creates
> service instances for these two profiles and connects them.
> It's uncertain that which of HSP and HFP eventually get connected
> and being used for SCO audio. And we start observing some problem
> because of this uncertainty:
>
> - For headset that supports WBS, we need HFP connect for codec
> negotiation. If HSP connects but not HFP, WBS cannot be used.

We should probably have a way to detect when one or the other is
connected and then don't attempt to connect again since they would
likely conflict when it comes to connecting SCO, that said is your
system setting AutoConnect for both HSP and HFP?

> - For WH-1000XM3, if BlueZ ever initiated HFP connection but failed,
> headset won't have working SCO audio even HSP is connected.

Hmm I guess this will need to treated separately, not sure if we can
exclude HSP to be connected once HFP fails, but then again I think the
system should not have them both set to be Auto Connected, HSP most
likely should only be set as as a fallback if HFP is not supported.

> Fix this at when device probes services, if HFP is in the uuid list,
> don't bother create one for HSP.
>
> Reviewed-by: Sonny Sasaka <[e-mail]>
> ---
>
> Changes in v1:
> - Initial change
>
>  src/device.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/src/device.c b/src/device.c
> index 7b0eb256e..4036bfd81 100644
> --- a/src/device.c
> +++ b/src/device.c
> @@ -4370,6 +4370,11 @@ static struct btd_service *probe_service(struct btd_device *device,
>         if (!device_match_profile(device, profile, uuids))
>                 return NULL;
>
> +       /* If device supports HFP, don't bother create service for HSP. */
> +       if (g_slist_find_custom(uuids, HFP_HS_UUID, bt_uuid_strcmp) &&
> +               bt_uuid_strcmp(profile->remote_uuid, HSP_HS_UUID) == 0)
> +               return NULL;
> +
>         l = find_service_with_profile(device->services, profile);
>         if (l)
>                 return l->data;
> --
> 2.27.0.212.ge8ba1cc988-goog
>

Pali Rohár July 3, 2020, 10:15 a.m. UTC | #2

On Tuesday 30 June 2020 15:49:34 Yu Liu wrote:
> From: Hsin-Yu Chao <[e-mail]>
> 
> For a BT headset that supports both HSP and HFP, BlueZ creates
> service instances for these two profiles and connects them.
> It's uncertain that which of HSP and HFP eventually get connected
> and being used for SCO audio. And we start observing some problem
> because of this uncertainty:
> 
> - For headset that supports WBS, we need HFP connect for codec
> negotiation. If HSP connects but not HFP, WBS cannot be used.
> - For WH-1000XM3, if BlueZ ever initiated HFP connection but failed,
> headset won't have working SCO audio even HSP is connected.
> 
> Fix this at when device probes services, if HFP is in the uuid list,
> don't bother create one for HSP.
> 
> Reviewed-by: Sonny Sasaka <[e-mail]>
> ---
> 
> Changes in v1:
> - Initial change
> 
>  src/device.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/src/device.c b/src/device.c
> index 7b0eb256e..4036bfd81 100644
> --- a/src/device.c
> +++ b/src/device.c
> @@ -4370,6 +4370,11 @@ static struct btd_service *probe_service(struct btd_device *device,
>  	if (!device_match_profile(device, profile, uuids))
>  		return NULL;
>  
> +	/* If device supports HFP, don't bother create service for HSP. */
> +	if (g_slist_find_custom(uuids, HFP_HS_UUID, bt_uuid_strcmp) &&
> +		bt_uuid_strcmp(profile->remote_uuid, HSP_HS_UUID) == 0)
> +		return NULL;
> +

This patch would completely break devices with buggy HFP profile
support.

Also this patch disallow usage of HSP when both HSP and HFP are
supported.

I'm against this change as it disallow to use HSP and HFP simultaneously
with hsphpfd daemon. With hsphfpd daemon it is working with non-buggy
headsets.

Also in pulseaudio we have more bug reports that HFP is does not work
(even when advertised), but HSP is working fine.

If you have a broken headset as you described in commit message, then
please add some quirk for it. But do not disallow usage of HSP / HFP for
all devices including non-broken ones, just because you have one broken
device.

>  	l = find_service_with_profile(device->services, profile);
>  	if (l)
>  		return l->data;
> -- 
> 2.27.0.212.ge8ba1cc988-goog
>

Patch

diff --git a/src/device.c b/src/device.c
index 7b0eb256e..4036bfd81 100644
--- a/src/device.c
+++ b/src/device.c
@@ -4370,6 +4370,11 @@  static struct btd_service *probe_service(struct btd_device *device,
 	if (!device_match_profile(device, profile, uuids))
 		return NULL;
 
+	/* If device supports HFP, don't bother create service for HSP. */
+	if (g_slist_find_custom(uuids, HFP_HS_UUID, bt_uuid_strcmp) &&
+		bt_uuid_strcmp(profile->remote_uuid, HSP_HS_UUID) == 0)
+		return NULL;
+
 	l = find_service_with_profile(device->services, profile);
 	if (l)
 		return l->data;