[PATCH] soundwire: qcom: allow multi-link on newer devices
Newer Qualcomm SoCs like X1E80100 might come with four speakers spread over two Soundwire controllers, thus they need a multi-link Soundwire stream runtime.
Cc: Mark Brown broonie@kernel.org Cc: alsa-devel@alsa-project.org Signed-off-by: Krzysztof Kozlowski krzysztof.kozlowski@linaro.org
---
This is an entirely different approach than my previous try here: https://lore.kernel.org/all/20231025144601.268645-1-krzysztof.kozlowski@lina... --- drivers/soundwire/qcom.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)
diff --git a/drivers/soundwire/qcom.c b/drivers/soundwire/qcom.c index 412b8e663a0a..57943724f0eb 100644 --- a/drivers/soundwire/qcom.c +++ b/drivers/soundwire/qcom.c @@ -920,6 +920,18 @@ static int qcom_swrm_init(struct qcom_swrm_ctrl *ctrl) return 0; }
+static int qcom_swrm_read_prop(struct sdw_bus *bus) +{ + struct qcom_swrm_ctrl *ctrl = to_qcom_sdw(bus); + + if (ctrl->version >= SWRM_VERSION_2_0_0) { + bus->multi_link = true; + bus->hw_sync_min_links = 3; + } + + return 0; +} + static enum sdw_command_response qcom_swrm_xfer_msg(struct sdw_bus *bus, struct sdw_msg *msg) { @@ -1078,6 +1090,7 @@ static const struct sdw_master_port_ops qcom_swrm_port_ops = { };
static const struct sdw_master_ops qcom_swrm_ops = { + .read_prop = qcom_swrm_read_prop, .xfer_msg = qcom_swrm_xfer_msg, .pre_bank_switch = qcom_swrm_pre_bank_switch, .post_bank_switch = qcom_swrm_post_bank_switch, @@ -1196,6 +1209,15 @@ static int qcom_swrm_stream_alloc_ports(struct qcom_swrm_ctrl *ctrl,
mutex_lock(&ctrl->port_lock); list_for_each_entry(m_rt, &stream->master_list, stream_node) { + /* + * For streams with multiple masters: + * Allocate ports only for devices connected to this master. + * Such devices will have ports allocated by their own master + * and its qcom_swrm_stream_alloc_ports() call. + */ + if (ctrl->bus.id != m_rt->bus->id) + continue; + if (m_rt->direction == SDW_DATA_DIR_RX) { maxport = ctrl->num_dout_ports; port_mask = &ctrl->dout_port_mask;
On 11/28/23 09:00, Krzysztof Kozlowski wrote:
Newer Qualcomm SoCs like X1E80100 might come with four speakers spread over two Soundwire controllers, thus they need a multi-link Soundwire stream runtime.
Cc: Mark Brown broonie@kernel.org Cc: alsa-devel@alsa-project.org Signed-off-by: Krzysztof Kozlowski krzysztof.kozlowski@linaro.org
This is an entirely different approach than my previous try here: https://lore.kernel.org/all/20231025144601.268645-1-krzysztof.kozlowski@lina...
drivers/soundwire/qcom.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)
diff --git a/drivers/soundwire/qcom.c b/drivers/soundwire/qcom.c index 412b8e663a0a..57943724f0eb 100644 --- a/drivers/soundwire/qcom.c +++ b/drivers/soundwire/qcom.c @@ -920,6 +920,18 @@ static int qcom_swrm_init(struct qcom_swrm_ctrl *ctrl) return 0; }
+static int qcom_swrm_read_prop(struct sdw_bus *bus) +{
- struct qcom_swrm_ctrl *ctrl = to_qcom_sdw(bus);
- if (ctrl->version >= SWRM_VERSION_2_0_0) {
bus->multi_link = true;
bus->hw_sync_min_links = 3;
- }
- return 0;
+}
static enum sdw_command_response qcom_swrm_xfer_msg(struct sdw_bus *bus, struct sdw_msg *msg) { @@ -1078,6 +1090,7 @@ static const struct sdw_master_port_ops qcom_swrm_port_ops = { };
static const struct sdw_master_ops qcom_swrm_ops = {
- .read_prop = qcom_swrm_read_prop,
nit-pick: read_prop() literally means "read platform properties".
The functionality implemented in this callback looks more like an initialization done in a probe, no?
.xfer_msg = qcom_swrm_xfer_msg, .pre_bank_switch = qcom_swrm_pre_bank_switch, .post_bank_switch = qcom_swrm_post_bank_switch, @@ -1196,6 +1209,15 @@ static int qcom_swrm_stream_alloc_ports(struct qcom_swrm_ctrl *ctrl,
mutex_lock(&ctrl->port_lock); list_for_each_entry(m_rt, &stream->master_list, stream_node) {
just realizing this now, are you sure the 'port_lock' is the proper means to protecting the stream->master_list? I don't see this used anywhere else in stream.c. I think you need to use bus_lock.
/*
* For streams with multiple masters:
* Allocate ports only for devices connected to this master.
* Such devices will have ports allocated by their own master
* and its qcom_swrm_stream_alloc_ports() call.
*/
if (ctrl->bus.id != m_rt->bus->id)
continue;
- if (m_rt->direction == SDW_DATA_DIR_RX) { maxport = ctrl->num_dout_ports; port_mask = &ctrl->dout_port_mask;
On 28/11/2023 16:35, Pierre-Louis Bossart wrote:
static enum sdw_command_response qcom_swrm_xfer_msg(struct sdw_bus *bus, struct sdw_msg *msg) { @@ -1078,6 +1090,7 @@ static const struct sdw_master_port_ops qcom_swrm_port_ops = { };
static const struct sdw_master_ops qcom_swrm_ops = {
- .read_prop = qcom_swrm_read_prop,
nit-pick: read_prop() literally means "read platform properties".
The functionality implemented in this callback looks more like an initialization done in a probe, no?
Yes, but multi_link is being set by sdw_bus_master_add() just before calling read_prop(). It looks a bit odd, because "bus" comes from the caller and is probably zero-ed already. Therefore I assumed the code did it on purpose - ignored multi_link set before sdw_bus_master_add(),
.xfer_msg = qcom_swrm_xfer_msg, .pre_bank_switch = qcom_swrm_pre_bank_switch, .post_bank_switch = qcom_swrm_post_bank_switch, @@ -1196,6 +1209,15 @@ static int qcom_swrm_stream_alloc_ports(struct qcom_swrm_ctrl *ctrl,
mutex_lock(&ctrl->port_lock); list_for_each_entry(m_rt, &stream->master_list, stream_node) {
just realizing this now, are you sure the 'port_lock' is the proper means to protecting the stream->master_list? I don't see this used anywhere else in stream.c. I think you need to use bus_lock.
This is from ctrl, internal driver structure:
struct qcom_swrm_ctrl *ctrl
Best regards, Krzysztof
On 11/29/23 10:43, Krzysztof Kozlowski wrote:
On 28/11/2023 16:35, Pierre-Louis Bossart wrote:
static enum sdw_command_response qcom_swrm_xfer_msg(struct sdw_bus *bus, struct sdw_msg *msg) { @@ -1078,6 +1090,7 @@ static const struct sdw_master_port_ops qcom_swrm_port_ops = { };
static const struct sdw_master_ops qcom_swrm_ops = {
- .read_prop = qcom_swrm_read_prop,
nit-pick: read_prop() literally means "read platform properties".
The functionality implemented in this callback looks more like an initialization done in a probe, no?
Yes, but multi_link is being set by sdw_bus_master_add() just before calling read_prop(). It looks a bit odd, because "bus" comes from the caller and is probably zero-ed already. Therefore I assumed the code did it on purpose - ignored multi_link set before sdw_bus_master_add(),
On the Intel side, there's a bit of luck here.
The caller intel_link_probe() does not set the multi-link property, but it's set in intel_link_startup() *AFTER* reading the properties - but we don't have any properties related to multi-link, only the ability to discard specific links.
.xfer_msg = qcom_swrm_xfer_msg, .pre_bank_switch = qcom_swrm_pre_bank_switch, .post_bank_switch = qcom_swrm_post_bank_switch, @@ -1196,6 +1209,15 @@ static int qcom_swrm_stream_alloc_ports(struct qcom_swrm_ctrl *ctrl,
mutex_lock(&ctrl->port_lock); list_for_each_entry(m_rt, &stream->master_list, stream_node) {
just realizing this now, are you sure the 'port_lock' is the proper means to protecting the stream->master_list? I don't see this used anywhere else in stream.c. I think you need to use bus_lock.
This is from ctrl, internal driver structure:
struct qcom_swrm_ctrl *ctrl
My point what that all other instances where list_for_each_entry() is used on stream->master list rely on the bus_lock.
You may be fine in this specific case with a QCOM-specific lock, not sure if there's any risk. At any rate that is not introduced by this patch, so for now
Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com
participants (2)
-
Krzysztof Kozlowski
-
Pierre-Louis Bossart