- 02 Jun, 2016 3 commits
-
-
Viresh Kumar authored
Mode switch takes just over a second to complete and the current timeout of one second isn't sufficient. Mode-switch logs from EVT 1.5: [ 56.709082] gb-firmware 1-3.3.1: Requested firmware package 'ara_00000126_00001002_fffe0001_ff980067_03.tftf' [ 57.003968] gb_control_mode_switch_operation: 176 [ 58.041616] greybus 1-3.3: mode switch detected Increase the timeout to two seconds. Tested with EVT 1.5 with gpbridge-test module. Signed-off-by:
Viresh Kumar <viresh.kumar@linaro.org> Tested-by:
Karthik Ravi Shankar <karthikrs@google.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@google.com>
-
Viresh Kumar authored
The operation layer allows only the 'core' operations on a connection, which is in its 'disconnecting' state. Mode switch is sent at the very end of interface-disable sequence, and the control connection is already in its 'disconnecting' state at that time. And so gb_operation_get_active() always fail with error -ENOTCONN. The operation core already has support for such 'core' operations, just that we were missing the needed flag while creating the mode switch operation. Fix that. Tested on EVT 1.5 with gpbridge-test module. Signed-off-by:
-
Viresh Kumar authored
This is the last step to required to finish the mode switch story. That is, call the hook provided by Interface layer to accomplish mode switch. Tested on EVT 1.5 with gpbridge-test module. Signed-off-by:
-
- 01 Jun, 2016 8 commits
-
-
Viresh Kumar authored
We should be using the generic version handling at bundle level, instead of at protocol level for bridged PHY devices as well. The bundle version handling is already in place, though it is *not* used today as we haven't bumped the version of control protocol yet. Remove protocol specific handling for bridged PHY devices. Tested on EVT 1.5 with gpbridge-test module. No nuttx changes are required with this. Signed-off-by:
Johan Hovold <johan@hovoldconsulting.com> Signed-off-by:
-
Viresh Kumar authored
We don't use these anymore. Drop them. Note that some macro's specific to bridged PHY devices aren't removed in the patch, as gbsim will break otherwise. They will be removed separately. Compile tested. Signed-off-by:
-
Axel Haslam authored
We should try to wait until all credits are accounted for before returning from shutdown. For this purpose add a helper function that will wait on completion, and call it form the shutdown. This helper will also be useful when "wait until sent" is implemented. Signed-off-by:
Axel Haslam <ahaslam@baylibre.com> Signed-off-by:
-
Axel Haslam authored
Data may be held pening in the hardware because of flow control mechanisms. When the port is closed, we need to flush all data that was not sent. For this, use the greybus message GB_UART_TYPE_FLUSH_FIFOS which will flush all data queued on the module but not yet sent on the data line. Suggested-by:
-
Axel Haslam authored
To avoid polling the firmware for space, a credit based system is implemented. The host will keep track of how many credits (bytes) it has sent to the firmware, and stop sending data when the quota is filled. The host will be informed that the firmware has more room for data when it handles the receive_credits request message from the firmware, and will continue to write data as credits become available. The firmware and the host may implement an algorithm to aggregate credits, and avoid extra greybus traffic. Suggested-by:
-
Axel Haslam authored
The firmware now buffers data instead of blocking while the transfer is sent, and the write operation cannot sleep. Instead of using gb_transfer_sync (which sleeps) in the write callback, buffer data in a fifo and send it from from a work queue. The write_room callback will will report 1 byte less that what is really available in the fifo, to leave space for extra characters that may be added by the tty layer. Signed-off-by:
-
Axel Haslam authored
Dtr_dts allows the tty layer to set the flow control lines to the correct state during open and close. Note that locking for newctrl is missing throughout the driver and will be addressed on a separate patch. Suggested-and-reviewed-by:
-
Axel Haslam authored
Handle the CRTSCTS flag in set_termios, so that auto flow control can be turned off. For this, add a new flag in the line coding request specifically for this purpose. Reviewed-by:
-
- 27 May, 2016 29 commits
-
-
Johan Hovold authored
The recently added gb_hd_release tracepoint was added to the wrong function. Signed-off-by:
-
Johan Hovold authored
Fix some whitespace issues introduced by the recent tracepoint changes. Signed-off-by:
-
Johan Hovold authored
Now that firmware has made the switch to the new interface boot sequence, we can remove support for the deprecated hotplug and hot-unplug operations. Reviewed-by:
-
Johan Hovold authored
Add a generic interface for bundle drivers to use to request that a mode switch is carried out on its behalf. Mode switching involves tearing down all connections to an interface, sending a unidirectional mode-switch request, and waiting for a mailbox event that triggers deferred control connection reset and re-enumeration of the interface. In case of a timeout waiting for the interface mailbox event, or on other errors, the interface is powered off. All of this needs to be done by core from work-queue context in order not to block incoming SVC requests and bundle-device tear down. Care must also be taken to serialise against concurrent module removal events and eject requests. Special handling of legacy mode-switching is also added in order to continue to support the ES3 bootrom. Signed-off-by:
-
Johan Hovold authored
Print an informative error message in case sending the mode-switch request fails. Signed-off-by:
-
Johan Hovold authored
Backport upstream reinit_completion() to pre-3.13 kernels. Reviewed-by:
-
Johan Hovold authored
Add a new host-device callback to flush any host-device queues, including any ongoing transfers, as part of connection tear down. The host-device buffers are flushed after the disconnecting operation have completed and the AP queues have been emptied. This can be used to flush already queued messages which upon reception would have been discarded by the remote end anyway. Note that this does not remove the need to flush any host-device queues as part of CPort disable which needs to make the CPort ready for reuse after deallocating all associated state and resetting the port. Suggested-by:
Mitchell Tasman <tasman@leaflabs.com> Signed-off-by:
-
Johan Hovold authored
Implement proper connection closure, which includes sending ping requests on the connection being tore down while coordinating with the remote interface as well as the SVC. This specifically implements the new ping operation, which in case of offloaded connections is handled by the host-device driver in an implementation-defined manner through a new callback. Note that the normal connection tear-down procedure is executed in case of failed connection establishment due to failed connected operation. Specifically, the disconnecting request is sent also in case the connected operation never succeeded. This is needed since the interface may have enabled FCT flow upon receiving the connected request. Signed-off-by:
-
Johan Hovold authored
Add a new core operation flag and a helper that core can use to create core operations. This will be used to implement the ping operations that core sends as part of connection tear down. Note that a new trace point is also added. Signed-off-by:
-
Johan Hovold authored
Restructure the operation activation state handling in preparation for a new disconnecting state. Signed-off-by:
-
Johan Hovold authored
Now that the legacy invalid state is gone, clean up the early connection-state check in the receive path and explicitly drop incoming messages for disabled connection. Signed-off-by:
-
Johan Hovold authored
Remove the unused legacy INVALID connection state. Signed-off-by:
-
Johan Hovold authored
The CPort Buffer configuration in the host-device needs to match the CPort feature flags set by the SVC, but they need not always be configured at the same point in time. This will be used when implementing proper connection tear down. Reviewed-by:
-
Johan Hovold authored
Rename an error label to make it more descriptive. Reviewed-by:
-
Johan Hovold authored
Make sure to reset CPorts at disable rather than enable as per specification. Reviewed-by:
-
Johan Hovold authored
Log failures to disable a host cport, and include the connection name in cport enable/disable error messages. Reviewed-by:
-
Johan Hovold authored
Add stubbed out connection-quiescing operation that is needed for proper connection tear down. Reviewed-by:
-
Johan Hovold authored
Implement the new disconnecting control operation needed for proper connection tear down. Reviewed-by:
-
Johan Hovold authored
Add control connection flag which will be set for control connections. Also add a helper function to test if the flag is set. Reviewed-by:
-
Johan Hovold authored
Add an interface quirk flag to suppress I/O during interface disable. This is needed to prevent I/O to the bootrom during controlled connection tear down (e.g. eject or driver unbind). This will be more obvious with the new connection tear-down procedure. Reviewed-by:
-
Johan Hovold authored
Clean up bootrom quirk handling in preparation for addition of further flags. Reviewed-by:
-
Johan Hovold authored
Add new helper to disable connections to interfaces that have already been disconnected (e.g. forcibly removed). The connection tear-down procedure differs enough depending on whether the interface is still present or already gone to warrant a dedicated helper. This will become more obvious with the new tear-down procedure, which involves I/O on the connection being tore down. This also simplifies handling of the legacy bootrom, which does not support the new tear-down operations. Specifically, this allows us to remove the early control-connection tear down during interface disable, and also avoids some error messages currently printed during legacy mode switch (i.e. bootrom boot-over-UniPro) and forcible removal. Reviewed-by:
-
Alex Elder authored
Coccinelle reports that gb_svc_pwrmon_debugfs_init() has a block of a for loop which is followed by an unnecessary semicolon. Get rid of that semicolon. Signed-off-by:
Alex Elder <elder@linaro.org> Reviewed-by:
-
Alex Elder authored
Coccinelle points out that the macro PTR_ERR_OR_ZERO() handles the frequent case of converting a pointer into either error code (if its value is an encoded error value) or 0 (otherwise). Switch some code in gb_power_supply_register() to use that macro. I have verified this is true of the kernel we're now working with (arche-6.0). Signed-off-by:
-
Alex Elder authored
Coccinelle points out that a call in gb_lights_channel_free() to flush_work() is passed which is always non-null. Prior to the call, there is an unnecessary check to see if that address is null. Get rid of the test. Signed-off-by:
-
Alex Elder authored
Coccinelle points out that debugfs_remove_recursive() handles a null argument properly, so there's no need to check for NULL before making the call. I have verified this is true of the kernel we're now working with (arche-6.0). Signed-off-by:
-
Alex Elder authored
Running "make coccicheck" on the Greybus code reports that gb_mmc_get_ro() and gb_mmc_get_cd() can return without releasing the mutex it acquired if there's an error. Fix this. Signed-off-by:
-
Alex Elder authored
Running "make coccicheck" on the Greybus code discovered that an error message in gb_camera_debugfs_init() was interpreting the wrong value in reporting the error code. Fix that. Signed-off-by:
-
Vaibhav Agarwal authored
Unique ids were reserved for CDSI0 and CDSI1 during _probe, however missed to release those ids during disconnect. This causes a memory leak of 128 bytes for each iteration of unipro_reset. Fix this. Signed-off-by:
Vaibhav Agarwal <vaibhav.agarwal@linaro.org> Reviewed-by:
-
