- 03 Mar, 2016 1 commit
-
-
Johan Hovold authored
Make sure not count errors during asynchronous tests twice (first in the timeout handler then again in the completion handler) to avoid obviously broken error stats such as: $ loopback_test -i 1000 -t transfer -p -o 200000 -c 64 -x -s 2000 1970-1-1 1:3:35 test: transfer path: gb_loopback0 size: 2000 iterations: 1000 errors: 1998 async: Enabled requests per-sec: min=0, max=0, average=0.310556, jitter=0 ap-throughput B/s: min=0 max=4026 average=1254.647461 jitter=4026 ap-latency usec: min=12803 max=12803 average=12803.000000 jitter=0 apbridge-latency usec: min=89 max=89 average=89.000000 jitter=0 gpbridge-latency usec: min=294 max=294 average=294.000000 jitter=0 where we supposedly have more errors than iterations (operations initiated). Signed-off-by:
Johan Hovold <johan@hovoldconsulting.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@google.com>
-
- 02 Mar, 2016 1 commit
-
-
Mark Greer authored
Provide finer-grained control of the audio streaming on APB1 by splitting the transmit/receive start and stop requests into prepare, start, stop, and shutdown. CC: Vaibhav Agarwal <vaibhav.agarwal@linaro.org> Signed-off-by:
Mark Greer <mgreer@animalcreek.com> Signed-off-by:
-
- 29 Feb, 2016 5 commits
-
-
Axel Haslam authored
Inotify does not handle sysfs events, so use poll instead. The loopback kernel driver will send a notification when the test is complete. So, open a poll file descriptor for every enabled device, and after starting the test, wait for an event from each device. After all events are received, read the total number of iterations and make sure the test is complete. Also, add missing stdint header which was included in inotify. Signed-off-by:
Axel Haslam <ahaslam@baylibre.com> Signed-off-by:
-
Axel Haslam authored
Currently, user space is notified for every message sent, but this is not really needed and does not work in the async case where all messages are sent from the start. Instead, notify userspace only when all the transfers are complete. This allows userspace to wait in a poll loop and wakeup only when the test is finished. Also, don't use the bundle kobj to send the notification it is the loopback device that contains the loopback attributes. Signed-off-by:
-
Viresh Kumar authored
WARN_ON() is a bit harsh here, as we just failed to power-off the HID device while it is getting removed. Replace it with dev_err(). Signed-off-by:
Viresh Kumar <viresh.kumar@linaro.org> Reviewed-by:
-
Viresh Kumar authored
hid_destroy_device() can potentially call callbacks defined in struct hid_ll_driver, which may initiate few greybus operations. And so connection (tx) should be kept enabled until the hid-device isn't destroyed. Reported-by:
Jiss Kuruvila <jkuruvila@google.com> Reported-by:
Laurent Pinchart <laurent.pinchart@linaro.org> Signed-off-by:
-
Mark Greer authored
Currently, it is assumed that all audio data CPorts registered on APB1 are used for transmitting audio data. That may not always be true like when a microphone is connected but no speakers. Also, the current special protocol lacks a way to tell APB1 whether the CPort being registered is for transmitting, receiving, or both. Fix by adding a 'direction' field to the register and unregister CPort requests and define bits indicating which direction (or both) audio data will go on that CPort. Signed-off-by:
-
- 27 Feb, 2016 1 commit
-
-
Greg Kroah-Hartman authored
Now that userspace is ready for all 32 bits of the vid/pid, take off our mask and send the full values. Signed-off-by:
-
- 26 Feb, 2016 19 commits
-
-
Alexandre Bailon authored
Currently, we are adding 0.5 to the average to round the average. But we are using the remainder to calculate the decimal, so we do not need to round the average. In addition, use a u64 type for the remainder to avoid overflow that might happen when stats->sum value is too big, usually for requests per seconds and the throughput. Signed-off-by:
Alexandre Bailon <abailon@baylibre.com> Reviewed-by:
-
Alexandre Bailon authored
Throughput and requests per second calculations are broken for asynchronous request. Instead of calculate the throughput for each iteration, calculate it once at the end of the test. In addition, update every seconds the min and the max for throughput and requests per second. Signed-off-by:
-
Alexandre Bailon authored
Currently, in case the case of error, statistics are updated for asynchronous but not for an asynchronous operation. Do not update the statistics in the case of error. Signed-off-by:
-
Alexandre Bailon authored
For the async ping transfer, statistics are counted twice, once after the after the gb_loopback_async_operation() and once in the callback. Only keep the one in the callback. Signed-off-by:
-
Rui Miguel Silva authored
lights can never be NULL at that point since lights_count must be different than zero, and we need only to validate the light_id. Reported-by:
Rui Miguel Silva <rui.silva@linaro.org> Signed-off-by:
-
Vaibhav Hiremath authored
HUB3613 configuration, either disable (Standby mode) or enable (HUB mode) is related to APB. So it makes perfect sense to put both of them together in one function. HUB3613 enable happens only at one place, in hub_conf_delayed_work() fn, but disable is initiated from multiple places. Move all calls to usb3613_hub_mode_ctrl(false) to apb_poweroff(). Testing Done: Tested on DB3.5 platform. Signed-off-by:
Vaibhav Hiremath <vaibhav.hiremath@linaro.org> Signed-off-by:
-
Vaibhav Hiremath authored
On first wake/detect pulse, everything works fine, as APB would be in poweroff state initially. But on subsequent wake/detect pulses, where APB is already in active state, internal function just returns doing nothing, as it finds that device is already in active state. So the solution would be to make sure that, whenever execution reaches to coldboot, make sure we power cycle it. Power off first, before powering on. Interrupt handler takes care of ignoring < 30msec pulses, so we should be safe here to power cycle APB. Testing Done: Testd on DB3.5 platform. Signed-off-by:
-
Johan Hovold authored
Add support for operations with short responses. So far we have assumed that the initiator of an operation always knows the exact size of the expected response. This is however not always the case and we've worked around this limitation in a couple of places by, for example, first requesting the size of a resource before fetching the actual data. To avoid such workarounds and simplify our protocols, add a short-response flag that can be set when allocating an operation. When this flag is set on an operation, core will accept a response that is shorter than the size of the (pre-allocated) response payload buffer. For now, we update the response-message payload_size field to reflect the actual length of the response received. Reviewed-by:
Laurent Pinchart <laurent.pinchart@ideasonboard.com> Tested-by:
-
Johan Hovold authored
As a preparatory clean up, add a temporary variable to point to the response message header. Reviewed-by:
-
Vaibhav Hiremath authored
In poweroff() fn, we are shutting down SVC, so disable interrupt as well. Testing Done: Tested on DB3.5 platform. Signed-off-by:
Michael Scott <michael.scott@linaro.org> Tested-by:
-
Vaibhav Hiremath authored
With support of interrupt based mechanism, gpio is not longer set to output mode, so gpio_set_value won't work. So use gpio_direction_output() fn in poweroff(), while setting value on wake/detect line. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
Since now driver supports interrupt based mechanism to read events from SVC over wake/detect line, no need to delay wake/detect assertion. We can assert wake/detect after SVC reset deassertion, so during boot itself SVC will start sending wake_out pulses. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
This patch enabled interrupt support on events received over wake/detect line. The driver follows below state machine, Default: wake/detect line is high (WD_STATE_IDLE) On Falling edge: SVC initiates boot (either cold/standby). On ES3, > 30msec = coldboot, else standby boot. Driver moves to WD_STATE_BOOT_INIT On rising edge (> 30msec): SVC expects APB to coldboot Driver wakes irq thread which kicks off APB coldboot (WD_STATE_COLDBOOT_TRIG) On rising edge (< 30msec): Driver ignores it, do nothing. After coldboot of APB, HUB configuration work is scheduled after 2 sec, allowing enough time for APB<->SVC/Switch to linkup (in multiple iterations) Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
If driver needs to process wake/detect events from SVC, by enabling interrupt support on wake/detect event, it becomes easier to maintain state of wake/detect line based on functionality. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
This is preparation of interrupt handling support, where APB coldboot and wake/detect handling will be handled as response to wake/detect interrupt. Due to slower I2C write operations in HUB configuration, it is important to separate HUB configuration, and probably delay it after APB is cold booted. Note that delayed work will be scheduled from interrupt handler, in following patches. To satisfy build (and bisect), remove apb_cold_boot() fn, which will be added back in the patch where it gets used again. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
With synchronization between SVC <=> AP over wake/detect line to bring APB's out of reset, we do not need any extra delays now. So remove it. Testing Done: Tested for DB3.5 and EVT1.2 platform. Signed-off-by:
-
Johan Hovold authored
Make example module 1-5 a 2x2 module by adding a second, dummy interface. This is both an example of how a 2x2 module would be represented and also suggests what a dummy interface may look like. A 2x2 module has two child interface devices and a num_interfaces value of two. In this example, the secondary interface 1-5.6, is a dummy interface and therefore lacks the normal identifying attributes (e.g. UniPro DDBL1 and Ara ids). We may eventually add an interface_type attribute to facilitate distinguishing various interface types (there may be more than two). In the following tree, the bundle attributes and child devices have been left out: greybus1/ ├── 1-2 │ ├── 1-2.2 │ │ ├── 1-2.2.1 │ │ ├── 1-2.2.2 │ │ ├── ddbl1_manufacturer_id │ │ ├── ddbl1_product_id │ │ ├── interface_id │ │ ├── product_id │ │ ├── serial_number │ │ ├── unique_id │ │ └── vendor_id │ ├── eject │ ├── module_id │ └── num_interfaces ├── 1-5 │ ├── 1-5.5 │ │ ├── 1-5.5.2 │ │ ├── ddbl1_manufacturer_id │ │ ├── ddbl1_product_id │ │ ├── interface_id │ │ ├── product_id │ │ ├── serial_number │ │ ├── unique_id │ │ └── vendor_id │ ├── 1-5.6 │ │ └── interface_id │ ├── eject │ ├── module_id │ └── num_interfaces └── 1-svc In this example there are two modules: 1-2 is a 1x2 module with one interface, and 1-5 is a 2x2 module with two interfaces of which the second (1-5.6) is a dummy interface. Signed-off-by:
-
Johan Hovold authored
Move example module 1-4 to position 5, effectively renaming it 1-5. Signed-off-by:
-
Johan Hovold authored
Introduce module devices and rename interface and bundle devices. Greybus module devices correspond to physical modules and have one or more interfaces. Modules have an id that is identical to the id of their primary interface, which in turn is the interface with lowest numbered id. The module name is constructed from the bus and module id: <bus_id>-<module_id> Interfaces and bundles are consequently renamed as <bus_id>-<module_id>.<interface_id> and <bus_id>-<module_id>.<interface_id>.<bundle_id> respectively. As before, interface ids (and therefore in a sense now also module ids) correspond to physical interface positions on the frame. Modules have the following attributes: eject module_id num_interfaces where module_id is the id of the module and num_interface the number of interfaces the module has. Note that the interface ids of a module's interfaces are expected to be <module_id>, <module_id + 1>, ..., <module_id + num_interfaces - 1>. Writing a non-zero argument to eject cleanly shuts down and unregisters all of the module interfaces before ejecting the module. The example sysfs tree now looks as follows with the second bus (APBridgeA) left out: greybus1/ ├── 1-2 │ ├── 1-2.2 │ │ ├── 1-2.2.1 │ │ │ ├── bundle_class │ │ │ ├── bundle_id │ │ │ └── state │ │ ├── 1-2.2.2 │ │ │ ├── bundle_class │ │ │ ├── bundle_id │ │ │ └── state │ │ ├── ddbl1_manufacturer_id │ │ ├── ddbl1_product_id │ │ ├── interface_id │ │ ├── product_id │ │ ├── serial_number │ │ ├── unique_id │ │ └── vendor_id │ ├── eject │ ├── module_id │ └── num_interfaces ├── 1-4 │ ├── 1-4.4 │ │ ├── 1-4.4.2 │ │ │ ├── bundle_class │ │ │ ├── bundle_id │ │ │ ├── gpbridge0 │ │ │ │ ├── gpio │ │ │ │ │ └── gpiochip490 │ │ │ │ └── i2c-4 │ │ │ └── state │ │ ├── ddbl1_manufacturer_id │ │ ├── ddbl1_product_id │ │ ├── interface_id │ │ ├── product_id │ │ ├── serial_number │ │ ├── unique_id │ │ └── vendor_id │ ├── eject │ ├── module_id │ └── num_interfaces └── 1-svc ├── ap_intf_id ├── eject └── endo_id where greybus1 is a bus; 1-svc the svc; 1-2, and 1-4 are modules; 1-2.2 and 1-4.4 are (primary) interfaces; and 1-2.2.1, 1-2.2.2, and 1-4.4.2 are bundles. Note that the svc eject attribute may eventually be renamed force_eject. Signed-off-by:
-
- 25 Feb, 2016 10 commits
-
-
Greg Kroah-Hartman authored
In kernel version 4.5, struct gpio_chip renamed the field 'dev' to 'parent' so handle this properly. Signed-off-by:
-
Greg Kroah-Hartman authored
Use the bundle device directly in gpio error messages instead of the gpio device, as they are the same pointer. This will make future gpio api changes much easier to handle. Signed-off-by:
-
Vaibhav Hiremath authored
If SVC coldboot fails or if of_platform_populate() fn fails, then state of device needs to be reverted. Importantly, if of_platform_populate() fails, then poweroff the SVC. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Johan Hovold authored
Fix incomplete receive-data sanity checks. The payload size was never verified before parsing the uart header and neither was the uart-header data size verified against the actual payload size, something which could lead to information leaks when passing data beyond the payload buffer to the tty layer. Also remove the incorrect check against the maximum (tx-buffer) payload size. Reviewed-by:
Bryan O'Donoghue <bryan.odonoghue@linaro.org> Signed-off-by:
-
Johan Hovold authored
Add dedicated serial-state request handler and add the missing sanity check on the incoming request. Reviewed-by:
-
Johan Hovold authored
Let's be well behaved and add a sanity check on the maximum greybus payload size to avoid underflow on the calculated buffer size. Reviewed-by:
-
Fabien Parent authored
The AP must enable the FCT flow of APBA once it has received the response from the AP that the connection between APBA and a module has been setted up. Disable the flow of FCT tokens when destroying connections. Signed-off-by:
Fabien Parent <fparent@baylibre.com> Reviewed-by:
-
Fabien Parent authored
Implement the control requests enabling/disabling the flow of FCT on APBA. Reviewed-by:
-
Fabien Parent authored
Add control requests to enable/disable the flow of unipro FCT tokens Reviewed-by:
-
Fabien Parent authored
In order to support mailbox-free control cport init on the bridges the AP must be able to enable/disable the flow of unipro fct tokens. Add a new API that will enable or disable on APBA the flow of fct tokens. Reviewed-by:
-
- 24 Feb, 2016 3 commits
-
-
Vaibhav Hiremath authored
Have a check inside all individual operational state change functions to check whether device is in same state, and if yes, then return immediately. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
If user switches from fw_flashing => off mode, then we do not need to do same things again, for example, clk_disable and wake/detect event, as while switching to fw_flashing, driver makes sure that device goes to off state and then brings back in fw_flashing state. Testing Done: Tested on DB3.5 platform. Signed-off-by:
-
Vaibhav Hiremath authored
Have a check inside all individual operational state change functions to check whether device is in same state, and if yes, then return immediately. Testing Done: Tested on DB3.5 platform Signed-off-by:
-
