GitLab

        RX error count is injected in the old shared memory in the field named
        "rx_bad_count". The old system requires to reset this counter if the
        rx_error count doesn't increase. Thus, wee need to compare the current
        rx_err counter to the previous one in order to publish the real number
        of errors or 0. This implementation was not supporting multiple segment
        in the same process, because the previous err counter was kept as a
        static variable. This counter is moved into masterfip device.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        This pacth fixes a bug in libmasterfip-irq.c: the library gets from
        mockturtle message one by one to avoid any packing effect (read comment
        in the source code)
        Library version is bumped to 1.1.3
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        This pacth fixes a bug in libmasterfip-irq.c preventing the library to
        be thread safe.
        Library version is pushed to 1.1.2
        README file explains where to get wbgen2 to make a full compilation
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signedoff-by: Michel Arruat <michel.arruat@cern.ch>

        When the macrocycle starts, it might happen that standard poll timeout
        (which is 1.1*macrocycle) is not enough because for instance the RT
        software waits for external trigger. So to avoid to leave poll on
        timeout and report an error, th efirst poll timeout = 2*macrocycle.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        When an error is detected, it is catched to prevent from application
        ignoring errors and continuing execution. This way the original error
        is returned and actions are not executed.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        update license:
                - RT + library software: LGPL-3.0-or-later
                - tools software: GPL-3

.0-or-later
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        This program should desappear when diamon is replaced by cosmos.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        This ual dependency was there just to do mmap of masterfip resources.
        Now a new library located in test/libs does the job.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        The aim is to provide a generic application to spare the developer of
        all teh code required to program a macrocycle. To this end, one has to
        provide a csv file describing the macrocycle to be executed and in
        terms of implementation focus mainly on data processing in user
        callbacks.
        The current cvs file and fiptest.c is provided as an example.
        Note: msg are not implemented but everything is ready to do so.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        In case of macrocycle switch at run time, it might happen that
        not all mockturtle messages were not consumed when the application
        decides to stop current macrocycle and start a new one.
        Therefore a purge is required to empty mockturtle buffers in the
        driver.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        makefiles has been changed to be compliant with the "repo-parent" concept
        in order to benefit from all the rules for cross compiler and install
        ...
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Federico Vaga <federico.vaga@cern.ch>

Signed-off-by: Federico Vaga <federico.vaga@cern.ch>

Signed-off-by: Federico Vaga <federico.vaga@cern.ch>

Signed-off-by: Federico Vaga <federico.vaga@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        Add masterfip_wbgen2_csr_20170717.h
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        Again a bug in offset for invalidating aperiodic message entry.
        Msg max size was used instead of real msg size when msg frame is
        received without error but the whole transaction fails on RP_FIN.
        Note: this patch has been submitted by Tom.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        Still a bug around aperiodic message, ack or not acked folowed or not
        by an RP_FIN. There were some bugs around error detection and
        invalidation of the message in HMQ.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        In case RP-FIN RX frame got a wrong CTRL byte, error was not properly
        reported.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        In case of not successful RX aperiodic message transaction, the
        invalidation of teh data-entry in the HMQ was not properly done due to
        the fact that the real payload size of the incoming aperiodic message
        was overwitten by the payload of the RF_FIN. Now the message payload
        size is cached to proceed to the roolback if necessary.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        Hard coded max aperiodic msg payload size has been replaced by couple
        of define.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

        For RP_ACK_MSG sent by an agent, master sends an ack only if the RX
        frame was OK. In case of wrong RX frame RP_ACK is not sent to agent and
        master doesn't wait for RP_FIN from the agent.
        For RP_MSG (no ack requested) sent by an agent, even if teh RX frame
        had errors, master waits for RP_FIN from agent.
Signed-off-by: Michel Arruat <michel.arruat@cern.ch>

Signed-off-by: Michel Arruat <michel.arruat@cern.ch>