We developed an acquisition system based on OMAP-L137, 32 Mbyte, 216 MHz.
The DSP core collects samples from an EPLD through EDMA McASP, bandpass filters them, then transfers them
in 1/64 Hz data packets to the ARM core, by means of DSPLINK SIO_issue/SIO_reclaim call pairs executed in a specific thread.
SIO_reclaim is blocking.
DSP core is running on DSPBIOS 5.41.03.17
DSPLINK is DSP/BIOS Link for Linux 1.65.00.01
The ARM core polls for pending data packets from DSP core through CHNL_Issue/CHNL_reclaim call pairs
executed in a specific thread.
CHNL_Issue/CHNL_reclaim are not blocking.
When a data packet is transferred, the ARM core delivers it to a client application through LAN or WLAN.
DSP core is running Arago Linux 2.6.37.
Generally our targets are safely running for days until the client application terminates the acquisition.
However, on a limited number of targets, typically after some minutes of acquisition, the following problem appears:
1)ARM core is completely frozen: the serial console does not respond any more.
2)On the ARM core polling thread, it appears that the CHNL_reclaim call is started, and no more completed (the function does not returns): no errors on previsous CHNL_issue/CHNL_recalin calls is found.
3)DSP core is generally running: EDMA McASP IRQ is regularly acknowledged, all threads are regularly runng, with the exception of the thread executing SIO calls: it appears to be blocked, as the SIO_reclaim call is entered and no more exited.
If the Linux watchdog is enabled, the target reboots, and acquisition can be restarted.
Heating or refrigerating the "troublesome" boards has no effect.
Any insight on such problem ? Could it be hardware/software ? Which is the best way to address it ?
Any suggestion is welcome.
Thank you for your attention.