OLPC Firmware q7b12

This page is monitored by the OLPC team.

OLPC Firmware Release q7b12

Works on XO-4 B1 and later only. Do not use on XO-4 A or other models.

device tree changes for GALCORE graphics accelerator,
a fix for <trac>12451</trac> (random hangs with big 01 on screen),
a manufacturing test fix for <trac>12394</trac> (DCON unfreeze glitch),
new embedded controller firmware EC 0.3.10,
a way for developers to configure the wireless network name and passphrase on test units, using the NN and PP tags,
an inbound telnet session is now mirrored to the local keyboard and display,
a fix for USB serial devices that allows re-use of the device by Open Firmware applications (such as the serial terminal),

3525 OLPC XO-1.75 XO-4 - review, default wireless configuration manufacturing tags, NN for network name, PP for passphrase.
3523 OLPC XO-1.75 XO-4 - feature, default wireless configuration manufacturing tags, NN for network name, EP for WEP, and PA for WPA.
3522 OLPC XO-4 - EC Code 0.3.10
3521 OLPC XO-1.75 XO-4 - tune DCON unfreeze timing. Ten values of resumeline are glitch free for each CPU speed, so a median value is chosen.
3520 OLPC XO-4 - Added device node for GALCORE graphics accelerator.
3519 OLPC XO-4 - Added device node for GALCORE graphics accelerator.
3518 telnetd - use the I/O multiplexor, so that the telnet client becomes a mirror of the keyboard, display, and serial console.
3517 usbserial - start the read again next time the device is opened, fixes re-use of USB serial devices by applications, without this change the read path yields no data on subsequent open because the bulk in is not set up.

526c26e OLPC trac <trac>12453</trac> - System randomly hangs up at big 01. Also "ghost ok" problem. Workaround by moving the interrupt stack from SRAM into TCM. For some reason, the saved PC value in the interrupt stack was getting corrupted. Sometimes it had the value 1, resulting in the "bit 01" syndrome, and sometimes it had the address after the correct one, resulting in the "ghost ok" syndrome. The problem does not happen when the interrupt stack is in TCM instead of SRAM.