TX1 crashing after trying to install CUDA 8 with low memory

I tried to install CUDA 8 on my TX1 when it had only 1.2 gb space. The cuda package was 1.1 gb space. I tried to install it and the TX got hanged. When restarted and tried to access it stays in first login page. With USB serial port access from laptop i am repeatedly getting “CPU: 2 PID: 30308 Comm: compiz Not tainted 4.4.38-tegra #1” error. It keeps on repeating its self or you can say blinking with the above error like below

[ 1322.194545] x3 : 00000000008faa60 x2 : 0000000000000000
[ 1322.199922] x1 : 0000000000001fc0 x0 : 0000007f801cd000
[ 1322.205314]
[ 1322.206802] Library at 0x7f735b921c: 0x7f7260d000 /usr/lib/aarch64-linux-gnu/tegra/libnvidia-glcore.so.28.2.0
[ 1322.216809] Library at 0x7f735b9200: 0x7f7260d000 /usr/lib/aarch64-linux-gnu/tegra/libnvidia-glcore.so.28.2.0
[ 1322.226716] vdso base = 0x7f81bf2000
[ 1323.173869] compiz[3708]: unhandled level 3 translation fault (7) at 0x7fa4512000, esr 0x92000047
[ 1323.182750] pgd = ffffffc059f90000
[ 1323.186357] [7fa4512000] *pgd=00000000d730f003, *pud=00000000d730f003, *pmd=00000000e10a2003, *pte=0000000000000000
[ 1323.196877]
[ 1323.198365] CPU: 3 PID: 3708 Comm: compiz Not tainted 4.4.38-tegra #1
[ 1323.205076] Hardware name: jetson_tx1 (DT)
[ 1323.209239] task: ffffffc0610ae400 ti: ffffffc0cee1c000 task.ti: ffffffc0cee1c000
[ 1323.220667] PC is at 0x7f975b921c
[ 1323.223984] LR is at 0x7f975b9200
[ 1323.227290] pc : [<0000007f975b921c>] lr : [<0000007f975b9200>] pstate: 40000000
[ 1323.234679] sp : 0000007ffbf61fb0
[ 1323.237995] x29: 0000007ffbf64230 x28: 00000000ffffffff
[ 1323.243325] x27: 0000000000000000 x26: 0000007f97ff5000
[ 1323.248649] x25: 0000007f97ff7000 x24: 0000007ffbf62050
[ 1323.253971] x23: 0000007ffbf62040 x22: 0000007f97ff5000
[ 1323.259296] x21: 0000007f97ff5000 x20: 0000000000002000
[ 1323.264624] x19: 0000000000000080 x18: 0000000000000008
[ 1323.269945] x17: 0000007fa5b6f988 x16: 0000007fa4148438
[ 1323.275266] x15: 0000007fa5f38cc0 x14: 0000000000000000
[ 1323.280596] x13: 000000001caad162 x12: 00000000000007e2
[ 1323.285917] x11: 000000000000001d x10: 0101010101010101
[ 1323.291238] x9 : 000000000092ead0 x8 : 00000000008fbd40
[ 1323.296567] x7 : 0000000000000000 x6 : 0000000000000000
[ 1323.301887] x5 : 0000007fa5c3e9b0 x4 : 00000000ffffffff
[ 1323.307208] x3 : 00000000008fbd40 x2 : 0000000000000000
[ 1323.312535] x1 : 0000000000001fc0 x0 : 0000007fa4512000
[ 1323.317858]
[ 1323.323278] Library at 0x7f975b921c: 0x7f9660d000 /usr/lib/aarch64-linux-gnu/tegra/libnvidia-glcore.so.28.2.0
[ 1323.333182] Library at 0x7f975b9200: 0x7f9660d000 /usr/lib/aarch64-linux-gnu/tegra/libnvidia-glcore.so.28.2.0
[ 1323.343087] vdso base = 0x7fa5f37000

libnvidia-glcore and compiz are associated with the GUI (GPU operations). If you have serial console, then you might be able to delete files and get more space. Once you have more space you can probably add back in the original set of files from the “/etc/nv_tegra_release” list (those files can be copied as a single package to the Jetson, and then unpacked from “/” to put them back in place). The key is to be able to get command line access. Without that you will be limited to cloning, removing items from the clone, and then flashing the clone back in place (using the same L4T release to flash as is on the unit).

For information on serial console, see:
https://www.jetsonhacks.com/2015/12/01/serial-console-nvidia-jetson-tx1/