[dpdk-dev] [PKTGEN] fixing weird termio issues that complicate debugging
keith.wiles at intel.com
Wed Jan 20 17:26:19 CET 2016
On 1/20/16, 12:32 AM, "dev on behalf of Matthew Hall" <dev-bounces at dpdk.org on behalf of mhall at mhcomputing.net> wrote:
>Since the pktgen code is reindented I am finding time to read through it
>and experiment and see if I can get it working.
>I have issues with the init process of pktgen. It is difficult to debug
>it because the init code does a lot of very scary stuff to the terminal
>control / TTY device at inconvenient times in an inconvenient order, and
>in the process damages the debug output and damages the screen of your
>GDB without doing weird things to run GDB on a different TTY.
>Of course I am willing to contribute patches and not just complain, but
>first I need some help to follow what is going on.
>Here is the problematic call-flow with some explanation what went wrong
>trying it on some community machines outside of its original environment:
>1) it calls printf("\n%s %s\n", wr_copyright_msg(), wr_powered_by());
>which dumps tons of weird boilerplate of licenses, copyrights, code
>It is open source and everybody that matters already knows who coded it,
>so is this stuff really that important? This gets in the way when you
>are trying to work on it and I just have to comment it out.
One problem is a number of people wanted to steal the code and use in a paid application, so the copyright is some what a requirement. As you may know I do a lot of debugging on Pktgen and I feel they are a nuisance. I can try to see if we can clean up these messages, but do not hold your breath on getting them to be removed.
>2) it calls wr_scrn_setw and tinkers with the windows size very early in
>the init which can make your terminal weird
>3) it calls rte_eal_init which produces a lot of nice debug output,
>which is fine
IMO most of the information from DPDK is not very useful as why do I need to see every lcore line, plus a lot of more useless information. Most of the information could be reduced a couple of lines or only report issues not just a bunch of useless information.
>4) it calls pktgen_init_screen, which calls wr_scrn_init, which calls
>wr_scrn_erase which destroys the valuable debug output just created in
>(c) which is a bad thing
The screen init should be scrolling the information off the screen to preserve that info, unless it was changed by mistake.
>5) it calls wr_print_copyright and dumps more boilerplate I am not sure
>6) it logs some helpful messages about the port / descriptor settings
>which is fine
>7) it calls the pktgen_config_ports function which can crash in ways you
>need the destroyed debug output to fix.
>For example in my case that function crashes here:
> if (pktgen.nb_ports == 0)
> pktgen_log_panic("*** Did not find any ports to use ***");
>8) Later it makes a logo and a splash screen (wr_log, wr_splash_screen).
>Is this stuff really needed? This is a ton of output for just starting
>up some test program.
>To fix this debug problem I propose some changes which I am happy to
>1) decide what of this output we really need here and greatly simplify
>how much gets printed out
>2) move wr_scrn_setw right before pktgen_init_screen and after
>rte_eal_init to prevent damaging that output
>3) consider how wr_scrn_init is called in pktgen_init_screen, because it
>calls wr_scrn_erase which damages output
Again it could be scrolling that information off the screen, just need a large screen scroll buffer.
>4) I think that pktgen_config_ports should be called before all this
>weird screen init stuff, so that if it fails you can actually see what
>One other random topic... on the long lines of code it looks like there
>are some gigantic tab-indents pushing things off to the right still. One
>example, maybe there are others or another setting which is needed to
>fix all of these:
Please use tab stop of 4 instead of 8. IMO tab stop of 8 is so 1970’s and we should not need tab stop of 8 as any system today will work. :-)
> info->seq_pkt = (pkt_seq_t *)rte_zmalloc_socket(buff,
>(sizeof(pkt_seq_t) * NUM_TOTAL_PKTS),
Improvement to Pktgen is always welcome and the copyright info is going to be a bit hard to remove as that was one of the requirements when I open sourced the code. I understand it maybe a bit of output. I do not think it is really a user issue causing users to stop using it as startup is only down once, in my case I may start Pktgen a few times a day for development and it does not seem to slow me down much. :-)
More information about the dev