On Tue, Jan 09, 2007 at 11:14:31PM -0500, Don Zickus wrote: > On Wed, Jan 10, 2007 at 09:28:12AM +0530, Vivek Goyal wrote: > > On Wed, Jan 10, 2007 at 11:21:02AM +0900, Ken'ichi Ohmichi wrote: > > > > > > Hi. > > > > > > Last year, we discussed about "dumping to a remote host by makedumpfile". > > > For this feature, I will add 2 options of STDOUT output to makedumpfile. > > > If this feature is implemented, makedumpfile can output the dump data to > > > the dump device directly also. > > > > > > > > > The first option "-S": > > > This option outputs the dump data to STDOUT. > > > The dump data output by this option is different from normal dumpfile. > > > Because this option outputs STDOUT with the fileoffsets by each data > > > instead of lseek(). The analysis tool (ex. crash) can not read this > > > dump data, and the next option "-R" creates the readable dumpfile from > > > this dump data. > > > > > > > I missed the discussion which happened last time. Why there are two options > > -S and -R? Why can't we generate the final output in one go on standard > > output? > > Makedumpfile doesn't output data sequentially to a file. It uses > lseek() alot. Because we cannot temporarily store a 4gig crash file > locally, Ken'ichi created a mechanism to send the jumble bits to > stdout using -S, which you then pipe/stream to the remote location. The > remote location then reads from stdin and assembles the jumble bits into a > file readable by crash/gdb. I believe that's correct. :) > > > What's the file format of this intermediate file? > > there is no intermediate file, it's just a stream of data. >
As per his mail he is creating a intermediate file dumpfile.tmp. And then at a later point of time he can invoke mkdumpfile again asynchronously to generate final output. > > > > Should we use -O instead of -S for outputting to standard output. > > "tar" utility uses -O for extracting output to standard output. > > > > > The second option "-R": > > > This option creates the readable dumpfile from STDIN (the dump data) > > > output by the option "-S". > > > > > > > What is readable dump file? Readable by crash? Can it be read by gdb? > > the resulting output file is no different than what makedumpfile creates > on the hard drive. the only thing these options acheive is how the data > arrives on to the remote disk. > Still don't understand why do I need -R? I think it comes from that fact that mkdumpfile needs a way to recognize whether input file is vmcore or its that intermediate file generated by mkdumpfile. Personally I think that mkdumpfile should automatically recognize what kind of file has been inputted (the way kexec does to differentiate between various kind of kernel images) and take the action accordingly. Thanks Vivek _______________________________________________ fastboot mailing list [email protected] https://lists.osdl.org/mailman/listinfo/fastboot
