On Fri, 4 Jul 2003 11:05:05 +0800, "Dong, Eddie" <[EMAIL PROTECTED]> wrote: >Keith: > I encountered an issue for ISR routine debug, the situation is: > 1: When kdb is entered from keyboard, probably the IRQ >is enabled when it enter kdb(). Then if I set bp at do_IRQ and leave >kdb, the debug may be re-entered as soon as kdba_restoreint(..) is >completed in kdb(). That will cause kdb to be hanged because of LEAVING >flag not cleared correctly.=20 > 2: Even when kdb() is entered from former debug event, >if I set bp at do_IRQ and leave, the debug may be re-entered within >local_bh_enable() due to its IRQ enabled within this function, thus same >problem existed like above. > > So what is your suggestion about this problem? If I don't make >mistake, probably we should move this 2 items to the end of function >kdb() and make a new release.
I am tempted to say "don't do that". There are some places in the kernel where it is not safe to put breakpoints, because kdb uses those paths as well. Instead of putting a breakpoint at do_IRQ, which affects the keyboard that kdb is using, put the breakpoint at the start of the interrupt handler that you really care about.
