Re: RFA: dwarf2read.c patch

[Daniel Jacobowitz <drow at mvista dot com>]

On Mon, May 05, 2003 at 06:09:14PM -0400, J. Johnston wrote:
> The following patch fixes a problem on ia64.  In dwarf2read.c,
> dwarf_decode_lines(), the function check_cu_functions() is called
> to check for a specific scenario caused by gcc.
> 
> The function takes an address and returns an address.  In some
> cases, it will return the lowpc value of the function rather than
> the address that was input.
> 
> On the ia64, this causes problems because the line table info spit
> out by the compiler is often specified with relative addresses.
> As well, ia64 addresses are special as they encode a slot number which goes
> 0, 1, 2, then skips to the next quadword boundary.  For example,
> 
> 0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x20, 0x21, 0x22, ...
> 
> Addresses such as 0x0e or 0x0f are invalid and cause an error to
> occur if, say, a breakpoint was attempted to be inserted there.
> 
> Now, in dwarf_decode_lines(), if we update the address counter when we call
> check_cu_functions(), adding relative offsets in subsequent entries often 
> results in
> invalid addresses.
> 
> This patch makes it so the check_cu_functions() call is only used to
> alter the address passed to record_line(); the calculated address is left 
> untouched
> so subsequent relative operations result in valid results.
> 
> Tested on ia64 and x86.
> 
> Ok to commit?

I can't approve this, but I can say that that's certainly what I meant
to do with the check_cu_functions.  I feel really dumb for not noticing
that I was corrupting the line state machine.


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer