RE: [RFA/RFC] blockframe.c Fix errorneous addr check

RE: [RFA/RFC] blockframe.c Fix errorneous addr check

[Andrew Volkov]

>On Thu, Jun 27, 2002 at 09:13:41PM +0400, Andrew Volkov wrote:
>> Hi,
>> 
>> This patch fix wrong detection of belonging addr to startup file:
>> if call main is last instuction in crt0, then 
>> inside_entry_file erroneously return false.
>> 
>> Ok to commit?
>> Andrey 
>> 
>> P.S. Possible this bug not in this func, but in symbol reader.
>
>I'm inclined to believe the bug is in the symbol reader.  Upper PC
>bounds are generally exclusive - first address after the function.
>
>-- 
>Daniel Jacobowitz                           Carnegie Mellon University
>MontaVista Software                         Debian GNU/Linux Developer
>

Daniel you're right, this bug in coffread.c(complete_symtab) function,
rather, in invalid calculating of current_source_end_addr:

	current_source_end_addr = start_addr + size;

but must be:

	current_source_end_addr = start_addr + size + <addr sizeof>;

As I understand, since machine address size is not known at time of parsing 
coff file, then we have big trouble :(.

Comments/suggestions?

Andrey

Re: [RFA/RFC] blockframe.c Fix errorneous addr check

[Kevin Buettner]

On Jun 27, 10:08pm, Andrew Volkov wrote:

> Daniel you're right, this bug in coffread.c(complete_symtab) function,
> rather, in invalid calculating of current_source_end_addr:
> 
> 	current_source_end_addr = start_addr + size;
> 
> but must be:
> 
> 	current_source_end_addr = start_addr + size + <addr sizeof>;
> 
> As I understand, since machine address size is not known at time of parsing 
> coff file, then we have big trouble :(.

Could you explain why you need to add in the size of an address?

Assuming it is needed, won't TARGET_PTR_BIT / TARGET_CHAR_BIT provide the
correct size?

Kevin

RE: [RFA/RFC] blockframe.c Fix errorneous addr check

[Andrew Volkov]

>On Jun 27, 10:58pm, Andrew Volkov wrote:
>
>> >> 	current_source_end_addr = start_addr + size;
>> >> 
>> >> but must be:
>> >> 
>> >> 	current_source_end_addr = start_addr + size + <addr sizeof>;
>> >> 
>> >> As I understand, since machine address size is not known at 
>> >time of parsing 
>> >> coff file, then we have big trouble :(.
>> >
>> >Could you explain why you need to add in the size of an address?
>> 
>> It's not necessarily must be addr sizeof, it may be 1, but, I think,
>> will be better, if current_source_end_addr will contain correct 
>> value for target.
>
>What I'd like to understand is why (start_addr + size) is insufficient.
>I.e, why do we need to add 1 or some other small value to compute the
>correct ending address?
>
>Kevin

Because start_addr + size is last_addr in block, hence all gdb parts, wich 
assume that highpc is first addr after block, will work improperly. 
Ex. this check always failed:

	if (addr >= lowpc && addr < highpc)
		....

when addr = last_addr_in_block.

While I bump with this problem when work with h8300 coff file.

Andrey




 

Re: [RFA/RFC] blockframe.c Fix errorneous addr check

[Kevin Buettner]

On Jun 28, 12:41pm, Andrew Volkov wrote:

> >On Jun 27, 10:58pm, Andrew Volkov wrote:
> >
> >> >> 	current_source_end_addr = start_addr + size;
> >> >> 
> >> >> but must be:
> >> >> 
> >> >> 	current_source_end_addr = start_addr + size + <addr sizeof>;
> >> >> 
> >> >> As I understand, since machine address size is not known at 
> >> >time of parsing 
> >> >> coff file, then we have big trouble :(.
> >> >
> >> >Could you explain why you need to add in the size of an address?
> >> 
> >> It's not necessarily must be addr sizeof, it may be 1, but, I think,
> >> will be better, if current_source_end_addr will contain correct 
> >> value for target.
> >
> >What I'd like to understand is why (start_addr + size) is insufficient.
> >I.e, why do we need to add 1 or some other small value to compute the
> >correct ending address?
> 
> Because start_addr + size is last_addr in block, hence all gdb parts, wich 
> assume that highpc is first addr after block, will work improperly. 

According to the comment just prior to complete_symtab(), ``size'' is
the number of bytes of text.  If that's really the case, then
start_addr + size should be the first address beyond the section in
question.

> Ex. this check always failed:
> 
> 	if (addr >= lowpc && addr < highpc)
> 		....
> 
> when addr = last_addr_in_block.

I see.

I think we need to look elsewhere for the bug though.  I think
``size'' is being computed incorrectly earlier on.  While it's
possible that the bug is in gdb or bfd, I'm now suspicious of the
value obtained from the object file.

> While I bump with this problem when work with h8300 coff file.

Can you post the output of ``objdump -h'' on one of these coff files?
Take a look at the size that it reports for .text and let us know if
it appears too small to you.

If it is too small, then the bug is in some other part of the toolchain,
possibly the linker.

Kevin

RE: [RFA/RFC] blockframe.c Fix errorneous addr check

[Andrew Volkov]

>
>On Jun 27, 10:08pm, Andrew Volkov wrote:
>
>> Daniel you're right, this bug in coffread.c(complete_symtab) 
>function,
>> rather, in invalid calculating of current_source_end_addr:
>> 
>> 	current_source_end_addr = start_addr + size;
>> 
>> but must be:
>> 
>> 	current_source_end_addr = start_addr + size + <addr sizeof>;
>> 
>> As I understand, since machine address size is not known at 
>time of parsing 
>> coff file, then we have big trouble :(.
>
>Could you explain why you need to add in the size of an address?

It's not necessarily must be addr sizeof, it may be 1, but, I think,
will be better, if current_source_end_addr will contain correct 
value for target.

>
>Assuming it is needed, won't TARGET_PTR_BIT / TARGET_CHAR_BIT 
>provide the
>correct size?
>
>Kevin
>
Sorry, I forget about this macros :)

Andrey

Re: [RFA/RFC] blockframe.c Fix errorneous addr check

[Kevin Buettner]

On Jun 27, 10:58pm, Andrew Volkov wrote:

> >> 	current_source_end_addr = start_addr + size;
> >> 
> >> but must be:
> >> 
> >> 	current_source_end_addr = start_addr + size + <addr sizeof>;
> >> 
> >> As I understand, since machine address size is not known at 
> >time of parsing 
> >> coff file, then we have big trouble :(.
> >
> >Could you explain why you need to add in the size of an address?
> 
> It's not necessarily must be addr sizeof, it may be 1, but, I think,
> will be better, if current_source_end_addr will contain correct 
> value for target.

What I'd like to understand is why (start_addr + size) is insufficient.
I.e, why do we need to add 1 or some other small value to compute the
correct ending address?

Kevin

[RFA/RFC] blockframe.c Fix errorneous addr check

[Andrew Volkov]

Hi,

This patch fix wrong detection of belonging addr to startup file:
if call main is last instuction in crt0, then 
inside_entry_file erroneously return false.

Ok to commit?
Andrey 

P.S. Possible this bug not in this func, but in symbol reader.


2002-06-27 Andrey Volkov <avolkov@transas.com>

    *blockframe.c (inside_entry_file): Fixed erroneous check that 
    addr is in startup file.
    
Index: blockframe.c
===================================================================
RCS file: /cvs/src/src/gdb/blockframe.c,v
retrieving revision 1.30
diff -u -r1.30 blockframe.c
--- blockframe.c	26 Jun 2002 15:28:46 -0000	1.30
+++ blockframe.c	27 Jun 2002 16:33:10 -0000
@@ -113,7 +113,7 @@
 	return 0;
     }
   return (addr >= symfile_objfile->ei.entry_file_lowpc &&
-	  addr < symfile_objfile->ei.entry_file_highpc);
+	  addr <= symfile_objfile->ei.entry_file_highpc);
 }
 
 /* Test a specified PC value to see if it is in the range of addresses

Re: [RFA/RFC] blockframe.c Fix errorneous addr check

[Daniel Jacobowitz]

On Thu, Jun 27, 2002 at 09:13:41PM +0400, Andrew Volkov wrote:
> Hi,
> 
> This patch fix wrong detection of belonging addr to startup file:
> if call main is last instuction in crt0, then 
> inside_entry_file erroneously return false.
> 
> Ok to commit?
> Andrey 
> 
> P.S. Possible this bug not in this func, but in symbol reader.

I'm inclined to believe the bug is in the symbol reader.  Upper PC
bounds are generally exclusive - first address after the function.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer