[RFC] problem with read_memory_string (reads 8 bytes at a time)

[RFC] problem with read_memory_string (reads 8 bytes at a time)

[Joel Brobecker]

Hello,

I did what I thought would be a nice cleanup in ada-lang.c last weekend,
by deleting a local function (extract_string), and replacing it by a call
to read_memory_string instead. This works fine most of the time, but
nightly testing did reveal a problem with that function on powerpc-elf.

    (gdb) start
    Cannot access memory at address 0x1002e8

Here is what happens:

    1. The "start" commands needs to know that the name of the "main"
       program is. As a result, we call ada_main_name, which finds
       the symbol that points us to the string containing that main
       name.

    2. To read that name, we call read_memory_string now, and this
       routine performs strings reads 8 bytes at a time.

The problem is that our string is at the end of our .rodata section.
The string is 8 bytes long, and sits at 0x1002e0. The .rodata section
ends at 0x001102ec. So here's what we end up doing:

    a. Read 8 bytes from 0x1002e0 - no problem.
       But we haven't read the \0 yet, so we keep going.

    b. Read 8 bytes from 0x1002e0 + 8 = 0x1002e8:
         i. The section ends at 0x001102ec, so bfd returns only 4 bytes read.
        ii. We find out that only 4 bytes were read, so we still need to
            read another 4 bytes for the 8byte read to be complete
       iii. the next 4byte read doesn't find a section from which to
            read the 4bytes, and so returns 0 signifying an error.

I don't really know how to fix this issue except by reading the string
one byte at a time :-(. Any suggestion?

Thanks,
-- 
Joel

Re: [RFC] problem with read_memory_string (reads 8 bytes at a time)

[Thiago Jung Bauermann]

El mié, 08-04-2009 a las 11:26 -0700, Joel Brobecker escribió:
>     b. Read 8 bytes from 0x1002e0 + 8 = 0x1002e8:
>          i. The section ends at 0x001102ec, so bfd returns only 4 bytes read.
>         ii. We find out that only 4 bytes were read, so we still need to
>             read another 4 bytes for the 8byte read to be complete
>        iii. the next 4byte read doesn't find a section from which to
>             read the 4bytes, and so returns 0 signifying an error.
> 
> I don't really know how to fix this issue except by reading the string
> one byte at a time :-(. Any suggestion?

I suggest using target_read_string. If I understand its code correctly,
it handles the corner case you hit.

I thought that valprint.c:read_string also handled that, but now that I
read the code again, it looks like it would fail too. :-(
-- 
[]'s
Thiago Jung Bauermann
IBM Linux Technology Center

Re: [RFC] problem with read_memory_string (reads 8 bytes at a time)

[Joel Brobecker]

> I suggest using target_read_string. If I understand its code correctly,
> it handles the corner case you hit.

Thanks for the tip!

It looks like it does, indeed - although a little ineffectively:
Once it failed to read a 4byte block, it tries reading 1 byte.
If that works, then it tries 3bytes, which won't work again,
and thus try 1 byte one more time. If that works, then it'll read
2 bytes, etc. I don't think this is a big deal, though.

That being said, I am wondering why we have more than 1 routine
to read strings... Is there something we should do?

(I will change my code to use target_read_string for now - this won't
help testing the case were we cross a boundary because the \0 is
byte 0 of the last 4 bytes of my section, and that section is 4bytes-
aligned).

-- 
Joel

Re: [RFC] problem with read_memory_string (reads 8 bytes at a time)

[Thiago Jung Bauermann]

El mié, 08-04-2009 a las 14:26 -0700, Joel Brobecker escribió:
> > I suggest using target_read_string. If I understand its code correctly,
> > it handles the corner case you hit.
> 
> Thanks for the tip!
> 
> It looks like it does, indeed - although a little ineffectively:
> Once it failed to read a 4byte block, it tries reading 1 byte.
> If that works, then it tries 3bytes, which won't work again,
> and thus try 1 byte one more time. If that works, then it'll read
> 2 bytes, etc. I don't think this is a big deal, though.

Well, it's not efficient for your case, but it's efficient for the case
where three bytes out of a 4 byte "bite" were good. The next try will
read those 3 bytes and off you go...

> That being said, I am wondering why we have more than 1 routine
> to read strings... Is there something we should do?

Yes, I wondered the same when I was implementing LA_GET_STRING. I
believe we can consolidate those routines. I just added a new item to
the ProjectIdeas wiki page, under the "Internals" section.
-- 
[]'s
Thiago Jung Bauermann
IBM Linux Technology Center