[PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[Remi Bernon]

Instead of using the file offset in the debug file.

This fixes a regression from 00a3423492bc90be99e529a64f13fdd80a0e8c0a,
causing incorrect symbol resolution when debug file have been stripped
from non-debug sections (in which case its .text section is empty and
doesn't have any file position).

The debug files could also be created with a different file alignment,
and have different file positions from the mmap-ed binary, or have the
section reordered.

This instead looks for the file image base, using the corresponding bfd
*ABS* symbols. As PE symbols only have 4 bytes, it also needs to keep
.text section vma high bits.

Signed-off-by: Remi Bernon <rbernon@codeweavers.com>
---

Hi!

As I'm not updating it often I only recently realized that perf had a
regression when using stripped debug info files, and all symbols from
PE files are off. This should make things better.

Cheers,

 tools/perf/util/symbol.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 77fc46ca07c0..0fc9a5410739 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -1581,10 +1581,6 @@ int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
 	if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
 		goto out_close;
 
-	section = bfd_get_section_by_name(abfd, ".text");
-	if (section)
-		dso->text_offset = section->vma - section->filepos;
-
 	symbols_size = bfd_get_symtab_upper_bound(abfd);
 	if (symbols_size == 0) {
 		bfd_close(abfd);
@@ -1602,6 +1598,22 @@ int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
 	if (symbols_count < 0)
 		goto out_free;
 
+	section = bfd_get_section_by_name(abfd, ".text");
+	if (section) {
+		for (i = 0; i < symbols_count; ++i) {
+			if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
+			    !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
+				break;
+		}
+		if (i < symbols_count) {
+			/* PE symbols can only have 4 bytes, so use .text high bits */
+			dso->text_offset = section->vma - (u32)section->vma;
+			dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
+		} else {
+			dso->text_offset = section->vma - section->filepos;
+		}
+	}
+
 	qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
 
 #ifdef bfd_get_section
-- 
2.33.0

Re: [PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[Arnaldo Carvalho de Melo]

Em Thu, Sep 09, 2021 at 09:26:36PM +0200, Remi Bernon escreveu:
> Instead of using the file offset in the debug file.
> 
> This fixes a regression from 00a3423492bc90be99e529a64f13fdd80a0e8c0a,
> causing incorrect symbol resolution when debug file have been stripped
> from non-debug sections (in which case its .text section is empty and
> doesn't have any file position).
> 
> The debug files could also be created with a different file alignment,
> and have different file positions from the mmap-ed binary, or have the
> section reordered.
> 
> This instead looks for the file image base, using the corresponding bfd
> *ABS* symbols. As PE symbols only have 4 bytes, it also needs to keep
> .text section vma high bits.

I added a:

Fixes: 00a3423492bc90be ("perf symbols: Make dso__load_bfd_symbols() load PE files from debug cache only")

To help stable@kernel.org to pick it, its on my local tree now.

It would be great to get a:

Reviewed-by: Nicholas Fraser <nfraser@codeweavers.com> 

Can we have it, please?

- Arnaldo

> Signed-off-by: Remi Bernon <rbernon@codeweavers.com>
> ---
> 
> Hi!
> 
> As I'm not updating it often I only recently realized that perf had a
> regression when using stripped debug info files, and all symbols from
> PE files are off. This should make things better.
> 
> Cheers,
> 
>  tools/perf/util/symbol.c | 20 ++++++++++++++++----
>  1 file changed, 16 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
> index 77fc46ca07c0..0fc9a5410739 100644
> --- a/tools/perf/util/symbol.c
> +++ b/tools/perf/util/symbol.c
> @@ -1581,10 +1581,6 @@ int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
>  	if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
>  		goto out_close;
>  
> -	section = bfd_get_section_by_name(abfd, ".text");
> -	if (section)
> -		dso->text_offset = section->vma - section->filepos;
> -
>  	symbols_size = bfd_get_symtab_upper_bound(abfd);
>  	if (symbols_size == 0) {
>  		bfd_close(abfd);
> @@ -1602,6 +1598,22 @@ int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
>  	if (symbols_count < 0)
>  		goto out_free;
>  
> +	section = bfd_get_section_by_name(abfd, ".text");
> +	if (section) {
> +		for (i = 0; i < symbols_count; ++i) {
> +			if (!strcmp(bfd_asymbol_name(symbols[i]), "__ImageBase") ||
> +			    !strcmp(bfd_asymbol_name(symbols[i]), "__image_base__"))
> +				break;
> +		}
> +		if (i < symbols_count) {
> +			/* PE symbols can only have 4 bytes, so use .text high bits */
> +			dso->text_offset = section->vma - (u32)section->vma;
> +			dso->text_offset += (u32)bfd_asymbol_value(symbols[i]);
> +		} else {
> +			dso->text_offset = section->vma - section->filepos;
> +		}
> +	}
> +
>  	qsort(symbols, symbols_count, sizeof(asymbol *), bfd_symbols__cmpvalue);
>  
>  #ifdef bfd_get_section
> -- 
> 2.33.0

-- 

- Arnaldo

Re: [PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[Rémi Bernon]

On 9/9/21 10:27 PM, Arnaldo Carvalho de Melo wrote:
> Em Thu, Sep 09, 2021 at 09:26:36PM +0200, Remi Bernon escreveu:
>> Instead of using the file offset in the debug file.
>>
>> This fixes a regression from 00a3423492bc90be99e529a64f13fdd80a0e8c0a,
>> causing incorrect symbol resolution when debug file have been stripped
>> from non-debug sections (in which case its .text section is empty and
>> doesn't have any file position).
>>
>> The debug files could also be created with a different file alignment,
>> and have different file positions from the mmap-ed binary, or have the
>> section reordered.
>>
>> This instead looks for the file image base, using the corresponding bfd
>> *ABS* symbols. As PE symbols only have 4 bytes, it also needs to keep
>> .text section vma high bits.
> 
> I added a:
> 
> Fixes: 00a3423492bc90be ("perf symbols: Make dso__load_bfd_symbols() load PE files from debug cache only")
> 
> To help stable@kernel.org to pick it, its on my local tree now.
> 
> It would be great to get a:
> 
> Reviewed-by: Nicholas Fraser <nfraser@codeweavers.com>
> 
> Can we have it, please?
> 

Well, Nicholas doesn't work with us anymore. I've reached them 
separately but didn't get any answer so I'm thinking they may not be 
interested.
-- 
Rémi Bernon <rbernon@codeweavers.com>

Re: [PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[Arnaldo Carvalho de Melo]

Em Sun, Sep 12, 2021 at 10:25:26AM +0200, Rémi Bernon escreveu:
> On 9/9/21 10:27 PM, Arnaldo Carvalho de Melo wrote:
> > Em Thu, Sep 09, 2021 at 09:26:36PM +0200, Remi Bernon escreveu:
> > > Instead of using the file offset in the debug file.
> > > 
> > > This fixes a regression from 00a3423492bc90be99e529a64f13fdd80a0e8c0a,
> > > causing incorrect symbol resolution when debug file have been stripped
> > > from non-debug sections (in which case its .text section is empty and
> > > doesn't have any file position).
> > > 
> > > The debug files could also be created with a different file alignment,
> > > and have different file positions from the mmap-ed binary, or have the
> > > section reordered.
> > > 
> > > This instead looks for the file image base, using the corresponding bfd
> > > *ABS* symbols. As PE symbols only have 4 bytes, it also needs to keep
> > > .text section vma high bits.
> > 
> > I added a:
> > 
> > Fixes: 00a3423492bc90be ("perf symbols: Make dso__load_bfd_symbols() load PE files from debug cache only")
> > 
> > To help stable@kernel.org to pick it, its on my local tree now.
> > 
> > It would be great to get a:
> > 
> > Reviewed-by: Nicholas Fraser <nfraser@codeweavers.com>
> > 
> > Can we have it, please?
> > 
> 
> Well, Nicholas doesn't work with us anymore. I've reached them separately
> but didn't get any answer so I'm thinking they may not be interested.

No problem, its already upstream now.

Thanks,

- Arnaldo

Re: [PATCH] perf symbol: Look for ImageBase in PE file to compute .text offset

[Rémi Bernon]

On 9/13/21 9:24 PM, Arnaldo Carvalho de Melo wrote:
> Em Sun, Sep 12, 2021 at 10:25:26AM +0200, Rémi Bernon escreveu:
>> On 9/9/21 10:27 PM, Arnaldo Carvalho de Melo wrote:
>>> Em Thu, Sep 09, 2021 at 09:26:36PM +0200, Remi Bernon escreveu:
>>>> Instead of using the file offset in the debug file.
>>>>
>>>> This fixes a regression from 00a3423492bc90be99e529a64f13fdd80a0e8c0a,
>>>> causing incorrect symbol resolution when debug file have been stripped
>>>> from non-debug sections (in which case its .text section is empty and
>>>> doesn't have any file position).
>>>>
>>>> The debug files could also be created with a different file alignment,
>>>> and have different file positions from the mmap-ed binary, or have the
>>>> section reordered.
>>>>
>>>> This instead looks for the file image base, using the corresponding bfd
>>>> *ABS* symbols. As PE symbols only have 4 bytes, it also needs to keep
>>>> .text section vma high bits.
>>>
>>> I added a:
>>>
>>> Fixes: 00a3423492bc90be ("perf symbols: Make dso__load_bfd_symbols() load PE files from debug cache only")
>>>
>>> To help stable@kernel.org to pick it, its on my local tree now.
>>>
>>> It would be great to get a:
>>>
>>> Reviewed-by: Nicholas Fraser <nfraser@codeweavers.com>
>>>
>>> Can we have it, please?
>>>
>>
>> Well, Nicholas doesn't work with us anymore. I've reached them separately
>> but didn't get any answer so I'm thinking they may not be interested.
> 
> No problem, its already upstream now.
> 
> Thanks,
> 
> - Arnaldo
> 

Awesome, thanks!
-- 
Rémi Bernon <rbernon@codeweavers.com>