Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Wei-min Pan]

On 11/2/2017 8:32 AM, Weimin Pan wrote:
> Running the test case with upstream gdb shows two failures:
>
> (1) Receiving different error messages when printing TLS variable before
>      program runs - because the ARM compiler does not emit dwarf attribute
>      DW_AT_location for TLS, the result is expected and the baseline may
>      need to be changed for aarch64.
>
> (2) Using "info address" command on C++ static TLS object resulted in
>      "symbol unresolved" error - below is a snippet from the test case:
>
> class K {
>   public:
>    static __thread int another_thread_local;
> };
>
> __thread int K::another_thread_local;
>
> (gdb) info address K::another_thread_local
> Symbol "K::another_thread_local" is unresolved.
>
> This patch contains fix for (2).
>
> Function info_address_command() handles the "info address" command and
> calls lookup_minimal_symbol_and_objfile() to find sym's symbol entry in
> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
> that function lookup_minimal_symbol_and_objfile() only looked up an
> objfile's minsym ordinary hash table, not its demangled hash table, which
> was the reason why the C++ name was not found.
>
> The fix is to call lookup_minimal_symbol(), which already looks up entries
> in both minsym's hash tables, to find names when traversing the object file
> list in lookup_minimal_symbol_and_objfile().
>
> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
> ---
>   gdb/ChangeLog |    5 +++++
>   gdb/minsyms.c |   17 +++--------------
>   2 files changed, 8 insertions(+), 14 deletions(-)
>
> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
> index 4b292e0..2f630bc 100644
> --- a/gdb/ChangeLog
> +++ b/gdb/ChangeLog
> @@ -1,3 +1,8 @@
> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
> +
> +	* minsyms.c (lookup_minimal_symbol_and_objfile): Use
> +	lookup_minimal_symbol() to find symbol entry.
> +
>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>   
>   	* breakpoint.c (print_breakpoint_location): Use the symbol saved
> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
> index 37edbd8..4edd8b1 100644
> --- a/gdb/minsyms.c
> +++ b/gdb/minsyms.c
> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const char *name)
>   {
>     struct bound_minimal_symbol result;
>     struct objfile *objfile;
> -  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
>   
>     ALL_OBJFILES (objfile)
>       {
> -      struct minimal_symbol *msym;
> -
> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
> -	   msym != NULL;
> -	   msym = msym->hash_next)
> -	{
> -	  if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
> -	    {
> -	      result.minsym = msym;
> -	      result.objfile = objfile;
> -	      return result;
> -	    }
> -	}
> +      result = lookup_minimal_symbol (name, NULL, objfile);
> +      if (result.minsym != NULL)
> +        return result;
>       }
>   
>     memset (&result, 0, sizeof (result));

Re: [PING 2][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

On 11/15/2017 5:51 PM, Wei-min Pan wrote:
>
> On 11/2/2017 8:32 AM, Weimin Pan wrote:
>> Running the test case with upstream gdb shows two failures:
>>
>> (1) Receiving different error messages when printing TLS variable before
>>      program runs - because the ARM compiler does not emit dwarf 
>> attribute
>>      DW_AT_location for TLS, the result is expected and the baseline may
>>      need to be changed for aarch64.
>>
>> (2) Using "info address" command on C++ static TLS object resulted in
>>      "symbol unresolved" error - below is a snippet from the test case:
>>
>> class K {
>>   public:
>>    static __thread int another_thread_local;
>> };
>>
>> __thread int K::another_thread_local;
>>
>> (gdb) info address K::another_thread_local
>> Symbol "K::another_thread_local" is unresolved.
>>
>> This patch contains fix for (2).
>>
>> Function info_address_command() handles the "info address" command and
>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol entry in
>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
>> that function lookup_minimal_symbol_and_objfile() only looked up an
>> objfile's minsym ordinary hash table, not its demangled hash table, 
>> which
>> was the reason why the C++ name was not found.
>>
>> The fix is to call lookup_minimal_symbol(), which already looks up 
>> entries
>> in both minsym's hash tables, to find names when traversing the 
>> object file
>> list in lookup_minimal_symbol_and_objfile().
>>
>> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
>> ---
>>   gdb/ChangeLog |    5 +++++
>>   gdb/minsyms.c |   17 +++--------------
>>   2 files changed, 8 insertions(+), 14 deletions(-)
>>
>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>> index 4b292e0..2f630bc 100644
>> --- a/gdb/ChangeLog
>> +++ b/gdb/ChangeLog
>> @@ -1,3 +1,8 @@
>> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
>> +
>> +    * minsyms.c (lookup_minimal_symbol_and_objfile): Use
>> +    lookup_minimal_symbol() to find symbol entry.
>> +
>>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>>         * breakpoint.c (print_breakpoint_location): Use the symbol saved
>> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
>> index 37edbd8..4edd8b1 100644
>> --- a/gdb/minsyms.c
>> +++ b/gdb/minsyms.c
>> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const char 
>> *name)
>>   {
>>     struct bound_minimal_symbol result;
>>     struct objfile *objfile;
>> -  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
>>       ALL_OBJFILES (objfile)
>>       {
>> -      struct minimal_symbol *msym;
>> -
>> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
>> -       msym != NULL;
>> -       msym = msym->hash_next)
>> -    {
>> -      if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
>> -        {
>> -          result.minsym = msym;
>> -          result.objfile = objfile;
>> -          return result;
>> -        }
>> -    }
>> +      result = lookup_minimal_symbol (name, NULL, objfile);
>> +      if (result.minsym != NULL)
>> +        return result;
>>       }
>>       memset (&result, 0, sizeof (result));
>

Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Yao Qi]

Wei-min Pan <weimin.pan@oracle.com> writes:

>> Function info_address_command() handles the "info address" command and
>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol entry in
>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
>> that function lookup_minimal_symbol_and_objfile() only looked up an
>> objfile's minsym ordinary hash table, not its demangled hash table, which
>> was the reason why the C++ name was not found.

lookup_minimal_symbol_and_objfile is documented as "only checks the
linkage name" in minsyms.h,

/* Find the minimal symbol named NAME, and return both the minsym
   struct and its objfile.  This only checks the linkage name.  */

struct bound_minimal_symbol lookup_minimal_symbol_and_objfile (const char *);

>>
>> The fix is to call lookup_minimal_symbol(), which already looks up entries
>> in both minsym's hash tables, to find names when traversing the object file
>> list in lookup_minimal_symbol_and_objfile().

so, it is incorrect to extend it to search demangled name.  If I set a
breakpoint on lookup_minimal_symbol_and_objfile,

(gdb) info address K::another_thread_local

Breakpoint 1, lookup_minimal_symbol_and_objfile (name=0x621000136e40 "K::another_thread_local") at gdb/minsyms.c:1012
1012	  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
(gdb) up
#1  0x0000000000af7690 in info_address_command (exp=0x604000023f5d "K::another_thread_local", from_tty=1)
    at gdb/printcmd.c:1567
1567		msym = lookup_minimal_symbol_and_objfile (SYMBOL_LINKAGE_NAME (sym));

The problem to me is that why SYMBOL_LINKAGE_NAME (sym) is
"K::another_thread_local", which is a demangled one.  The symbol's name
is set in dwarf2read.c:new_symbol_full,

      /* Cache this symbol's name and the name's demangled form (if any).  */
      SYMBOL_SET_LANGUAGE (sym, cu->language, &objfile->objfile_obstack);
      linkagename = dwarf2_physname (name, die, cu);
      SYMBOL_SET_NAMES (sym, linkagename, strlen (linkagename), 0, objfile);

however, dwarf2_physname doesn't return the linkage name, so the
symbol's linkagename is set incorrectly.  I think the right fix would be
call other function to get linkagename, maybe, dw2_linkage_name?  I
don't know.

-- 
Yao (齐尧)

Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

On 1/16/2018 9:12 AM, Yao Qi wrote:
> Wei-min Pan <weimin.pan@oracle.com> writes:
>
>>> Function info_address_command() handles the "info address" command and
>>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol entry in
>>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
>>> that function lookup_minimal_symbol_and_objfile() only looked up an
>>> objfile's minsym ordinary hash table, not its demangled hash table, which
>>> was the reason why the C++ name was not found.
> lookup_minimal_symbol_and_objfile is documented as "only checks the
> linkage name" in minsyms.h,
>
> /* Find the minimal symbol named NAME, and return both the minsym
>     struct and its objfile.  This only checks the linkage name.  */
>
> struct bound_minimal_symbol lookup_minimal_symbol_and_objfile (const char *);
>
>>> The fix is to call lookup_minimal_symbol(), which already looks up entries
>>> in both minsym's hash tables, to find names when traversing the object file
>>> list in lookup_minimal_symbol_and_objfile().
> so, it is incorrect to extend it to search demangled name.  If I set a
> breakpoint on lookup_minimal_symbol_and_objfile,
>
> (gdb) info address K::another_thread_local
>
> Breakpoint 1, lookup_minimal_symbol_and_objfile (name=0x621000136e40 "K::another_thread_local") at gdb/minsyms.c:1012
> 1012	  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
> (gdb) up
> #1  0x0000000000af7690 in info_address_command (exp=0x604000023f5d "K::another_thread_local", from_tty=1)
>      at gdb/printcmd.c:1567
> 1567		msym = lookup_minimal_symbol_and_objfile (SYMBOL_LINKAGE_NAME (sym));
>
> The problem to me is that why SYMBOL_LINKAGE_NAME (sym) is
> "K::another_thread_local", which is a demangled one.  The symbol's name
> is set in dwarf2read.c:new_symbol_full,
>
>        /* Cache this symbol's name and the name's demangled form (if any).  */
>        SYMBOL_SET_LANGUAGE (sym, cu->language, &objfile->objfile_obstack);
>        linkagename = dwarf2_physname (name, die, cu);
>        SYMBOL_SET_NAMES (sym, linkagename, strlen (linkagename), 0, objfile);
>
> however, dwarf2_physname doesn't return the linkage name, so the
> symbol's linkagename is set incorrectly.  I think the right fix would be
> call other function to get linkagename, maybe, dw2_linkage_name?  I
> don't know.
dwarf2_physname() does call dw2_linkage_name(). But since the class member,
i.e.  another_thread_local, does not contain either DW_AT_linkage_name or
DW_AT_MIPS_linkage_name attribute, its canonicalized name gets returned.

Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Yao Qi]

On Wed, Jan 17, 2018 at 1:05 AM, Weimin Pan <weimin.pan@oracle.com> wrote:
>
> dwarf2_physname() does call dw2_linkage_name(). But since the class member,
> i.e.  another_thread_local, does not contain either DW_AT_linkage_name or
> DW_AT_MIPS_linkage_name attribute, its canonicalized name gets returned.
>

dwarf2_physname is called twice with the parameter NAME is
"another_thread_local".
The first is about DW_TAG_member and the second is about DW_TAG_variable.

 <2><37>: Abbrev Number: 3 (DW_TAG_member)
    <38>   DW_AT_name        : (indirect string, offset: 0x6e):
another_thread_local
    <3c>   DW_AT_decl_file   : 1
    <3d>   DW_AT_decl_line   : 3
    <3e>   DW_AT_type        : <0x44>
    <42>   DW_AT_external    : 1
    <42>   DW_AT_accessibility: 1 (public)
    <43>   DW_AT_declaration : 1
....
 <1><68>: Abbrev Number: 6 (DW_TAG_variable)
    <69>   DW_AT_specification: <0x37>
    <6d>   DW_AT_decl_line   : 6
    <6e>   DW_AT_linkage_name: (indirect string, offset: 0x0):
_ZN1K20another_thread_localE

The second one matters, not the first one.  If I set this breakpoint,
it will hit twice,

(gdb) b dwarf2_physname if strcmp (name, "another_thread_local") == 0

in the 2nd breakpoint hit, you can see dwarf2_physname is called by
new_symbol_full,

(gdb) bt 5
#0  dwarf2_physname (name=0x60d0000096fe "another_thread_local",
die=0x621000156ce0, cu=0x612000018340)
    at gdb/dwarf2read.c:11183
#1  0x00000000008f20a9 in new_symbol_full (die=0x621000156ce0,
type=0x0, cu=0x612000018340, space=0x0)
    at gdb/dwarf2read.c:21416
#2  0x00000000008c80a6 in read_variable (die=0x621000156ce0,
cu=0x612000018340) at gdb/dwarf2read.c:14439

it returns "K::another_thread_local", and use it to set symbol name.
That symbol
is used in lookup_minimal_symbol_and_objfile (SYMBOL_LINKAGE_NAME (sym))
later when you type command "info address K::another_thread_local".

-- 
Yao (齐尧)

Re: [PING][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Wei-min Pan]

On 1/17/2018 12:55 AM, Yao Qi wrote:
> On Wed, Jan 17, 2018 at 1:05 AM, Weimin Pan <weimin.pan@oracle.com> wrote:
>> dwarf2_physname() does call dw2_linkage_name(). But since the class member,
>> i.e.  another_thread_local, does not contain either DW_AT_linkage_name or
>> DW_AT_MIPS_linkage_name attribute, its canonicalized name gets returned.
>>
> dwarf2_physname is called twice with the parameter NAME is
> "another_thread_local".
> The first is about DW_TAG_member and the second is about DW_TAG_variable.
>
>   <2><37>: Abbrev Number: 3 (DW_TAG_member)
>      <38>   DW_AT_name        : (indirect string, offset: 0x6e):
> another_thread_local
>      <3c>   DW_AT_decl_file   : 1
>      <3d>   DW_AT_decl_line   : 3
>      <3e>   DW_AT_type        : <0x44>
>      <42>   DW_AT_external    : 1
>      <42>   DW_AT_accessibility: 1 (public)
>      <43>   DW_AT_declaration : 1
> ....
>   <1><68>: Abbrev Number: 6 (DW_TAG_variable)
>      <69>   DW_AT_specification: <0x37>
>      <6d>   DW_AT_decl_line   : 6
>      <6e>   DW_AT_linkage_name: (indirect string, offset: 0x0):
> _ZN1K20another_thread_localE
>
> The second one matters, not the first one.  If I set this breakpoint,
> it will hit twice,
>
> (gdb) b dwarf2_physname if strcmp (name, "another_thread_local") == 0
>
> in the 2nd breakpoint hit, you can see dwarf2_physname is called by
> new_symbol_full,
>
> (gdb) bt 5
> #0  dwarf2_physname (name=0x60d0000096fe "another_thread_local",
> die=0x621000156ce0, cu=0x612000018340)
>      at gdb/dwarf2read.c:11183
> #1  0x00000000008f20a9 in new_symbol_full (die=0x621000156ce0,
> type=0x0, cu=0x612000018340, space=0x0)
>      at gdb/dwarf2read.c:21416
> #2  0x00000000008c80a6 in read_variable (die=0x621000156ce0,
> cu=0x612000018340) at gdb/dwarf2read.c:14439
>
> it returns "K::another_thread_local", and use it to set symbol name.
> That symbol
> is used in lookup_minimal_symbol_and_objfile (SYMBOL_LINKAGE_NAME (sym))
> later when you type command "info address K::another_thread_local".

Yes, the 2nd breakpoint hit in dwarf2_physname() was to handle
the DW_TAG_variable die for another_thread_local. The function
did fetch the correct mangled name from dw2_linkage_name() and
set it in local variable "mangled". Problem is "mangled" got
demangled with gdb_demangle(), for the purpose of suppressing
C++ template function's return type which I don't quite follow,
and the demangled name gets returned.

Re: [PING 3][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

While this problem shows up in aarch64, the fix is actually
in gdb core. Pedro, would you please review the patch when
you get a chance?

Thanks.

On 1/11/2018 4:32 PM, Wei-min Pan wrote:
>
> On 11/15/2017 5:51 PM, Wei-min Pan wrote:
>>
>> On 11/2/2017 8:32 AM, Weimin Pan wrote:
>>> Running the test case with upstream gdb shows two failures:
>>>
>>> (1) Receiving different error messages when printing TLS variable 
>>> before
>>>      program runs - because the ARM compiler does not emit dwarf 
>>> attribute
>>>      DW_AT_location for TLS, the result is expected and the baseline 
>>> may
>>>      need to be changed for aarch64.
>>>
>>> (2) Using "info address" command on C++ static TLS object resulted in
>>>      "symbol unresolved" error - below is a snippet from the test case:
>>>
>>> class K {
>>>   public:
>>>    static __thread int another_thread_local;
>>> };
>>>
>>> __thread int K::another_thread_local;
>>>
>>> (gdb) info address K::another_thread_local
>>> Symbol "K::another_thread_local" is unresolved.
>>>
>>> This patch contains fix for (2).
>>>
>>> Function info_address_command() handles the "info address" command and
>>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol entry in
>>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
>>> that function lookup_minimal_symbol_and_objfile() only looked up an
>>> objfile's minsym ordinary hash table, not its demangled hash table, 
>>> which
>>> was the reason why the C++ name was not found.
>>>
>>> The fix is to call lookup_minimal_symbol(), which already looks up 
>>> entries
>>> in both minsym's hash tables, to find names when traversing the 
>>> object file
>>> list in lookup_minimal_symbol_and_objfile().
>>>
>>> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
>>> ---
>>>   gdb/ChangeLog |    5 +++++
>>>   gdb/minsyms.c |   17 +++--------------
>>>   2 files changed, 8 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>>> index 4b292e0..2f630bc 100644
>>> --- a/gdb/ChangeLog
>>> +++ b/gdb/ChangeLog
>>> @@ -1,3 +1,8 @@
>>> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
>>> +
>>> +    * minsyms.c (lookup_minimal_symbol_and_objfile): Use
>>> +    lookup_minimal_symbol() to find symbol entry.
>>> +
>>>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>>>         * breakpoint.c (print_breakpoint_location): Use the symbol 
>>> saved
>>> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
>>> index 37edbd8..4edd8b1 100644
>>> --- a/gdb/minsyms.c
>>> +++ b/gdb/minsyms.c
>>> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const char 
>>> *name)
>>>   {
>>>     struct bound_minimal_symbol result;
>>>     struct objfile *objfile;
>>> -  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
>>>       ALL_OBJFILES (objfile)
>>>       {
>>> -      struct minimal_symbol *msym;
>>> -
>>> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
>>> -       msym != NULL;
>>> -       msym = msym->hash_next)
>>> -    {
>>> -      if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
>>> -        {
>>> -          result.minsym = msym;
>>> -          result.objfile = objfile;
>>> -          return result;
>>> -        }
>>> -    }
>>> +      result = lookup_minimal_symbol (name, NULL, objfile);
>>> +      if (result.minsym != NULL)
>>> +        return result;
>>>       }
>>>       memset (&result, 0, sizeof (result));
>>
>

Re: [PING 4][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

On 1/23/2018 5:11 PM, Weimin Pan wrote:
> While this problem shows up in aarch64, the fix is actually
> in gdb core. Pedro, would you please review the patch when
> you get a chance?
>
> Thanks.
>
> On 1/11/2018 4:32 PM, Wei-min Pan wrote:
>>
>> On 11/15/2017 5:51 PM, Wei-min Pan wrote:
>>>
>>> On 11/2/2017 8:32 AM, Weimin Pan wrote:
>>>> Running the test case with upstream gdb shows two failures:
>>>>
>>>> (1) Receiving different error messages when printing TLS variable 
>>>> before
>>>>      program runs - because the ARM compiler does not emit dwarf 
>>>> attribute
>>>>      DW_AT_location for TLS, the result is expected and the 
>>>> baseline may
>>>>      need to be changed for aarch64.
>>>>
>>>> (2) Using "info address" command on C++ static TLS object resulted in
>>>>      "symbol unresolved" error - below is a snippet from the test 
>>>> case:
>>>>
>>>> class K {
>>>>   public:
>>>>    static __thread int another_thread_local;
>>>> };
>>>>
>>>> __thread int K::another_thread_local;
>>>>
>>>> (gdb) info address K::another_thread_local
>>>> Symbol "K::another_thread_local" is unresolved.
>>>>
>>>> This patch contains fix for (2).
>>>>
>>>> Function info_address_command() handles the "info address" command and
>>>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol 
>>>> entry in
>>>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. Problem is
>>>> that function lookup_minimal_symbol_and_objfile() only looked up an
>>>> objfile's minsym ordinary hash table, not its demangled hash table, 
>>>> which
>>>> was the reason why the C++ name was not found.
>>>>
>>>> The fix is to call lookup_minimal_symbol(), which already looks up 
>>>> entries
>>>> in both minsym's hash tables, to find names when traversing the 
>>>> object file
>>>> list in lookup_minimal_symbol_and_objfile().
>>>>
>>>> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
>>>> ---
>>>>   gdb/ChangeLog |    5 +++++
>>>>   gdb/minsyms.c |   17 +++--------------
>>>>   2 files changed, 8 insertions(+), 14 deletions(-)
>>>>
>>>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>>>> index 4b292e0..2f630bc 100644
>>>> --- a/gdb/ChangeLog
>>>> +++ b/gdb/ChangeLog
>>>> @@ -1,3 +1,8 @@
>>>> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
>>>> +
>>>> +    * minsyms.c (lookup_minimal_symbol_and_objfile): Use
>>>> +    lookup_minimal_symbol() to find symbol entry.
>>>> +
>>>>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>>>>         * breakpoint.c (print_breakpoint_location): Use the symbol 
>>>> saved
>>>> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
>>>> index 37edbd8..4edd8b1 100644
>>>> --- a/gdb/minsyms.c
>>>> +++ b/gdb/minsyms.c
>>>> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const char 
>>>> *name)
>>>>   {
>>>>     struct bound_minimal_symbol result;
>>>>     struct objfile *objfile;
>>>> -  unsigned int hash = msymbol_hash (name) % MINIMAL_SYMBOL_HASH_SIZE;
>>>>       ALL_OBJFILES (objfile)
>>>>       {
>>>> -      struct minimal_symbol *msym;
>>>> -
>>>> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
>>>> -       msym != NULL;
>>>> -       msym = msym->hash_next)
>>>> -    {
>>>> -      if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
>>>> -        {
>>>> -          result.minsym = msym;
>>>> -          result.objfile = objfile;
>>>> -          return result;
>>>> -        }
>>>> -    }
>>>> +      result = lookup_minimal_symbol (name, NULL, objfile);
>>>> +      if (result.minsym != NULL)
>>>> +        return result;
>>>>       }
>>>>       memset (&result, 0, sizeof (result));
>>>
>>
>

Re: [PING 5][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

On 2/5/2018 5:21 PM, Weimin Pan wrote:
> On 1/23/2018 5:11 PM, Weimin Pan wrote:
>> While this problem shows up in aarch64, the fix is actually
>> in gdb core. Pedro, would you please review the patch when
>> you get a chance?
>>
>> Thanks.
>>
>> On 1/11/2018 4:32 PM, Wei-min Pan wrote:
>>>
>>> On 11/15/2017 5:51 PM, Wei-min Pan wrote:
>>>>
>>>> On 11/2/2017 8:32 AM, Weimin Pan wrote:
>>>>> Running the test case with upstream gdb shows two failures:
>>>>>
>>>>> (1) Receiving different error messages when printing TLS variable 
>>>>> before
>>>>>      program runs - because the ARM compiler does not emit dwarf 
>>>>> attribute
>>>>>      DW_AT_location for TLS, the result is expected and the 
>>>>> baseline may
>>>>>      need to be changed for aarch64.
>>>>>
>>>>> (2) Using "info address" command on C++ static TLS object resulted in
>>>>>      "symbol unresolved" error - below is a snippet from the test 
>>>>> case:
>>>>>
>>>>> class K {
>>>>>   public:
>>>>>    static __thread int another_thread_local;
>>>>> };
>>>>>
>>>>> __thread int K::another_thread_local;
>>>>>
>>>>> (gdb) info address K::another_thread_local
>>>>> Symbol "K::another_thread_local" is unresolved.
>>>>>
>>>>> This patch contains fix for (2).
>>>>>
>>>>> Function info_address_command() handles the "info address" command 
>>>>> and
>>>>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol 
>>>>> entry in
>>>>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. 
>>>>> Problem is
>>>>> that function lookup_minimal_symbol_and_objfile() only looked up an
>>>>> objfile's minsym ordinary hash table, not its demangled hash 
>>>>> table, which
>>>>> was the reason why the C++ name was not found.
>>>>>
>>>>> The fix is to call lookup_minimal_symbol(), which already looks up 
>>>>> entries
>>>>> in both minsym's hash tables, to find names when traversing the 
>>>>> object file
>>>>> list in lookup_minimal_symbol_and_objfile().
>>>>>
>>>>> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
>>>>> ---
>>>>>   gdb/ChangeLog |    5 +++++
>>>>>   gdb/minsyms.c |   17 +++--------------
>>>>>   2 files changed, 8 insertions(+), 14 deletions(-)
>>>>>
>>>>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>>>>> index 4b292e0..2f630bc 100644
>>>>> --- a/gdb/ChangeLog
>>>>> +++ b/gdb/ChangeLog
>>>>> @@ -1,3 +1,8 @@
>>>>> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
>>>>> +
>>>>> +    * minsyms.c (lookup_minimal_symbol_and_objfile): Use
>>>>> +    lookup_minimal_symbol() to find symbol entry.
>>>>> +
>>>>>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>>>>>         * breakpoint.c (print_breakpoint_location): Use the symbol 
>>>>> saved
>>>>> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
>>>>> index 37edbd8..4edd8b1 100644
>>>>> --- a/gdb/minsyms.c
>>>>> +++ b/gdb/minsyms.c
>>>>> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const 
>>>>> char *name)
>>>>>   {
>>>>>     struct bound_minimal_symbol result;
>>>>>     struct objfile *objfile;
>>>>> -  unsigned int hash = msymbol_hash (name) % 
>>>>> MINIMAL_SYMBOL_HASH_SIZE;
>>>>>       ALL_OBJFILES (objfile)
>>>>>       {
>>>>> -      struct minimal_symbol *msym;
>>>>> -
>>>>> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
>>>>> -       msym != NULL;
>>>>> -       msym = msym->hash_next)
>>>>> -    {
>>>>> -      if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
>>>>> -        {
>>>>> -          result.minsym = msym;
>>>>> -          result.objfile = objfile;
>>>>> -          return result;
>>>>> -        }
>>>>> -    }
>>>>> +      result = lookup_minimal_symbol (name, NULL, objfile);
>>>>> +      if (result.minsym != NULL)
>>>>> +        return result;
>>>>>       }
>>>>>       memset (&result, 0, sizeof (result));
>>>>
>>>
>>
>

Re: [PING 6][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Weimin Pan]

On 3/5/2018 5:19 PM, Weimin Pan wrote:
>
>
> On 2/5/2018 5:21 PM, Weimin Pan wrote:
>> On 1/23/2018 5:11 PM, Weimin Pan wrote:
>>> While this problem shows up in aarch64, the fix is actually
>>> in gdb core. Pedro, would you please review the patch when
>>> you get a chance?
>>>
>>> Thanks.
>>>
>>> On 1/11/2018 4:32 PM, Wei-min Pan wrote:
>>>>
>>>> On 11/15/2017 5:51 PM, Wei-min Pan wrote:
>>>>>
>>>>> On 11/2/2017 8:32 AM, Weimin Pan wrote:
>>>>>> Running the test case with upstream gdb shows two failures:
>>>>>>
>>>>>> (1) Receiving different error messages when printing TLS variable 
>>>>>> before
>>>>>>      program runs - because the ARM compiler does not emit dwarf 
>>>>>> attribute
>>>>>>      DW_AT_location for TLS, the result is expected and the 
>>>>>> baseline may
>>>>>>      need to be changed for aarch64.
>>>>>>
>>>>>> (2) Using "info address" command on C++ static TLS object 
>>>>>> resulted in
>>>>>>      "symbol unresolved" error - below is a snippet from the test 
>>>>>> case:
>>>>>>
>>>>>> class K {
>>>>>>   public:
>>>>>>    static __thread int another_thread_local;
>>>>>> };
>>>>>>
>>>>>> __thread int K::another_thread_local;
>>>>>>
>>>>>> (gdb) info address K::another_thread_local
>>>>>> Symbol "K::another_thread_local" is unresolved.
>>>>>>
>>>>>> This patch contains fix for (2).
>>>>>>
>>>>>> Function info_address_command() handles the "info address" 
>>>>>> command and
>>>>>> calls lookup_minimal_symbol_and_objfile() to find sym's symbol 
>>>>>> entry in
>>>>>> mininal symbol table if SYMBOL_COMPUTED_OPS (sym) is false. 
>>>>>> Problem is
>>>>>> that function lookup_minimal_symbol_and_objfile() only looked up an
>>>>>> objfile's minsym ordinary hash table, not its demangled hash 
>>>>>> table, which
>>>>>> was the reason why the C++ name was not found.
>>>>>>
>>>>>> The fix is to call lookup_minimal_symbol(), which already looks 
>>>>>> up entries
>>>>>> in both minsym's hash tables, to find names when traversing the 
>>>>>> object file
>>>>>> list in lookup_minimal_symbol_and_objfile().
>>>>>>
>>>>>> Tested in both aarch64-linux-gnu and amd64-linux-gnu. No 
>>>>>> regressions.
>>>>>> ---
>>>>>>   gdb/ChangeLog |    5 +++++
>>>>>>   gdb/minsyms.c |   17 +++--------------
>>>>>>   2 files changed, 8 insertions(+), 14 deletions(-)
>>>>>>
>>>>>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>>>>>> index 4b292e0..2f630bc 100644
>>>>>> --- a/gdb/ChangeLog
>>>>>> +++ b/gdb/ChangeLog
>>>>>> @@ -1,3 +1,8 @@
>>>>>> +2017-11-01  Weimin Pan  <weimin.pan@oracle.com>
>>>>>> +
>>>>>> +    * minsyms.c (lookup_minimal_symbol_and_objfile): Use
>>>>>> +    lookup_minimal_symbol() to find symbol entry.
>>>>>> +
>>>>>>   2017-10-27  Keith Seitz  <keiths@redhat.com>
>>>>>>         * breakpoint.c (print_breakpoint_location): Use the 
>>>>>> symbol saved
>>>>>> diff --git a/gdb/minsyms.c b/gdb/minsyms.c
>>>>>> index 37edbd8..4edd8b1 100644
>>>>>> --- a/gdb/minsyms.c
>>>>>> +++ b/gdb/minsyms.c
>>>>>> @@ -881,23 +881,12 @@ lookup_minimal_symbol_and_objfile (const 
>>>>>> char *name)
>>>>>>   {
>>>>>>     struct bound_minimal_symbol result;
>>>>>>     struct objfile *objfile;
>>>>>> -  unsigned int hash = msymbol_hash (name) % 
>>>>>> MINIMAL_SYMBOL_HASH_SIZE;
>>>>>>       ALL_OBJFILES (objfile)
>>>>>>       {
>>>>>> -      struct minimal_symbol *msym;
>>>>>> -
>>>>>> -      for (msym = objfile->per_bfd->msymbol_hash[hash];
>>>>>> -       msym != NULL;
>>>>>> -       msym = msym->hash_next)
>>>>>> -    {
>>>>>> -      if (strcmp (MSYMBOL_LINKAGE_NAME (msym), name) == 0)
>>>>>> -        {
>>>>>> -          result.minsym = msym;
>>>>>> -          result.objfile = objfile;
>>>>>> -          return result;
>>>>>> -        }
>>>>>> -    }
>>>>>> +      result = lookup_minimal_symbol (name, NULL, objfile);
>>>>>> +      if (result.minsym != NULL)
>>>>>> +        return result;
>>>>>>       }
>>>>>>       memset (&result, 0, sizeof (result));
>>>>>
>>>>
>>>
>>
>

Re: [PING 6][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Simon Marchi]

Hi Weimin,

This is just to tell you that I started looking into this, since Yao 
will likely not have the time to do it (though I don't mind if he 
does!).  But it might take a bit of time since I don't know much about 
the particular case of TLS on aarch64, so I'll need to ramp up on it.

Simon

Re: [PING 6][PATCH PR gdb/18071] TLS variables can't be resolved on aarch64-linux-gnu

[Wei-min Pan]

Hi Simon,

Thanks for taking this up. Yao did take a look at the patch late last 
year but didn't have time to finish it up.
Please let me know if you have comments/questions.

Weimin

On 3/14/2018 2:06 PM, Simon Marchi wrote:
> Hi Weimin,
>
> This is just to tell you that I started looking into this, since Yao 
> will likely not have the time to do it (though I don't mind if he 
> does!).  But it might take a bit of time since I don't know much about 
> the particular case of TLS on aarch64, so I'll need to ramp up on it.
>
> Simon