Re: [PATCH] Don't drop static function bp locations w/o debug info

[Luis Machado <luis.machado@linaro.org>]

On 9/24/20 12:16 PM, Pedro Alves wrote:
> On 9/22/20 9:20 PM, Luis Machado wrote:
>> The new tests are failing for aarch64-linux-gnu on Ubuntu 18.04 (GCC 7.5.0).
>>
>> The breakpoint location patterns don't seem to be matching. Here's what I see...
>>
>> info breakpoint^M
>> Num     Type           Disp Enb Address            What^M
>> 1       breakpoint     keep y   <MULTIPLE>         ^M
>> 1.1                         y   0x0000000000000794 <foo@plt+4>^M
>> 1.2                         y   0x00000000000008bc <foo>^M
> 
> The test is:
> 
> 	    test_info_break_2 \
> 		"<foo@plt.*>" \
> 		"<foo\\+$decimal>"
> 
> So it should be trivial to adjust with
> 
>   -	"<foo\\+$decimal>"
>   +	"<foo(\\+$decimal)?>"
> 
> 
> However, I wonder whether we're actually seeing a bug here.
> It seems like on x86, the breakpoint is set past the
> prologue (thus the "+$decimal"):
> 
>   Num     Type           Disp Enb Address            What
>   1       breakpoint     keep y   <MULTIPLE>
>   1.1                         y   0x0000000000001050 <foo@plt>
>   1.2                         y   0x0000000000001165 <foo+8>
> 
> while it is set exactly at the function's address on Aarch64.
> 
> Why didn't GDB skip the prologue of <foo> on Aarch64?
> Is there really no instruction to skip?
> 
> Here's what I see on x86-64, without debug info
>   (outputs/gdb.base/msym-bp-shl/msym-bp-shl-0):
> 
> (gdb) info breakpoints
> Num     Type           Disp Enb Address            What
> 1       breakpoint     keep y   <MULTIPLE>
> 1.1                         y   0x0000000000001050 <foo@plt>
> 1.2                         y   0x0000000000001165 <foo+8>
> 
> and on x86-64, with debug info
>   (outputs/gdb.base/msym-bp-shl/msym-bp-shl-1):
> 
> (gdb) info breakpoints
> Num     Type           Disp Enb Address            What
> 1       breakpoint     keep y   <MULTIPLE>
> 1.1                         y   0x0000000000001050 <foo@plt>
> 1.2                         y   0x0000000000001165 in foo at src/gdb/testsuite/gdb.base/msym-bp-shl-main-2.c:21
> 
> Note the location addresses are the same without vs with
> debug info: 0x1050 and 0x1165.  Is that also the case on Aarch64?

It is the same case for AArch64. The reason for not skipping a prologue 
is because there doesn't seem to be one. The generated code for aarch64 
is the following:

(gdb) disass foo
    0x00000000000008bc <+0>:     nop
    0x00000000000008c0 <+4>:     ret
End of assembler dump.

I guess the assumptions of the test are valid for x86, but they're not 
valid for aarch64.