Re: [Crash-utility] [PATCH] add -s option for struct command
At 2012-4-18 21:21, Dave Anderson wrote:
In our original discussions, i thought that I had made it clear that
the introduction of a new option paradigm with submembers could be
avoided by using, for example, "page._mapcounter" instead of having
to enter "page._mapcount.counter"? This option makes the struct
command seemingly violate its own rules, and really confuses things.
For example, with your patch, a user would see things like this:
The most important reason why I insisted this option is the performance.
Both original struct and print command are very slow. When kernel
debugger wants to parse a bit amount of data, the performance of
original struct and print command is not ideal.
crash> page._mapcount.counter ffffea0000000508 -s
-1
crash> page._mapcount.counter ffffea0000000508
struct: invalid format: page._mapcount.counter
crash> page._mapcount ffffea0000000508
_mapcount = {
counter = -1
}
crash> page._mapcount ffffea0000000508 -s
struct: invalid data structure reference page._mapcount
crash>
An idea of solving this confusion is changing the error information.
When users uses "-s" option and error happens, error information
suggests to use struct command without "-s" option if it is valid. And
vice versa, when error happens without specified "-s" option.
I had suggested that you look into the get_member_data() function
in to the gdb/symtab.c file to access the member offset and size
values.
Actually, the function need to be changed a lot to support what I want.
I need the information of submember, and I need the position and size of
bitfield. After investigation, function print_command_1 hides the data
that I want. I know it is not a good idea of modifying this function.
But what if a new function which has the similar mechanism with function
print_command_1?
I also don't like the idea of modifying the prototype of
the stalwart print_command_1() gdb function, and the creation
of a new gdb command. Whenever there is a need to update the
embedded gdb version, patches like this can be problematic.
I would prefer that you create a new "GNU_XXXX" #define,
similar to GNU_GET_SYMBOL_TYPE, pass the request through
the gdb_command_funnel switch statement, and write a new
standalone function to accomplish what you have done in the
print_command_1() function.
--
--
Regards
Qiao Nuohan