Hi Ive heard that one can not build the entire Linux kernel with -O0 option. why is that ? being a compiler developer, i can not think of reasons why that is the case. Thanks, Xin
On Wed, 30 Jul 2014 09:03:38 -0500, Xin Tong said:
Ive heard that one can not build the entire Linux kernel with -O0 option. why is that ? being a compiler developer, i can not think of reasons why that is the case.
The short answer: -O0 completely suppresses function inlining, and there are several places where the kernel depends on inlining for correct operation (most notably with things like __builtin_return_address() and friends for introspecting the stack, but there's a few other corner cases I can't remember at the moment...)
why can not __builtin_return_address() be made *never* inline and use current level+1 to get the return address of the function of interest. For any stack introspection, having 1 more level will not hurt functionality. given its explanation below — Built-in Function: void * *__builtin_return_address* (unsigned int level) This function returns the return address of the current function, or of one of its callers. The level argument is number of frames to scan up the call stack. A value of 0 yields the return address of the current function, a value of 1 yields the return address of the caller of the current function, and so forth. When inlining the expected behavior is that the function returns the address of the function that is returned to. To work around this behavior use the noinline function attribute. On Wed, Jul 30, 2014 at 11:18 AM, <Valdis.Kletnieks@vt.edu> wrote:
On Wed, 30 Jul 2014 09:03:38 -0500, Xin Tong said:
Ive heard that one can not build the entire Linux kernel with -O0 option. why is that ? being a compiler developer, i can not think of reasons why that is the case.
The short answer: -O0 completely suppresses function inlining, and there are several places where the kernel depends on inlining for correct operation (most notably with things like __builtin_return_address() and friends for introspecting the stack, but there's a few other corner cases I can't remember at the moment...)
On Thu, Jul 31, 2014 at 12:59 AM, Xin Tong <trent.tong@gmail.com> wrote:
why can not __builtin_return_address() be made *never* inline and use current level+1 to get the return address of the function of interest. For any stack introspection, having 1 more level will not hurt functionality.
Actually, the answer for your remark is "impossible" - in the case when the kernel is compiled without frame pointer. (CONFIG_FRAME_POINTER=n) which is true for certain variant of RHEL / CentOS. Without the availability of EBP on the stack, there is no way to know when to stop reading the stack to retrieve the previous stackframe. Of course u can statically walk the disassembly of the function and see how much stack space the particular function has allocated. But that requires implementing a disassembler in the kernel.
given its explanation below
— Built-in Function: void * *__builtin_return_address* (unsigned int level )
This function returns the return address of the current function, or of one of its callers. The level argument is number of frames to scan up the call stack. A value of 0 yields the return address of the current function, a value of 1 yields the return address of the caller of the current function, and so forth. When inlining the expected behavior is that the function returns the address of the function that is returned to. To work around this behavior use the noinline function attribute.
-- Regards, Peter Teoh
In that case, the __builtin_return_address(level) level > 1 is not possible either ? what if the kernel uses this ? Xin On Wed, Jul 30, 2014 at 8:00 PM, Peter Teoh <htmldeveloper@gmail.com> wrote:
On Thu, Jul 31, 2014 at 12:59 AM, Xin Tong <trent.tong@gmail.com> wrote:
why can not __builtin_return_address() be made *never* inline and use current level+1 to get the return address of the function of interest. For any stack introspection, having 1 more level will not hurt functionality.
Actually, the answer for your remark is "impossible" - in the case when the kernel is compiled without frame pointer. (CONFIG_FRAME_POINTER=n) which is true for certain variant of RHEL / CentOS. Without the availability of EBP on the stack, there is no way to know when to stop reading the stack to retrieve the previous stackframe. Of course u can statically walk the disassembly of the function and see how much stack space the particular function has allocated. But that requires implementing a disassembler in the kernel.
given its explanation below
— Built-in Function: void * *__builtin_return_address* (unsigned int level)
This function returns the return address of the current function, or of one of its callers. The level argument is number of frames to scan up the call stack. A value of 0 yields the return address of the current function, a value of 1 yields the return address of the caller of the current function, and so forth. When inlining the expected behavior is that the function returns the address of the function that is returned to. To work around this behavior use the noinline function attribute.
-- Regards, Peter Teoh
If function is built with framepointer, then EBP + 4 == return address of the caller of then present function. Because by convention, the entire function usually don't touch the EBP's value, so with respect to that, u can always retrieve the return address of the caller. (which is what this function does). and u ask if is not compiled inline? Then __builtin_return_address() become a function itself? Then u are getting the caller of "__builtin_return_address". That was not the original intention. Its purpose is to get the caller address of the current function. On Thu, Jul 31, 2014 at 9:05 AM, Xin Tong <trent.tong@gmail.com> wrote:
In that case, the __builtin_return_address(level) level > 1 is not possible either ? what if the kernel uses this ?
Xin
On Wed, Jul 30, 2014 at 8:00 PM, Peter Teoh <htmldeveloper@gmail.com> wrote:
On Thu, Jul 31, 2014 at 12:59 AM, Xin Tong <trent.tong@gmail.com> wrote:
why can not __builtin_return_address() be made *never* inline and use current level+1 to get the return address of the function of interest. For any stack introspection, having 1 more level will not hurt functionality.
Actually, the answer for your remark is "impossible" - in the case when the kernel is compiled without frame pointer. (CONFIG_FRAME_POINTER=n) which is true for certain variant of RHEL / CentOS. Without the availability of EBP on the stack, there is no way to know when to stop reading the stack to retrieve the previous stackframe. Of course u can statically walk the disassembly of the function and see how much stack space the particular function has allocated. But that requires implementing a disassembler in the kernel.
given its explanation below
— Built-in Function: void * *__builtin_return_address* (unsigned int level)
This function returns the return address of the current function, or of one of its callers. The level argument is number of frames to scan up the call stack. A value of 0 yields the return address of the current function, a value of 1 yields the return address of the caller of the current function, and so forth. When inlining the expected behavior is that the function returns the address of the function that is returned to. To work around this behavior use the noinline function attribute.
-- Regards, Peter Teoh
-- Regards, Peter Teoh
participants (3)
-
Peter Teoh -
Valdis.Kletnieks＠vt.edu -
Xin Tong