Fix behavior around jumps with immediate values > 0x7fffffff in the interpreter #447
Conversation
Signed-off-by: Alan Jowett <[email protected]>
Signed-off-by: Alan Jowett <[email protected]>
Signed-off-by: Alan Jowett <[email protected]>
external/bpf_conformance
Outdated
There was a problem hiding this comment.
Should these be a separate commit?
| * @return The sign extended 64-bit value. | ||
| */ | ||
| static int64_t sign_extend_immediate(int32_t immediate) { | ||
| return (int64_t)immediate; |
There was a problem hiding this comment.
I am currently working through whether or not this is guaranteed to give the proper behavior according to the C standard. In newer versions of the C standard, twos complement representation of signed integers is specified. In older versions, it is not. Therefore, we will (I think) need to be very careful about this. See, e.g., https://www.open-std.org/jtc1/sc22/wg14/www/docs/n2218.htm.
There was a problem hiding this comment.
Further research:
- GCC represents integers as twos complement: https://gcc.gnu.org/onlinedocs/gcc-5.3.0/gcc/Integers-implementation.html#Integers-implementation
- MSVC represents integers as twos complement: https://learn.microsoft.com/en-us/cpp/c-language/range-of-integer-values?view=msvc-170
There does not appear to be a place where Clang (llvm) defines its implementation-specific behavior. See llvm/llvm-project#11644 for the status of such a document.
There was a problem hiding this comment.
In the yet-to-be standardized version of C, twos complement is specified as the representation to use for signed integers. See https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3220.pdf
Signed-off-by: Alan Jowett <[email protected]>
| * @return The sign extended 64-bit value. | ||
| */ | ||
| static int64_t sign_extend_immediate(int32_t immediate) { | ||
| return (int64_t)immediate; |
There was a problem hiding this comment.
In the yet-to-be standardized version of C, twos complement is specified as the representation to use for signed integers. See https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3220.pdf
Resolves: #444
Interpreter wasn't correctly sign extending the immediate value before comparing it to the register.