Here are a couple of replacement functions that make slime disassemble work with cmucl/sparc. The disassembly works for me on cmucl/sparc. The foreign source location appears to work, but the only foreign function I've been able to test with is call_into_lisp which doesn't have any line info. But slime recognizes that and prints a message to that affect.
Ray
(defun frame-ip (frame) "Return the (absolute) instruction pointer and the relative pc of FRAME." (if (not frame) (sys:int-sap (ldb (byte vm:word-bits 0) -1)) (let ((debug-fun (di::frame-debug-function frame))) (etypecase debug-fun (di::compiled-debug-function (let* ((code-loc (di:frame-code-location frame)) (component (di::compiled-debug-function-component debug-fun)) (pc (di::compiled-code-location-pc code-loc)) (ip (sys:without-gcing (sys:sap-int (sys:sap+ (kernel:code-instructions component) pc))))) (values ip pc))) (di::interpreted-debug-function (sys:int-sap (ldb (byte vm:word-bits 0) -1))) (di::bogus-debug-function #-x86 (let* ((real (di::frame-real-frame (di::frame-up frame))) (fp (di::frame-pointer real))) #+(or) (progn (format *debug-io* "Frame-real-frame = ~S~%" real) (format *debug-io* "fp = ~S~%" fp) (format *debug-io* "lra = ~S~%" (kernel:stack-ref fp vm::lra-save-offset))) (values (sys:int-sap (kernel:get-lisp-obj-address (kernel:stack-ref fp vm::lra-save-offset))) 0)) #+x86 (let ((fp (di::frame-pointer (di:frame-up frame)))) (multiple-value-bind (ra ofp) (di::x86-call-context fp) (declare (ignore ofp)) (values ra 0))))))))
(defun foreign-frame-p (frame) #-x86 (let ((ip (frame-ip frame))) (and (sys:system-area-pointer-p ip) (typep (di::frame-debug-function frame) 'di::bogus-debug-function))) #+x86 (let ((ip (frame-ip frame))) (and (sys:system-area-pointer-p ip) (multiple-value-bind (pc code) (di::compute-lra-data-from-pc ip) (declare (ignore pc)) (not code)))))
(defun gdb-exec (cmd) (with-temporary-file (file filename) (write-string cmd file) (force-output file) (let* ((output (make-string-output-stream)) ;; gdb on sparc needs to know the executable to find the ;; symbols. Without this, gdb can't disassemble anything. ;; NOTE: We assume that the first entry in ;; lisp::*cmucl-lib* is the bin directory where lisp is ;; located. If this is not true, we'll have to do ;; something better to find the lisp executable. (lisp-path #+sparc (list (namestring (probe-file (merge-pathnames "lisp" (car (lisp::parse-unix-search-path lisp::*cmucl-lib*)))))) #-sparc nil) (proc (ext:run-program "gdb" `(,@lisp-path "-batch" "-x" ,filename) :wait t :output output))) (assert (eq (ext:process-status proc) :exited)) (assert (eq (ext:process-exit-code proc) 0)) (get-output-stream-string output))))
On 8/26/10 3:45 PM, Raymond Toy wrote:
Here are a couple of replacement functions that make slime disassemble work with cmucl/sparc. The disassembly works for me on cmucl/sparc. The foreign source location appears to work, but the only foreign function I've been able to test with is call_into_lisp which doesn't have any line info. But slime recognizes that and prints a message to that affect.
I've updated the code a little bit. frame-ip will return -1 (as before) if it can't compute the frame ip. This will probably cause an error later on, but I guess that's ok. There's also a slight cleanup on what frame-ip returns for a bogus-function. We try to adjust the lra object to the actual return address. This isn't really necessary since the only use of that is (I think) the address inside a C function that we want gdb to disassemble. Hence any address in the function is ok.
Ray
Raymond Toy toy.raymond@gmail.com writes:
On 8/26/10 3:45 PM, Raymond Toy wrote:
Here are a couple of replacement functions that make slime disassemble work with cmucl/sparc. The disassembly works for me on cmucl/sparc. The foreign source location appears to work, but the only foreign function I've been able to test with is call_into_lisp which doesn't have any line info. But slime recognizes that and prints a message to that affect.
I've updated the code a little bit. frame-ip will return -1 (as before) if it can't compute the frame ip. This will probably cause an error later on, but I guess that's ok. There's also a slight cleanup on what frame-ip returns for a bogus-function. We try to adjust the lra object to the actual return address. This isn't really necessary since the only use of that is (I think) the address inside a C function that we want gdb to disassemble. Hence any address in the function is ok.
Committed, thanks!
-
Raymond Toy -
Stas Boukarev