If the primary X11 screen has a non-zero offset (e.g. X=2560,Y=0) as returned by SDL_GetDisplayBounds(), mouse coordinates returned by SDL_GetGlobalMouseState() will be incorrect (the said offset of the primary screen will be added to them).

The patch below (also mailed to sdl@lists.libsdl.org - http://lists.libsdl.org/pipermail/sdl-libsdl.org/2014-October/096764.html) fixes that:

diff -r bfdc18891a60 src/video/x11/SDL_x11mouse.c
--- a/src/video/x11/SDL_x11mouse.c	Mon Oct 27 19:53:44 2014 -0400
+++ b/src/video/x11/SDL_x11mouse.c	Wed Oct 29 16:33:25 2014 -0400
@@ -382,8 +382,14 @@
                 retval |= (mask & Button1Mask) ? SDL_BUTTON_LMASK : 0;
                 retval |= (mask & Button2Mask) ? SDL_BUTTON_MMASK : 0;
                 retval |= (mask & Button3Mask) ? SDL_BUTTON_RMASK : 0;
-                *x = data->x + rootx;
-                *y = data->y + rooty;
+                /* SDL_DisplayData->x,y point to screen origin, and adding them to mouse coordinates relative to root window doesn't do the right thing
+                 * (observed on dual monitor setup with primary display being the rightmost one - mouse was offset to the right).
+                 *
+                 * Adding root position to root-relative coordinates seems to be a better way to get absolute position. */
+                XWindowAttributes root_attrs;
+                X11_XGetWindowAttributes(display, root, &root_attrs);
+                *x = root_attrs.x + rootx;
+                *y = root_attrs.y + rooty;
                 return retval;
             }
         }