arch/tile: add Tilera's <arch/sim.h> header as an open-source header

This change adds one of the Tilera standard <arch> headers to the set
of headers shipped with Linux.  The <arch/sim.h> header provides
methods for programmatically interacting with the Tilera simulator.

The current <arch/sim.h> provides inline assembly for the _sim_syscall
function, so the declaration and definition previously provided
manually in Linux are no longer needed.  We now use the standard
sim_validate_lines_evicted() method from <arch/sim.h> rather than
rolling our own direct call to sim_syscall().

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>

1 files changed, 0 insertions, 22 deletions

diff --git a/arch/tile/kernel/entry.S b/arch/tile/kernel/entry.S
index 3d01383b1b0e..80d13f013bb2 100644
--- a/arch/tile/kernel/entry.S
+++ b/arch/tile/kernel/entry.S
@@ -25,28 +25,6 @@ STD_ENTRY(current_text_addr)
         { move r0, lr; jrp lr }
         STD_ENDPROC(current_text_addr)
 
-STD_ENTRY(_sim_syscall)
-        /*
-         * Wait for r0-r9 to be ready (and lr on the off chance we
-         * want the syscall to locate its caller), then make a magic
-         * simulator syscall.
-         *
-         * We carefully stall until the registers are readable in case they
-         * are the target of a slow load, etc. so that tile-sim will
-         * definitely be able to read all of them inside the magic syscall.
-         *
-         * Technically this is wrong for r3-r9 and lr, since an interrupt
-         * could come in and restore the registers with a slow load right
-         * before executing the mtspr. We may need to modify tile-sim to
-         * explicitly stall for this case, but we do not yet have
-         * a way to implement such a stall.
-         */
-        { and zero, lr, r9 ; and zero, r8, r7 }
-        { and zero, r6, r5 ; and zero, r4, r3 }
-        { and zero, r2, r1 ; mtspr SIM_CONTROL, r0 }
-        { jrp lr }
-        STD_ENDPROC(_sim_syscall)
-
 /*
  * Implement execve().  The i386 code has a note that forking from kernel
  * space results in no copy on write until the execve, so we should be