tile: convert uses of "inv" to "finv"

The "inv" (invalidate) instruction is generally less safe than "finv"
(flush and invalidate), as it will drop dirty data from the cache.
It turns out we have almost no need for "inv" (other than for the
older 32-bit architecture in some limited cases), so convert to
"finv" where possible and delete the extra "inv" infrastructure.

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

1 files changed, 8 insertions, 5 deletions

diff --git a/arch/tile/lib/cacheflush.c b/arch/tile/lib/cacheflush.c
index 8f8ad814b139..2238b40abf3c 100644
--- a/arch/tile/lib/cacheflush.c
+++ b/arch/tile/lib/cacheflush.c
@@ -147,18 +147,21 @@ void finv_buffer_remote(void *buffer, size_t size, int hfh)
                 force_load(p);
 
         /*
-         * Repeat, but with inv's instead of loads, to get rid of the
+         * Repeat, but with finv's instead of loads, to get rid of the
          * data we just loaded into our own cache and the old home L3.
-         * No need to unroll since inv's don't target a register.
+         * No need to unroll since finv's don't target a register.
+         * The finv's are guaranteed not to actually flush the data in
+         * the buffer back to their home, since we just read it, so the
+         * lines are clean in cache; we will only invalidate those lines.
          */
         p = (char *)buffer + size - 1;
-        __insn_inv(p);
+        __insn_finv(p);
         p -= step_size;
         p = (char *)((unsigned long)p | (step_size - 1));
         for (; p >= base; p -= step_size)
-                __insn_inv(p);
+                __insn_finv(p);
 
-        /* Wait for the load+inv's (and thus finvs) to have completed. */
+        /* Wait for these finv's (and thus the first finvs) to be done. */
         __insn_mf();
 
 #ifdef __tilegx__