# --- T2-COPYRIGHT-BEGIN ---
# t2/package/*/xf86-video-rendition/02-sync-actually-waits.patch
# Copyright (C) 2026 The T2 SDE Project
# SPDX-License-Identifier: GPL-2.0 or patched project license
# --- T2-COPYRIGHT-END ---

--- a/src/accelX.c	2026-07-12 13:21:07.730171541 +0200
+++ b/src/accelX.c	2026-07-12 13:22:26.878897395 +0200
@@ -47,6 +47,25 @@
                           } \
                         } while (0)
 
+/* Drain whatever is currently sitting in the output fifo.  This throws data
+   away, so it is only used to get rid of stale results before a sync. */
+#define flushfifoout()  do { int _c=0; \
+                          while ((_c++<0xfffff)&&((verite_in8(iob+FIFOOUTVALID)&0x7)>0)) \
+                              (void)verite_in32(iob); \
+                        } while (0)
+
+/* Block until the RISC has pushed a result word, then consume it.  Unlike a
+   plain drain this really waits: the ucode only produces the word once it has
+   retired every command queued ahead of it, which is what makes it a sync. */
+#define waitfifoout(res) do { int _c=0; \
+                          while ((_c++<0xfffff)&&((verite_in8(iob+FIFOOUTVALID)&0x7)==0)) /* spin */; \
+                          if ((verite_in8(iob+FIFOOUTVALID)&0x7)==0) { \
+                              ErrorF("RENDITION: RISC synchronization failed (%s)\n", res); \
+                              return; \
+                          } \
+                          (void)verite_in32(iob); \
+                        } while (0)
+
 #define P1(x)           ((vu32)x)
 #define P2(x, y)        ((((vu16)x)<<16)+((vu16)y))
 
@@ -324,6 +343,7 @@
     ErrorF("#InitUcode(2)# FIFOIN_FREE 0x%x -- \n",verite_in8(iob+FIFOINFREE));
 #endif
 
+    waitfifo2(4, 1);
     verite_out32(iob, 0);     /* a0 - ucode init command */
     verite_out32(iob, 0);     /* a1 - 1024 byte context store area */
     verite_out32(iob, 0);     /* a2 */
@@ -393,6 +413,7 @@
 
     v1k_flushicache(pScreenInfo);
     v1k_start(pScreenInfo, pRendition->board.csucode_base);
+    waitfifo(4);
     verite_out32(iob, 0);     /* a0 - ucode init command */
     verite_out32(iob, 0);     /* a1 - 1024 byte context store area */
     verite_out32(iob, 0);     /* a2 */
@@ -447,8 +468,6 @@
     renditionPtr pRendition = RENDITIONPTR(pScreenInfo);
     unsigned long iob = pRendition->board.io_base;
 
-    int c;
-
 #ifdef DEBUG
     ErrorF("RENDITION: RENDITIONSyncV1000 called\n");
 
@@ -456,57 +475,25 @@
     ErrorF("#Sync (1)# FIFO_OUTVALID 0x%x -- \n",verite_in8(iob+FIFOOUTVALID));
 #endif
 
-    c=0;
-    /* empty output fifo,
-       i.e. if there is any valid data in the output fifo then read it */
-
-    while ((c++<0xfffff) && ((verite_in8(iob+FIFOOUTVALID)&0x7)>0))
-      (void)verite_in32(iob);
-
-/*       if(!(c%0xffff))ErrorF("#F1# !0x%x! -- ",verite_in8(iob+FIFOOUTVALID)); */
-
-    if (c >= 0xfffff) { 
-        ErrorF("RENDITION: RISC synchronization failed (1) FIFO out == %d!\n",
-	       verite_in8(iob+FIFOOUTVALID)&0x1f);
-        return; 
-    }
+    /* discard stale results so the readbacks below can't pick up an old word */
+    flushfifoout();
 
-    /* sync RISC */
+    /* Sync the RISC: CMD_GET_PIXEL is retired in order behind every command
+       already queued, so once its result word shows up the ucode is idle and
+       it is safe to halt it. */
     waitfifo(2);
     verite_out32(iob, CMD_GET_PIXEL);
     verite_out32(iob, 0);
-
-    c=0;
-    while ((c++<0xfffff) && ((verite_in8(iob+FIFOOUTVALID)&0x7)>0))
-      (void)verite_in32(iob);
-
-/*       if(!(c%0xffff))ErrorF("#F2# !0x%x! -- ",verite_in8(iob+FIFOOUTVALID)); */
-
-
-    if (c >= 0xfffff) { 
-      ErrorF ("Rendition: RISC synchronization failed (2) FIFO out == %d!\n",
-	      verite_in8(iob+FIFOOUTVALID)&0x1f);
-        return; 
-    }
+    waitfifoout("RISC idle");
 
     /* sync pixel engine using csucode -- I suppose this is quite slow <ml> */
     v1k_stop(pScreenInfo);
     v1k_start(pScreenInfo, pRendition->board.csucode_base);
+    waitfifo(1);
     verite_out32(iob, 2);     /* a0 - sync command */
 
-    c=0;
-    while ((c++<0xfffff) && ((verite_in8(iob+FIFOOUTVALID)&0x7)>0))
-      (void)verite_in32(iob);
-
-/*      if(!(c%0xffff))ErrorF("#F3# !0x%x! -- ",verite_in8(iob+FIFOOUTVALID)); */
-
-    if (c == 0xfffff) { 
-        ErrorF ("Rendition: Pixel engine synchronization failed FIFO out == %d!\n",
-		verite_in8(iob+FIFOOUTVALID)&0x1f);
-        return;
-    }
-
     /* restart the ucode */
+    waitfifo(4);
     verite_out32(iob, 0);     /* a0 - ucode init command */
     verite_out32(iob, 0);     /* a1 - 1024 byte context store area */
     verite_out32(iob, 0);     /* a2 */
@@ -524,7 +511,16 @@
     verite_out32(iob, pRendition->board.mode.virtualwidth  *
 	    (pRendition->board.mode.bitsperpixel>>3));
     verite_out32(iob, (pRendition->board.mode.stride1<<12) |
-	    (pRendition->board.mode.stride0<<8)); 
+	    (pRendition->board.mode.stride0<<8));
+
+    /* The csucode's pixel engine sync (a0=2) produces no result word, so it
+       can only be observed indirectly: the RISC executes strictly in order, so
+       a readback that retires *after* the restart proves everything queued
+       before it -- including the pixel engine sync -- has completed. */
+    waitfifo(2);
+    verite_out32(iob, CMD_GET_PIXEL);
+    verite_out32(iob, 0);
+    waitfifoout("pixel engine idle");
 }