|
|
@@ -418,3 +418,20 @@ static void __dma_memcpy(u32 daddr, s16 dmod, u32 saddr, s16 smod, size_t cnt, u
|
|
|
* without the anomaly version (because we are atomic already).
|
|
|
*/
|
|
|
__builtin_bfin_ssync();
|
|
|
+
|
|
|
+ if (bfin_read_MDMA_S_CONFIG())
|
|
|
+ while (!(bfin_read_MDMA_D_IRQ_STATUS() & DMA_DONE))
|
|
|
+ continue;
|
|
|
+
|
|
|
+ if (conf & DMA2D) {
|
|
|
+ /* For larger bit sizes, we've already divided down cnt so it
|
|
|
+ * is no longer a multiple of 64k. So we have to break down
|
|
|
+ * the limit here so it is a multiple of the incoming size.
|
|
|
+ * There is no limitation here in terms of total size other
|
|
|
+ * than the hardware though as the bits lost in the shift are
|
|
|
+ * made up by MODIFY (== we can hit the whole address space).
|
|
|
+ * X: (2^(16 - 0)) * 1 == (2^(16 - 1)) * 2 == (2^(16 - 2)) * 4
|
|
|
+ */
|
|
|
+ u32 shift = abs(dmod) >> 1;
|
|
|
+ size_t ycnt = cnt >> (16 - shift);
|
|
|
+ cnt = 1 << (16 - shift);
|
