net/mlx4_en: Set statistics bitmap at port init
[firefly-linux-kernel-4.4.55.git] / drivers / dma / sh / rcar-dmac.c
index 711da01a200b615c4ece273c28febd12a1cee006..a18d16cc479559e2ca7719eebe8755c7a64d242b 100644 (file)
@@ -1593,6 +1593,7 @@ static int rcar_dmac_probe(struct platform_device *pdev)
                DMA_SLAVE_BUSWIDTH_2_BYTES | DMA_SLAVE_BUSWIDTH_4_BYTES |
                DMA_SLAVE_BUSWIDTH_8_BYTES | DMA_SLAVE_BUSWIDTH_16_BYTES |
                DMA_SLAVE_BUSWIDTH_32_BYTES | DMA_SLAVE_BUSWIDTH_64_BYTES;
+       unsigned int channels_offset = 0;
        struct dma_device *engine;
        struct rcar_dmac *dmac;
        struct resource *mem;
@@ -1612,6 +1613,19 @@ static int rcar_dmac_probe(struct platform_device *pdev)
        if (ret < 0)
                return ret;
 
+       /*
+        * A still unconfirmed hardware bug prevents the IPMMU microTLB 0 to be
+        * flushed correctly, resulting in memory corruption. DMAC 0 channel 0
+        * is connected to microTLB 0 on currently supported platforms, so we
+        * can't use it with the IPMMU. As the IOMMU API operates at the device
+        * level we can't disable it selectively, so ignore channel 0 for now if
+        * the device is part of an IOMMU group.
+        */
+       if (pdev->dev.iommu_group) {
+               dmac->n_channels--;
+               channels_offset = 1;
+       }
+
        dmac->channels = devm_kcalloc(&pdev->dev, dmac->n_channels,
                                      sizeof(*dmac->channels), GFP_KERNEL);
        if (!dmac->channels)
@@ -1662,7 +1676,8 @@ static int rcar_dmac_probe(struct platform_device *pdev)
        INIT_LIST_HEAD(&dmac->engine.channels);
 
        for (i = 0; i < dmac->n_channels; ++i) {
-               ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i], i);
+               ret = rcar_dmac_chan_probe(dmac, &dmac->channels[i],
+                                          i + channels_offset);
                if (ret < 0)
                        goto error;
        }