layerscape: add patches-5.4
[openwrt/staging/wigyori.git] / target / linux / layerscape / patches-5.4 / 806-dma-0016-MLK-19931-1-dmaengine-fsl-edma-v3-fix-potential-kern.patch
diff --git a/target/linux/layerscape/patches-5.4/806-dma-0016-MLK-19931-1-dmaengine-fsl-edma-v3-fix-potential-kern.patch b/target/linux/layerscape/patches-5.4/806-dma-0016-MLK-19931-1-dmaengine-fsl-edma-v3-fix-potential-kern.patch
new file mode 100644 (file)
index 0000000..28a59d4
--- /dev/null
@@ -0,0 +1,33 @@
+From 4967ec844a52498af050b1176fedeca37aca111e Mon Sep 17 00:00:00 2001
+From: Robin Gong <yibin.gong@nxp.com>
+Date: Tue, 16 Oct 2018 01:06:29 +0800
+Subject: [PATCH] MLK-19931-1: dmaengine: fsl-edma-v3: fix potential kernel
+ crash in cyclic
+
+  There is one potential race condition in virt-dma framework as below:
+terminate dma channel after the last dma done interrupt, but before
+vchan_complete tasklet scheduled, thus the free-ed 'vd' (free in
+fsl_edma3_terminate_all) maybe still be touched in vchan_complete()
+which cause NULL pointer crash.
+  Kernel community noticed this issue and fix it at virt-dma level:
+https://patchwork.kernel.org/patch/10057791/. To avoid backport too
+much patches, set 'vc->cyclic = NULL' in terminate dma channel
+interfaces to fix such issue easily.
+
+Signed-off-by: Robin Gong <yibin.gong@nxp.com>
+Acked-by: Fugang Duan <fugang.duan@nxp.com>
+(cherry picked from commit 18c9083826400a2ef731496391a0b5e71d461a5f)
+---
+ drivers/dma/fsl-edma-v3.c | 1 +
+ 1 file changed, 1 insertion(+)
+
+--- a/drivers/dma/fsl-edma-v3.c
++++ b/drivers/dma/fsl-edma-v3.c
+@@ -285,6 +285,7 @@ static int fsl_edma3_terminate_all(struc
+       fsl_chan->edesc = NULL;
+       fsl_chan->idle = true;
+       fsl_chan->used = false;
++      fsl_chan->vchan.cyclic = NULL;
+       vchan_get_all_descriptors(&fsl_chan->vchan, &head);
+       spin_unlock_irqrestore(&fsl_chan->vchan.lock, flags);
+       vchan_dma_desc_free_list(&fsl_chan->vchan, &head);