The Dynamic DMA Window (DDW) mechanism relies on EEH to obtain the
configuration address of devices. For example, the functions query_ddw()
and create_ddw() make use of eeh_dev struct. So, the dependency is
intrinsic - DDW mechanism will fail if EEH is not enabled.
Despite this dependency, no check for EEH availability is performed in DDW
code. This patch adds a check based on eeh_enabled() function, so if EEH is
not enabled before eeh_dev struct use, DDW will fallback to default iommu
mechanism and won't fail.
One use case for this patch is when we disable EEH globally via kernel
command-line ("eeh=off") - without the patch, a device probe can hit a kernel
oops because EEH is disabled but DDW will try to use it.
Signed-off-by: Guilherme G. Piccoli <gpicc...@linux.vnet.ibm.com>
---
arch/powerpc/platforms/pseries/iommu.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/platforms/pseries/iommu.c
b/arch/powerpc/platforms/pseries/iommu.c
index bd98ce2..1ff55cc 100644
--- a/arch/powerpc/platforms/pseries/iommu.c
+++ b/arch/powerpc/platforms/pseries/iommu.c
@@ -1224,8 +1224,10 @@ static int dma_set_mask_pSeriesLP(struct device *dev,
u64 dma_mask)
pdev = to_pci_dev(dev);
- /* only attempt to use a new window if 64-bit DMA is requested */
- if (!disable_ddw && dma_mask == DMA_BIT_MASK(64)) {
+ /* We should check if EEH is enabled here, since DDW mechanism has
+ * an intrinsic dependency of EEH config addr information. Also, we
+ * only attempt to use a new window if 64-bit DMA is requested */
+ if (eeh_enabled() && !disable_ddw && dma_mask == DMA_BIT_MASK(64)) {
dn = pci_device_to_OF_node(pdev);
dev_dbg(dev, "node is %s\n", dn->full_name);
--
2.1.0
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev
