Thread (45 messages) 45 messages, 7 authors, 2015-02-05

[PATCH v4 3/6] of: fix size when dma-range is not used

From: Rob Herring <hidden>
Date: 2015-01-29 02:51:31
Also in: linux-devicetree, linux-iommu, linux-pci, lkml

On Wed, Jan 28, 2015 at 5:05 AM, Catalin Marinas
[off-list ref] wrote:
quoted hunk ↗ jump to hunk
On Tue, Jan 27, 2015 at 06:55:15PM +0000, Murali Karicheri wrote:
quoted
On 01/27/2015 06:27 AM, Robin Murphy wrote:
quoted
On 23/01/15 22:32, Murali Karicheri wrote:
quoted
Fix the dma-range size when the DT attribute is missing. i.e set size to
dev->coherent_dma_mask + 1 instead of dev->coherent_dma_mask. To detect
overflow when mask is set to max of u64, add a check, log error and
return.
Some platform use mask format for size in DTS. So add a work around to
catch this and fix.

Cc: Joerg Roedel <joro@8bytes.org>
Cc: Grant Likely <redacted>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Will Deacon <redacted>
Cc: Russell King <redacted>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>

Signed-off-by: Murali Karicheri <redacted>
---
drivers/of/device.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/drivers/of/device.c b/drivers/of/device.c
index 2de320d..0a5ff54 100644
--- a/drivers/of/device.c
+++ b/drivers/of/device.c
@@ -105,12 +105,24 @@ void of_dma_configure(struct device *dev, struct
device_node *np)
ret = of_dma_get_range(np, &dma_addr, &paddr, &size);
if (ret < 0) {
dma_addr = offset = 0;
- size = dev->coherent_dma_mask;
+ size = dev->coherent_dma_mask + 1;
} else {
offset = PFN_DOWN(paddr - dma_addr);
+ /*
+ * Add a work around to treat the size as mask + 1 in case
+ * it is defined in DT as a mask.
+ */
+ if (size & 1)
+ size = size + 1;
dev_dbg(dev, "dma_pfn_offset(%#08lx)\n", offset);
}

+ /* if size is 0, we have an overflow of u64 */
+ if (!size) {
+ dev_err(dev, "invalid size\n");
+ return;
+ }
+
This seems potentially fragile to dodgy DTs given that we might also be
using size to make a mask later. Would it make sense to double-up a
sanity check as mask-format detection? Something like:

if is_power_of_2(size)
// use size
else if is_power_of_2(size + 1)
// use size + 1
else
// cry
How about having the logic like this?

      ret = of_dma_get_range(np, &dma_addr, &paddr, &size);
      if (ret < 0) {
              dma_addr = offset = 0;
              size = dev->coherent_dma_mask + 1;
      } else {
              offset = PFN_DOWN(paddr - dma_addr);
              dev_dbg(dev, "dma_pfn_offset(%#08lx)\n", offset);
      }

      if (is_power_of_2(size + 1))
              size = size + 1;
      else if (!is_power_of_2(size))
      {
              dev_err(dev, "invalid size\n");
              return;
      }
In of_dma_configure(), we currently assume that the default coherent
mask is 32-bit. In this thread:

http://article.gmane.org/gmane.linux.kernel/1835096

we talked about setting the coherent mask based on size automatically.
I'm not sure about the size but I think we can assume is 32-bit mask + 1
if it is not specified in the DT. Instead of just assuming a default
mask, let's assume a default size and create the mask based on this
(untested):
diff --git a/drivers/of/platform.c b/drivers/of/platform.c
index 5b33c6a21807..9ff8d1286b44 100644
--- a/drivers/of/platform.c
+++ b/drivers/of/platform.c
@@ -170,10 +170,10 @@ static void of_dma_configure(struct device *dev)
        struct iommu_ops *iommu;

        /*
-        * Set default dma-mask to 32 bit. Drivers are expected to setup
-        * the correct supported dma_mask.
+        * Set default size to cover the 32-bit. Drivers are expected to setup
+        * the correct size and dma_mask.
         */
-       dev->coherent_dma_mask = DMA_BIT_MASK(32);
+       size = 1ULL << 32;

        /*
         * Set it to coherent_dma_mask by default if the architecture
@@ -185,13 +185,24 @@ static void of_dma_configure(struct device *dev)
        ret = of_dma_get_range(dev->of_node, &dma_addr, &paddr, &size);
        if (ret < 0) {
                dma_addr = offset = 0;
-               size = dev->coherent_dma_mask;
Are we assuming dma_addr, paddr and size are not touched on error? If
so, we can get rid of this clause entirely.
        } else {
                offset = PFN_DOWN(paddr - dma_addr);
                dev_dbg(dev, "dma_pfn_offset(%#08lx)\n", dev->dma_pfn_offset);
        }
        dev->dma_pfn_offset = offset;

+       /*
+        * Workaround for DTs setting the size to a mask or 0.
+        */
+       if (is_power_of_2(size + 1))
+               size += 1;
As I mentioned, I think power of 2 is too restrictive (from a DT
perspective even though the kernel may require a power of 2 here for
the mask). Just checking bit0 set should be enough.

Also, we need a WARN here so DTs get fixed.
+
+       /*
+        * Coherent DMA masks larger than 32-bit must be explicitly set by the
+        * driver.
+        */
+       dev->coherent_dma_mask = min(DMA_BIT_MASK(32), DMA_BIT_MASK(ilog2(size)));
+
        coherent = of_dma_is_coherent(dev->of_node);
        dev_dbg(dev, "device is%sdma coherent\n",
                coherent ? " " : " not ");

--
Catalin
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help