[PATCH] of/irq: Fix msi-map calculation for nonzero rid-base

-----Original Message-----
From: Mark Rutland [mailto:mark.rutland at arm.com]
Sent: Tuesday, February 09, 2016 10:08 AM
To: Stuart Yoder <redacted>
Cc: Marc Zyngier <redacted>; Robin Murphy <robin.murphy@arm.com>;
robh+dt at kernel.org; frowand.list at gmail.com; grant.likely at linaro.org;
devicetree at vger.kernel.org; david.daney at cavium.com; linux-arm-
kernel at lists.infradead.org; linux-kernel at vger.kernel.org; stable at vger.kernel.org
Subject: Re: [PATCH] of/irq: Fix msi-map calculation for nonzero rid-base

On Tue, Feb 09, 2016 at 03:56:55PM +0000, Stuart Yoder wrote:

quoted

quoted

-----Original Message-----
From: Marc Zyngier [mailto:marc.zyngier at arm.com]
Sent: Tuesday, February 09, 2016 6:06 AM
To: Robin Murphy <robin.murphy@arm.com>; robh+dt at kernel.org; frowand.list at gmail.com;
grant.likely at linaro.org; devicetree at vger.kernel.org
Cc: mark.rutland at arm.com; david.daney at cavium.com; Stuart Yoder

[off-list ref];

quoted

quoted

linux-arm-kernel at lists.infradead.org; linux-kernel at vger.kernel.org;
stable at vger.kernel.org
Subject: Re: [PATCH] of/irq: Fix msi-map calculation for nonzero rid-base

Hi Robin,

On 09/02/16 11:04, Robin Murphy wrote:

quoted

The existing msi-map code is fine for shifting the entire RID space
upwards, but attempting finer-grained remapping reveals a bug. It turns
out that we are mistakenly treating the msi-base part as an offset, not
as a new base to remap onto, so things get squiffy when rid-base is
nonzero. Fix this, and at the same time add a sanity check against
having msi-map-mask clash with a nonzero rid-base, as that's another
thing one can easily get wrong.

CC: <redacted>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>

Looks like Stuart and you both found the same bug at the same time:
https://lkml.org/lkml/2016/2/8/1066

but yours seem more correct to me (the rid_base masking in Stuart's
version seems odd).

quoted

---
 drivers/of/irq.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/of/irq.c b/drivers/of/irq.c
index 7ee21ae..e7bfc17 100644
--- a/drivers/of/irq.c
+++ b/drivers/of/irq.c

@@ -635,6 +635,13 @@ static u32 __of_msi_map_rid(struct device *dev, struct

device_node **np,

quoted

 		msi_base = be32_to_cpup(msi_map + 2);
 		rid_len = be32_to_cpup(msi_map + 3);

+		if (rid_base & ~map_mask) {
+			dev_err(parent_dev,
+				"Invalid msi-map translation - msi-map-mask (0x%x) ignores

rid-

quoted

quoted

base (0x%x)\n",

quoted

+				map_mask, rid_base);
+			return rid_out;
+		}
+
 		msi_controller_node = of_find_node_by_phandle(phandle);

 		matched = (masked_rid >= rid_base &&

@@ -654,7 +661,7 @@ static u32 __of_msi_map_rid(struct device *dev, struct

device_node

quoted

quoted

**np,

quoted

 	if (!matched)
 		return rid_out;

-	rid_out = masked_rid + msi_base;
+	rid_out = masked_rid - rid_base + msi_base;
 	dev_dbg(dev,
 		"msi-map at: %s, using mask %08x, rid-base: %08x, msi-base: %08x,

length:

quoted

quoted

%08x, rid: %08x -> %08x\n",

quoted

 		dev_name(parent_dev), map_mask, rid_base, msi_base,

This computation:  masked_rid - rid_base

...doesn't seem right to me.  We are taking a rid that
has been already masked and subtracting a rid base that has
not been masked.

The binding only mentions that the input RID is masked, not the base, so
that seems correct to me.

quoted

I don't see how you can combine masked and unmasked values in the same
calculation.

Say I have this msi mapping:

                   msi-map = <0x0100 &its 0x11 0x1>;
                   msi-map-mask = <0xff>;

I'd say that this is an inconsistent set of properties, and it's
probably worth warning if we encounter this. There is no possible way
that rid-base can be encountered.

quoted

masked_rid = 0x0
rid_base = 0x0100
msi_base = 0x11

masked_rid - rid_base is 0x0 - 0x0100...which does not
give the msi index/offset we want.

Correct final answer should be 0x11.

You can unambiguously describe this with:

	msi-map = <0x00 &its 0x11 0x1>;
	msi-map-mask = <0xff>;

This is exactly the pattern we follow in example 2 in the binding
document.

quoted

In my patch I masked the rid_base so it can be subtracted
from the masked_rid.

masked_rid_base = 0x00

   msi_base + (masked_rid - masked_rid_base) = 0x11

As above, I think that this is an inconsistent DT, and we should
warn/fail in that case.

Thanks...understand now.  I'll test Robin's patch and confirm
that it works as is for me.

Thanks,
Stuart