Re: [PCI] BUG: unable to handle kernel

From: Murali Karicheri
Date: Mon Mar 09 2015 - 13:05:25 EST


On 03/09/2015 12:07 PM, Guenter Roeck wrote:
On 03/09/2015 08:53 AM, Murali Karicheri wrote:
On 03/09/2015 10:44 AM, Bjorn Helgaas wrote:
On Mon, Mar 9, 2015 at 9:17 AM, Murali Karicheri<m-karicheri2@xxxxxx>
wrote:
On 03/06/2015 12:58 PM, Murali Karicheri wrote:

On 03/06/2015 11:55 AM, Guenter Roeck wrote:

On Fri, Mar 06, 2015 at 10:48:59AM -0500, Murali Karicheri wrote:
[ ... ]

From 098b4f5e4ab9407fbdbfcca3a91785c17e25cf03 Mon Sep 17
00:00:00 2001
From: Murali Karicheri<m-karicheri2@xxxxxx>
Date: Fri, 6 Mar 2015 10:23:08 -0500
Subject: [PATCH] pci: of : fix kernel crash

This is a debug patch to root cause the kernel crash

commit 0b2af171520e5d5e7d5b5f479b90a6a5014d9df6

PCI: Update DMA configuration from DT

Signed-off-by: Murali Karicheri<m-karicheri2@xxxxxx>
---
drivers/of/of_pci.c | 8 ++++++++
drivers/pci/host-bridge.c | 5 +++++
2 files changed, 13 insertions(+)

diff --git a/drivers/of/of_pci.c b/drivers/of/of_pci.c
index 86d3c38..5a59fb8 100644
--- a/drivers/of/of_pci.c
+++ b/drivers/of/of_pci.c
@@ -129,6 +129,14 @@ void of_pci_dma_configure(struct pci_dev
*pci_dev)
struct device *dev =&pci_dev->dev;
struct device *bridge = pci_get_host_bridge_device(pci_dev);

+ if (!bridge || !bridge->parent) {
+ if (!bridge)
+ pr_err("PCI bridge not found\n");
+ if (!bridge->parent)
+ pr_err("PCI bridge parent not found\n");


You'll see a crash here if bridge is NULL. Maybe add an else before
the second
if statement ? Also, dev_err might be a bit more useful and would be
available.

Fixed and attached.

Murali

Thanks,
Guenter

+ return;
+ }
+
of_dma_configure(dev, bridge->parent->of_node);
pci_put_host_bridge_device(bridge);
}
diff --git a/drivers/pci/host-bridge.c b/drivers/pci/host-bridge.c
index 3e5bbf9..ef2ab51 100644
--- a/drivers/pci/host-bridge.c
+++ b/drivers/pci/host-bridge.c
@@ -28,6 +28,11 @@ struct device *pci_get_host_bridge_device(struct
pci_dev *dev)
struct pci_bus *root_bus = find_pci_root_bus(dev->bus);
struct device *bridge = root_bus->bridge;

+ if (!bridge) {
+ pr_err("PCI: bridge not found\n");
+ return NULL;
+ }
+
kobject_get(&bridge->kobj);
return bridge;
}
--
1.7.9.5



BJorn,

Any chance of applying the attached debug patch to see if this fixes
and
provide some additional information on this BUG? Not sure who will
pick this
one and apply.

The change that caused the oops (0b2af171520e ("PCI: Update DMA
configuration from DT")) only exists on my pci/iommu branch, so I'm
the one to apply it.

It's much easier for me to deal with plain text patches (not
attachments).

I'm hesitating because I don't want to encourage use of the 0-day
testing robot as a tool at which we can just throw debug patches. The
robot is a service that costs somebody real money, and I want to be a
good neighbor when using it.

Thanks for the clarification as I don't have much information on the
testing robot. At the same time the question is how similar incidence
in the past have been handled. I am a newbie w.r.t to this. This is
first time I have introduced a patch that impacts multiple arch/machines.


Was the information in the robot's report enough to reproduce the
oops? If not, is there additional information we could add to the
report that would enable you to reproduce it? Even if we can't
reproduce the oops, the report seems detailed enough that we should be
able to deduce the problem and produce a fix in which we have high
confidence.

The BUG report essentially indicates the crash happened in
of_pci_dma_configure(). The machine specific log make sense to a
person familiar with this arch and I am not familiar with the same. So
anyone can help narrow down the root cause of this?

Looking at the code, there are two ptr variables that are accessed
without checking for NULL as initial thinking was that these can never
be NULL. So the debug patch is just adding addition check before
accessing the ptr. I can send this patch without debug prints if that
make sense. I was thinking to get confirmation that this is indeed the
case before adding the check. What do you think the right approach
here? Send a patch for this to the ML for adding the check as a
potential fix? Or someone can help me investigate the crash dump and
root cause it? or if we can use test robot to confirm this, I can
re-send the patch ASIS to the list. Please clarify.

If the assumption is that the pointers can never be NULL,
wouldn't it be important to see a call trace and to find out
if the NULL pointers can actually be seen by design,
or if there is some other bug ?

Call trace shows

[ 0.576666] [<7976c1ac>] pci_device_add+0xbc/0x820
[ 0.576666] [<7976c1ac>] pci_device_add+0xbc/0x820


And BUG seems to be in of_pci_dma_configure() as shown in the BUG report.

of_pci_dma_configure() calls newly added API call to pci_get_host_bridge_device(). Seems like this has succeeded which means bridge is non NULL IMO. However in this function it passes bridge->parent->of_node to of_dma_configure(). So I suspect bridge->parent is NULL for some reason. Is there a chance for parent being NULL in this or any other platform?


I am a bit concerned that adding those NULL pointer checks
might end up hiding some other bug, ie that they just hide
the real bug without fixing it.


I agree with you as well. That is the reason I had added the debug prints in the attached patch to check what is NULL here and that can help us dig into this more.

If BJorn can accept this debug patch for finding this, that will help. I can re-send it to the list as a debug patch if everyone agrees. Otherwise I don't know how to proceed from here.

Thanks and regards,

Murali
Thanks,
Guenter



--
Murali Karicheri
Linux Kernel, Texas Instruments
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/