On 03/06/2017 03:48 PM, Benjamin Herrenschmidt wrote:
On Mon, 2017-03-06 at 13:46 -0600, Larry Finger wrote:
I was able to create a PPC emulation with debian-8.7.1-powerpc-CD-1.iso
following the instructions in https://gmplib.org/~tege/qemu.html. My only
problem was that "-net tap" fails and I did not find any way to get networking
working.

After looking at the screen through a number of crashes, I have determined that
the top entry in the traceback comes from dmam_alloc_coherent(). I have not been
able to see the offset to determine which BUG_ON call in that routine is being
triggered.

I tried to modify panic() to see if I could keep the screen on longer after the
failure, but no joy so far.

I think the problem is this code in drivers/macintosh/macio_asic.c:

#ifdef CONFIG_PCI
        /* Set the DMA ops to the ones from the PCI device, this could be
         * fishy if we didn't know that on PowerMac it's always direct ops
         * or iommu ops that will work fine
         *
         * To get all the fields, copy all archdata
         */
        dev->ofdev.dev.archdata = chip->lbus.pdev->dev.archdata;
#endif /* CONFIG_PCI */

This is definitely bad. A quick fix is to copy the new dev->dma_ops field
(as well, there is still stuff in archdata that we need too).

A better long term fix is to have a set of macio_dma_ops wrappers that do
"the right thing".

Ben,

Attached is a patch that fixes the crash. At the moment, it has my s-o-b, but I do not feel it right to claim authorship. My role should be a Reported-and-tested-by. Please advise.

Larry


>From fe51bd0233e10d0eed64cfe177f5d43ca95642a7 Mon Sep 17 00:00:00 2001
From: Larry Finger <[email protected]>
Date: Thu, 9 Mar 2017 16:06:21 -0600
Subject: [PATCH] powerpc: Fix crash introduced with commit 5657933dbb6e

Code inserted during the git merges before kernel 4.11-rc1 caused an
early panic quickly followed by a complete shutdown. The traceback was
not displayed long enough to read or photograph.

The problem was bisected to commit 5657933dbb6e ("treewide: Move dma_ops
from struct dev_archdata into struct device"). The problem was a missing
copy of critical information from struct dev_archdata into struct device,
leading to the dereference of a null pointer.

Fixes: commit 5657933dbb6e ("treewide: Move dma_ops from struct dev_archdata into struct device")
Signed-off-by: Larry Finger <[email protected]>
---
 drivers/macintosh/macio_asic.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/macintosh/macio_asic.c b/drivers/macintosh/macio_asic.c
index 3f041b1..f757cef2 100644
--- a/drivers/macintosh/macio_asic.c
+++ b/drivers/macintosh/macio_asic.c
@@ -392,6 +392,7 @@ static struct macio_dev * macio_add_one_device(struct macio_chip *chip,
 	 * To get all the fields, copy all archdata
 	 */
 	dev->ofdev.dev.archdata = chip->lbus.pdev->dev.archdata;
+	dev->ofdev.dev.dma_ops = chip->lbus.pdev->dev.dma_ops;
 #endif /* CONFIG_PCI */
 
 #ifdef DEBUG
-- 
2.10.2

Reply via email to