Re: [REGRESSION next-20170426] Commit 09515ef5ddad ("of/acpi: Configure dma operations at probe time for platform/amba/pci bus devices") causes oops in mvneta

[Sricharan R]

Hi Ralph,

<snip..>

>>>>>
>>>>> Commit 09515ef5ddad ("of/acpi: Configure dma operations at probe
>>>>> time for platform/amba/pci bus devices") causes a kernel panic as
>>>>> in the log below on an armada-385. Reverting the commit fixes the
>>>>> issue.
>>>>>
>>>>> Regards
>>>>> Ralph    
>>>>
>>>> Somehow not getting a obvious clue on whats going wrong with the
>>>> logs below. From the log and looking in to dts, the drivers seems
>>>> to the one for "marvell,armada-370-neta".  
>>>
>>> Correct.
>>>   
>>>> Issue looks the data from the dma
>>>> has gone bad and subsequently referring the wrong data has resulted
>>>> in the crash. Looks like the dma_masks is the one going wrong.
>>>> Can i get some logs from mvneta_probe, about dev->dma_mask,
>>>> dev->coherent_dma_mask and dev->dma_ops with and without the patch
>>>> to see whats the difference ?  
>>>
>>> Not sure I understood what exactly you are after. Might be faster to
>>> just send me a patch with all debug print statements you like to
>>> see. 
>>
>> Attached the patch with debug prints.
>>
>> Regards,
>>  Sricharan
>>
> 
> Hi Sricharan
> 
> With commit 09515ef5ddad
> 
> [    1.288962] mvneta f1070000.ethernet: dev->dma_mask 0xffffffff
> [    1.294827] mvneta f1070000.ethernet: dev->coherent_dma_mask 0xffffffff
> [    1.301472] mvneta f1070000.ethernet: dev->dma_ops 0x40b00c0601460
> 
> [    1.322047] mvneta f1034000.ethernet: dev->dma_mask 0xffffffff
> [    1.327904] mvneta f1034000.ethernet: dev->coherent_dma_mask 0xffffffff
> [    1.334549] mvneta f1034000.ethernet: dev->dma_ops 0x40b00c0601460
> 
> 
> With the patch reverted, the build that works
> 
> [    1.289001] mvneta f1070000.ethernet: dev->dma_mask 0xffffffff
> [    1.294866] mvneta f1070000.ethernet: dev->coherent_dma_mask 0xffffffff
> [    1.301511] mvneta f1070000.ethernet: dev->dma_ops 0x40b00c06014a8
> 
> [    1.317005] mvneta f1034000.ethernet: dev->dma_mask 0xffffffff
> [    1.322867] mvneta f1034000.ethernet: dev->coherent_dma_mask 0xffffffff
> [    1.329508] mvneta f1034000.ethernet: dev->dma_ops 0x40b00c06014a8
> 

My bad, i think it is this patch missing [1], attached it as well.
Infact, this was in the series initially and got acked to get merged
separately well before the series. I should have sent this to Russell.
I will do this now. If this fixes up the issue,
i will take this patch separately, while this series gets tested
on -next.

[1] https://patchwork.kernel.org/patch/9362113/

-- 
"QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
From be36ea5f2c7d1c28dc8f829b5d2c817826481086 Mon Sep 17 00:00:00 2001
From: Laurent Pinchart <laurent.pinchart+renesas@xxxxxxxxxxxxxxxx>
Date: Fri, 15 May 2015 02:00:02 +0300
Subject: [PATCH] arm: dma-mapping: Don't override dma_ops in
 arch_setup_dma_ops()

The arch_setup_dma_ops() function is in charge of setting dma_ops with a
call to set_dma_ops(). set_dma_ops() is also called from

- highbank and mvebu bus notifiers
- dmabounce (to be replaced with swiotlb)
- arm_iommu_attach_device

(arm_iommu_attach_device is itself called from IOMMU and bus master
device drivers)

To allow the arch_setup_dma_ops() call to be moved from device add time
to device probe time we must ensure that dma_ops already setup by any of
the above callers will not be overriden.

Aftering replacing dmabounce with swiotlb, converting IOMMU drivers to
of_xlate and taking care of highbank and mvebu, the workaround should be
removed.

[Rebased on top of 4.11-rc8]
Signed-off-by: Sricharan R <sricharan@xxxxxxxxxxxxxx>
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@xxxxxxxxxxxxxxxx>
---
 arch/arm/mm/dma-mapping.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c
index 0268584..c742dfd 100644
--- a/arch/arm/mm/dma-mapping.c
+++ b/arch/arm/mm/dma-mapping.c
@@ -2408,6 +2408,15 @@ void arch_setup_dma_ops(struct device *dev, u64 dma_base, u64 size,
 	const struct dma_map_ops *dma_ops;
 
 	dev->archdata.dma_coherent = coherent;
+
+	/*
+	 * Don't override the dma_ops if they have already been set. Ideally
+	 * this should be the only location where dma_ops are set, remove this
+	 * check when all other callers of set_dma_ops will have disappeared.
+	 */
+	if (dev->dma_ops)
+		return;
+
 	if (arm_setup_iommu_dma_ops(dev, dma_base, size, iommu))
 		dma_ops = arm_get_iommu_dma_map_ops(coherent);
 	else
-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation