Re: [PATCH] ARM: dts: bcm2711: fix soc's node dma-ranges

From: Phil Elwell <phil@raspberrypi.org>
To: Florian Fainelli <f.fainelli@gmail.com>,
	Nicolas Saenz Julienne <nsaenzjulienne@suse.de>,
	Rob Herring <robh+dt@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Eric Anholt <eric@anholt.net>, Stefan Wahren <wahrenst@gmx.net>
Cc: mbrugger@suse.com, devicetree@vger.kernel.org,
	bcm-kernel-feedback-list@broadcom.com,
	linux-rpi-kernel@lists.infradead.org,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] ARM: dts: bcm2711: fix soc's node dma-ranges
Date: Fri, 6 Dec 2019 10:16:07 +0000	[thread overview]
Message-ID: <e72de603-2ad9-5a3b-109e-8ee14bf3293c@raspberrypi.org> (raw)
In-Reply-To: <711470d3-e683-69d4-8f4e-791a76faab29@gmail.com>

Hi Nicolas,

On 06/12/2019 00:08, Florian Fainelli wrote:
> On 12/4/19 4:56 AM, Nicolas Saenz Julienne wrote:
>> Raspberry Pi's firmware has a feature to select how much memory to
>> reserve for its GPU called 'gpu_mem'. The possible values go from 16MB
>> to 944MB, with a default of 64MB. This memory resides in the topmost
>> part of the lower 1GB memory area and grows bigger expanding towards the
>> begging of memory.
>>
>> It turns out that with low 'gpu_mem' values (16MB and 32MB) the size of
>> the memory available to the system in the lower 1GB area can outgrow the
>> interconnect's dma-range as its size was selected based on the maximum
>> system memory available given the default gpu_mem configuration. This
>> makes that memory slice unavailable for DMA. And may cause nasty kernel
>> warnings if CMA happens to include it.
>>
>> Change soc's dma-ranges to really reflect it's HW limitation, which is
>> being able to only DMA to the lower 1GB area.
>>
>> Fixes: 7dbe8c62ceeb ("ARM: dts: Add minimal Raspberry Pi 4 support")
>> Signed-off-by: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
>> ---
>>
>> NOTE: I'd appreciate if someone from the RPi foundation commented on
>> this as it's something that I'll propose to be backported to their tree.

The 0x3c000000 size was a mistake that arose from c0000000 + 3c000000 = 
fc000000, but that is mixing apples and oranges (actually DMA addresses
and host physical addresses). Please correct it as you are proposing.

> 
> I don't think our additional DTS changes will be merged until -rc1 is
> cut, so we have some time to figure this one out. Thanks
> 
>>
>>   arch/arm/boot/dts/bcm2711.dtsi | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/arm/boot/dts/bcm2711.dtsi b/arch/arm/boot/dts/bcm2711.dtsi
>> index 5b61cd915f2b..d6a0e350b7b4 100644
>> --- a/arch/arm/boot/dts/bcm2711.dtsi
>> +++ b/arch/arm/boot/dts/bcm2711.dtsi
>> @@ -43,7 +43,7 @@ soc {
>>   			 <0x7c000000  0x0 0xfc000000  0x02000000>,
>>   			 <0x40000000  0x0 0xff800000  0x00800000>;
>>   		/* Emulate a contiguous 30-bit address range for DMA */
>> -		dma-ranges = <0xc0000000  0x0 0x00000000  0x3c000000>;
>> +		dma-ranges = <0xc0000000  0x0 0x00000000  0x40000000>;
>>   
>>   		/*
>>   		 * This node is the provider for the enable-method for
>>
> 
> 