[PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[Dmitry Osipenko]

DMA config is incorrect, because of it DMA transfer is never issued and
tegra20_fuse_read() always returns 0x0.

Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 drivers/soc/tegra/fuse/fuse-tegra.c   | 1 +
 drivers/soc/tegra/fuse/fuse-tegra20.c | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/soc/tegra/fuse/fuse-tegra.c b/drivers/soc/tegra/fuse/fuse-tegra.c
index b7c552e3133c..73a3a2c74021 100644
--- a/drivers/soc/tegra/fuse/fuse-tegra.c
+++ b/drivers/soc/tegra/fuse/fuse-tegra.c
@@ -132,6 +132,7 @@ static int tegra_fuse_probe(struct platform_device *pdev)
 
 	/* take over the memory region from the early initialization */
 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	fuse->phys = res->start;
 	fuse->base = devm_ioremap_resource(&pdev->dev, res);
 	if (IS_ERR(fuse->base))
 		return PTR_ERR(fuse->base);
diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c b/drivers/soc/tegra/fuse/fuse-tegra20.c
index 294413a969a0..a33f48c06771 100644
--- a/drivers/soc/tegra/fuse/fuse-tegra20.c
+++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
@@ -59,7 +59,7 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse, unsigned int offset)
 
 	mutex_lock(&fuse->apbdma.lock);
 
-	fuse->apbdma.config.src_addr = fuse->apbdma.phys + FUSE_BEGIN + offset;
+	fuse->apbdma.config.src_addr = fuse->phys + FUSE_BEGIN + offset;
 
 	err = dmaengine_slave_config(fuse->apbdma.chan, &fuse->apbdma.config);
 	if (err)
@@ -119,6 +119,7 @@ static int tegra20_fuse_probe(struct tegra_fuse *fuse)
 	fuse->apbdma.config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
 	fuse->apbdma.config.src_maxburst = 1;
 	fuse->apbdma.config.dst_maxburst = 1;
+	fuse->apbdma.config.direction = DMA_DEV_TO_MEM;
 
 	init_completion(&fuse->apbdma.wait);
 	mutex_init(&fuse->apbdma.lock);
-- 
2.14.1

[PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Dmitry Osipenko]

Currently efuse driver requests DMA channel from an arbitrary DMA device,
it is not a problem since there is only one DMA provider for Tegra20 yet,
but it will become troublesome once another provider would be added.

Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 arch/arm/boot/dts/tegra20.dtsi | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
index fb485a5e63d7..f1579c9a7ef4 100644
--- a/arch/arm/boot/dts/tegra20.dtsi
+++ b/arch/arm/boot/dts/tegra20.dtsi
@@ -600,6 +600,8 @@
 		clock-names = "fuse";
 		resets = <&tegra_car 39>;
 		reset-names = "fuse";
+		dmas = <&apbdma 0>;
+		dma-names = "fuse";
 	};
 
 	pcie@80003000 {
-- 
2.14.1

[PATCH v1 3/3] soc/tegra: fuse: Explicitly request DMA channel from APB DMA driver

[Dmitry Osipenko]

Currently efuse driver requests DMA channel from an arbitrary DMA device,
it is not a problem since there is only one DMA provider for Tegra20 yet,
but it will become troublesome once another provider would be added.

Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 drivers/soc/tegra/fuse/fuse-tegra20.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c b/drivers/soc/tegra/fuse/fuse-tegra20.c
index a33f48c06771..06ec3f9573e6 100644
--- a/drivers/soc/tegra/fuse/fuse-tegra20.c
+++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
@@ -98,14 +98,9 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse, unsigned int offset)
 
 static int tegra20_fuse_probe(struct tegra_fuse *fuse)
 {
-	dma_cap_mask_t mask;
-
-	dma_cap_zero(mask);
-	dma_cap_set(DMA_SLAVE, mask);
-
-	fuse->apbdma.chan = dma_request_channel(mask, NULL, NULL);
-	if (!fuse->apbdma.chan)
-		return -EPROBE_DEFER;
+	fuse->apbdma.chan = dma_request_chan(fuse->dev, "fuse");
+	if (IS_ERR(fuse->apbdma.chan))
+		return PTR_ERR(fuse->apbdma.chan);
 
 	fuse->apbdma.virt = dma_alloc_coherent(fuse->dev, sizeof(u32),
 					       &fuse->apbdma.phys,
-- 
2.14.1

Re: [PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[Jon Hunter]

Hi Dmitry,

On 25/09/17 23:35, Dmitry Osipenko wrote:
> DMA config is incorrect, because of it DMA transfer is never issued and
> tegra20_fuse_read() always returns 0x0.
> 
> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> ---
>  drivers/soc/tegra/fuse/fuse-tegra.c   | 1 +
>  drivers/soc/tegra/fuse/fuse-tegra20.c | 3 ++-
>  2 files changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/soc/tegra/fuse/fuse-tegra.c b/drivers/soc/tegra/fuse/fuse-tegra.c
> index b7c552e3133c..73a3a2c74021 100644
> --- a/drivers/soc/tegra/fuse/fuse-tegra.c
> +++ b/drivers/soc/tegra/fuse/fuse-tegra.c
> @@ -132,6 +132,7 @@ static int tegra_fuse_probe(struct platform_device *pdev)
>  
>  	/* take over the memory region from the early initialization */
>  	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	fuse->phys = res->start;
>  	fuse->base = devm_ioremap_resource(&pdev->dev, res);
>  	if (IS_ERR(fuse->base))
>  		return PTR_ERR(fuse->base);
> diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c b/drivers/soc/tegra/fuse/fuse-tegra20.c
> index 294413a969a0..a33f48c06771 100644
> --- a/drivers/soc/tegra/fuse/fuse-tegra20.c
> +++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
> @@ -59,7 +59,7 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse, unsigned int offset)
>  
>  	mutex_lock(&fuse->apbdma.lock);
>  
> -	fuse->apbdma.config.src_addr = fuse->apbdma.phys + FUSE_BEGIN + offset;
> +	fuse->apbdma.config.src_addr = fuse->phys + FUSE_BEGIN + offset;
>  
>  	err = dmaengine_slave_config(fuse->apbdma.chan, &fuse->apbdma.config);
>  	if (err)
> @@ -119,6 +119,7 @@ static int tegra20_fuse_probe(struct tegra_fuse *fuse)
>  	fuse->apbdma.config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>  	fuse->apbdma.config.src_maxburst = 1;
>  	fuse->apbdma.config.dst_maxburst = 1;
> +	fuse->apbdma.config.direction = DMA_DEV_TO_MEM;
>  
>  	init_completion(&fuse->apbdma.wait);
>  	mutex_init(&fuse->apbdma.lock);

Thanks for the fix.

When booting the mainline on Tegra20 Trimslice, I only see the
tegra20_fuse_read_early() called and not the tegra20_fuse_read(). That's
not to say it is not needed, but I am wondering if we really need this
complex tegra20_fuse_read() using DMA and whether we should just have
the normal fuse->read() call tegra20_fuse_read_early() as well to
simplify matters?

Maybe Thierry or Stephen know the history here?

As far as the fix is concerned ...

Reviewed-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[Stephen Warren]

On 09/26/2017 08:08 AM, Jon Hunter wrote:
> Hi Dmitry,
> 
> On 25/09/17 23:35, Dmitry Osipenko wrote:
>> DMA config is incorrect, because of it DMA transfer is never issued and
>> tegra20_fuse_read() always returns 0x0.
>>
>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>> ---
>>   drivers/soc/tegra/fuse/fuse-tegra.c   | 1 +
>>   drivers/soc/tegra/fuse/fuse-tegra20.c | 3 ++-
>>   2 files changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra.c b/drivers/soc/tegra/fuse/fuse-tegra.c
>> index b7c552e3133c..73a3a2c74021 100644
>> --- a/drivers/soc/tegra/fuse/fuse-tegra.c
>> +++ b/drivers/soc/tegra/fuse/fuse-tegra.c
>> @@ -132,6 +132,7 @@ static int tegra_fuse_probe(struct platform_device *pdev)
>>   
>>   	/* take over the memory region from the early initialization */
>>   	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>> +	fuse->phys = res->start;
>>   	fuse->base = devm_ioremap_resource(&pdev->dev, res);
>>   	if (IS_ERR(fuse->base))
>>   		return PTR_ERR(fuse->base);
>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c b/drivers/soc/tegra/fuse/fuse-tegra20.c
>> index 294413a969a0..a33f48c06771 100644
>> --- a/drivers/soc/tegra/fuse/fuse-tegra20.c
>> +++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
>> @@ -59,7 +59,7 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse, unsigned int offset)
>>   
>>   	mutex_lock(&fuse->apbdma.lock);
>>   
>> -	fuse->apbdma.config.src_addr = fuse->apbdma.phys + FUSE_BEGIN + offset;
>> +	fuse->apbdma.config.src_addr = fuse->phys + FUSE_BEGIN + offset;
>>   
>>   	err = dmaengine_slave_config(fuse->apbdma.chan, &fuse->apbdma.config);
>>   	if (err)
>> @@ -119,6 +119,7 @@ static int tegra20_fuse_probe(struct tegra_fuse *fuse)
>>   	fuse->apbdma.config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>>   	fuse->apbdma.config.src_maxburst = 1;
>>   	fuse->apbdma.config.dst_maxburst = 1;
>> +	fuse->apbdma.config.direction = DMA_DEV_TO_MEM;
>>   
>>   	init_completion(&fuse->apbdma.wait);
>>   	mutex_init(&fuse->apbdma.lock);
> 
> Thanks for the fix.
> 
> When booting the mainline on Tegra20 Trimslice, I only see the
> tegra20_fuse_read_early() called and not the tegra20_fuse_read(). That's
> not to say it is not needed, but I am wondering if we really need this
> complex tegra20_fuse_read() using DMA and whether we should just have
> the normal fuse->read() call tegra20_fuse_read_early() as well to
> simplify matters?
> 
> Maybe Thierry or Stephen know the history here?

There's some HW bug related to reading the fuse registers from the CPU. 
The fix was implemented long ago by Olof; see the git commit description 
pasted below. IIRC, the code directly reads the fuse registers before 
the point where DMA is available (the argument being we can't do 
anything else, and this period of time is short so the risk hopefully 
low), but once DMA is available, it is used to avoid the HW bug. The bug 
was apparently fixed in Tegra30; see the other commit description pasted 
below.

> commit e2f91578b35347341482f6af9e4fcf3174531efd
> Author: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
> Date:   Wed Oct 12 23:52:29 2011 -0700
> 
>     ARM: tegra: use APB DMA for accessing APB devices
>     
>     Tegra2 hangs if APB registers are accessed from the cpu during an
>     apb dma operation. The workaround is to use apb dma to read/write the
>     registers instead.
>     
>     There is a dependency loop between fuses, clocks, and APBDMA.  If dma
>     is enabled, fuse reads must go through APBDMA to avoid corruption due
>     to a hw bug.  APBDMA requires a clock to be enabled.  Clocks must read
>     a fuse to determine allowable cpu frequencies.
>     
>     Separate out the fuse DMA initialization, and allow the fuse read
>     and write functions to be called without using DMA before the DMA
>     initialization has been completed.  Access to the fuses before APBDMA
>     is initialized won't hit the hardware bug because nothing else can be
>     using DMA.
>     
>     Original fuse registar access code from Varun Wadekar
>     <vwadekar-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>, improved by Colin Cross <ccross-z5hGa2qSFaRBDgjK7y7TUQ@public.gmane.org>
>     and later moved to separate driver by Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>     
>     Major refactoring/cleanup by Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>.
>     
>     Changes since v1:
>     
>     * fix 'return false' on error condition
>     * dequeue dma ops in case of timeout
>     
>     From: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>     Signed-off-by: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>     Signed-off-by: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
>     Acked-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

> commit b861c275ea5cfeab32241c3c92a203579d5699ff
> Author: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> Date:   Wed Jun 20 18:06:34 2012 +0530
> 
>     ARM: tegra: apbio access using dma for tegra20 only
>     
>     The Tegra20 HW issue with accessing APBIO registers (such
>     as fuse registers) directly from the CPU concurrently with
>     APB DMA accesses has been fixed in Tegra30 and later chips.
>     
>     Access these registers directly from the CPU on Tegra30
>     and later, and apply the workaround only for Tegra20.
>     
>     Signed-off-by: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>     Tested-by: Chaitanya Bandi <bandik-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>     Signed-off-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Re: [PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[Dmitry Osipenko]

On 26.09.2017 19:08, Stephen Warren wrote:
> On 09/26/2017 08:08 AM, Jon Hunter wrote:
>> Hi Dmitry,
>>
>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>> DMA config is incorrect, because of it DMA transfer is never issued and
>>> tegra20_fuse_read() always returns 0x0.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>> ---
>>>   drivers/soc/tegra/fuse/fuse-tegra.c   | 1 +
>>>   drivers/soc/tegra/fuse/fuse-tegra20.c | 3 ++-
>>>   2 files changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra.c
>>> b/drivers/soc/tegra/fuse/fuse-tegra.c
>>> index b7c552e3133c..73a3a2c74021 100644
>>> --- a/drivers/soc/tegra/fuse/fuse-tegra.c
>>> +++ b/drivers/soc/tegra/fuse/fuse-tegra.c
>>> @@ -132,6 +132,7 @@ static int tegra_fuse_probe(struct platform_device *pdev)
>>>         /* take over the memory region from the early initialization */
>>>       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>>> +    fuse->phys = res->start;
>>>       fuse->base = devm_ioremap_resource(&pdev->dev, res);
>>>       if (IS_ERR(fuse->base))
>>>           return PTR_ERR(fuse->base);
>>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c
>>> b/drivers/soc/tegra/fuse/fuse-tegra20.c
>>> index 294413a969a0..a33f48c06771 100644
>>> --- a/drivers/soc/tegra/fuse/fuse-tegra20.c
>>> +++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
>>> @@ -59,7 +59,7 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse,
>>> unsigned int offset)
>>>         mutex_lock(&fuse->apbdma.lock);
>>>   -    fuse->apbdma.config.src_addr = fuse->apbdma.phys + FUSE_BEGIN + offset;
>>> +    fuse->apbdma.config.src_addr = fuse->phys + FUSE_BEGIN + offset;
>>>         err = dmaengine_slave_config(fuse->apbdma.chan, &fuse->apbdma.config);
>>>       if (err)
>>> @@ -119,6 +119,7 @@ static int tegra20_fuse_probe(struct tegra_fuse *fuse)
>>>       fuse->apbdma.config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>>>       fuse->apbdma.config.src_maxburst = 1;
>>>       fuse->apbdma.config.dst_maxburst = 1;
>>> +    fuse->apbdma.config.direction = DMA_DEV_TO_MEM;
>>>         init_completion(&fuse->apbdma.wait);
>>>       mutex_init(&fuse->apbdma.lock);
>>
>> Thanks for the fix.
>>
>> When booting the mainline on Tegra20 Trimslice, I only see the
>> tegra20_fuse_read_early() called and not the tegra20_fuse_read(). That's
>> not to say it is not needed, but I am wondering if we really need this
>> complex tegra20_fuse_read() using DMA and whether we should just have
>> the normal fuse->read() call tegra20_fuse_read_early() as well to
>> simplify matters?
>>
>> Maybe Thierry or Stephen know the history here?
> 
> There's some HW bug related to reading the fuse registers from the CPU. The fix
> was implemented long ago by Olof; see the git commit description pasted below.
> IIRC, the code directly reads the fuse registers before the point where DMA is
> available (the argument being we can't do anything else, and this period of time
> is short so the risk hopefully low), but once DMA is available, it is used to
> avoid the HW bug. The bug was apparently fixed in Tegra30; see the other commit
> description pasted below.
> 
>> commit e2f91578b35347341482f6af9e4fcf3174531efd
>> Author: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
>> Date:   Wed Oct 12 23:52:29 2011 -0700
>>
>>     ARM: tegra: use APB DMA for accessing APB devices
>>         Tegra2 hangs if APB registers are accessed from the cpu during an
>>     apb dma operation. The workaround is to use apb dma to read/write the
>>     registers instead.
>>         There is a dependency loop between fuses, clocks, and APBDMA.  If dma
>>     is enabled, fuse reads must go through APBDMA to avoid corruption due
>>     to a hw bug.  APBDMA requires a clock to be enabled.  Clocks must read
>>     a fuse to determine allowable cpu frequencies.
>>         Separate out the fuse DMA initialization, and allow the fuse read
>>     and write functions to be called without using DMA before the DMA
>>     initialization has been completed.  Access to the fuses before APBDMA
>>     is initialized won't hit the hardware bug because nothing else can be
>>     using DMA.
>>         Original fuse registar access code from Varun Wadekar
>>     <vwadekar-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>, improved by Colin Cross <ccross-z5hGa2qSFaRBDgjK7y7TUQ@public.gmane.org>
>>     and later moved to separate driver by Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>         Major refactoring/cleanup by Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>.
>>         Changes since v1:
>>         * fix 'return false' on error condition
>>     * dequeue dma ops in case of timeout
>>         From: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>     Signed-off-by: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>     Signed-off-by: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
>>     Acked-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> 
>> commit b861c275ea5cfeab32241c3c92a203579d5699ff
>> Author: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>> Date:   Wed Jun 20 18:06:34 2012 +0530
>>
>>     ARM: tegra: apbio access using dma for tegra20 only
>>         The Tegra20 HW issue with accessing APBIO registers (such
>>     as fuse registers) directly from the CPU concurrently with
>>     APB DMA accesses has been fixed in Tegra30 and later chips.
>>         Access these registers directly from the CPU on Tegra30
>>     and later, and apply the workaround only for Tegra20.
>>         Signed-off-by: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>     Tested-by: Chaitanya Bandi <bandik-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>     Signed-off-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Thank you very much for the clarification.

-- 
Dmitry

Re: [PATCH v1 1/3] soc/tegra: fuse: Fix reading registers using DMA on Tegra20

[Jon Hunter]

On 26/09/17 18:31, Dmitry Osipenko wrote:
> On 26.09.2017 19:08, Stephen Warren wrote:
>> On 09/26/2017 08:08 AM, Jon Hunter wrote:
>>> Hi Dmitry,
>>>
>>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>>> DMA config is incorrect, because of it DMA transfer is never issued and
>>>> tegra20_fuse_read() always returns 0x0.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>> ---
>>>>   drivers/soc/tegra/fuse/fuse-tegra.c   | 1 +
>>>>   drivers/soc/tegra/fuse/fuse-tegra20.c | 3 ++-
>>>>   2 files changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra.c
>>>> b/drivers/soc/tegra/fuse/fuse-tegra.c
>>>> index b7c552e3133c..73a3a2c74021 100644
>>>> --- a/drivers/soc/tegra/fuse/fuse-tegra.c
>>>> +++ b/drivers/soc/tegra/fuse/fuse-tegra.c
>>>> @@ -132,6 +132,7 @@ static int tegra_fuse_probe(struct platform_device *pdev)
>>>>         /* take over the memory region from the early initialization */
>>>>       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>>>> +    fuse->phys = res->start;
>>>>       fuse->base = devm_ioremap_resource(&pdev->dev, res);
>>>>       if (IS_ERR(fuse->base))
>>>>           return PTR_ERR(fuse->base);
>>>> diff --git a/drivers/soc/tegra/fuse/fuse-tegra20.c
>>>> b/drivers/soc/tegra/fuse/fuse-tegra20.c
>>>> index 294413a969a0..a33f48c06771 100644
>>>> --- a/drivers/soc/tegra/fuse/fuse-tegra20.c
>>>> +++ b/drivers/soc/tegra/fuse/fuse-tegra20.c
>>>> @@ -59,7 +59,7 @@ static u32 tegra20_fuse_read(struct tegra_fuse *fuse,
>>>> unsigned int offset)
>>>>         mutex_lock(&fuse->apbdma.lock);
>>>>   -    fuse->apbdma.config.src_addr = fuse->apbdma.phys + FUSE_BEGIN + offset;
>>>> +    fuse->apbdma.config.src_addr = fuse->phys + FUSE_BEGIN + offset;
>>>>         err = dmaengine_slave_config(fuse->apbdma.chan, &fuse->apbdma.config);
>>>>       if (err)
>>>> @@ -119,6 +119,7 @@ static int tegra20_fuse_probe(struct tegra_fuse *fuse)
>>>>       fuse->apbdma.config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
>>>>       fuse->apbdma.config.src_maxburst = 1;
>>>>       fuse->apbdma.config.dst_maxburst = 1;
>>>> +    fuse->apbdma.config.direction = DMA_DEV_TO_MEM;
>>>>         init_completion(&fuse->apbdma.wait);
>>>>       mutex_init(&fuse->apbdma.lock);
>>>
>>> Thanks for the fix.
>>>
>>> When booting the mainline on Tegra20 Trimslice, I only see the
>>> tegra20_fuse_read_early() called and not the tegra20_fuse_read(). That's
>>> not to say it is not needed, but I am wondering if we really need this
>>> complex tegra20_fuse_read() using DMA and whether we should just have
>>> the normal fuse->read() call tegra20_fuse_read_early() as well to
>>> simplify matters?
>>>
>>> Maybe Thierry or Stephen know the history here?
>>
>> There's some HW bug related to reading the fuse registers from the CPU. The fix
>> was implemented long ago by Olof; see the git commit description pasted below.
>> IIRC, the code directly reads the fuse registers before the point where DMA is
>> available (the argument being we can't do anything else, and this period of time
>> is short so the risk hopefully low), but once DMA is available, it is used to
>> avoid the HW bug. The bug was apparently fixed in Tegra30; see the other commit
>> description pasted below.
>>
>>> commit e2f91578b35347341482f6af9e4fcf3174531efd
>>> Author: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
>>> Date:   Wed Oct 12 23:52:29 2011 -0700
>>>
>>>     ARM: tegra: use APB DMA for accessing APB devices
>>>         Tegra2 hangs if APB registers are accessed from the cpu during an
>>>     apb dma operation. The workaround is to use apb dma to read/write the
>>>     registers instead.
>>>         There is a dependency loop between fuses, clocks, and APBDMA.  If dma
>>>     is enabled, fuse reads must go through APBDMA to avoid corruption due
>>>     to a hw bug.  APBDMA requires a clock to be enabled.  Clocks must read
>>>     a fuse to determine allowable cpu frequencies.
>>>         Separate out the fuse DMA initialization, and allow the fuse read
>>>     and write functions to be called without using DMA before the DMA
>>>     initialization has been completed.  Access to the fuses before APBDMA
>>>     is initialized won't hit the hardware bug because nothing else can be
>>>     using DMA.
>>>         Original fuse registar access code from Varun Wadekar
>>>     <vwadekar-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>, improved by Colin Cross <ccross-z5hGa2qSFaRBDgjK7y7TUQ@public.gmane.org>
>>>     and later moved to separate driver by Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>>         Major refactoring/cleanup by Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>.
>>>         Changes since v1:
>>>         * fix 'return false' on error condition
>>>     * dequeue dma ops in case of timeout
>>>         From: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>>     Signed-off-by: Jon Mayo <jmayo-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>.
>>>     Signed-off-by: Olof Johansson <olof-nZhT3qVonbNeoWH0uzbU5w@public.gmane.org>
>>>     Acked-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>
>>> commit b861c275ea5cfeab32241c3c92a203579d5699ff
>>> Author: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>> Date:   Wed Jun 20 18:06:34 2012 +0530
>>>
>>>     ARM: tegra: apbio access using dma for tegra20 only
>>>         The Tegra20 HW issue with accessing APBIO registers (such
>>>     as fuse registers) directly from the CPU concurrently with
>>>     APB DMA accesses has been fixed in Tegra30 and later chips.
>>>         Access these registers directly from the CPU on Tegra30
>>>     and later, and apply the workaround only for Tegra20.
>>>         Signed-off-by: Laxman Dewangan <ldewangan-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>>     Tested-by: Chaitanya Bandi <bandik-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>>     Signed-off-by: Stephen Warren <swarren-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> 
> Thank you very much for the clarification.

Ditto, being lazy I had only looked back as far as when it move to
drivers/soc/tegra, should have looked further!

Great, so looks like a good fix.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Jon Hunter]

On 25/09/17 23:35, Dmitry Osipenko wrote:
> Currently efuse driver requests DMA channel from an arbitrary DMA device,
> it is not a problem since there is only one DMA provider for Tegra20 yet,
> but it will become troublesome once another provider would be added.
> 
> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> ---
>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
> index fb485a5e63d7..f1579c9a7ef4 100644
> --- a/arch/arm/boot/dts/tegra20.dtsi
> +++ b/arch/arm/boot/dts/tegra20.dtsi
> @@ -600,6 +600,8 @@
>  		clock-names = "fuse";
>  		resets = <&tegra_car 39>;
>  		reset-names = "fuse";
> +		dmas = <&apbdma 0>;
> +		dma-names = "fuse";
>  	};
>  
>  	pcie@80003000 {
> 

Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Jon Hunter]

On 26/09/17 21:54, Jon Hunter wrote:
> 
> On 25/09/17 23:35, Dmitry Osipenko wrote:
>> Currently efuse driver requests DMA channel from an arbitrary DMA device,
>> it is not a problem since there is only one DMA provider for Tegra20 yet,
>> but it will become troublesome once another provider would be added.
>>
>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>> ---
>>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>>  1 file changed, 2 insertions(+)
>>
>> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
>> index fb485a5e63d7..f1579c9a7ef4 100644
>> --- a/arch/arm/boot/dts/tegra20.dtsi
>> +++ b/arch/arm/boot/dts/tegra20.dtsi
>> @@ -600,6 +600,8 @@
>>  		clock-names = "fuse";
>>  		resets = <&tegra_car 39>;
>>  		reset-names = "fuse";
>> +		dmas = <&apbdma 0>;
>> +		dma-names = "fuse";
>>  	};
>>  
>>  	pcie@80003000 {
>>
> 
> Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Actually, request-id '0' is a valid request. Does this work ok?

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Dmitry Osipenko]

On 27.09.2017 00:25, Jon Hunter wrote:
> 
> On 26/09/17 21:54, Jon Hunter wrote:
>>
>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>> Currently efuse driver requests DMA channel from an arbitrary DMA device,
>>> it is not a problem since there is only one DMA provider for Tegra20 yet,
>>> but it will become troublesome once another provider would be added.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>> ---
>>>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>>>  1 file changed, 2 insertions(+)
>>>
>>> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
>>> index fb485a5e63d7..f1579c9a7ef4 100644
>>> --- a/arch/arm/boot/dts/tegra20.dtsi
>>> +++ b/arch/arm/boot/dts/tegra20.dtsi
>>> @@ -600,6 +600,8 @@
>>>  		clock-names = "fuse";
>>>  		resets = <&tegra_car 39>;
>>>  		reset-names = "fuse";
>>> +		dmas = <&apbdma 0>;
>>> +		dma-names = "fuse";
>>>  	};
>>>  
>>>  	pcie@80003000 {
>>>
>>
>> Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> 
> Actually, request-id '0' is a valid request. Does this work ok?
> 

It works fine, I have verified that reading on CPU == reading by DMA. The
REQ_SEL 0 is "Not Assigned" and seems acts as DRQ=1. I know that it is not
entirely correct, but APB DMA driver is hardcoded to the master mode, while we
need slave mode.

-- 
Dmitry

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Jon Hunter]

On 26/09/17 22:54, Dmitry Osipenko wrote:
> On 27.09.2017 00:25, Jon Hunter wrote:
>>
>> On 26/09/17 21:54, Jon Hunter wrote:
>>>
>>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>>> Currently efuse driver requests DMA channel from an arbitrary DMA device,
>>>> it is not a problem since there is only one DMA provider for Tegra20 yet,
>>>> but it will become troublesome once another provider would be added.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>> ---
>>>>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>>>>  1 file changed, 2 insertions(+)
>>>>
>>>> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
>>>> index fb485a5e63d7..f1579c9a7ef4 100644
>>>> --- a/arch/arm/boot/dts/tegra20.dtsi
>>>> +++ b/arch/arm/boot/dts/tegra20.dtsi
>>>> @@ -600,6 +600,8 @@
>>>>  		clock-names = "fuse";
>>>>  		resets = <&tegra_car 39>;
>>>>  		reset-names = "fuse";
>>>> +		dmas = <&apbdma 0>;
>>>> +		dma-names = "fuse";
>>>>  	};
>>>>  
>>>>  	pcie@80003000 {
>>>>
>>>
>>> Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>
>> Actually, request-id '0' is a valid request. Does this work ok?
>>
> 
> It works fine, I have verified that reading on CPU == reading by DMA. The
> REQ_SEL 0 is "Not Assigned" and seems acts as DRQ=1. I know that it is not
> entirely correct, but APB DMA driver is hardcoded to the master mode, while we
> need slave mode.

Looking at the TRM I see that it is 'CNTR_REQ' so I am not sure if this
is a timer/counter that is driving this.

Jon

-- 
nvpublic

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Dmitry Osipenko]

On 27.09.2017 01:10, Jon Hunter wrote:
> 
> On 26/09/17 22:54, Dmitry Osipenko wrote:
>> On 27.09.2017 00:25, Jon Hunter wrote:
>>>
>>> On 26/09/17 21:54, Jon Hunter wrote:
>>>>
>>>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>>>> Currently efuse driver requests DMA channel from an arbitrary DMA device,
>>>>> it is not a problem since there is only one DMA provider for Tegra20 yet,
>>>>> but it will become troublesome once another provider would be added.
>>>>>
>>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>> ---
>>>>>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>>>>>  1 file changed, 2 insertions(+)
>>>>>
>>>>> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
>>>>> index fb485a5e63d7..f1579c9a7ef4 100644
>>>>> --- a/arch/arm/boot/dts/tegra20.dtsi
>>>>> +++ b/arch/arm/boot/dts/tegra20.dtsi
>>>>> @@ -600,6 +600,8 @@
>>>>>  		clock-names = "fuse";
>>>>>  		resets = <&tegra_car 39>;
>>>>>  		reset-names = "fuse";
>>>>> +		dmas = <&apbdma 0>;
>>>>> +		dma-names = "fuse";
>>>>>  	};
>>>>>  
>>>>>  	pcie@80003000 {
>>>>>
>>>>
>>>> Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>>
>>> Actually, request-id '0' is a valid request. Does this work ok?
>>>
>>
>> It works fine, I have verified that reading on CPU == reading by DMA. The
>> REQ_SEL 0 is "Not Assigned" and seems acts as DRQ=1. I know that it is not
>> entirely correct, but APB DMA driver is hardcoded to the master mode, while we
>> need slave mode.
> 
> Looking at the TRM I see that it is 'CNTR_REQ' so I am not sure if this
> is a timer/counter that is driving this.
> 

Oh, wow. Indeed it's TRIG_SEL is NA. Good catch! So it works because counter = 0.

Then APB DMA driver needs to be changed to support master mode. Seems that
should be simple to change, I'll look into it.

-- 
Dmitry

Re: [PATCH v1 2/3] ARM: dts: tegra20: Add DMA phandle to 'fuse' node

[Dmitry Osipenko]

On 27.09.2017 01:31, Dmitry Osipenko wrote:
> On 27.09.2017 01:10, Jon Hunter wrote:
>>
>> On 26/09/17 22:54, Dmitry Osipenko wrote:
>>> On 27.09.2017 00:25, Jon Hunter wrote:
>>>>
>>>> On 26/09/17 21:54, Jon Hunter wrote:
>>>>>
>>>>> On 25/09/17 23:35, Dmitry Osipenko wrote:
>>>>>> Currently efuse driver requests DMA channel from an arbitrary DMA device,
>>>>>> it is not a problem since there is only one DMA provider for Tegra20 yet,
>>>>>> but it will become troublesome once another provider would be added.
>>>>>>
>>>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>>> ---
>>>>>>  arch/arm/boot/dts/tegra20.dtsi | 2 ++
>>>>>>  1 file changed, 2 insertions(+)
>>>>>>
>>>>>> diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
>>>>>> index fb485a5e63d7..f1579c9a7ef4 100644
>>>>>> --- a/arch/arm/boot/dts/tegra20.dtsi
>>>>>> +++ b/arch/arm/boot/dts/tegra20.dtsi
>>>>>> @@ -600,6 +600,8 @@
>>>>>>  		clock-names = "fuse";
>>>>>>  		resets = <&tegra_car 39>;
>>>>>>  		reset-names = "fuse";
>>>>>> +		dmas = <&apbdma 0>;
>>>>>> +		dma-names = "fuse";
>>>>>>  	};
>>>>>>  
>>>>>>  	pcie@80003000 {
>>>>>>
>>>>>
>>>>> Acked-by: Jon Hunter <jonathanh-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>>>
>>>> Actually, request-id '0' is a valid request. Does this work ok?
>>>>
>>>
>>> It works fine, I have verified that reading on CPU == reading by DMA. The
>>> REQ_SEL 0 is "Not Assigned" and seems acts as DRQ=1. I know that it is not
>>> entirely correct, but APB DMA driver is hardcoded to the master mode, while we
>>> need slave mode.
>>
>> Looking at the TRM I see that it is 'CNTR_REQ' so I am not sure if this
>> is a timer/counter that is driving this.
>>
> 
> Oh, wow. Indeed it's TRIG_SEL is NA. Good catch! So it works because counter = 0.
> 
> Then APB DMA driver needs to be changed to support master mode. Seems that
> should be simple to change, I'll look into it.
> 

s/support master/support slave/ of course. Thank you for the review ;)

-- 
Dmitry