Using PTDMA driver for generic DMA

Using PTDMA driver for generic DMA

[Tadeusz Struk]

Hello,
I'm trying to use the AMD PTDMA engine for generic data transfers,
but I can't get it to work.
It fails for me with "ptdma 0000:52:00.2: PTDMA error: (null) (0x27)"
which, I'm guessing, translates to "ERR 39: ODMA0_AXI_DECERR"
After some research I've found a note in drivers/dma/ptdma/Kconfig,
which says that "...this DMA controller is intended to be used with
AMD Non-Transparent Bridge devices and not for general purpose 
peripheral DMA."
My question is, what is the "ERR 39: ODMA0_AXI_DECERR", and is it even
possible to use the PTDMA engine to talk to anything else other than
AMD NTB?

Thanks,
Tadeusz

Re: Using PTDMA driver for generic DMA

[Raju Rangoju]

Hello Tadeusz,

Sorry for the delay in response.

The "ERR 39: ODMA0_AXI_DECERR" is a Decode Error relating to destination.

Please note that, the PTDMA DMA controller is intended to be used with 
AMD Non-Transparent Bridge devices and not for general purpose 
peripheral DMA use cases. PTDMA engine can not talk to anything other 
than NTB device.

Thanks,
Raju

On 2/2/2024 10:27 PM, Tadeusz Struk wrote:
> Hello,
> I'm trying to use the AMD PTDMA engine for generic data transfers,
> but I can't get it to work.
> It fails for me with "ptdma 0000:52:00.2: PTDMA error: (null) (0x27)"
> which, I'm guessing, translates to "ERR 39: ODMA0_AXI_DECERR"
> After some research I've found a note in drivers/dma/ptdma/Kconfig,
> which says that "...this DMA controller is intended to be used with
> AMD Non-Transparent Bridge devices and not for general purpose 
> peripheral DMA."
> My question is, what is the "ERR 39: ODMA0_AXI_DECERR", and is it even
> possible to use the PTDMA engine to talk to anything else other than
> AMD NTB?
> 
> Thanks,
> Tadeusz
> 

Re: Using PTDMA driver for generic DMA

[Eric Pilmore]

On Wed, Feb 7, 2024 at 5:26 AM Raju Rangoju <Raju.Rangoju@amd.com> wrote:
>
> Hello Tadeusz,
>
> Sorry for the delay in response.
>
> The "ERR 39: ODMA0_AXI_DECERR" is a Decode Error relating to destination.
>
> Please note that, the PTDMA DMA controller is intended to be used with
> AMD Non-Transparent Bridge devices and not for general purpose
> peripheral DMA use cases. PTDMA engine can not talk to anything other
> than NTB device.
>
> Thanks,
> Raju

Hi Raju,

Are you able to share why the PTDMA has that limitation? How does it
even know whether it's talking to an (AMD) NTB device? Is there some
special handshaking or flow control happening that only works in
conjunction with the AMD NTB?

I understand if that may be a company confidential thing and you can't
disclose, but it just seems odd that a standalone DMA engine would
only function if its destination address represented only one specific
type of device.

Thanks,
Eric

Re: Using PTDMA driver for generic DMA

[Tadeusz Struk]

Hi Raju,
On 2/7/24 14:26, Raju Rangoju wrote:
> Hello Tadeusz,
> 
> Sorry for the delay in response.
> 
> The "ERR 39: ODMA0_AXI_DECERR" is a Decode Error relating to destination.

Could you please send a patch that adds short comments explaining these 
error codes similar to this one?

Thanks and regards,
Tadeusz

[PATCH] dmaengine: ptdma: use consistent DMA masks

[Tadeusz Struk]

The PTDMA driver sets DMA masks in two different places for the same
device inconsistently. First call is in pt_pci_probe(), where it uses
48bit mask. The second call is in pt_dmaengine_register(), where it
uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
on DMA transfers between main memory and other devices.
Without the extra call it works fine. Additionally the second call
doesn't check the return value so it can silently fail.
Remove the superfluous dma_set_mask() call and only use 48bit mask.

Cc: stable@vger.kernel.org
Fixes: b0b4a6b10577 ("dmaengine: ptdma: register PTDMA controller as a DMA resource")

Signed-off-by: Tadeusz Struk <tstruk@gigaio.com>
---
 drivers/dma/ptdma/ptdma-dmaengine.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/dma/ptdma/ptdma-dmaengine.c b/drivers/dma/ptdma/ptdma-dmaengine.c
index 1aa65e5de0f3..f79240734807 100644
--- a/drivers/dma/ptdma/ptdma-dmaengine.c
+++ b/drivers/dma/ptdma/ptdma-dmaengine.c
@@ -385,8 +385,6 @@ int pt_dmaengine_register(struct pt_device *pt)
 	chan->vc.desc_free = pt_do_cleanup;
 	vchan_init(&chan->vc, dma_dev);
 
-	dma_set_mask_and_coherent(pt->dev, DMA_BIT_MASK(64));
-
 	ret = dma_async_device_register(dma_dev);
 	if (ret)
 		goto err_reg;
-- 
2.43.2

Re: [PATCH] dmaengine: ptdma: use consistent DMA masks

[Basavaraj Natikar]

On 2/20/2024 1:40 AM, Tadeusz Struk wrote:
> The PTDMA driver sets DMA masks in two different places for the same
> device inconsistently. First call is in pt_pci_probe(), where it uses
> 48bit mask. The second call is in pt_dmaengine_register(), where it
> uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
> on DMA transfers between main memory and other devices.
> Without the extra call it works fine. Additionally the second call
> doesn't check the return value so it can silently fail.
> Remove the superfluous dma_set_mask() call and only use 48bit mask.
>
> Cc: stable@vger.kernel.org
> Fixes: b0b4a6b10577 ("dmaengine: ptdma: register PTDMA controller as a DMA resource")
>
> Signed-off-by: Tadeusz Struk <tstruk@gigaio.com>
> ---
>  drivers/dma/ptdma/ptdma-dmaengine.c | 2 --
>  1 file changed, 2 deletions(-)
>
> diff --git a/drivers/dma/ptdma/ptdma-dmaengine.c b/drivers/dma/ptdma/ptdma-dmaengine.c
> index 1aa65e5de0f3..f79240734807 100644
> --- a/drivers/dma/ptdma/ptdma-dmaengine.c
> +++ b/drivers/dma/ptdma/ptdma-dmaengine.c
> @@ -385,8 +385,6 @@ int pt_dmaengine_register(struct pt_device *pt)
>  	chan->vc.desc_free = pt_do_cleanup;
>  	vchan_init(&chan->vc, dma_dev);
>  
> -	dma_set_mask_and_coherent(pt->dev, DMA_BIT_MASK(64));
> -

Looks good to me
Reviewed-by: Basavaraj Natikar <Basavaraj.Natikar@amd.com>

>  	ret = dma_async_device_register(dma_dev);
>  	if (ret)
>  		goto err_reg;

Re: [PATCH] dmaengine: ptdma: use consistent DMA masks

[Vinod Koul]

On 19-02-24, 21:10, Tadeusz Struk wrote:
> The PTDMA driver sets DMA masks in two different places for the same
> device inconsistently. First call is in pt_pci_probe(), where it uses
> 48bit mask. The second call is in pt_dmaengine_register(), where it
> uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
> on DMA transfers between main memory and other devices.
> Without the extra call it works fine. Additionally the second call
> doesn't check the return value so it can silently fail.
> Remove the superfluous dma_set_mask() call and only use 48bit mask.
> 
> Cc: stable@vger.kernel.org
> Fixes: b0b4a6b10577 ("dmaengine: ptdma: register PTDMA controller as a DMA resource")
> 
No empty line here please

> Signed-off-by: Tadeusz Struk <tstruk@gigaio.com>

I cant pick this, it was sent by email which this patch was not
signed-off by, please either resend from same id as sob or sign with
both

> ---
>  drivers/dma/ptdma/ptdma-dmaengine.c | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/drivers/dma/ptdma/ptdma-dmaengine.c b/drivers/dma/ptdma/ptdma-dmaengine.c
> index 1aa65e5de0f3..f79240734807 100644
> --- a/drivers/dma/ptdma/ptdma-dmaengine.c
> +++ b/drivers/dma/ptdma/ptdma-dmaengine.c
> @@ -385,8 +385,6 @@ int pt_dmaengine_register(struct pt_device *pt)
>  	chan->vc.desc_free = pt_do_cleanup;
>  	vchan_init(&chan->vc, dma_dev);
>  
> -	dma_set_mask_and_coherent(pt->dev, DMA_BIT_MASK(64));
> -
>  	ret = dma_async_device_register(dma_dev);
>  	if (ret)
>  		goto err_reg;
> -- 
> 2.43.2

-- 
~Vinod

Re: [PATCH] dmaengine: ptdma: use consistent DMA masks

[Tadeusz Struk]

On 2/22/24 14:56, Vinod Koul wrote:
>> Signed-off-by: Tadeusz Struk<tstruk@gigaio.com>
> I cant pick this, it was sent by email which this patch was not
> signed-off by, please either resend from same id as sob or sign with
> both

I will resend with a proper "From" line now.
Thanks,
Tadeusz

[PATCH RESEND] dmaengine: ptdma: use consistent DMA masks

[Tadeusz Struk]

From: Tadeusz Struk <tstruk@gigaio.com>

The PTDMA driver sets DMA masks in two different places for the same
device inconsistently. First call is in pt_pci_probe(), where it uses
48bit mask. The second call is in pt_dmaengine_register(), where it
uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
on DMA transfers between main memory and other devices.
Without the extra call it works fine. Additionally the second call
doesn't check the return value so it can silently fail.
Remove the superfluous dma_set_mask() call and only use 48bit mask.

Cc: stable@vger.kernel.org
Fixes: b0b4a6b10577 ("dmaengine: ptdma: register PTDMA controller as a DMA resource")
Reviewed-by: Basavaraj Natikar <Basavaraj.Natikar@amd.com>
Signed-off-by: Tadeusz Struk <tstruk@gigaio.com>
---
 drivers/dma/ptdma/ptdma-dmaengine.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/dma/ptdma/ptdma-dmaengine.c b/drivers/dma/ptdma/ptdma-dmaengine.c
index 1aa65e5de0f3..f79240734807 100644
--- a/drivers/dma/ptdma/ptdma-dmaengine.c
+++ b/drivers/dma/ptdma/ptdma-dmaengine.c
@@ -385,8 +385,6 @@ int pt_dmaengine_register(struct pt_device *pt)
 	chan->vc.desc_free = pt_do_cleanup;
 	vchan_init(&chan->vc, dma_dev);
 
-	dma_set_mask_and_coherent(pt->dev, DMA_BIT_MASK(64));
-
 	ret = dma_async_device_register(dma_dev);
 	if (ret)
 		goto err_reg;
-- 
2.43.2

Re: [PATCH] dmaengine: ptdma: use consistent DMA masks

[Vinod Koul]

On Mon, 19 Feb 2024 21:10:39 +0100, Tadeusz Struk wrote:
> The PTDMA driver sets DMA masks in two different places for the same
> device inconsistently. First call is in pt_pci_probe(), where it uses
> 48bit mask. The second call is in pt_dmaengine_register(), where it
> uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
> on DMA transfers between main memory and other devices.
> Without the extra call it works fine. Additionally the second call
> doesn't check the return value so it can silently fail.
> Remove the superfluous dma_set_mask() call and only use 48bit mask.
> 
> [...]

Applied, thanks!

[1/1] dmaengine: ptdma: use consistent DMA masks
      commit: df2515a17914ecfc2a0594509deaf7fcb8d191ac

Best regards,
-- 
~Vinod

Re: [PATCH RESEND] dmaengine: ptdma: use consistent DMA masks

[Vinod Koul]

On Thu, 22 Feb 2024 17:30:53 +0100, Tadeusz Struk wrote:
> The PTDMA driver sets DMA masks in two different places for the same
> device inconsistently. First call is in pt_pci_probe(), where it uses
> 48bit mask. The second call is in pt_dmaengine_register(), where it
> uses a 64bit mask. Using 64bit dma mask causes IO_PAGE_FAULT errors
> on DMA transfers between main memory and other devices.
> Without the extra call it works fine. Additionally the second call
> doesn't check the return value so it can silently fail.
> Remove the superfluous dma_set_mask() call and only use 48bit mask.
> 
> [...]

Applied, thanks!

[1/1] dmaengine: ptdma: use consistent DMA masks
      commit: df2515a17914ecfc2a0594509deaf7fcb8d191ac

Best regards,
-- 
~Vinod