Re: [edk2-devel] [PATCH v4 edk2-platforms 1/1] Silicon/DesignWare/Driver: DwEmacSnpDxe: Add DesignWare EMAC driver

["Ooi, Tzy Way" <tzy.way.ooi@intel.com>]

Hi Bieshuevel,

In the DmaMap function, there is one line where my code look like this:

"Status = DmaMap (MapOperationBusMasterCommonBuffer, Snp->MacDriver.TxdescRing[0], &BufferSize, &Snp->MacDriver.TxdescRingMap[0], &Snp->MappingTxdesc);"

You asked me why the [0] and how many descriptor I am mapping. In my code, there is total 10 descriptors. The way I did in current code was I use DmaAllocateBuffer to allocate a memory size which is big enough for 10 descriptors. After that, I just map the first descriptor and I access the rest of the descriptor by using the first descriptor address plus the size for each descriptor to get the next descriptor address. Wonder if this method is ok? Or do I need to use DmaAllocateBuffer and DmaMap for 10 times for total 10 descriptors? 

Thank you

Best regards,
Tzy Way

-----Original Message-----
From: Ard Biesheuvel <ard.biesheuvel@linaro.org> 
Sent: Friday, May 31, 2019 5:20 PM
To: Ooi, Tzy Way <tzy.way.ooi@intel.com>
Cc: edk2-devel-groups-io <devel@edk2.groups.io>; Kinney, Michael D <michael.d.kinney@intel.com>; Loh, Tien Hock <tien.hock.loh@intel.com>; leif.lindholm@linaro.org
Subject: Re: [edk2-devel] [PATCH v4 edk2-platforms 1/1] Silicon/DesignWare/Driver: DwEmacSnpDxe: Add DesignWare EMAC driver

On Fri, 31 May 2019 at 11:14, Ooi, Tzy Way <tzy.way.ooi@intel.com> wrote:
>
> Thanks Biesheuvel.
>
> I would like to confirm with you on regards of one of the comment. There is one comment where you mentioned that:
> "I am aware that TX is usually the hot path in UEFI, but we should still try to map a buffer directly rather than copy the contents into uncached memory"
>
> I would like to confirm with you if do you mean I should do the DmaMap here instead of doing it at the beginning of the code while DmaAllocate is executed?

Yes.

Each time you receive a buffer, you DmaMap() it with the MapOperationBusMasterRead attribute, and hand it to the device. After the device is finished, you DmaUnmap() it again. This removes the need for uncached memory.

On the RX path, you can allocate the buffers via AllocatePages() rather than via DmaAllocateBuffer(), and map the for DMA using
DmaMap() with the MapOperationBusMasterWrite attribute. Once the buffer is filled by the hardware, you can DmaUnmap() it and copy the contents back to the caller, after which you can recycle the buffer or free it.

> Or do you mean I shouldn’t copy it into a uncached memory region?
>

DmaAllocateBuffer() gives you an uncached region, so you should only use it for data structures that are accessed and modified by the CPU and the hardware at the same time, e.g., the descriptor rings. The actual data buffers only need directional DMA, so using uncached memory results in an unnecessary performance hit.


> Thank you
> -----Original Message-----
> From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> Sent: Wednesday, May 22, 2019 3:22 PM
> To: edk2-devel-groups-io <devel@edk2.groups.io>; Ooi, Tzy Way 
> <tzy.way.ooi@intel.com>
> Cc: Kinney, Michael D <michael.d.kinney@intel.com>; Loh, Tien Hock 
> <tien.hock.loh@intel.com>; leif.lindholm@linaro.org
> Subject: Re: [edk2-devel] [PATCH v4 edk2-platforms 1/1] 
> Silicon/DesignWare/Driver: DwEmacSnpDxe: Add DesignWare EMAC driver
>
> On Wed, 22 May 2019 at 03:37, Ooi, Tzy Way <tzy.way.ooi@intel.com> wrote:
> >
> > Hi Maintainers,
> >
> > Please could you help to review this patch? Thanks.
> >
>
> I already did, but my email response got rejected twice
>
> https://edk2.groups.io/g/devel/message/40829?p=,,,20,0,0,0::Created,,e
> mac,20,2,0,31541649