[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <BY5PR07MB811989E3A3568B0B787A610CB0929@BY5PR07MB8119.namprd07.prod.outlook.com>
Date: Tue, 9 Mar 2021 10:49:27 +0000
From: Sanket Parmar <sparmar@...ence.com>
To: Christoph Hellwig <hch@...radead.org>
CC: "peter.chen@...nel.org" <peter.chen@...nel.org>,
Pawel Laszczak <pawell@...ence.com>,
"a-govindraju@...com" <a-govindraju@...com>,
"linux-usb@...r.kernel.org" <linux-usb@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Rahul Kumar <kurahul@...ence.com>,
"gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
"kishon@...com" <kishon@...com>
Subject: RE: [PATCH 2/2] usb: cdns3: Optimize DMA request buffer allocation
> On Tue, Mar 09, 2021 at 10:18:43AM +0000, Sanket Parmar wrote:
> > > On Tue, Mar 09, 2021 at 06:19:40AM +0100, Sanket Parmar wrote:
> > > > dma_alloc_coherent() might fail on the platform with a small DMA
> region.
> > > >
> > > > To avoid such failure in cdns3_prepare_aligned_request_buf(),
> > > > dma_alloc_coherent() is replaced with kmalloc and dma_map API to
> > > > allocate aligned request buffer of dynamic length.
> > >
> > > dma_alloc_noncoherent is the proper API instead of using kmalloc, which
> > > can lead to unaddressable memory that might require bounce buffering.
> >
> > cdns3 device required DMA coherent buffer to perform operations. So
> > dma_alloc_noncoherent will not help here.
> >
> > Also all gadget classes(except g_ether) use kmalloc to allocated request
> buffer,
> > and device driver uses usb_gadget_map_request_by_dev to map the
> request
> > buffer. Similar approach is used to allocate aligned buffer.
>
> If you can use kmalloc and dma_map_single you can use
> dma_alloc_noncoherent per definition.
Okay. I was not aware of it. I will test it.
Thank you for your feedback.
--
Sanket
Powered by blists - more mailing lists