From: "NG, TZE YEE" <tze.yee.ng@altera.com>
To: Frank Li <Frank.li@nxp.com>
Cc: Olivier Dautricourt <olivierdautricourt@gmail.com>,
Stefan Roese <sr@denx.de>, Vinod Koul <vkoul@kernel.org>,
Frank Li <Frank.Li@kernel.org>,
"dmaengine@vger.kernel.org" <dmaengine@vger.kernel.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"NG, ADRIAN HO YIN" <adrian.ho.yin.ng@altera.com>,
"Nazle Asmade,
Muhammad Nazim Amirul"
<muhammad.nazim.amirul.nazle.asmade@altera.com>
Subject: Re: [PATCH] dma: altera-msgdma: Replace memcpy with io32write in msgdma_copy_one
Date: Mon, 25 May 2026 06:17:36 +0000 [thread overview]
Message-ID: <8d9bd7a0-4927-40d8-a747-c3d7cf34cf67@altera.com> (raw)
In-Reply-To: <ag4n0ycl1KB3p0hP@lizhi-Precision-Tower-5810>
On 21/5/2026 5:29 am, Frank Li wrote:
> [Some people who received this message don't often get email from frank.li@nxp.com. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
>
> On Mon, May 18, 2026 at 11:47:20PM -0700, tze.yee.ng@altera.com wrote:
>> From: Adrian Ng Ho Yin <adrianhoyin.ng@altera.com>
>>
>> The descriptor FIFO requires that all words of a descriptor are written
>> in order, with the control word written last to flush it into the DMA
>> engine. Using memcpy() is unsafe since it does not guarantee ordering of
>> MMIO writes across all architectures.
>>
>> Replace memcpy() with an explicit iowrite32() loop for each 32-bit word
>> (except the control word). The control word is still written separately,
>> with write barriers, to ensure it is always the final word pushed into
>> the FIFO.
>>
>> This makes the programming of descriptors fully deterministic and
>> portable across different architectures.
>>
>> Signed-off-by: Adrian Ng Ho Yin <adrianhoyin.ng@altera.com>
>> Signed-off-by: Tze Yee Ng <tze.yee.ng@altera.com>
>> ---
>> drivers/dma/altera-msgdma.c | 21 ++++++++++++---------
>> 1 file changed, 12 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/dma/altera-msgdma.c b/drivers/dma/altera-msgdma.c
>> index b46999c81df0..5816973d2c70 100644
>> --- a/drivers/dma/altera-msgdma.c
>> +++ b/drivers/dma/altera-msgdma.c
>> @@ -495,6 +495,9 @@ static void msgdma_copy_one(struct msgdma_device *mdev,
>> struct msgdma_sw_desc *desc)
>> {
>> void __iomem *hw_desc = mdev->desc;
>> + const u32 *src = (const u32 *)&desc->hw_desc;
>> + unsigned int i, nwords = offsetof(struct msgdma_extended_desc, control) /
>> + sizeof(u32);
>>
>> /*
>> * Check if the DESC FIFO it not full. If its full, we need to wait
>> @@ -505,16 +508,16 @@ static void msgdma_copy_one(struct msgdma_device *mdev,
>> mdelay(1);
>>
>> /*
>> - * The descriptor needs to get copied into the descriptor FIFO
>> - * of the DMA controller. The descriptor will get flushed to the
>> - * FIFO, once the last word (control word) is written. Since we
>> - * are not 100% sure that memcpy() writes all word in the "correct"
>> - * order (address from low to high) on all architectures, we make
>> - * sure this control word is written last by single coding it and
>> - * adding some write-barriers here.
>> + * The descriptor must be written into the descriptor FIFO of the DMA
>> + * controller. The FIFO is flushed and the descriptor becomes valid once
>> + * the last word (the control word) is written. To guarantee the ordering
>> + * of MMIO writes across all architectures, we write each 32-bit word
>> + * individually using iowrite32(), and handle the control word separately
>> + * at the end. This ensures the control word is always written last and
>> + * prevents memcpy() or the compiler from reordering accesses.
>> */
>> - memcpy((void __force *)hw_desc, &desc->hw_desc,
>> - sizeof(desc->hw_desc) - sizeof(u32));
>> + for (i = 0; i < nwords; i++)
>> + iowrite32(src[i], hw_desc + i * sizeof(u32));
>
> why not use memcpy_toio()?
>
> Frank
>>
>> /* Write control word last to flush this descriptor into the FIFO */
>> mdev->idle = false;
>> --
>> 2.43.7
>>
Hi Frank,
Good point — the issue with the original code was memcpy() with __force
to __iomem, not bulk MMIO copies in general. memcpy_toio() is the right
API for the descriptor body as long as we still omit the control word
and write it last with wmb() + iowrite32(), which the hardware uses to
commit the descriptor into the FIFO.
I'll submit the v2 patch with memcpy_toio().
Thanks,
Tze Yee
prev parent reply other threads:[~2026-05-25 6:17 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-05-19 6:47 tze.yee.ng
2026-05-20 21:29 ` Frank Li
2026-05-25 6:17 ` NG, TZE YEE [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=8d9bd7a0-4927-40d8-a747-c3d7cf34cf67@altera.com \
--to=tze.yee.ng@altera.com \
--cc=Frank.Li@kernel.org \
--cc=Frank.li@nxp.com \
--cc=adrian.ho.yin.ng@altera.com \
--cc=dmaengine@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=muhammad.nazim.amirul.nazle.asmade@altera.com \
--cc=olivierdautricourt@gmail.com \
--cc=sr@denx.de \
--cc=vkoul@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®