From: Lucas Stach <l.stach@pengutronix.de>
To: Michael Tretter <m.tretter@pengutronix.de>, barebox@lists.infradead.org
Subject: Re: [PATCH 3/3] firmware: zynqmp-fpga: do not use DMA coherent memory for bitstream
Date: Wed, 18 Aug 2021 15:47:02 +0200 [thread overview]
Message-ID: <8719b93227095011ebf0a83b5aed202acc8f8d23.camel@pengutronix.de> (raw)
In-Reply-To: <20210818133505.582698-4-m.tretter@pengutronix.de>
Am Mittwoch, dem 18.08.2021 um 15:35 +0200 schrieb Michael Tretter:
> Trying to do unaligned access of coherent memory on AArch64 will lead to
> an abort. This can happen when the FPGA loader copies the bitstream to
> the temporary buffer for the transfer to the FPGA.
>
> Convert the driver to use regular memory for the temporary buffer to
> prevent the issue.
>
> Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
> ---
> drivers/firmware/zynqmp-fpga.c | 20 +++++++++++++-------
> 1 file changed, 13 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/firmware/zynqmp-fpga.c b/drivers/firmware/zynqmp-fpga.c
> index 667910479aa7..0a0e7e880849 100644
> --- a/drivers/firmware/zynqmp-fpga.c
> +++ b/drivers/firmware/zynqmp-fpga.c
> @@ -203,7 +203,7 @@ static int fpgamgr_program_finish(struct firmware_handler *fh)
> size_t body_length;
> int header_length = 0;
> enum xilinx_byte_order byte_order;
> - u64 addr;
> + dma_addr_t addr;
> int status = 0;
> u8 flags = ZYNQMP_FPGA_BIT_ONLY_BIN;
>
> @@ -240,13 +240,19 @@ static int fpgamgr_program_finish(struct firmware_handler *fh)
> * memory. Allocate some extra space at the end of the buffer for the
> * bitstream size.
> */
> - buf_aligned = dma_alloc_coherent(body_length + sizeof(buf_size),
> - DMA_ADDRESS_BROKEN);
> + buf_aligned = dma_alloc(body_length + sizeof(u32));
> if (!buf_aligned) {
> status = -ENOBUFS;
> goto err_free;
> }
>
> + addr = dma_map_single(&mgr->dev, buf_aligned,
> + body_length + sizeof(u32), DMA_TO_DEVICE);
> + if (dma_mapping_error(&mgr->dev, addr)) {
> + status = -EFAULT;
> + goto err_free;
> + }
> +
Usage of both dma_map_single and explicit dma_sync_single_for_* for a
single transfer looks odd. dma_map_single already does the cache sync,
which you then do a second time in the sync calls.
Instead you should move this dma_map_single call to the place where you
added the dma_sync_single_for_device and replace the
dma_sync_single_for_cpu with a dma_unmap_single.
Regards,
Lucas
> if (!(mgr->features & ZYNQMP_PM_FEATURE_BYTE_ORDER_IRREL) &&
> byte_order == XILINX_BYTE_ORDER_BIN)
> copy_words_swapped((u32 *)buf_aligned, body,
> @@ -254,8 +260,6 @@ static int fpgamgr_program_finish(struct firmware_handler *fh)
> else
> memcpy((u32 *)buf_aligned, body, body_length);
>
> - addr = (u64)buf_aligned;
> -
> if (mgr->features & ZYNQMP_PM_FEATURE_SIZE_NOT_NEEDED) {
> buf_size = body_length;
> } else {
> @@ -263,11 +267,13 @@ static int fpgamgr_program_finish(struct firmware_handler *fh)
> buf_size = addr + body_length;
> }
>
> - status = mgr->eemi_ops->fpga_load(addr, buf_size, flags);
> + dma_sync_single_for_device(addr, body_length + sizeof(u32), DMA_TO_DEVICE);
> + status = mgr->eemi_ops->fpga_load((u64)addr, buf_size, flags);
> + dma_sync_single_for_cpu(addr, body_length + sizeof(u32), DMA_TO_DEVICE);
> if (status < 0)
> dev_err(&mgr->dev, "unable to load fpga\n");
>
> - dma_free_coherent(buf_aligned, 0, body_length + sizeof(buf_size));
> + dma_free(buf_aligned);
>
> err_free:
> free(mgr->buf);
_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox
next prev parent reply other threads:[~2021-08-18 13:48 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-08-18 13:35 [PATCH 0/3] firmware: zynqmp-fpga: fix fpga loading with optimized string functions Michael Tretter
2021-08-18 13:35 ` [PATCH 1/3] firmware: zynqmp-fpga: initialize flags at function start Michael Tretter
2021-08-18 13:35 ` [PATCH 2/3] firmware: zynqmp-fpga: avoid additional buffer for size argument Michael Tretter
2021-08-18 13:35 ` [PATCH 3/3] firmware: zynqmp-fpga: do not use DMA coherent memory for bitstream Michael Tretter
2021-08-18 13:47 ` Lucas Stach [this message]
2021-08-19 8:14 ` Michael Tretter
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=8719b93227095011ebf0a83b5aed202acc8f8d23.camel@pengutronix.de \
--to=l.stach@pengutronix.de \
--cc=barebox@lists.infradead.org \
--cc=m.tretter@pengutronix.de \
/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