From: linux@arm.linux.org.uk (Russell King - ARM Linux)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH V3 08/12] dmaengine/dw_dmac: Unmap all memory buffers after completion of non-slave transfers
Date: Thu, 2 Feb 2012 12:15:23 +0000 [thread overview]
Message-ID: <20120202121523.GG889@n2100.arm.linux.org.uk> (raw)
In-Reply-To: <c098ca5959c947b3634555bc7bf4a8b3af82a2ae.1328091915.git.viresh.kumar@st.com>
On Wed, Feb 01, 2012 at 04:12:24PM +0530, Viresh Kumar wrote:
> Currently, after completion of transfer, source address or destination address
> of only the first LLI descriptor is unmapped. And length passed for unmap is
> total length of all descriptors in the list. Which means unmapping code assumed
> that the memory buffers pointed to by the descriptors will be physically
> contiguous, which might not be the case.
>
> This patch intends to fix this wrong expectation of dw_dmac. Now, first desc
> will not contain total length of transfer. But individual descriptors will
> contain their individual lengths. Finally, we will call unmap for all
> descriptors.
I'm still unconvinced that this is anywhere close to correct.
Look. DMA engines only automatically unmap when they're being used by the
async_tx stuff for memcpy's, and RAID computation offload, and the like.
These work with individual single buffers, not scatterlists. These will
not be fragmented.
If the caller mapped it with a single or page mapping, then the DMA engine
ABSOLUTELY MUST unmap it with the equivalent unmapping function _with the
same arguments_ which were used to map it - and certainly not doing it
piece meal on multiple small bits of the buffer. The whole buffer entirely
in one go. Even if the DMA engine driver decides internally to split the
buffer into smaller chunks.
The DMA API demands that the following will always be satisfied:
dma = dma_map_single(dev, addr, len, dir);
dma_unmap_single(dev, dma, len, dir);
where those arguments passed into dma_unmap_single() are those which were
passed into/outof dma_map_single() _totally_ _unmodified_.
The same is true for dma_map_page()..dma_unmap_page().
>
> Signed-off-by: Viresh Kumar <viresh.kumar@st.com>
> ---
> drivers/dma/dw_dmac.c | 49 +++++++++++++++++++++++++++++++------------------
> 1 files changed, 31 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/dma/dw_dmac.c b/drivers/dma/dw_dmac.c
> index 5d7b199..49d477c 100644
> --- a/drivers/dma/dw_dmac.c
> +++ b/drivers/dma/dw_dmac.c
> @@ -231,6 +231,14 @@ static void dwc_dostart(struct dw_dma_chan *dwc, struct dw_desc *first)
>
> /*----------------------------------------------------------------------*/
>
> +#define dw_dmac_unmap(utype, _desc, _child, _buf, _dir) \
> +do { \
> + list_for_each_entry(_child, &_desc->tx_list, desc_node) \
> + dma_unmap_##utype(parent, _child->lli._buf, \
> + _child->len, _dir); \
> + dma_unmap_##utype(parent, _desc->lli._buf, _desc->len, _dir); \
> +} while (0)
> +
> static void
> dwc_descriptor_complete(struct dw_dma_chan *dwc, struct dw_desc *desc,
> bool callback_required)
> @@ -264,19 +272,19 @@ dwc_descriptor_complete(struct dw_dma_chan *dwc, struct dw_desc *desc,
> struct device *parent = chan2parent(&dwc->chan);
> if (!(txd->flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
> if (txd->flags & DMA_COMPL_DEST_UNMAP_SINGLE)
> - dma_unmap_single(parent, desc->lli.dar,
> - desc->len, DMA_FROM_DEVICE);
> + dw_dmac_unmap(single, desc, child, dar,
> + DMA_FROM_DEVICE);
> else
> - dma_unmap_page(parent, desc->lli.dar,
> - desc->len, DMA_FROM_DEVICE);
> + dw_dmac_unmap(page, desc, child, dar,
> + DMA_FROM_DEVICE);
> }
> if (!(txd->flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
> if (txd->flags & DMA_COMPL_SRC_UNMAP_SINGLE)
> - dma_unmap_single(parent, desc->lli.sar,
> - desc->len, DMA_TO_DEVICE);
> + dw_dmac_unmap(single, desc, child, sar,
> + DMA_TO_DEVICE);
> else
> - dma_unmap_page(parent, desc->lli.sar,
> - desc->len, DMA_TO_DEVICE);
> + dw_dmac_unmap(page, desc, child, sar,
> + DMA_TO_DEVICE);
> }
> }
>
> @@ -676,6 +684,7 @@ dwc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> desc->lli.dar = dest + offset;
> desc->lli.ctllo = ctllo;
> desc->lli.ctlhi = xfer_count;
> + desc->len = xfer_count << src_width;
>
> if (!first) {
> first = desc;
> @@ -701,7 +710,6 @@ dwc_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> DMA_TO_DEVICE);
>
> first->txd.flags = flags;
> - first->len = len;
>
> return &first->txd;
>
> @@ -725,7 +733,6 @@ dwc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
> unsigned int mem_width;
> unsigned int i;
> struct scatterlist *sg;
> - size_t total_len = 0;
>
> dev_vdbg(chan2dev(chan), "prep_dma_slave\n");
>
> @@ -774,6 +781,7 @@ slave_sg_todev_fill_desc:
> }
>
> desc->lli.ctlhi = dlen >> mem_width;
> + desc->len = dlen;
>
> if (!first) {
> first = desc;
> @@ -787,7 +795,6 @@ slave_sg_todev_fill_desc:
> &first->tx_list);
> }
> prev = desc;
> - total_len += dlen;
>
> if (len)
> goto slave_sg_todev_fill_desc;
> @@ -831,6 +838,7 @@ slave_sg_fromdev_fill_desc:
> len = 0;
> }
> desc->lli.ctlhi = dlen >> reg_width;
> + desc->len = dlen;
>
> if (!first) {
> first = desc;
> @@ -844,7 +852,6 @@ slave_sg_fromdev_fill_desc:
> &first->tx_list);
> }
> prev = desc;
> - total_len += dlen;
>
> if (len)
> goto slave_sg_fromdev_fill_desc;
> @@ -863,8 +870,6 @@ slave_sg_fromdev_fill_desc:
> prev->txd.phys, sizeof(prev->lli),
> DMA_TO_DEVICE);
>
> - first->len = total_len;
> -
> return &first->txd;
>
> err_desc_get:
> @@ -950,11 +955,19 @@ dwc_tx_status(struct dma_chan *chan,
> ret = dma_async_is_complete(cookie, last_complete, last_used);
> }
>
> - if (ret != DMA_SUCCESS)
> - dma_set_tx_state(txstate, last_complete, last_used,
> - dwc_first_active(dwc)->len);
> - else
> + if (ret != DMA_SUCCESS) {
> + struct dw_desc *desc, *child;
> + unsigned int len;
> +
> + desc = dwc_first_active(dwc);
> + len = desc->len;
> + list_for_each_entry(child, &desc->tx_list, desc_node)
> + len += child->len;
> +
> + dma_set_tx_state(txstate, last_complete, last_used, len);
> + } else {
> dma_set_tx_state(txstate, last_complete, last_used, 0);
> + }
>
> if (dwc->paused)
> return DMA_PAUSED;
> --
> 1.7.8.110.g4cb5d
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
next prev parent reply other threads:[~2012-02-02 12:15 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-02-01 10:42 [PATCH V3 00/12] dmaengine: Pl08x and dw_dmac updates Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 01/12] dmaengine/dw_dmac: Hibernation support in dw_dmac Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 02/12] dmaengine: Add flow controller information to dma_slave_config Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 03/12] dmaengine: Pass dma_slave_config .device_fc = NULL for all existing users Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 04/12] dmaengine/amba-pl08x: Take flow controller info from DMA_SLAVE_CONFIG Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 05/12] dmaengine/dw_dmac: Don't use magic number for total number of channels Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 06/12] dmaengine/dw_dmac: Use dev_get_platdata() instead of accessing dev directly Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 07/12] dmaengine/dw_dmac: Don't handle block interrupts Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 08/12] dmaengine/dw_dmac: Unmap all memory buffers after completion of non-slave transfers Viresh Kumar
2012-02-02 12:15 ` Russell King - ARM Linux [this message]
2012-02-06 8:11 ` Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 09/12] dmaengine/dw_dmac: Add 64 bit access width support for slave xfers on mem side Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 10/12] dmaengine/dw_dmac: Add support for DMA_SLAVE_CONFIG Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 11/12] dmaengine/dw_dmac: Fix dw_dmac user drivers to adapt to slave_config changes Viresh Kumar
2012-02-01 10:42 ` [PATCH V3 12/12] dmaengine/dw_dmac: Remove unused fields in struct dw_dma_slave Viresh Kumar
2012-02-22 12:51 ` [PATCH V3 00/12] dmaengine: Pl08x and dw_dmac updates Vinod Koul
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=20120202121523.GG889@n2100.arm.linux.org.uk \
--to=linux@arm.linux.org.uk \
--cc=linux-arm-kernel@lists.infradead.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;
as well as URLs for NNTP newsgroup(s).