| Message ID | 1307345414-26872-1-git-send-email-shawn.guo@linaro.org |
|---|---|
| State | New |
| Headers | show |
On Mon, 6 Jun 2011 15:30:12 +0800 Shawn Guo <shawn.guo@linaro.org> wrote: > Like dma_set(get)_max_seg_size for max_segment_size, the patch adds > max_segment_number into device_dma_parameters and creates the > corresponding dmaengine API dma_set(get)_max_seg_number for it. As I wrote in another mail, dma_get|set_max_seg_size and dma_get|set_seg_boundary are added to the dma mapping API. That is, they enable the drivers (or subsystems) to tell IOMMUs about the device dma limitations. When IOMMUs merge scatter list entries, they look at the limitations. > Here is the user story that tells the need of the new api. The > mxs-mmc is the mmc host controller for Freescale MXS architecture. > There are a pair of mmc host specific parameters max_seg_size and > max_segs that mxs-mmc host driver needs to tell mmc core, so that > mmc core can know how big each data segment could be and how many > segments could be handled one time in a scatter list by host driver. > > The mxs-mmc driver is one user of dmaengine mxs-dma, and it will call > mxs-dma to transfer data in scatter list. That is to say mxs-mmc has > no idea of what max_seg_size and max_segs should be, because they are > all mxs-dma capability parameters, and mxs-mmc needs to query them > from mxs-dma. max_segs isn't unrelated with the dma mapping API. I explained above, IOMMUs doesn't increase the number of segments (could decrease the number of segments by merging). The limitation about the number of segment already lives elsewhere (e.g. queue's limits.max_segments).
On Mon, Jun 6, 2011 at 12:30 AM, Shawn Guo <shawn.guo@linaro.org> wrote: > Like dma_set(get)_max_seg_size for max_segment_size, the patch adds > max_segment_number into device_dma_parameters and creates the > corresponding dmaengine API dma_set(get)_max_seg_number for it. > > Here is the user story that tells the need of the new api. The > mxs-mmc is the mmc host controller for Freescale MXS architecture. > There are a pair of mmc host specific parameters max_seg_size and > max_segs that mxs-mmc host driver needs to tell mmc core, so that > mmc core can know how big each data segment could be and how many > segments could be handled one time in a scatter list by host driver. > > The mxs-mmc driver is one user of dmaengine mxs-dma, and it will call > mxs-dma to transfer data in scatter list. That is to say mxs-mmc has > no idea of what max_seg_size and max_segs should be, because they are > all mxs-dma capability parameters, and mxs-mmc needs to query them > from mxs-dma. > > Right now, there is well defined dma api (dma_get_max_seg_size) for > mmc to query max_seg_size from dma driver, but the one for max_segs > is missing. That's why mxs-mmc driver has to hard-code it. > > The mxs-mmc is just one example to demonstrate the need of the new > api, and there are other mmc host drivers (mxcmmc on imx-dma is > another example) and possibly even other dmaengine users need this > new api to know the maximum segments that dma driver can handle per > dma call. > > Signed-off-by: Shawn Guo <shawn.guo@linaro.org> > --- > Changes since v1: > * Update commit message to explain why the new api is needed > > include/linux/device.h | 1 + > include/linux/dma-mapping.h | 15 +++++++++++++++ > 2 files changed, 16 insertions(+), 0 deletions(-) > > diff --git a/include/linux/device.h b/include/linux/device.h > index c66111a..44cb2528 100644 > --- a/include/linux/device.h > +++ b/include/linux/device.h > @@ -487,6 +487,7 @@ struct device_dma_parameters { > * sg limitations. > */ Given the discussion seems like this patch is missing an update to the documentation of the struct to clarify the definition of dma provider. I.e. that this info belongs to the device doing the dma, and that is is not necessarily the same as the device that is requesting dma service. Other than that seems like a natural extension of the existing usage of device_dma_parameters in drivers/dma/.
On Tue, Jun 07, 2011 at 03:35:51PM -0700, Dan Williams wrote: > On Mon, Jun 6, 2011 at 12:30 AM, Shawn Guo <shawn.guo@linaro.org> wrote: > > Like dma_set(get)_max_seg_size for max_segment_size, the patch adds > > max_segment_number into device_dma_parameters and creates the > > corresponding dmaengine API dma_set(get)_max_seg_number for it. > > > > Here is the user story that tells the need of the new api. The > > mxs-mmc is the mmc host controller for Freescale MXS architecture. > > There are a pair of mmc host specific parameters max_seg_size and > > max_segs that mxs-mmc host driver needs to tell mmc core, so that > > mmc core can know how big each data segment could be and how many > > segments could be handled one time in a scatter list by host driver. > > > > The mxs-mmc driver is one user of dmaengine mxs-dma, and it will call > > mxs-dma to transfer data in scatter list. That is to say mxs-mmc has > > no idea of what max_seg_size and max_segs should be, because they are > > all mxs-dma capability parameters, and mxs-mmc needs to query them > > from mxs-dma. > > > > Right now, there is well defined dma api (dma_get_max_seg_size) for > > mmc to query max_seg_size from dma driver, but the one for max_segs > > is missing. That's why mxs-mmc driver has to hard-code it. > > > > The mxs-mmc is just one example to demonstrate the need of the new > > api, and there are other mmc host drivers (mxcmmc on imx-dma is > > another example) and possibly even other dmaengine users need this > > new api to know the maximum segments that dma driver can handle per > > dma call. > > > > Signed-off-by: Shawn Guo <shawn.guo@linaro.org> > > --- > > Changes since v1: > > * Update commit message to explain why the new api is needed > > > > include/linux/device.h | 1 + > > include/linux/dma-mapping.h | 15 +++++++++++++++ > > 2 files changed, 16 insertions(+), 0 deletions(-) > > > > diff --git a/include/linux/device.h b/include/linux/device.h > > index c66111a..44cb2528 100644 > > --- a/include/linux/device.h > > +++ b/include/linux/device.h > > @@ -487,6 +487,7 @@ struct device_dma_parameters { > > * sg limitations. > > */ > > Given the discussion seems like this patch is missing an update to the > documentation of the struct to clarify the definition of dma provider. > I.e. that this info belongs to the device doing the dma, and that is > is not necessarily the same as the device that is requesting dma > service. > What about the following? /* * device_dma_parameters is a property of DMA provider, and it belongs to the * 'struct device' that actually provides DMA service, typically the drivers * under drivers/dma, although in some cases the DMA provider and block device * uses DMA service happen to be the same 'struct device'. * * It's not necessary for every single DMA providers to have this structure, * because some DMA providers simply do not have these parameters/limitations. * For those do have, the DMA providers should be responsible for setting the * parameters up. */ struct device_dma_parameters { unsigned int max_segment_size; unsigned int max_segment_number; unsigned long segment_boundary_mask; };
diff --git a/include/linux/device.h b/include/linux/device.h index c66111a..44cb2528 100644 --- a/include/linux/device.h +++ b/include/linux/device.h @@ -487,6 +487,7 @@ struct device_dma_parameters { * sg limitations. */ unsigned int max_segment_size; + unsigned int max_segment_number; unsigned long segment_boundary_mask; }; diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h index ba8319a..fd314f4 100644 --- a/include/linux/dma-mapping.h +++ b/include/linux/dma-mapping.h @@ -131,6 +131,21 @@ static inline unsigned int dma_set_max_seg_size(struct device *dev, return -EIO; } +static inline unsigned int dma_get_max_seg_number(struct device *dev) +{ + return dev->dma_parms ? dev->dma_parms->max_segment_number : 1; +} + +static inline unsigned int dma_set_max_seg_number(struct device *dev, + unsigned int number) +{ + if (dev->dma_parms) { + dev->dma_parms->max_segment_number = number; + return 0; + } else + return -EIO; +} + static inline unsigned long dma_get_seg_boundary(struct device *dev) { return dev->dma_parms ?
Like dma_set(get)_max_seg_size for max_segment_size, the patch adds max_segment_number into device_dma_parameters and creates the corresponding dmaengine API dma_set(get)_max_seg_number for it. Here is the user story that tells the need of the new api. The mxs-mmc is the mmc host controller for Freescale MXS architecture. There are a pair of mmc host specific parameters max_seg_size and max_segs that mxs-mmc host driver needs to tell mmc core, so that mmc core can know how big each data segment could be and how many segments could be handled one time in a scatter list by host driver. The mxs-mmc driver is one user of dmaengine mxs-dma, and it will call mxs-dma to transfer data in scatter list. That is to say mxs-mmc has no idea of what max_seg_size and max_segs should be, because they are all mxs-dma capability parameters, and mxs-mmc needs to query them from mxs-dma. Right now, there is well defined dma api (dma_get_max_seg_size) for mmc to query max_seg_size from dma driver, but the one for max_segs is missing. That's why mxs-mmc driver has to hard-code it. The mxs-mmc is just one example to demonstrate the need of the new api, and there are other mmc host drivers (mxcmmc on imx-dma is another example) and possibly even other dmaengine users need this new api to know the maximum segments that dma driver can handle per dma call. Signed-off-by: Shawn Guo <shawn.guo@linaro.org> --- Changes since v1: * Update commit message to explain why the new api is needed include/linux/device.h | 1 + include/linux/dma-mapping.h | 15 +++++++++++++++ 2 files changed, 16 insertions(+), 0 deletions(-)