On Thu, 2016-09-15 at 16:14 +0300, Eugeniy Paltsev wrote:
> This patch is to address a proposal by Andy in this thread:
> http://www.spinics.net/lists/dmaengine/msg10754.html
> Split platform data to actual hardware properties, and platform
> quirks.
> Now we able to use quirks and hardware properties separately from
> different sources (pdata, device tree or autoconfig registers)
>
My comments below.
Sorry for delayed answer.
> Signed-off-by: Eugeniy Paltsev <[email protected]>
> ---
> drivers/dma/dw/core.c | 31 +++++++++++++++-----------
> drivers/dma/dw/platform.c | 42 +++++++++++++++++++++----
> -----------
> include/linux/platform_data/dma-dw.h | 20 +++++++++++------
> 3 files changed, 57 insertions(+), 36 deletions(-)
>
> diff --git a/drivers/dma/dw/core.c b/drivers/dma/dw/core.c
> index c2c0a61..9352735 100644
> --- a/drivers/dma/dw/core.c
> +++ b/drivers/dma/dw/core.c
> @@ -1451,10 +1451,25 @@ int dw_dma_probe(struct dw_dma_chip *chip)
>
> dw->regs = chip->regs;
> chip->dw = dw;
+ empty line.
> + /* Reassign the platform data pointer */
> + pdata = dw->pdata;
>
> pm_runtime_get_sync(chip->dev);
>
> - if (!chip->pdata) {
> + if ((!chip->pdata) ||
> + (chip->pdata && test_bit(QUIRKS_ONLY_USED, &chip->pdata-
I don't think you need atomic test / set of those bits.
> >quirks))) {
> +
> + /*
> + * Fill quirks with the default values in case of
> pdata absence
/*
* Multi-line comments should include full sentences
* (punctuation matters).
*/
> + */
> + if (!chip->pdata) {
> + set_bit(QUIRKS_IS_PRIVATE, &pdata->quirks);
> + set_bit(QUIRKS_IS_MEMCPY, &pdata->quirks);
> + set_bit(QUIRKS_IS_NOLLP, &pdata->quirks);
> + } else {
> + pdata->quirks = chip->pdata->quirks;
> + }
> +
> dw_params = dma_readl(dw, DW_PARAMS);
> dev_dbg(chip->dev, "DW_PARAMS: 0x%08x\n", dw_params);
>
> @@ -1464,9 +1479,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> goto err_pdata;
> }
>
> - /* Reassign the platform data pointer */
> - pdata = dw->pdata;
> -
> /* Get hardware configuration parameters */
> pdata->nr_channels = (dw_params >> DW_PARAMS_NR_CHAN
> & 7) + 1;
> pdata->nr_masters = (dw_params >> DW_PARAMS_NR_MASTER
> & 3) + 1;
> @@ -1477,8 +1489,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> pdata->block_size = dma_readl(dw, MAX_BLK_SIZE);
>
> /* Fill platform data with the default values */
> - pdata->is_private = true;
> - pdata->is_memcpy = true;
> pdata->chan_allocation_order =
> CHAN_ALLOCATION_ASCENDING;
> pdata->chan_priority = CHAN_PRIORITY_ASCENDING;
> } else if (chip->pdata->nr_channels > DW_DMA_MAX_NR_CHANNELS)
> {
> @@ -1486,9 +1496,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> goto err_pdata;
> } else {
> memcpy(dw->pdata, chip->pdata, sizeof(*dw->pdata));
> -
> - /* Reassign the platform data pointer */
> - pdata = dw->pdata;
> }
>
> dw->chan = devm_kcalloc(chip->dev, pdata->nr_channels,
> sizeof(*dw->chan),
> @@ -1569,7 +1576,7 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> (dwc_params >> DWC_PARAMS_MBLK_EN &
> 0x1) == 0;
> } else {
> dwc->block_size = pdata->block_size;
> - dwc->nollp = pdata->is_nollp;
> + dwc->nollp = test_bit(QUIRKS_IS_NOLLP,
> &pdata->quirks);
Perhaps you need another patch which actually moves nollp to dwc->flags.
> }
> }
>
> @@ -1582,9 +1589,9 @@ int dw_dma_probe(struct dw_dma_chip *chip)
>
> /* Set capabilities */
> dma_cap_set(DMA_SLAVE, dw->dma.cap_mask);
> - if (pdata->is_private)
> + if (test_bit(QUIRKS_IS_PRIVATE, &pdata->quirks))
> dma_cap_set(DMA_PRIVATE, dw->dma.cap_mask);
> - if (pdata->is_memcpy)
> + if (test_bit(QUIRKS_IS_MEMCPY, &pdata->quirks))
> dma_cap_set(DMA_MEMCPY, dw->dma.cap_mask);
>
> dw->dma.dev = chip->dev;
> diff --git a/drivers/dma/dw/platform.c b/drivers/dma/dw/platform.c
> index 5bda0eb..308b977 100644
> --- a/drivers/dma/dw/platform.c
> +++ b/drivers/dma/dw/platform.c
> @@ -12,6 +12,7 @@
> * published by the Free Software Foundation.
> */
>
> +#include <linux/bitops.h>
> #include <linux/module.h>
> #include <linux/device.h>
> #include <linux/clk.h>
> @@ -111,41 +112,48 @@ dw_dma_parse_dt(struct platform_device *pdev)
> return NULL;
> }
>
> - if (of_property_read_u32(np, "dma-masters", &nr_masters))
> - return NULL;
> - if (nr_masters < 1 || nr_masters > DW_DMA_MAX_NR_MASTERS)
> - return NULL;
> -
> - if (of_property_read_u32(np, "dma-channels", &nr_channels))
> - return NULL;
> -
> pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
> if (!pdata)
> return NULL;
>
> + set_bit(QUIRKS_ONLY_USED, &pdata->quirks);
> +
> + if (of_property_read_bool(np, "is-private"))
> + set_bit(QUIRKS_IS_PRIVATE, &pdata->quirks);
> +
> + if (of_property_read_bool(np, "is-memcpy"))
> + set_bit(QUIRKS_IS_MEMCPY, &pdata->quirks);
> +
> + if (of_property_read_bool(np, "is-nollp"))
> + set_bit(QUIRKS_IS_NOLLP, &pdata->quirks);
I would suggest to convert to unified device property API first.
> +
> + if (of_property_read_u32(np, "dma-masters", &nr_masters))
> + return pdata;
> + if (nr_masters < 1 || nr_masters > DW_DMA_MAX_NR_MASTERS)
> + return pdata;
> +
> pdata->nr_masters = nr_masters;
> - pdata->nr_channels = nr_channels;
>
> - if (of_property_read_bool(np, "is_private"))
> - pdata->is_private = true;
> + if (of_property_read_u32(np, "dma-channels", &nr_channels))
> + return pdata;
>
> - if (!of_property_read_u32(np, "chan_allocation_order", &tmp))
> + pdata->nr_channels = nr_channels;
> +
> + if (!of_property_read_u32(np, "chan-allocation-order", &tmp))
> pdata->chan_allocation_order = (unsigned char)tmp;
>
> - if (!of_property_read_u32(np, "chan_priority", &tmp))
> + if (!of_property_read_u32(np, "chan-priority", &tmp))
> pdata->chan_priority = tmp;
>
> - if (!of_property_read_u32(np, "block_size", &tmp))
> + if (!of_property_read_u32(np, "block-size", &tmp))
> pdata->block_size = tmp;
>
> if (!of_property_read_u32_array(np, "data-width", arr,
> nr_masters)) {
> for (tmp = 0; tmp < nr_masters; tmp++)
> pdata->data_width[tmp] = arr[tmp];
> - } else if (!of_property_read_u32_array(np, "data_width", arr,
> nr_masters)) {
> - for (tmp = 0; tmp < nr_masters; tmp++)
> - pdata->data_width[tmp] = BIT(arr[tmp] &
> 0x07);
> }
>
> + clear_bit(QUIRKS_ONLY_USED, &pdata->quirks);
> return pdata;
> }
> #else
> diff --git a/include/linux/platform_data/dma-dw.h
> b/include/linux/platform_data/dma-dw.h
> index 5f0e11e..9cd8199 100644
> --- a/include/linux/platform_data/dma-dw.h
> +++ b/include/linux/platform_data/dma-dw.h
> @@ -37,10 +37,7 @@ struct dw_dma_slave {
> /**
> * struct dw_dma_platform_data - Controller configuration parameters
> * @nr_channels: Number of channels supported by hardware (max 8)
> - * @is_private: The device channels should be marked as private and
> not for
> - * by the general purpose DMA channel allocator.
> - * @is_memcpy: The device channels do support memory-to-memory
> transfers.
> - * @is_nollp: The device channels does not support multi block
> transfers.
> + * @quirks: Bit field with platform quirks
> * @chan_allocation_order: Allocate channels starting from 0 or 7
> * @chan_priority: Set channel priority increasing from 0 to 7 or 7
> to 0.
> * @block_size: Maximum block size supported by the controller
> @@ -50,9 +47,18 @@ struct dw_dma_slave {
> */
> struct dw_dma_platform_data {
> unsigned int nr_channels;
> - bool is_private;
> - bool is_memcpy;
> - bool is_nollp;
> +/* Only use quirks from platform data structure */
> +#define QUIRKS_ONLY_USED 0
> +/*
> + * The device channels should be marked as private and not for
> + * by the general purpose DMA channel allocator.
> + */
> +#define QUIRKS_IS_PRIVATE 1
> +/* The device channels do support memory-to-memory transfers. */
> +#define QUIRKS_IS_MEMCPY 2
> +/* The device channels do not support multi block transfers. */
> +#define QUIRKS_IS_NOLLP 3
You may move descriptions to above struct description field. It will be
consolidated and moreover visible in kernel-doc processed documents.
> + unsigned long quirks;
> #define CHAN_ALLOCATION_ASCENDING 0 /* zero to seven */
> #define CHAN_ALLOCATION_DESCENDING 1 /* seven to zero
> */
> unsigned char chan_allocation_order;
--
Andy Shevchenko <[email protected]>
Intel Finland Oy
On Thu, 2016-10-20 at 13:48 +0300, Andy Shevchenko wrote:
> On Thu, 2016-09-15 at 16:14 +0300, Eugeniy Paltsev wrote:
> >
> > This patch is to address a proposal by Andy in this thread:
> > http://www.spinics.net/lists/dmaengine/msg10754.html
> > Split platform data to actual hardware properties, and platform
> > quirks.
> > Now we able to use quirks and hardware properties separately from
> > different sources (pdata, device tree or autoconfig registers)
> >
> My comments below.
>
> Sorry for delayed answer.
No problems.
>
> >
> > Signed-off-by: Eugeniy Paltsev <[email protected]>
> > ---
> > drivers/dma/dw/core.c | 31 +++++++++++++++---------
> > --
> > drivers/dma/dw/platform.c | 42 +++++++++++++++++++++---
> > -
> > -----------
> > include/linux/platform_data/dma-dw.h | 20 +++++++++++------
> > 3 files changed, 57 insertions(+), 36 deletions(-)
> >
> > diff --git a/drivers/dma/dw/core.c b/drivers/dma/dw/core.c
> > index c2c0a61..9352735 100644
> > --- a/drivers/dma/dw/core.c
> > +++ b/drivers/dma/dw/core.c
> > @@ -1451,10 +1451,25 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> >
> > dw->regs = chip->regs;
> > chip->dw = dw;
> + empty line.
>
> >
> > + /* Reassign the platform data pointer */
> > + pdata = dw->pdata;
> >
> > pm_runtime_get_sync(chip->dev);
> >
> > - if (!chip->pdata) {
> > + if ((!chip->pdata) ||
> > + (chip->pdata && test_bit(QUIRKS_ONLY_USED, &chip-
> > >pdata-
> I don't think you need atomic test / set of those bits.
I don't need atomic bit operations here, I just used standard bit API
to make code more clear.
> >
> > >
> > > quirks))) {
> > +
> > + /*
> > + * Fill quirks with the default values in case of
> > pdata absence
> /*
> * Multi-line comments should include full sentences
> * (punctuation matters).
> */
>
> >
> > + */
> > + if (!chip->pdata) {
> > + set_bit(QUIRKS_IS_PRIVATE, &pdata-
> > >quirks);
> > + set_bit(QUIRKS_IS_MEMCPY, &pdata->quirks);
> > + set_bit(QUIRKS_IS_NOLLP, &pdata->quirks);
> > + } else {
> > + pdata->quirks = chip->pdata->quirks;
> > + }
> > +
> > dw_params = dma_readl(dw, DW_PARAMS);
> > dev_dbg(chip->dev, "DW_PARAMS: 0x%08x\n",
> > dw_params);
> >
> > @@ -1464,9 +1479,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> > goto err_pdata;
> > }
> >
> > - /* Reassign the platform data pointer */
> > - pdata = dw->pdata;
> > -
> > /* Get hardware configuration parameters */
> > pdata->nr_channels = (dw_params >>
> > DW_PARAMS_NR_CHAN
> > & 7) + 1;
> > pdata->nr_masters = (dw_params >>
> > DW_PARAMS_NR_MASTER
> > & 3) + 1;
> > @@ -1477,8 +1489,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> > pdata->block_size = dma_readl(dw, MAX_BLK_SIZE);
> >
> > /* Fill platform data with the default values */
> > - pdata->is_private = true;
> > - pdata->is_memcpy = true;
> > pdata->chan_allocation_order =
> > CHAN_ALLOCATION_ASCENDING;
> > pdata->chan_priority = CHAN_PRIORITY_ASCENDING;
> > } else if (chip->pdata->nr_channels >
> > DW_DMA_MAX_NR_CHANNELS)
> > {
> > @@ -1486,9 +1496,6 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> > goto err_pdata;
> > } else {
> > memcpy(dw->pdata, chip->pdata, sizeof(*dw-
> > >pdata));
> > -
> > - /* Reassign the platform data pointer */
> > - pdata = dw->pdata;
> > }
> >
> > dw->chan = devm_kcalloc(chip->dev, pdata->nr_channels,
> > sizeof(*dw->chan),
> > @@ -1569,7 +1576,7 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> > (dwc_params >> DWC_PARAMS_MBLK_EN
> > &
> > 0x1) == 0;
> > } else {
> > dwc->block_size = pdata->block_size;
> > - dwc->nollp = pdata->is_nollp;
> > + dwc->nollp = test_bit(QUIRKS_IS_NOLLP,
> > &pdata->quirks);
> Perhaps you need another patch which actually moves nollp to dwc-
> >flags.
As I can see, we already have DW_DMA_IS_SOFT_LLP flag in "dwc->flags"
with same functionality, which is set if "dwc->nollp" is true. Probably
we can use this flag and get rid of "dwc->nollp".
But I'm a bit confused why we clear DW_DMA_IS_SOFT_LLP bit in
"dwc_scan_descriptors" and "dwc_terminate_all" functions.
Any ideas about that?
> >
> > }
> > }
> >
> > @@ -1582,9 +1589,9 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> >
> > /* Set capabilities */
> > dma_cap_set(DMA_SLAVE, dw->dma.cap_mask);
> > - if (pdata->is_private)
> > + if (test_bit(QUIRKS_IS_PRIVATE, &pdata->quirks))
> > dma_cap_set(DMA_PRIVATE, dw->dma.cap_mask);
> > - if (pdata->is_memcpy)
> > + if (test_bit(QUIRKS_IS_MEMCPY, &pdata->quirks))
> > dma_cap_set(DMA_MEMCPY, dw->dma.cap_mask);
> >
> > dw->dma.dev = chip->dev;
> > diff --git a/drivers/dma/dw/platform.c b/drivers/dma/dw/platform.c
> > index 5bda0eb..308b977 100644
> > --- a/drivers/dma/dw/platform.c
> > +++ b/drivers/dma/dw/platform.c
> > @@ -12,6 +12,7 @@
> > * published by the Free Software Foundation.
> > */
> >
> > +#include <linux/bitops.h>
> > #include <linux/module.h>
> > #include <linux/device.h>
> > #include <linux/clk.h>
> > @@ -111,41 +112,48 @@ dw_dma_parse_dt(struct platform_device *pdev)
> > return NULL;
> > }
> >
> > - if (of_property_read_u32(np, "dma-masters", &nr_masters))
> > - return NULL;
> > - if (nr_masters < 1 || nr_masters > DW_DMA_MAX_NR_MASTERS)
> > - return NULL;
> > -
> > - if (of_property_read_u32(np, "dma-channels",
> > &nr_channels))
> > - return NULL;
> > -
> > pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata),
> > GFP_KERNEL);
> > if (!pdata)
> > return NULL;
> >
> > + set_bit(QUIRKS_ONLY_USED, &pdata->quirks);
> > +
> > + if (of_property_read_bool(np, "is-private"))
> > + set_bit(QUIRKS_IS_PRIVATE, &pdata->quirks);
> > +
> > + if (of_property_read_bool(np, "is-memcpy"))
> > + set_bit(QUIRKS_IS_MEMCPY, &pdata->quirks);
> > +
> > + if (of_property_read_bool(np, "is-nollp"))
> > + set_bit(QUIRKS_IS_NOLLP, &pdata->quirks);
> I would suggest to convert to unified device property API first.
I'll do that.
>
> >
> > +
> > + if (of_property_read_u32(np, "dma-masters", &nr_masters))
> > + return pdata;
> > + if (nr_masters < 1 || nr_masters > DW_DMA_MAX_NR_MASTERS)
> > + return pdata;
> > +
> > pdata->nr_masters = nr_masters;
> > - pdata->nr_channels = nr_channels;
> >
> > - if (of_property_read_bool(np, "is_private"))
> > - pdata->is_private = true;
> > + if (of_property_read_u32(np, "dma-channels",
> > &nr_channels))
> > + return pdata;
> >
> > - if (!of_property_read_u32(np, "chan_allocation_order",
> > &tmp))
> > + pdata->nr_channels = nr_channels;
> > +
> > + if (!of_property_read_u32(np, "chan-allocation-order",
> > &tmp))
> > pdata->chan_allocation_order = (unsigned char)tmp;
> >
> > - if (!of_property_read_u32(np, "chan_priority", &tmp))
> > + if (!of_property_read_u32(np, "chan-priority", &tmp))
> > pdata->chan_priority = tmp;
> >
> > - if (!of_property_read_u32(np, "block_size", &tmp))
> > + if (!of_property_read_u32(np, "block-size", &tmp))
> > pdata->block_size = tmp;
> >
> > if (!of_property_read_u32_array(np, "data-width", arr,
> > nr_masters)) {
> > for (tmp = 0; tmp < nr_masters; tmp++)
> > pdata->data_width[tmp] = arr[tmp];
> > - } else if (!of_property_read_u32_array(np, "data_width",
> > arr,
> > nr_masters)) {
> > - for (tmp = 0; tmp < nr_masters; tmp++)
> > - pdata->data_width[tmp] = BIT(arr[tmp] &
> > 0x07);
> > }
> >
> > + clear_bit(QUIRKS_ONLY_USED, &pdata->quirks);
> > return pdata;
> > }
> > #else
> > diff --git a/include/linux/platform_data/dma-dw.h
> > b/include/linux/platform_data/dma-dw.h
> > index 5f0e11e..9cd8199 100644
> > --- a/include/linux/platform_data/dma-dw.h
> > +++ b/include/linux/platform_data/dma-dw.h
> > @@ -37,10 +37,7 @@ struct dw_dma_slave {
> > /**
> > * struct dw_dma_platform_data - Controller configuration
> > parameters
> > * @nr_channels: Number of channels supported by hardware (max 8)
> > - * @is_private: The device channels should be marked as private
> > and
> > not for
> > - * by the general purpose DMA channel allocator.
> > - * @is_memcpy: The device channels do support memory-to-memory
> > transfers.
> > - * @is_nollp: The device channels does not support multi block
> > transfers.
> > + * @quirks: Bit field with platform quirks
> > * @chan_allocation_order: Allocate channels starting from 0 or 7
> > * @chan_priority: Set channel priority increasing from 0 to 7 or
> > 7
> > to 0.
> > * @block_size: Maximum block size supported by the controller
> > @@ -50,9 +47,18 @@ struct dw_dma_slave {
> > */
> > struct dw_dma_platform_data {
> > unsigned int nr_channels;
> > - bool is_private;
> > - bool is_memcpy;
> > - bool is_nollp;
> > +/* Only use quirks from platform data structure */
> > +#define QUIRKS_ONLY_USED 0
> > +/*
> > + * The device channels should be marked as private and not for
> > + * by the general purpose DMA channel allocator.
> > + */
> > +#define QUIRKS_IS_PRIVATE 1
> > +/* The device channels do support memory-to-memory transfers. */
> > +#define QUIRKS_IS_MEMCPY 2
> > +/* The device channels do not support multi block transfers. */
> > +#define QUIRKS_IS_NOLLP 3
> You may move descriptions to above struct description field. It will
> be
> consolidated and moreover visible in kernel-doc processed documents.
Good idea.
> >
> > + unsigned long quirks;
> > #define CHAN_ALLOCATION_ASCENDING 0 /* zero to seven
> > */
> > #define CHAN_ALLOCATION_DESCENDING 1 /* seven to
> > zero
> > */
> > unsigned char chan_allocation_order;
--
Paltsev Eugeniy
On Thu, 2016-10-20 at 13:16 +0000, Eugeniy Paltsev wrote:
> On Thu, 2016-10-20 at 13:48 +0300, Andy Shevchenko wrote:
> > On Thu, 2016-09-15 at 16:14 +0300, Eugeniy Paltsev wrote:
> > >
> > > This patch is to address a proposal by Andy in this thread:
> > > http://www.spinics.net/lists/dmaengine/msg10754.html
> > > Split platform data to actual hardware properties, and platform
> > > quirks.
> > > Now we able to use quirks and hardware properties separately from
> > > different sources (pdata, device tree or autoconfig registers)
> > - if (!chip->pdata) {
> > > + if ((!chip->pdata) ||
> > > + (chip->pdata && test_bit(QUIRKS_ONLY_USED, &chip-
> > > > pdata-
> >
> > I don't think you need atomic test / set of those bits.
>
> I don't need atomic bit operations here, I just used standard bit API
> to make code more clear.
I gave more thoughts to that and I think we would leave boolean
variables in platform data. It would be cleaner for users how they
defined quirks. Besides that any additional quirk or modification of the
existing one will produce simple and readable change.
>
> > > + */
> > > + if (!chip->pdata) {
> > > + set_bit(QUIRKS_IS_PRIVATE, &pdata-
> > > > quirks);
> > >
> > > + set_bit(QUIRKS_IS_MEMCPY, &pdata-
> > > >quirks);
> > > + set_bit(QUIRKS_IS_NOLLP, &pdata->quirks);
>>> + } else {
> > > + pdata->quirks = chip->pdata->quirks;
> > > + }
Ditto.
> > > @@ -1569,7 +1576,7 @@ int dw_dma_probe(struct dw_dma_chip *chip)
> > > (dwc_params >> DWC_PARAMS_MBLK_EN
> > > &
> > > 0x1) == 0;
> > > } else {
> > > dwc->block_size = pdata->block_size;
> > > - dwc->nollp = pdata->is_nollp;
> > > + dwc->nollp = test_bit(QUIRKS_IS_NOLLP,
> > > &pdata->quirks);
> >
> > Perhaps you need another patch which actually moves nollp to dwc-
> > > flags.
>
> As I can see, we already have DW_DMA_IS_SOFT_LLP flag in "dwc->flags"
> with same functionality, which is set if "dwc->nollp" is true.
> Probably
> we can use this flag and get rid of "dwc->nollp".
> But I'm a bit confused why we clear DW_DMA_IS_SOFT_LLP bit in
> "dwc_scan_descriptors" and "dwc_terminate_all" functions.
> Any ideas about that?
Sounds like a plan. I think you are right, the property quite unlikely
can be changed run-time (though I have one patch regarding Braswell and
CherryView platforms that have a chicken bit to enable / disable the
feature). So, please, get rid of an additional boolean field.
--
Andy Shevchenko <[email protected]>
Intel Finland Oy