[alsa-devel] [RFC PATCH 00/11] ARM: s3c64xx: Let amba-pl08x driver handle DMA
One of the biggest roadblocks on the way of S3C64xx to DeviceTree support is its DMA driver, which is completely platform-specific and provides private API (s3c-dma), not even saying that its design is completely against multiplatform-awareness.
The DMA controller present on this SoC series is a custom variant of ARM PrimeCell PL080 modified by Samsung to add some extra features. It is mostly compatible with original PL080, except: - CH_CONTROL2 register is added between CH_CONTROL and CH_CONFIG, - offset of CH_CONFIG register is different, - transfer size field is moved from CH_CONTROL to CH_CONTROL2, - transfer size field is extended to 24 bits, allowing much bigger single transfer, - LLI consists of one more word, to account for CH_CONTROL2 register.
Since all the rest is fully compatible with standard PL080 there is no point in having separate driver just for this single variant, so I decided to look into adding support for it to the amba-pl08x driver.
There was already some attempt to achieve this before, but this was before Russel's big rework of the driver to use virtual channels, making the old patches being not much of use.
This RFC series is a proof of concept that I managed to make during last days of hacking. Except one patch adding clkdev lookup to clock driver (which is being replaced with a CCF-compliant driver ATM), this is enough to get memcpy and slave transfers to work on S3C64xx.
I have tested this on Mini6410 and SMDK6410 boards using dmatest for memcpy and Samsung I2S with madplay/aplay for slave transfers. Unfortunately I do not have access to other platforms with PL08x so I could not test for any regressions introduced on them.
Credits for two patches go to Alban Bedel, who made a series fixing this driver to make it usable with audio drivers. I rebased his patches on top of mine and corrected coding style a bit.
OK, that's all. Any comments are welcome. Feel free to start throwing eggs and tomatoes if you find this awful, but I won't be upset if I get some Tested-by or Acked-by as well. ;)
Alban Bedel (2): dmaengine: PL08x: Fix reading the byte count in cctl dmaengine: PL08x: Add cyclic transfer support
Tomasz Figa (9): dma: amba-pl08x: Use bitmap to pass variant specific quirks dma: amba-pl08x: Refactor pl08x_getbytes_chan() to lower indentation dma: amba-pl08x: Add support for different offset of CONFIG register dma: amba-pl08x: Add support for PL080S variant dma: amba-pl08x: Add support for different maximum transfer size dma: amba-pl08x: Keep LLIs aligned to 4-word boundary spi: s3c64xx: Do not require legacy DMA API in case of S3C64XX ASoC: samsung: Do not require legacy DMA API in case of S3C64XX ARM: s3c64xx: Add support for DMA using generic amba-pl08x driver
arch/arm/Kconfig | 1 + arch/arm/mach-s3c64xx/Kconfig | 8 +- arch/arm/mach-s3c64xx/Makefile | 1 + arch/arm/mach-s3c64xx/common.h | 5 + arch/arm/mach-s3c64xx/include/mach/dma.h | 65 +++++ arch/arm/mach-s3c64xx/pl080.c | 244 +++++++++++++++++++ drivers/dma/amba-pl08x.c | 399 ++++++++++++++++++++++--------- drivers/spi/Kconfig | 2 +- include/linux/amba/pl080.h | 1 + sound/soc/samsung/Kconfig | 2 +- 10 files changed, 613 insertions(+), 115 deletions(-) create mode 100644 arch/arm/mach-s3c64xx/pl080.c
Instead of defining new bool field in vendor_data struct for each quirk, it is more reasonable to use a single flags field and make each quirk use single bits.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 31 +++++++++++++++++-------------- 1 file changed, 17 insertions(+), 14 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index 8bad254..d443a68 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -93,18 +93,22 @@ static struct amba_driver pl08x_amba_driver; struct pl08x_driver_data;
+/** Controller supports dual AHB masters. */ +#define PL08X_IS_DUALMASTER (1 << 0) +/** + * Controller has Nomadik security extension bits that need to be checked + * for permission before use and some registers are missing. + */ +#define PL08X_IS_NOMADIK (1 << 1) + /** * struct vendor_data - vendor-specific config parameters for PL08x derivatives * @channels: the number of channels available in this variant - * @dualmaster: whether this version supports dual AHB masters or not. - * @nomadik: whether the channels have Nomadik security extension bits - * that need to be checked for permission before use and some registers are - * missing + * @flags: Vendor-specific flags, see PL08X_IS_* */ struct vendor_data { u8 channels; - bool dualmaster; - bool nomadik; + u32 flags; };
/* @@ -1391,7 +1395,7 @@ static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy( /* Both to be incremented or the code will break */ txd->cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
- if (pl08x->vd->dualmaster) + if (pl08x->vd->flags & PL08X_IS_DUALMASTER) txd->cctl |= pl08x_select_bus(pl08x->mem_buses, pl08x->mem_buses);
@@ -1612,7 +1616,7 @@ bool pl08x_filter_id(struct dma_chan *chan, void *chan_id) static void pl08x_ensure_on(struct pl08x_driver_data *pl08x) { /* The Nomadik variant does not have the config register */ - if (pl08x->vd->nomadik) + if (pl08x->vd->flags & PL08X_IS_NOMADIK) return; writel(PL080_CONFIG_ENABLE, pl08x->base + PL080_CONFIG); } @@ -1897,7 +1901,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id) /* By default, AHB1 only. If dualmaster, from platform */ pl08x->lli_buses = PL08X_AHB1; pl08x->mem_buses = PL08X_AHB1; - if (pl08x->vd->dualmaster) { + if (pl08x->vd->flags & PL08X_IS_DUALMASTER) { pl08x->lli_buses = pl08x->pd->lli_buses; pl08x->mem_buses = pl08x->pd->mem_buses; } @@ -1954,7 +1958,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id) * down for the secure world only. Lock up these channels * by perpetually serving a dummy virtual channel. */ - if (vd->nomadik) { + if (vd->flags & PL08X_IS_NOMADIK) { u32 val;
val = readl(ch->base + PL080_CH_CONFIG); @@ -2039,18 +2043,17 @@ out_no_pl08x: /* PL080 has 8 channels and the PL080 have just 2 */ static struct vendor_data vendor_pl080 = { .channels = 8, - .dualmaster = true, + .flags = PL08X_IS_DUALMASTER, };
static struct vendor_data vendor_nomadik = { .channels = 8, - .dualmaster = true, - .nomadik = true, + .flags = PL08X_IS_DUALMASTER | PL08X_IS_NOMADIK, };
static struct vendor_data vendor_pl081 = { .channels = 2, - .dualmaster = false, + .flags = 0, };
static struct amba_id pl08x_ids[] = {
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
Instead of defining new bool field in vendor_data struct for each quirk, it is more reasonable to use a single flags field and make each quirk use single bits.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Not that I see the point, but I have no strong preference so: Acked-by: Linus Walleij linus.walleij@linaro.org
Yours, Linus Walleij
On Sun, Jun 16, 2013 at 10:54:08PM +0200, Tomasz Figa wrote:
Instead of defining new bool field in vendor_data struct for each quirk, it is more reasonable to use a single flags field and make each quirk use single bits.
Please explain why this is better over the existing system, and why it is not just churn for code modification's sake.
On Monday 17 of June 2013 19:48:56 Russell King - ARM Linux wrote:
On Sun, Jun 16, 2013 at 10:54:08PM +0200, Tomasz Figa wrote:
Instead of defining new bool field in vendor_data struct for each quirk, it is more reasonable to use a single flags field and make each quirk use single bits.
Please explain why this is better over the existing system, and why it is not just churn for code modification's sake.
It isn't anything important. I just thought that this is a better solution to store more than just two flags. Initially I had more of them added in further patches, but final version ended with just one, so this can be dropped.
Best regards, Tomasz
On Mon, Jun 17, 2013 at 8:56 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
On Monday 17 of June 2013 19:48:56 Russell King - ARM Linux wrote:
On Sun, Jun 16, 2013 at 10:54:08PM +0200, Tomasz Figa wrote:
Instead of defining new bool field in vendor_data struct for each quirk, it is more reasonable to use a single flags field and make each quirk use single bits.
Please explain why this is better over the existing system, and why it is not just churn for code modification's sake.
It isn't anything important. I just thought that this is a better solution to store more than just two flags. Initially I had more of them added in further patches, but final version ended with just one, so this can be dropped.
I thought about it yesterday and it basically looks to me that it tries to outsmart the compiler on how to stuff the flags to save struct space on the heap and hint that they can be determined quickly by a & operation, this is like premature optimization.
So please go back to the old system.
Yours, Linus Walleij
On Tuesday 18 June 2013, Linus Walleij wrote:
I thought about it yesterday and it basically looks to me that it tries to outsmart the compiler on how to stuff the flags to save struct space on the heap and hint that they can be determined quickly by a & operation, this is like premature optimization.
So please go back to the old system.
Note that a 'bool' is typically stored as a 32-bit field, so there would be some space saving if there are a lot of flags.
I agree with the conclusion though, the previous state was more readable, which outweighs any benefits.
Arnd
Further patch will introduce support for PL080S, which requires some things to be done conditionally, thus increasing indentation level of some functions even more.
This patch reduces indentation level of pl08x_getbytes_chan() function by inverting several conditions and returning from function wherever possible.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 51 ++++++++++++++++++++++++++---------------------- 1 file changed, 28 insertions(+), 23 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index d443a68..1e57ded 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -473,47 +473,52 @@ static inline u32 get_bytes_in_cctl(u32 cctl) /* The channel should be paused when calling this */ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) { + struct pl08x_lli *llis_va; struct pl08x_phy_chan *ch; + dma_addr_t llis_bus; struct pl08x_txd *txd; size_t bytes = 0; + int index; + u32 clli;
ch = plchan->phychan; txd = plchan->at;
+ if (!ch || !txd) + return bytes; + /* * Follow the LLIs to get the number of remaining * bytes in the currently active transaction. */ - if (ch && txd) { - u32 clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2; + clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2; + + /* First get the remaining bytes in the active transfer */ + bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
- /* First get the remaining bytes in the active transfer */ - bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL)); + if (!clli) + return bytes;
- if (clli) { - struct pl08x_lli *llis_va = txd->llis_va; - dma_addr_t llis_bus = txd->llis_bus; - int index; + llis_va = txd->llis_va; + llis_bus = txd->llis_bus;
- BUG_ON(clli < llis_bus || clli >= llis_bus + + BUG_ON(clli < llis_bus || clli >= llis_bus + sizeof(struct pl08x_lli) * MAX_NUM_TSFR_LLIS);
- /* - * Locate the next LLI - as this is an array, - * it's simple maths to find. - */ - index = (clli - llis_bus) / sizeof(struct pl08x_lli); + /* + * Locate the next LLI - as this is an array, + * it's simple maths to find. + */ + index = (clli - llis_bus) / sizeof(struct pl08x_lli);
- for (; index < MAX_NUM_TSFR_LLIS; index++) { - bytes += get_bytes_in_cctl(llis_va[index].cctl); + for (; index < MAX_NUM_TSFR_LLIS; index++) { + bytes += get_bytes_in_cctl(llis_va[index].cctl);
- /* - * A LLI pointer of 0 terminates the LLI list - */ - if (!llis_va[index].lli) - break; - } - } + /* + * A LLI pointer of 0 terminates the LLI list + */ + if (!llis_va[index].lli) + break; }
return bytes;
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
Further patch will introduce support for PL080S, which requires some things to be done conditionally, thus increasing indentation level of some functions even more.
This patch reduces indentation level of pl08x_getbytes_chan() function by inverting several conditions and returning from function wherever possible.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Nice refactoring and a patch that should be applied immediately no matter what happens with the rest. Reviewed-by: Linus Walleij linus.walleij@linaro.org
PS:
if (!ch || !txd)
return bytes;
Just return 0; is even clearer in this case but who cares.
Yours, Linus Walleij
Some variants of PL08x (namely PL080S, found in Samsung S3C64xx SoCs) have CONFIG register at different offset. This patch makes the driver use offset from vendor data struct.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 51 ++++++++++++++++++++++++++++-------------------- 1 file changed, 30 insertions(+), 21 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index 1e57ded..93913b4 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -107,6 +107,7 @@ struct pl08x_driver_data; * @flags: Vendor-specific flags, see PL08X_IS_* */ struct vendor_data { + u8 config_offset; u8 channels; u32 flags; }; @@ -334,11 +335,12 @@ static void pl08x_release_mux(struct pl08x_dma_chan *plchan) */
/* Whether a certain channel is busy or not */ -static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch) +static int pl08x_phy_channel_busy(struct pl08x_driver_data *pl08x, + struct pl08x_phy_chan *ch) { unsigned int val;
- val = readl(ch->base + PL080_CH_CONFIG); + val = readl(ch->base + pl08x->vd->config_offset); return val & PL080_CONFIG_ACTIVE; }
@@ -362,7 +364,7 @@ static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan) plchan->at = txd;
/* Wait for channel inactive */ - while (pl08x_phy_channel_busy(phychan)) + while (pl08x_phy_channel_busy(pl08x, phychan)) cpu_relax();
lli = &txd->llis_va[0]; @@ -377,7 +379,7 @@ static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan) writel(lli->dst, phychan->base + PL080_CH_DST_ADDR); writel(lli->lli, phychan->base + PL080_CH_LLI); writel(lli->cctl, phychan->base + PL080_CH_CONTROL); - writel(txd->ccfg, phychan->base + PL080_CH_CONFIG); + writel(txd->ccfg, phychan->base + pl08x->vd->config_offset);
/* Enable the DMA channel */ /* Do not access config register until channel shows as disabled */ @@ -385,11 +387,13 @@ static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan) cpu_relax();
/* Do not access config register until channel shows as inactive */ - val = readl(phychan->base + PL080_CH_CONFIG); + + val = readl(phychan->base + pl08x->vd->config_offset); while ((val & PL080_CONFIG_ACTIVE) || (val & PL080_CONFIG_ENABLE)) - val = readl(phychan->base + PL080_CH_CONFIG); + val = readl(phychan->base + pl08x->vd->config_offset);
- writel(val | PL080_CONFIG_ENABLE, phychan->base + PL080_CH_CONFIG); + writel(val | PL080_CONFIG_ENABLE, + phychan->base + pl08x->vd->config_offset); }
/* @@ -402,34 +406,36 @@ static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan) * For P->M transfers, disable the peripheral first to stop it filling * the DMAC FIFO, and then pause the DMAC. */ -static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch) +static void pl08x_pause_phy_chan(struct pl08x_driver_data *pl08x, + struct pl08x_phy_chan *ch) { u32 val; int timeout;
/* Set the HALT bit and wait for the FIFO to drain */ - val = readl(ch->base + PL080_CH_CONFIG); + val = readl(ch->base + pl08x->vd->config_offset); val |= PL080_CONFIG_HALT; - writel(val, ch->base + PL080_CH_CONFIG); + writel(val, ch->base + pl08x->vd->config_offset);
/* Wait for channel inactive */ for (timeout = 1000; timeout; timeout--) { - if (!pl08x_phy_channel_busy(ch)) + if (!pl08x_phy_channel_busy(pl08x, ch)) break; udelay(1); } - if (pl08x_phy_channel_busy(ch)) + if (pl08x_phy_channel_busy(pl08x, ch)) pr_err("pl08x: channel%u timeout waiting for pause\n", ch->id); }
-static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch) +static void pl08x_resume_phy_chan(struct pl08x_driver_data *pl08x, + struct pl08x_phy_chan *ch) { u32 val;
/* Clear the HALT bit */ - val = readl(ch->base + PL080_CH_CONFIG); + val = readl(ch->base + pl08x->vd->config_offset); val &= ~PL080_CONFIG_HALT; - writel(val, ch->base + PL080_CH_CONFIG); + writel(val, ch->base + pl08x->vd->config_offset); }
/* @@ -441,12 +447,12 @@ static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch) static void pl08x_terminate_phy_chan(struct pl08x_driver_data *pl08x, struct pl08x_phy_chan *ch) { - u32 val = readl(ch->base + PL080_CH_CONFIG); + u32 val = readl(ch->base + pl08x->vd->config_offset);
val &= ~(PL080_CONFIG_ENABLE | PL080_CONFIG_ERR_IRQ_MASK | PL080_CONFIG_TC_IRQ_MASK);
- writel(val, ch->base + PL080_CH_CONFIG); + writel(val, ch->base + pl08x->vd->config_offset);
writel(1 << ch->id, pl08x->base + PL080_ERR_CLEAR); writel(1 << ch->id, pl08x->base + PL080_TC_CLEAR); @@ -1576,11 +1582,11 @@ static int pl08x_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd, pl08x_free_txd_list(pl08x, plchan); break; case DMA_PAUSE: - pl08x_pause_phy_chan(plchan->phychan); + pl08x_pause_phy_chan(pl08x, plchan->phychan); plchan->state = PL08X_CHAN_PAUSED; break; case DMA_RESUME: - pl08x_resume_phy_chan(plchan->phychan); + pl08x_resume_phy_chan(pl08x, plchan->phychan); plchan->state = PL08X_CHAN_RUNNING; break; default: @@ -1966,7 +1972,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id) if (vd->flags & PL08X_IS_NOMADIK) { u32 val;
- val = readl(ch->base + PL080_CH_CONFIG); + val = readl(ch->base + vd->config_offset); if (val & (PL080N_CONFIG_ITPROT | PL080N_CONFIG_SECPROT)) { dev_info(&adev->dev, "physical channel %d reserved for secure access only\n", i); ch->locked = true; @@ -1974,7 +1980,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id) }
dev_dbg(&adev->dev, "physical channel %d is %s\n", - i, pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE"); + i, pl08x_phy_channel_busy(pl08x, ch) ? "BUSY" : "FREE"); }
/* Register as many memcpy channels as there are physical channels */ @@ -2049,16 +2055,19 @@ out_no_pl08x: static struct vendor_data vendor_pl080 = { .channels = 8, .flags = PL08X_IS_DUALMASTER, + .config_offset = PL080_CH_CONFIG, };
static struct vendor_data vendor_nomadik = { .channels = 8, .flags = PL08X_IS_DUALMASTER | PL08X_IS_NOMADIK, + .config_offset = PL080_CH_CONFIG, };
static struct vendor_data vendor_pl081 = { .channels = 2, .flags = 0, + .config_offset = PL080_CH_CONFIG, };
static struct amba_id pl08x_ids[] = {
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
Some variants of PL08x (namely PL080S, found in Samsung S3C64xx SoCs) have CONFIG register at different offset. This patch makes the driver use offset from vendor data struct.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Looks good to me: Reviewed-by: Linus Walleij linus.walleij@linaro.org
Yours, Linus Walleij
On Sun, Jun 16, 2013 at 10:54:10PM +0200, Tomasz Figa wrote:
Some variants of PL08x (namely PL080S, found in Samsung S3C64xx SoCs) have CONFIG register at different offset. This patch makes the driver use offset from vendor data struct.
I'd suggest doing this a different way. Instead of having to pass around two pointers everywhere in order to access this register, add to struct pl08x_phy_chan a void __iomem *reg_config; member, and initialize that to base + vd->config_offset. Then use ch->reg_cfg instead of ch->base + PL080_CH_CONFIG.
This has the benefit that you won't have to modify a whole load of functions to pass another argument, which costs not only an additional register, but also storage to keep it around.
On Monday 17 of June 2013 19:52:23 Russell King - ARM Linux wrote:
On Sun, Jun 16, 2013 at 10:54:10PM +0200, Tomasz Figa wrote:
Some variants of PL08x (namely PL080S, found in Samsung S3C64xx SoCs) have CONFIG register at different offset. This patch makes the driver use offset from vendor data struct.
I'd suggest doing this a different way. Instead of having to pass around two pointers everywhere in order to access this register, add to struct pl08x_phy_chan a void __iomem *reg_config; member, and initialize that to base + vd->config_offset. Then use ch->reg_cfg instead of ch->base + PL080_CH_CONFIG.
This has the benefit that you won't have to modify a whole load of functions to pass another argument, which costs not only an additional register, but also storage to keep it around.
OK. Let me do it this way and see how it turns out. However keep in mind that next patch adds further dependencies on access to vendor_data struct, so there is nothing sure.
Best regards, Tomasz
PL080S is a modified version of PL080 that can be found on Samsung SoCs, such as S3C6400 and S3C6410.
It has different offset of CONFIG register, separate CONTROL1 register that holds transfer size and larger maximum transfer size.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 68 +++++++++++++++++++++++++++++++++++++++------- include/linux/amba/pl080.h | 1 + 2 files changed, 59 insertions(+), 10 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index 93913b4..d1f1333 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -100,6 +100,8 @@ struct pl08x_driver_data; * for permission before use and some registers are missing. */ #define PL08X_IS_NOMADIK (1 << 1) +/** Controller is PL080S (PL080 modified by Samsung) */ +#define PL08X_IS_PL080S (1 << 2)
/** * struct vendor_data - vendor-specific config parameters for PL08x derivatives @@ -123,6 +125,7 @@ struct pl08x_lli { u32 dst; u32 lli; u32 cctl; + u32 cctl1; };
/** @@ -381,6 +384,9 @@ static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan) writel(lli->cctl, phychan->base + PL080_CH_CONTROL); writel(txd->ccfg, phychan->base + pl08x->vd->config_offset);
+ if (pl08x->vd->flags & PL08X_IS_PL080S) + writel(lli->cctl1, phychan->base + PL080S_CH_CONTROL2); + /* Enable the DMA channel */ /* Do not access config register until channel shows as disabled */ while (readl(pl08x->base + PL080_EN_CHAN) & (1 << phychan->id)) @@ -476,9 +482,28 @@ static inline u32 get_bytes_in_cctl(u32 cctl) return bytes; }
+static inline u32 get_bytes_in_cctl_pl080s(u32 cctl, u32 cctl1) +{ + /* The source width defines the number of bytes */ + u32 bytes = cctl1 & PL080S_CONTROL_TRANSFER_SIZE_MASK; + + switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) { + case PL080_WIDTH_8BIT: + break; + case PL080_WIDTH_16BIT: + bytes *= 2; + break; + case PL080_WIDTH_32BIT: + bytes *= 4; + break; + } + return bytes; +} + /* The channel should be paused when calling this */ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) { + struct pl08x_driver_data *pl08x = plchan->host; struct pl08x_lli *llis_va; struct pl08x_phy_chan *ch; dma_addr_t llis_bus; @@ -500,7 +525,12 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2;
/* First get the remaining bytes in the active transfer */ - bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL)); + if (pl08x->vd->flags & PL08X_IS_PL080S) + bytes = get_bytes_in_cctl_pl080s( + readl(ch->base + PL080_CH_CONTROL), + readl(ch->base + PL080S_CH_CONTROL2)); + else + bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
if (!clli) return bytes; @@ -518,7 +548,11 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) index = (clli - llis_bus) / sizeof(struct pl08x_lli);
for (; index < MAX_NUM_TSFR_LLIS; index++) { - bytes += get_bytes_in_cctl(llis_va[index].cctl); + if (pl08x->vd->flags & PL08X_IS_PL080S) + bytes += get_bytes_in_cctl_pl080s(llis_va[index].cctl, + llis_va[index].cctl1); + else + bytes += get_bytes_in_cctl(llis_va[index].cctl);
/* * A LLI pointer of 0 terminates the LLI list @@ -780,7 +814,7 @@ static void pl08x_choose_master_bus(struct pl08x_lli_build_data *bd, * Fills in one LLI for a certain transfer descriptor and advance the counter */ static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd, - int num_llis, int len, u32 cctl) + int num_llis, int len, u32 cctl, u32 cctl1) { struct pl08x_lli *llis_va = bd->txd->llis_va; dma_addr_t llis_bus = bd->txd->llis_bus; @@ -788,6 +822,7 @@ static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd, BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
llis_va[num_llis].cctl = cctl; + llis_va[num_llis].cctl1 = cctl1; llis_va[num_llis].src = bd->srcbus.addr; llis_va[num_llis].dst = bd->dstbus.addr; llis_va[num_llis].lli = llis_bus + (num_llis + 1) * @@ -804,11 +839,12 @@ static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd, bd->remainder -= len; }
-static inline void prep_byte_width_lli(struct pl08x_lli_build_data *bd, +static inline void prep_byte_width_lli(struct pl08x_driver_data *pl08x, + struct pl08x_lli_build_data *bd, u32 *cctl, u32 len, int num_llis, size_t *total_bytes) { *cctl = pl08x_cctl_bits(*cctl, 1, 1, len); - pl08x_fill_lli_for_desc(bd, num_llis, len, *cctl); + pl08x_fill_lli_for_desc(bd, num_llis, len, *cctl, len); (*total_bytes) += len; }
@@ -912,7 +948,7 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth, bd.dstbus.buswidth, 0); - pl08x_fill_lli_for_desc(&bd, num_llis++, 0, cctl); + pl08x_fill_lli_for_desc(&bd, num_llis++, 0, cctl, 0); break; }
@@ -934,7 +970,7 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, dev_vdbg(&pl08x->adev->dev, "%s byte width LLIs (remain 0x%08x)\n", __func__, bd.remainder); - prep_byte_width_lli(&bd, &cctl, early_bytes, num_llis++, + prep_byte_width_lli(pl08x, &bd, &cctl, early_bytes, num_llis++, &total_bytes); }
@@ -992,7 +1028,7 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth, bd.dstbus.buswidth, tsize); pl08x_fill_lli_for_desc(&bd, num_llis++, - lli_len, cctl); + lli_len, cctl, tsize); total_bytes += lli_len; }
@@ -1003,8 +1039,8 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, dev_vdbg(&pl08x->adev->dev, "%s align with boundary, send odd bytes (remain %zu)\n", __func__, bd.remainder); - prep_byte_width_lli(&bd, &cctl, bd.remainder, - num_llis++, &total_bytes); + prep_byte_width_lli(pl08x, &bd, &cctl, + bd.remainder, num_llis++, &total_bytes); } }
@@ -2064,6 +2100,12 @@ static struct vendor_data vendor_nomadik = { .config_offset = PL080_CH_CONFIG, };
+static struct vendor_data vendor_pl080s = { + .channels = 8, + .flags = PL08X_IS_DUALMASTER | PL08X_IS_PL080S, + .config_offset = PL080S_CH_CONFIG, +}; + static struct vendor_data vendor_pl081 = { .channels = 2, .flags = 0, @@ -2071,6 +2113,12 @@ static struct vendor_data vendor_pl081 = { };
static struct amba_id pl08x_ids[] = { + /* Samsung PL080S variant */ + { + .id = 0x0a141080, + .mask = 0xffffffff, + .data = &vendor_pl080s, + }, /* PL080 */ { .id = 0x00041080, diff --git a/include/linux/amba/pl080.h b/include/linux/amba/pl080.h index 3e7b62f..ef36a0a 100644 --- a/include/linux/amba/pl080.h +++ b/include/linux/amba/pl080.h @@ -87,6 +87,7 @@ #define PL080_CONTROL_SB_SIZE_MASK (0x7 << 12) #define PL080_CONTROL_SB_SIZE_SHIFT (12) #define PL080_CONTROL_TRANSFER_SIZE_MASK (0xfff << 0) +#define PL080S_CONTROL_TRANSFER_SIZE_MASK (0xffffff << 0) #define PL080_CONTROL_TRANSFER_SIZE_SHIFT (0)
#define PL080_BSIZE_1 (0x0)
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
PL080S is a modified version of PL080 that can be found on Samsung SoCs, such as S3C6400 and S3C6410.
It has different offset of CONFIG register, separate CONTROL1 register that holds transfer size and larger maximum transfer size.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Acked-by: Linus Walleij linus.walleij@linaro.org
(I would have done some smallish things differently but it is only a matter of taste.)
static struct amba_id pl08x_ids[] = {
/* Samsung PL080S variant */
{
.id = 0x0a141080,
.mask = 0xffffffff,
.data = &vendor_pl080s,
},
Interesting that Samsung claim to be vendor 0x41 (ARM). Well ARM never made revision no 0x0a so this works.
Yours, Linus Walleij
On Monday 17 of June 2013 15:39:34 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
PL080S is a modified version of PL080 that can be found on Samsung SoCs, such as S3C6400 and S3C6410.
It has different offset of CONFIG register, separate CONTROL1 register that holds transfer size and larger maximum transfer size.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Acked-by: Linus Walleij linus.walleij@linaro.org
(I would have done some smallish things differently but it is only a matter of taste.)
static struct amba_id pl08x_ids[] = {
/* Samsung PL080S variant */
{
.id = 0x0a141080,
.mask = 0xffffffff,
.data = &vendor_pl080s,
},
Interesting that Samsung claim to be vendor 0x41 (ARM).
Yeah, I had identical thought when I saw it first time.
Well ARM never made revision no 0x0a so this works.
Good, this is exactly what I wanted to read! ;)
I was afraid that this ID might conflict with some revision of normal PL080, but fortunately it is not the case.
Best refards, Tomasz
PL080S has separate register to store transfer size in, allowing single transfer to be much larger than in standard PL080.
This patch makes the amba-pl08x driver aware of this and removes writing transfer size to reserved bits of CH_CONTROL register on PL080S, which was not a problem witn transfer sizes fitting the original bitfield of PL080, but now would overwrite other fields.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index d1f1333..eb10eb8 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -112,6 +112,7 @@ struct vendor_data { u8 config_offset; u8 channels; u32 flags; + u32 max_transfer_size; };
/* @@ -843,7 +844,10 @@ static inline void prep_byte_width_lli(struct pl08x_driver_data *pl08x, struct pl08x_lli_build_data *bd, u32 *cctl, u32 len, int num_llis, size_t *total_bytes) { - *cctl = pl08x_cctl_bits(*cctl, 1, 1, len); + if (pl08x->vd->flags & PL08X_IS_PL080S) + *cctl = pl08x_cctl_bits(*cctl, 1, 1, 0); + else + *cctl = pl08x_cctl_bits(*cctl, 1, 1, len); pl08x_fill_lli_for_desc(bd, num_llis, len, *cctl, len); (*total_bytes) += len; } @@ -992,7 +996,7 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, * MIN(buswidths) */ max_bytes_per_lli = bd.srcbus.buswidth * - PL080_CONTROL_TRANSFER_SIZE_MASK; + pl08x->vd->max_transfer_size; dev_vdbg(&pl08x->adev->dev, "%s max bytes per lli = %zu\n", __func__, max_bytes_per_lli); @@ -1025,8 +1029,14 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, "size 0x%08zx (remainder 0x%08zx)\n", __func__, lli_len, bd.remainder);
- cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth, - bd.dstbus.buswidth, tsize); + if (pl08x->vd->flags & PL08X_IS_PL080S) + cctl = pl08x_cctl_bits(cctl, + bd.srcbus.buswidth, + bd.dstbus.buswidth, 0); + else + cctl = pl08x_cctl_bits(cctl, + bd.srcbus.buswidth, + bd.dstbus.buswidth, tsize); pl08x_fill_lli_for_desc(&bd, num_llis++, lli_len, cctl, tsize); total_bytes += lli_len; @@ -2092,24 +2102,28 @@ static struct vendor_data vendor_pl080 = { .channels = 8, .flags = PL08X_IS_DUALMASTER, .config_offset = PL080_CH_CONFIG, + .max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK, };
static struct vendor_data vendor_nomadik = { .channels = 8, .flags = PL08X_IS_DUALMASTER | PL08X_IS_NOMADIK, .config_offset = PL080_CH_CONFIG, + .max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK, };
static struct vendor_data vendor_pl080s = { .channels = 8, .flags = PL08X_IS_DUALMASTER | PL08X_IS_PL080S, .config_offset = PL080S_CH_CONFIG, + .max_transfer_size = PL080S_CONTROL_TRANSFER_SIZE_MASK, };
static struct vendor_data vendor_pl081 = { .channels = 2, .flags = 0, .config_offset = PL080_CH_CONFIG, + .max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK, };
static struct amba_id pl08x_ids[] = {
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
PL080S has separate register to store transfer size in, allowing single transfer to be much larger than in standard PL080.
This patch makes the amba-pl08x driver aware of this and removes writing transfer size to reserved bits of CH_CONTROL register on PL080S, which was not a problem witn transfer sizes fitting the original bitfield of PL080, but now would overwrite other fields.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Very straight-forward. Reviewed-by: Linus Walleij linus.walleij@linaro.org
Yours, Linus Walleij
On Monday 17 of June 2013 15:42:15 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
PL080S has separate register to store transfer size in, allowing single transfer to be much larger than in standard PL080.
This patch makes the amba-pl08x driver aware of this and removes writing transfer size to reserved bits of CH_CONTROL register on PL080S, which was not a problem witn transfer sizes fitting the original bitfield of PL080, but now would overwrite other fields.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Very straight-forward. Reviewed-by: Linus Walleij linus.walleij@linaro.org
Thanks.
Actually today I found another way of doing this, which wouldn't require checking PL08X_IS_PL080S flag at all.
Simply masking the transfer size in pl08x_cctl_bits() with mask of the original bitfield would be enough I think, as writes to those bits seem to be ignored on PL080S.
Best regards, Tomasz
PL080 reference manual states that to LLI entries should be aligned to 4-word boundary to make LLI fetches more efficient. This patch adds a 3-word padding to the LLi struct to make this condition true.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index eb10eb8..0da5539 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -127,6 +127,7 @@ struct pl08x_lli { u32 lli; u32 cctl; u32 cctl1; + u32 dummy[3]; };
/**
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
PL080 reference manual states that to LLI entries should be aligned to 4-word boundary to make LLI fetches more efficient. This patch adds a 3-word padding to the LLi struct to make this condition true.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
drivers/dma/amba-pl08x.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index eb10eb8..0da5539 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -127,6 +127,7 @@ struct pl08x_lli { u32 lli; u32 cctl; u32 cctl1;
u32 dummy[3];
Atleast put a comment into the code explaining what this is all about. Or someone will add another member to the struct and all is lost. Call it "padding" rather than dummy.
};
So it used to be like this before you added cctl1:
struct pl08x_lli { u32 src; u32 dst; u32 lli; u32 cctl; };
Meaning it was 3 words.
And now you make it take 8 words for everyone.
Atleast this patch should be squashed into the patch adding cctl1.
But I really don't like this fragile way of casting structs right into memory, and I don't like that teh other PL080's also have to waste 8 words when their LLIs fit so nicely into 4.
I would have solved this problem by creating a marshalling function that just allocate the number of bytes the LLI entry shall have and fill it in by assigning directly to the precise target memory cell. This way the LLIs will take 4 words on the original variants and you can use some nice logic to pad out to 8 words on the PL080S variant.
Yours, Linus Walleij
On Monday 17 of June 2013 15:51:20 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
PL080 reference manual states that to LLI entries should be aligned to 4-word boundary to make LLI fetches more efficient. This patch adds a 3-word padding to the LLi struct to make this condition true.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
drivers/dma/amba-pl08x.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index eb10eb8..0da5539 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -127,6 +127,7 @@ struct pl08x_lli {
u32 lli; u32 cctl; u32 cctl1;
u32 dummy[3];
Atleast put a comment into the code explaining what this is all about. Or someone will add another member to the struct and all is lost. Call it "padding" rather than dummy.
};
So it used to be like this before you added cctl1:
struct pl08x_lli { u32 src; u32 dst; u32 lli; u32 cctl; };
Meaning it was 3 words.
And now you make it take 8 words for everyone.
Atleast this patch should be squashed into the patch adding cctl1.
But I really don't like this fragile way of casting structs right into memory, and I don't like that teh other PL080's also have to waste 8 words when their LLIs fit so nicely into 4.
I would have solved this problem by creating a marshalling function that just allocate the number of bytes the LLI entry shall have and fill it in by assigning directly to the precise target memory cell. This way the LLIs will take 4 words on the original variants and you can use some nice logic to pad out to 8 words on the PL080S variant.
Definitely a valid point. I'll see what I can do about it.
I was thinking about it originally, but I couldn't find any really good solution for this so I just went with this extremely simple approach as a proof of concept and to show the problem. :)
Best regards, Tomasz
On Mon, Jun 17, 2013 at 03:51:20PM +0200, Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index eb10eb8..0da5539 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -127,6 +127,7 @@ struct pl08x_lli { u32 lli; u32 cctl; u32 cctl1;
u32 dummy[3];
Atleast put a comment into the code explaining what this is all about. Or someone will add another member to the struct and all is lost. Call it "padding" rather than dummy.
Another solution to that is to use __attribute__((__aligned__(16))) if you want it aligned to 16 byte boundaries.
From: Alban Bedel alban.bedel@avionic-design.de
There are more fields than just SWIDTH in CH_CONTROL register, so read register value must be masked in addition to shifting.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 4 ++++ 1 file changed, 4 insertions(+)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index 0da5539..bb3b36b 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -471,6 +471,8 @@ static inline u32 get_bytes_in_cctl(u32 cctl) /* The source width defines the number of bytes */ u32 bytes = cctl & PL080_CONTROL_TRANSFER_SIZE_MASK;
+ cctl &= PL080_CONTROL_SWIDTH_MASK; + switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) { case PL080_WIDTH_8BIT: break; @@ -489,6 +491,8 @@ static inline u32 get_bytes_in_cctl_pl080s(u32 cctl, u32 cctl1) /* The source width defines the number of bytes */ u32 bytes = cctl1 & PL080S_CONTROL_TRANSFER_SIZE_MASK;
+ cctl &= PL080_CONTROL_SWIDTH_MASK; + switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) { case PL080_WIDTH_8BIT: break;
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
From: Alban Bedel alban.bedel@avionic-design.de
There are more fields than just SWIDTH in CH_CONTROL register, so read register value must be masked in addition to shifting.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Acked-by: Linus Walleij linus.walleij@linaro.org
Are we just lucky on current variants such that all unmasked bits happen to be zero on them?
Yours, Linus Walleij
On Monday 17 of June 2013 15:53:14 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
From: Alban Bedel alban.bedel@avionic-design.de
There are more fields than just SWIDTH in CH_CONTROL register, so read register value must be masked in addition to shifting.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Acked-by: Linus Walleij linus.walleij@linaro.org
Are we just lucky on current variants such that all unmasked bits happen to be zero on them?
This is really interesting, because if you look at the bit layout, there is a lot of other bitfields above the SWIDTH field, like DWIDTH, src and dest AHB master selection, src and dest incerement setting, protection and terminal count interrupt enable bits.
Best regards, Tomasz
On Mon, Jun 17, 2013 at 03:53:14PM +0200, Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
From: Alban Bedel alban.bedel@avionic-design.de
There are more fields than just SWIDTH in CH_CONTROL register, so read register value must be masked in addition to shifting.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Acked-by: Linus Walleij linus.walleij@linaro.org
Are we just lucky on current variants such that all unmasked bits happen to be zero on them?
It's probably that all the places which this gets used, the transfers are 8-bit (like the UART) so it "just works" irrespective of that.
From: Alban Bedel alban.bedel@avionic-design.de
Many audio interface drivers require support of cyclic transfers to work correctly, for example Samsung ASoC DMA driver. This patch adds support for cyclic transfers to the amba-pl08x driver.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/dma/amba-pl08x.c | 181 +++++++++++++++++++++++++++++++++++------------ 1 file changed, 136 insertions(+), 45 deletions(-)
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index bb3b36b..210a893 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -184,6 +184,7 @@ struct pl08x_sg { * @ccfg: config reg values for current txd * @done: this marks completed descriptors, which should not have their * mux released. + * @cyclic: indicate cyclic transfers */ struct pl08x_txd { struct virt_dma_desc vd; @@ -198,6 +199,8 @@ struct pl08x_txd { */ u32 ccfg; bool done; + + bool cyclic; };
/** @@ -561,9 +564,9 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) bytes += get_bytes_in_cctl(llis_va[index].cctl);
/* - * A LLI pointer of 0 terminates the LLI list ++ * A LLI pointer going backward terminates the LLI list */ - if (!llis_va[index].lli) + if (llis_va[index].lli <= clli) break; }
@@ -1075,10 +1078,14 @@ static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x, }
llis_va = txd->llis_va; - /* The final LLI terminates the LLI. */ - llis_va[num_llis - 1].lli = 0; - /* The final LLI element shall also fire an interrupt. */ - llis_va[num_llis - 1].cctl |= PL080_CONTROL_TC_IRQ_EN; + if (txd->cyclic) { + /* Link back to the first LLI. */ + llis_va[num_llis - 1].lli = txd->llis_bus | bd.lli_bus; + } else { + /* The final LLI terminates the LLI. */ + llis_va[num_llis - 1].lli = 0; + llis_va[num_llis - 1].cctl |= PL080_CONTROL_TC_IRQ_EN; + }
#ifdef VERBOSE_DEBUG { @@ -1470,25 +1477,19 @@ static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy( return vchan_tx_prep(&plchan->vc, &txd->vd, flags); }
-static struct dma_async_tx_descriptor *pl08x_prep_slave_sg( - struct dma_chan *chan, struct scatterlist *sgl, - unsigned int sg_len, enum dma_transfer_direction direction, - unsigned long flags, void *context) +static struct pl08x_txd *pl08x_init_txd( + struct dma_chan *chan, + enum dma_transfer_direction direction, + dma_addr_t *slave_addr) { struct pl08x_dma_chan *plchan = to_pl08x_chan(chan); struct pl08x_driver_data *pl08x = plchan->host; struct pl08x_txd *txd; - struct pl08x_sg *dsg; - struct scatterlist *sg; enum dma_slave_buswidth addr_width; - dma_addr_t slave_addr; int ret, tmp; u8 src_buses, dst_buses; u32 maxburst, cctl;
- dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n", - __func__, sg_dma_len(sgl), plchan->name); - txd = pl08x_get_txd(plchan); if (!txd) { dev_err(&pl08x->adev->dev, "%s no txd\n", __func__); @@ -1502,14 +1503,14 @@ static struct dma_async_tx_descriptor *pl08x_prep_slave_sg( */ if (direction == DMA_MEM_TO_DEV) { cctl = PL080_CONTROL_SRC_INCR; - slave_addr = plchan->cfg.dst_addr; + *slave_addr = plchan->cfg.dst_addr; addr_width = plchan->cfg.dst_addr_width; maxburst = plchan->cfg.dst_maxburst; src_buses = pl08x->mem_buses; dst_buses = plchan->cd->periph_buses; } else if (direction == DMA_DEV_TO_MEM) { cctl = PL080_CONTROL_DST_INCR; - slave_addr = plchan->cfg.src_addr; + *slave_addr = plchan->cfg.src_addr; addr_width = plchan->cfg.src_addr_width; maxburst = plchan->cfg.src_maxburst; src_buses = plchan->cd->periph_buses; @@ -1558,24 +1559,107 @@ static struct dma_async_tx_descriptor *pl08x_prep_slave_sg( else txd->ccfg |= plchan->signal << PL080_CONFIG_SRC_SEL_SHIFT;
+ return txd; +} + +static int pl08x_tx_add_sg(struct pl08x_txd *txd, + enum dma_transfer_direction direction, + dma_addr_t slave_addr, + dma_addr_t buf_addr, + unsigned int len) +{ + struct pl08x_sg *dsg; + + dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT); + if (!dsg) + return -ENOMEM; + + list_add_tail(&dsg->node, &txd->dsg_list); + + dsg->len = len; + if (direction == DMA_MEM_TO_DEV) { + dsg->src_addr = buf_addr; + dsg->dst_addr = slave_addr; + } else { + dsg->src_addr = slave_addr; + dsg->dst_addr = buf_addr; + } + + return 0; +} + +static struct dma_async_tx_descriptor *pl08x_prep_slave_sg( + struct dma_chan *chan, struct scatterlist *sgl, + unsigned int sg_len, enum dma_transfer_direction direction, + unsigned long flags, void *context) +{ + struct pl08x_dma_chan *plchan = to_pl08x_chan(chan); + struct pl08x_driver_data *pl08x = plchan->host; + struct pl08x_txd *txd; + struct scatterlist *sg; + int ret, tmp; + dma_addr_t slave_addr; + + dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n", + __func__, sg_dma_len(sgl), plchan->name); + + txd = pl08x_init_txd(chan, direction, &slave_addr); + if (!txd) + return NULL; + for_each_sg(sgl, sg, sg_len, tmp) { - dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT); - if (!dsg) { + ret = pl08x_tx_add_sg(txd, direction, slave_addr, + sg_dma_address(sg), + sg_dma_len(sg)); + if (ret) { pl08x_release_mux(plchan); pl08x_free_txd(pl08x, txd); dev_err(&pl08x->adev->dev, "%s no mem for pl080 sg\n", __func__); return NULL; } - list_add_tail(&dsg->node, &txd->dsg_list); + }
- dsg->len = sg_dma_len(sg); - if (direction == DMA_MEM_TO_DEV) { - dsg->src_addr = sg_dma_address(sg); - dsg->dst_addr = slave_addr; - } else { - dsg->src_addr = slave_addr; - dsg->dst_addr = sg_dma_address(sg); + ret = pl08x_fill_llis_for_desc(plchan->host, txd); + if (!ret) { + pl08x_release_mux(plchan); + pl08x_free_txd(pl08x, txd); + return NULL; + } + + return vchan_tx_prep(&plchan->vc, &txd->vd, flags); +} + +static struct dma_async_tx_descriptor *pl08x_prep_dma_cyclic( + struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len, + size_t period_len, enum dma_transfer_direction direction, + unsigned long flags, void *context) +{ + struct pl08x_dma_chan *plchan = to_pl08x_chan(chan); + struct pl08x_driver_data *pl08x = plchan->host; + struct pl08x_txd *txd; + int ret, tmp; + dma_addr_t slave_addr; + + dev_dbg(&pl08x->adev->dev, + "%s prepare cyclic transaction of %d/%d bytes %s %s\n", + __func__, period_len, buf_len, + direction == DMA_MEM_TO_DEV ? "to" : "from", + plchan->name); + + txd = pl08x_init_txd(chan, direction, &slave_addr); + if (!txd) + return NULL; + + txd->cyclic = true; + txd->cctl |= PL080_CONTROL_TC_IRQ_EN; + for (tmp = 0 ; tmp < buf_len ; tmp += period_len) { + ret = pl08x_tx_add_sg(txd, direction, slave_addr, + buf_addr + tmp, period_len); + if (ret) { + pl08x_release_mux(plchan); + pl08x_free_txd(pl08x, txd); + return NULL; } }
@@ -1719,23 +1803,28 @@ static irqreturn_t pl08x_irq(int irq, void *dev) spin_lock(&plchan->vc.lock); tx = plchan->at; if (tx) { - plchan->at = NULL; - /* - * This descriptor is done, release its mux - * reservation. - */ - pl08x_release_mux(plchan); - tx->done = true; - vchan_cookie_complete(&tx->vd); - - /* - * And start the next descriptor (if any), - * otherwise free this channel. - */ - if (vchan_next_desc(&plchan->vc)) - pl08x_start_next_txd(plchan); - else - pl08x_phy_free(plchan); + if (tx->cyclic) { + vchan_cyclic_callback(&tx->vd); + } else { + plchan->at = NULL; + /* + * This descriptor is done, release + * its mux reservation. + */ + pl08x_release_mux(plchan); + tx->done = true; + vchan_cookie_complete(&tx->vd); + + /* + * And start the next descriptor + * (if any), otherwise free this + * channel. + */ + if (vchan_next_desc(&plchan->vc)) + pl08x_start_next_txd(plchan); + else + pl08x_phy_free(plchan); + } } spin_unlock(&plchan->vc.lock);
@@ -1939,6 +2028,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id)
/* Initialize slave engine */ dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask); + dma_cap_set(DMA_CYCLIC, pl08x->slave.cap_mask); pl08x->slave.dev = &adev->dev; pl08x->slave.device_alloc_chan_resources = pl08x_alloc_chan_resources; pl08x->slave.device_free_chan_resources = pl08x_free_chan_resources; @@ -1946,6 +2036,7 @@ static int pl08x_probe(struct amba_device *adev, const struct amba_id *id) pl08x->slave.device_tx_status = pl08x_dma_tx_status; pl08x->slave.device_issue_pending = pl08x_issue_pending; pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg; + pl08x->slave.device_prep_dma_cyclic = pl08x_prep_dma_cyclic; pl08x->slave.device_control = pl08x_control;
/* Get the platform data */
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
From: Alban Bedel alban.bedel@avionic-design.de
Many audio interface drivers require support of cyclic transfers to work correctly, for example Samsung ASoC DMA driver. This patch adds support for cyclic transfers to the amba-pl08x driver.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Nice! This is useful for others as well.
@@ -198,6 +199,8 @@ struct pl08x_txd { */ u32 ccfg; bool done;
bool cyclic;
};
You can never have enough whitespace right?
/** @@ -561,9 +564,9 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) bytes += get_bytes_in_cctl(llis_va[index].cctl);
/*
* A LLI pointer of 0 terminates the LLI list
++ * A LLI pointer going backward terminates the LLI list
++? Really?
This looks like a merge leftover being merged in.
if (!llis_va[index].lli)
if (llis_va[index].lli <= clli)
This was clever.
The rest of the code looks nice.
Yours, Linus Walleij
On Monday 17 of June 2013 15:57:38 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
From: Alban Bedel alban.bedel@avionic-design.de
Many audio interface drivers require support of cyclic transfers to work correctly, for example Samsung ASoC DMA driver. This patch adds support for cyclic transfers to the amba-pl08x driver.
Signed-off-by: Alban Bedel alban.bedel@avionic-design.de Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Nice! This is useful for others as well.
@@ -198,6 +199,8 @@ struct pl08x_txd {
*/ u32 ccfg; bool done;
bool cyclic;
};
You can never have enough whitespace right?
/**
@@ -561,9 +564,9 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)> bytes += get_bytes_in_cctl(llis_va[index].cctl);
/*
* A LLI pointer of 0 terminates the LLI list
++ * A LLI pointer going backward terminates the LLI list
++? Really?
This looks like a merge leftover being merged in.
I think it's a copy/paste error when applying some hunks of the original patch manually.
if (!llis_va[index].lli)
if (llis_va[index].lli <= clli)
This was clever.
The rest of the code looks nice.
OK. I will fix the issues you mentioned in next version.
Best regards, Tomasz
On Sun, Jun 16, 2013 at 10:54:15PM +0200, Tomasz Figa wrote:
@@ -561,9 +564,9 @@ static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan) bytes += get_bytes_in_cctl(llis_va[index].cctl);
/*
* A LLI pointer of 0 terminates the LLI list
++ * A LLI pointer going backward terminates the LLI list
Spot the unfixed up merge error...
With support for amba-pl08x driver, on S3C64xx the generic DMA engine API can be used instead of the private s3c-dma interface.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- drivers/spi/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/spi/Kconfig b/drivers/spi/Kconfig index 2015897..8da1c22 100644 --- a/drivers/spi/Kconfig +++ b/drivers/spi/Kconfig @@ -366,7 +366,7 @@ config SPI_S3C24XX_FIQ config SPI_S3C64XX tristate "Samsung S3C64XX series type SPI" depends on (ARCH_S3C24XX || ARCH_S3C64XX || ARCH_S5P64X0 || ARCH_EXYNOS) - select S3C64XX_DMA if ARCH_S3C64XX + select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080 help SPI driver for Samsung S3C64XX and newer SoCs.
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
With support for amba-pl08x driver, on S3C64xx the generic DMA engine API can be used instead of the private s3c-dma interface.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
(...)
select S3C64XX_DMA if ARCH_S3C64XX
select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080
I guess this means the SPI driver does not work without DMA. But is the SPI driver already augmented to handle generic DMA engine? Anyway:
Acked-by: Linus Walleij linus.walleij@linaro.org
Yours, Linus Walleij
On Mon, Jun 17, 2013 at 03:59:39PM +0200, Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
select S3C64XX_DMA if ARCH_S3C64XX
select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080
I guess this means the SPI driver does not work without DMA. But is the SPI driver already augmented to handle generic DMA engine? Anyway:
Yes, the driver currently relies on DMA. There are some pending patches for non-DMA operation that I need to test. To be honest for both this and the ASoC driver I'd be more inclined to just either select or depend on the DMA controller rather than leaving the option of using the old code.
On Monday 17 of June 2013 17:06:30 Mark Brown wrote:
On Mon, Jun 17, 2013 at 03:59:39PM +0200, Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
select S3C64XX_DMA if ARCH_S3C64XX
select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080
I guess this means the SPI driver does not work without DMA. But is the SPI driver already augmented to handle generic DMA engine?
Anyway:
Yes, the driver currently relies on DMA. There are some pending patches for non-DMA operation that I need to test. To be honest for both this and the ASoC driver I'd be more inclined to just either select or depend on the DMA controller rather than leaving the option of using the old code.
Well, I just left this for the purpose of this RFC, so testing things on both drivers would be possible (to see if any problems are caused by new driver or just by things that don't work).
If this driver turns out to be working correctly, I will happily remove the old private one and after s3c24xx moves to Heiko's DMA engine driver (or as it already moved?) we will be able to completely drop the private s3c-dma API.
Best regards, Tomasz
With support for amba-pl08x driver, on S3C64xx the generic DMA engine API can be used instead of the private s3c-dma interface.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- sound/soc/samsung/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/sound/soc/samsung/Kconfig b/sound/soc/samsung/Kconfig index ae0ea87..dbced1e 100644 --- a/sound/soc/samsung/Kconfig +++ b/sound/soc/samsung/Kconfig @@ -1,7 +1,7 @@ config SND_SOC_SAMSUNG tristate "ASoC support for Samsung" depends on PLAT_SAMSUNG - select S3C64XX_DMA if ARCH_S3C64XX + select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080 select S3C2410_DMA if ARCH_S3C24XX help Say Y or M if you want to add support for codecs attached to
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
With support for amba-pl08x driver, on S3C64xx the generic DMA engine API can be used instead of the private s3c-dma interface.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
(...)
select S3C64XX_DMA if ARCH_S3C64XX
select S3C64XX_DMA if ARCH_S3C64XX && !S3C64XX_PL080 select S3C2410_DMA if ARCH_S3C24XX
I hope it will look nicer (like just selecting the real PL080 driver) in the end. OK can live with this: Acked-by: Linus Walleij linus.walleij@linaro.org
Yours, Linus Walleij
This patch adds all required platform-specific data and initialization code to support the generic amba-pl08x driver on S3C64xx SoCs.
Also some compatibility definitions are added to make the transition from legacy API to DMA engine easier. The biggest hack here is passing const char * pointers through DMA resource, casted to unsigned long, but this is how Samsung DMA wrappers (used to support both s3c-dma and DMA engine in drivers) is designed.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com --- arch/arm/Kconfig | 1 + arch/arm/mach-s3c64xx/Kconfig | 8 +- arch/arm/mach-s3c64xx/Makefile | 1 + arch/arm/mach-s3c64xx/common.h | 5 + arch/arm/mach-s3c64xx/include/mach/dma.h | 65 ++++++++ arch/arm/mach-s3c64xx/pl080.c | 244 +++++++++++++++++++++++++++++++ 6 files changed, 323 insertions(+), 1 deletion(-) create mode 100644 arch/arm/mach-s3c64xx/pl080.c
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index 42d6ea2..fab8f3c 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -714,6 +714,7 @@ config ARCH_S3C64XX bool "Samsung S3C64XX" select ARCH_HAS_CPUFREQ select ARCH_REQUIRE_GPIOLIB + select ARM_AMBA select ARM_VIC select CLKDEV_LOOKUP select CLKSRC_MMIO diff --git a/arch/arm/mach-s3c64xx/Kconfig b/arch/arm/mach-s3c64xx/Kconfig index 2057853..704c5e4 100644 --- a/arch/arm/mach-s3c64xx/Kconfig +++ b/arch/arm/mach-s3c64xx/Kconfig @@ -28,9 +28,15 @@ config CPU_S3C6410 help Enable S3C6410 CPU support
+config S3C64XX_PL080 + bool "S3C64XX DMA using generic PL08x driver" + select AMBA_PL08X + select SAMSUNG_DMADEV + config S3C64XX_DMA - bool "S3C64XX DMA" + bool "S3C64XX DMA using legacy S3C DMA API" select S3C_DMA + depends on !S3C64XX_PL080
config S3C64XX_SETUP_SDHCI bool diff --git a/arch/arm/mach-s3c64xx/Makefile b/arch/arm/mach-s3c64xx/Makefile index 31d0c91..4e3326a 100644 --- a/arch/arm/mach-s3c64xx/Makefile +++ b/arch/arm/mach-s3c64xx/Makefile @@ -27,6 +27,7 @@ obj-$(CONFIG_CPU_IDLE) += cpuidle.o # DMA support
obj-$(CONFIG_S3C64XX_DMA) += dma.o +obj-$(CONFIG_S3C64XX_PL080) += pl080.o
# Device support
diff --git a/arch/arm/mach-s3c64xx/common.h b/arch/arm/mach-s3c64xx/common.h index 6cfc99b..60a667a 100644 --- a/arch/arm/mach-s3c64xx/common.h +++ b/arch/arm/mach-s3c64xx/common.h @@ -58,4 +58,9 @@ int __init s3c64xx_pm_late_initcall(void); static inline int s3c64xx_pm_late_initcall(void) { return 0; } #endif
+#ifdef CONFIG_S3C64XX_PL080 +extern struct pl08x_platform_data s3c64xx_dma0_plat_data; +extern struct pl08x_platform_data s3c64xx_dma1_plat_data; +#endif + #endif /* __ARCH_ARM_MACH_S3C64XX_COMMON_H */ diff --git a/arch/arm/mach-s3c64xx/include/mach/dma.h b/arch/arm/mach-s3c64xx/include/mach/dma.h index fe1a98c..6f88965 100644 --- a/arch/arm/mach-s3c64xx/include/mach/dma.h +++ b/arch/arm/mach-s3c64xx/include/mach/dma.h @@ -11,6 +11,8 @@ #ifndef __ASM_ARCH_DMA_H #define __ASM_ARCH_DMA_H __FILE__
+#ifdef CONFIG_S3C64XX_DMA + #define S3C_DMA_CHANNELS (16)
/* see mach-s3c2410/dma.h for notes on dma channel numbers */ @@ -128,4 +130,67 @@ struct s3c2410_dma_chan {
#include <plat/dma-core.h>
+#else + +#define S3C64XX_DMA_CHAN(name) ((unsigned long)(name)) + +/* DMA0/SDMA0 */ +#define DMACH_UART0 S3C64XX_DMA_CHAN("uart0_tx") +#define DMACH_UART0_SRC2 S3C64XX_DMA_CHAN("uart0_rx") +#define DMACH_UART1 S3C64XX_DMA_CHAN("uart1_tx") +#define DMACH_UART1_SRC2 S3C64XX_DMA_CHAN("uart1_rx") +#define DMACH_UART2 S3C64XX_DMA_CHAN("uart2_tx") +#define DMACH_UART2_SRC2 S3C64XX_DMA_CHAN("uart2_rx") +#define DMACH_UART3 S3C64XX_DMA_CHAN("uart3_tx") +#define DMACH_UART3_SRC2 S3C64XX_DMA_CHAN("uart3_rx") +#define DMACH_PCM0_TX S3C64XX_DMA_CHAN("pcm0_tx") +#define DMACH_PCM0_RX S3C64XX_DMA_CHAN("pcm0_rx") +#define DMACH_I2S0_OUT S3C64XX_DMA_CHAN("i2s0_tx") +#define DMACH_I2S0_IN S3C64XX_DMA_CHAN("i2s0_rx") +#define DMACH_SPI0_TX S3C64XX_DMA_CHAN("spi0_tx") +#define DMACH_SPI0_RX S3C64XX_DMA_CHAN("spi0_rx") +#define DMACH_HSI_I2SV40_TX S3C64XX_DMA_CHAN("i2s2_tx") +#define DMACH_HSI_I2SV40_RX S3C64XX_DMA_CHAN("i2s2_rx") + +/* DMA1/SDMA1 */ +#define DMACH_PCM1_TX S3C64XX_DMA_CHAN("pcm1_tx") +#define DMACH_PCM1_RX S3C64XX_DMA_CHAN("pcm1_rx") +#define DMACH_I2S1_OUT S3C64XX_DMA_CHAN("i2s1_tx") +#define DMACH_I2S1_IN S3C64XX_DMA_CHAN("i2s1_rx") +#define DMACH_SPI1_TX S3C64XX_DMA_CHAN("spi1_tx") +#define DMACH_SPI1_RX S3C64XX_DMA_CHAN("spi1_rx") +#define DMACH_AC97_PCMOUT S3C64XX_DMA_CHAN("ac97_out") +#define DMACH_AC97_PCMIN S3C64XX_DMA_CHAN("ac97_in") +#define DMACH_AC97_MICIN S3C64XX_DMA_CHAN("ac97_mic") +#define DMACH_PWM S3C64XX_DMA_CHAN("pwm") +#define DMACH_IRDA S3C64XX_DMA_CHAN("irda") +#define DMACH_EXTERNAL S3C64XX_DMA_CHAN("external") +#define DMACH_SECURITY_RX S3C64XX_DMA_CHAN("sec_rx") +#define DMACH_SECURITY_TX S3C64XX_DMA_CHAN("sec_tx") + +enum dma_ch { + DMACH_MAX = 32 +}; + +struct s3c2410_dma_client { + char *name; +}; + +static inline bool samsung_dma_has_circular(void) +{ + return true; +} + +static inline bool samsung_dma_is_dmadev(void) +{ + return true; +} + +#define pl330_filter pl08x_filter_id + +#include <linux/amba/pl08x.h> +#include <plat/dma-ops.h> + +#endif + #endif /* __ASM_ARCH_IRQ_H */ diff --git a/arch/arm/mach-s3c64xx/pl080.c b/arch/arm/mach-s3c64xx/pl080.c new file mode 100644 index 0000000..734b79f --- /dev/null +++ b/arch/arm/mach-s3c64xx/pl080.c @@ -0,0 +1,244 @@ +/* + * Samsung's S3C64XX generic DMA support using amba-pl08x driver. + * + * Copyright (c) 2013 Tomasz Figa tomasz.figa@gmail.com + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 as + * published by the Free Software Foundation. + */ + +#include <linux/kernel.h> +#include <linux/amba/bus.h> +#include <linux/amba/pl080.h> +#include <linux/amba/pl08x.h> +#include <linux/of.h> + +#include <mach/irqs.h> +#include <mach/map.h> + +#include "regs-sys.h" + +static int pl08x_get_signal(const struct pl08x_channel_data *cd) +{ + return cd->min_signal; +} + +static void pl08x_put_signal(const struct pl08x_channel_data *cd, int ch) +{ +} + +/* + * DMA0 + */ + +static struct pl08x_channel_data s3c64xx_dma0_info[] = { + { + .bus_id = "uart0_tx", + .min_signal = 0, + .max_signal = 0, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart0_rx", + .min_signal = 1, + .max_signal = 1, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart1_tx", + .min_signal = 2, + .max_signal = 2, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart1_rx", + .min_signal = 3, + .max_signal = 3, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart2_tx", + .min_signal = 4, + .max_signal = 4, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart2_rx", + .min_signal = 5, + .max_signal = 5, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart3_tx", + .min_signal = 6, + .max_signal = 6, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "uart3_rx", + .min_signal = 7, + .max_signal = 7, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "pcm0_tx", + .min_signal = 8, + .max_signal = 8, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "pcm0_rx", + .min_signal = 9, + .max_signal = 9, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s0_tx", + .min_signal = 10, + .max_signal = 10, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s0_rx", + .min_signal = 11, + .max_signal = 11, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "spi0_tx", + .min_signal = 12, + .max_signal = 12, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "spi0_rx", + .min_signal = 13, + .max_signal = 13, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s2_tx", + .min_signal = 14, + .max_signal = 14, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s2_rx", + .min_signal = 15, + .max_signal = 15, + .periph_buses = PL08X_AHB2, + } +}; + +struct pl08x_platform_data s3c64xx_dma0_plat_data = { + .memcpy_channel = { + .bus_id = "memcpy", + .cctl_memcpy = + (PL080_BSIZE_16 << PL080_CONTROL_SB_SIZE_SHIFT | \ + PL080_BSIZE_16 << PL080_CONTROL_DB_SIZE_SHIFT | \ + PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT | \ + PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT | \ + PL080_CONTROL_PROT_BUFF | PL080_CONTROL_PROT_CACHE | \ + PL080_CONTROL_PROT_SYS), + }, + .lli_buses = PL08X_AHB1, + .mem_buses = PL08X_AHB1, + .get_signal = pl08x_get_signal, + .put_signal = pl08x_put_signal, + .slave_channels = s3c64xx_dma0_info, + .num_slave_channels = ARRAY_SIZE(s3c64xx_dma0_info), +}; + +static AMBA_AHB_DEVICE(s3c64xx_dma0, "dma-pl080s.0", 0x0a141080, + 0x75000000, {IRQ_DMA0}, &s3c64xx_dma0_plat_data); + +/* + * DMA1 + */ + +static struct pl08x_channel_data s3c64xx_dma1_info[] = { + { + .bus_id = "pcm1_tx", + .min_signal = 0, + .max_signal = 0, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "pcm1_rx", + .min_signal = 1, + .max_signal = 1, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s1_tx", + .min_signal = 2, + .max_signal = 2, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "i2s1_rx", + .min_signal = 3, + .max_signal = 3, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "spi1_tx", + .min_signal = 4, + .max_signal = 4, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "spi1_rx", + .min_signal = 5, + .max_signal = 5, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "ac97_out", + .min_signal = 6, + .max_signal = 6, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "ac97_in", + .min_signal = 7, + .max_signal = 7, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "ac97_mic", + .min_signal = 8, + .max_signal = 8, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "pwm", + .min_signal = 9, + .max_signal = 9, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "irda", + .min_signal = 10, + .max_signal = 10, + .periph_buses = PL08X_AHB2, + }, { + .bus_id = "external", + .min_signal = 11, + .max_signal = 11, + .periph_buses = PL08X_AHB2, + }, +}; + +struct pl08x_platform_data s3c64xx_dma1_plat_data = { + .memcpy_channel = { + .bus_id = "memcpy", + .cctl_memcpy = + (PL080_BSIZE_16 << PL080_CONTROL_SB_SIZE_SHIFT | \ + PL080_BSIZE_16 << PL080_CONTROL_DB_SIZE_SHIFT | \ + PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT | \ + PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT | \ + PL080_CONTROL_PROT_BUFF | PL080_CONTROL_PROT_CACHE | \ + PL080_CONTROL_PROT_SYS), + }, + .lli_buses = PL08X_AHB1, + .mem_buses = PL08X_AHB1, + .get_signal = pl08x_get_signal, + .put_signal = pl08x_put_signal, + .slave_channels = s3c64xx_dma1_info, + .num_slave_channels = ARRAY_SIZE(s3c64xx_dma1_info), +}; + +static AMBA_AHB_DEVICE(s3c64xx_dma1, "dma-pl080s.1", 0x0a141080, + 0x75100000, {IRQ_DMA1}, &s3c64xx_dma1_plat_data); + +static int __init s3c64xx_pl080_init(void) +{ + /* Set all DMA configuration to be DMA, not SDMA */ + writel(0xffffff, S3C64XX_SDMA_SEL); + + if (of_have_populated_dt()) + return 0; + + amba_device_register(&s3c64xx_dma0_device, &iomem_resource); + amba_device_register(&s3c64xx_dma1_device, &iomem_resource); + + return 0; +} +arch_initcall(s3c64xx_pl080_init);
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
This patch adds all required platform-specific data and initialization code to support the generic amba-pl08x driver on S3C64xx SoCs.
Also some compatibility definitions are added to make the transition from legacy API to DMA engine easier. The biggest hack here is passing const char * pointers through DMA resource, casted to unsigned long, but this is how Samsung DMA wrappers (used to support both s3c-dma and DMA engine in drivers) is designed.
Signed-off-by: Tomasz Figa tomasz.figa@gmail.com
Patch looks like an OK stepping stone.
+static AMBA_AHB_DEVICE(s3c64xx_dma0, "dma-pl080s.0", 0x0a141080,
0x75000000, {IRQ_DMA0}, &s3c64xx_dma0_plat_data);
(...)
+static AMBA_AHB_DEVICE(s3c64xx_dma1, "dma-pl080s.1", 0x0a141080,
0x75100000, {IRQ_DMA1}, &s3c64xx_dma1_plat_data);
Ah, does this mean the device does not have any AMBA PrimeCell ID contents?
Then use the ID 0x00053080 instead of 0x0a141080
0x53 is "S" for Samsung. Variant 0.
Yours, Linus Walleij
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com wrote:
One of the biggest roadblocks on the way of S3C64xx to DeviceTree support is its DMA driver, which is completely platform-specific and provides private API (s3c-dma), not even saying that its design is completely against multiplatform-awareness.
The DMA controller present on this SoC series is a custom variant of ARM PrimeCell PL080 modified by Samsung to add some extra features. It is mostly compatible with original PL080, except:
- CH_CONTROL2 register is added between CH_CONTROL and CH_CONFIG,
- offset of CH_CONFIG register is different,
- transfer size field is moved from CH_CONTROL to CH_CONTROL2,
- transfer size field is extended to 24 bits, allowing much bigger single transfer,
- LLI consists of one more word, to account for CH_CONTROL2 register.
Since all the rest is fully compatible with standard PL080 there is no point in having separate driver just for this single variant, so I decided to look into adding support for it to the amba-pl08x driver.
There was already some attempt to achieve this before, but this was before Russel's big rework of the driver to use virtual channels, making the old patches being not much of use.
This RFC series is a proof of concept that I managed to make during last days of hacking. Except one patch adding clkdev lookup to clock driver (which is being replaced with a CCF-compliant driver ATM), this is enough to get memcpy and slave transfers to work on S3C64xx.
I have tested this on Mini6410 and SMDK6410 boards using dmatest for memcpy and Samsung I2S with madplay/aplay for slave transfers. Unfortunately I do not have access to other platforms with PL08x so I could not test for any regressions introduced on them.
Credits for two patches go to Alban Bedel, who made a series fixing this driver to make it usable with audio drivers. I rebased his patches on top of mine and corrected coding style a bit.
OK, that's all. Any comments are welcome. Feel free to start throwing eggs and tomatoes if you find this awful, but I won't be upset if I get some Tested-by or Acked-by as well. ;)
I've waited for a long time for this to happen, thank you very much for doing this.
Yours, Linus Walleij
On Monday 17 of June 2013 16:06:25 Linus Walleij wrote:
On Sun, Jun 16, 2013 at 10:54 PM, Tomasz Figa tomasz.figa@gmail.com
wrote:
One of the biggest roadblocks on the way of S3C64xx to DeviceTree support is its DMA driver, which is completely platform-specific and provides private API (s3c-dma), not even saying that its design is completely against multiplatform-awareness.
The DMA controller present on this SoC series is a custom variant of ARM PrimeCell PL080 modified by Samsung to add some extra features.
It is mostly compatible with original PL080, except:
CH_CONTROL2 register is added between CH_CONTROL and CH_CONFIG,
offset of CH_CONFIG register is different,
transfer size field is moved from CH_CONTROL to CH_CONTROL2,
transfer size field is extended to 24 bits, allowing much bigger
single transfer,
LLI consists of one more word, to account for CH_CONTROL2 register.
Since all the rest is fully compatible with standard PL080 there is no point in having separate driver just for this single variant, so I decided to look into adding support for it to the amba-pl08x driver.
There was already some attempt to achieve this before, but this was before Russel's big rework of the driver to use virtual channels, making the old patches being not much of use.
This RFC series is a proof of concept that I managed to make during last days of hacking. Except one patch adding clkdev lookup to clock driver (which is being replaced with a CCF-compliant driver ATM), this is enough to get memcpy and slave transfers to work on S3C64xx.
I have tested this on Mini6410 and SMDK6410 boards using dmatest for memcpy and Samsung I2S with madplay/aplay for slave transfers. Unfortunately I do not have access to other platforms with PL08x so I could not test for any regressions introduced on them.
Credits for two patches go to Alban Bedel, who made a series fixing this driver to make it usable with audio drivers. I rebased his patches on top of mine and corrected coding style a bit.
OK, that's all. Any comments are welcome. Feel free to start throwing eggs and tomatoes if you find this awful, but I won't be upset if I get some Tested-by or Acked-by as well. ;)
I've waited for a long time for this to happen, thank you very much for doing this.
You're welcome. Thanks for your review.
Best regards, Tomasz
On Sun, Jun 16, 2013 at 10:54:07PM +0200, Tomasz Figa wrote:
One of the biggest roadblocks on the way of S3C64xx to DeviceTree support is its DMA driver, which is completely platform-specific and provides private API (s3c-dma), not even saying that its design is completely against multiplatform-awareness.
I tried to test this on my s3c64xx based system but it gave me a kernel that didn't boot far enough to give console output (there's some early init stuff that uses SPI...). That said, I needed:
diff --git a/drivers/dma/amba-pl08x.c b/drivers/dma/amba-pl08x.c index 210a893..0f49707 100644 --- a/drivers/dma/amba-pl08x.c +++ b/drivers/dma/amba-pl08x.c @@ -313,7 +313,7 @@ static int pl08x_request_mux(struct pl08x_dma_chan *plchan) int ret;
if (plchan->mux_use++ == 0 && pd->get_signal) { - ret = pd->get_signal(plchan->cd); + ret = (pd->get_signal)(plchan->cd); if (ret < 0) { plchan->mux_use = 0; return ret;
to get it to build which makes me suspect the compiler a bit as well... the system has audio, SPI and MMC enabled.
I was applying this to -next, are there any other dependencies I need or anything?
participants (5)
-
Arnd Bergmann
-
Linus Walleij
-
Mark Brown
-
Russell King - ARM Linux
-
Tomasz Figa