diff mbox series

[V3,2/2] dmaengine: fsl-dpaa2-qdma: Add NXP dpaa2 qDMA controller driver for Layerscape SoCs

Message ID 20190409072212.15860-2-peng.ma@nxp.com (mailing list archive)
State Changes Requested
Headers show
Series [V3,1/2] dmaengine: fsl-dpaa2-qdma: Add the DPDMAI(Data Path DMA Interface) support | expand

Commit Message

Peng Ma April 9, 2019, 7:22 a.m. UTC 
DPPA2(Data Path Acceleration Architecture 2) qDMA
The qDMA supports channel virtualization by allowing DMA jobs to be enqueued
into different frame queues. Core can initiate a DMA transaction by preparing
a frame descriptor(FD) for each DMA job and enqueuing this job to a frame queue.
through a hardware portal. The qDMA prefetches DMA jobs from the frame queues.
It then schedules and dispatches to internal DMA hardware engines, which
generate read and write requests. Both qDMA source data and destination data can
be either contiguous or non-contiguous using one or more scatter/gather tables.
The qDMA supports global bandwidth flow control where all DMA transactions are
stalled if the bandwidth threshold has been reached. Also supported are
transaction based read throttling.

Add NXP dppa2 qDMA to support some of Layerscape SoCs.
such as: LS1088A, LS208xA, LX2, etc.

Signed-off-by: Peng Ma <peng.ma@nxp.com>
---
changed for v3:
	- Add depends on arm64 for dpaa2 qdma driver 
	- The dpaa2_io_service_[de]register functions have a new parameter
	So update all calls to some functions

 drivers/dma/Kconfig                     |    2 +
 drivers/dma/Makefile                    |    1 +
 drivers/dma/fsl-dpaa2-qdma/Kconfig      |    9 +
 drivers/dma/fsl-dpaa2-qdma/Makefile     |    3 +
 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c |  782 +++++++++++++++++++++++++++++++
 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h |  152 ++++++
 6 files changed, 949 insertions(+), 0 deletions(-)
 create mode 100644 drivers/dma/fsl-dpaa2-qdma/Kconfig
 create mode 100644 drivers/dma/fsl-dpaa2-qdma/Makefile
 create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
 create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h

Comments

Vinod Koul April 29, 2019, 5:32 a.m. UTC | #1 
On 09-04-19, 15:22, Peng Ma wrote:
> DPPA2(Data Path Acceleration Architecture 2) qDMA
> The qDMA supports channel virtualization by allowing DMA jobs to be enqueued
> into different frame queues. Core can initiate a DMA transaction by preparing
> a frame descriptor(FD) for each DMA job and enqueuing this job to a frame queue.
> through a hardware portal. The qDMA prefetches DMA jobs from the frame queues.
> It then schedules and dispatches to internal DMA hardware engines, which
> generate read and write requests. Both qDMA source data and destination data can
> be either contiguous or non-contiguous using one or more scatter/gather tables.
> The qDMA supports global bandwidth flow control where all DMA transactions are
> stalled if the bandwidth threshold has been reached. Also supported are
> transaction based read throttling.
> 
> Add NXP dppa2 qDMA to support some of Layerscape SoCs.
> such as: LS1088A, LS208xA, LX2, etc.
> 
> Signed-off-by: Peng Ma <peng.ma@nxp.com>
> ---
> changed for v3:
> 	- Add depends on arm64 for dpaa2 qdma driver 
> 	- The dpaa2_io_service_[de]register functions have a new parameter
> 	So update all calls to some functions
> 
>  drivers/dma/Kconfig                     |    2 +
>  drivers/dma/Makefile                    |    1 +
>  drivers/dma/fsl-dpaa2-qdma/Kconfig      |    9 +
>  drivers/dma/fsl-dpaa2-qdma/Makefile     |    3 +
>  drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c |  782 +++++++++++++++++++++++++++++++
>  drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h |  152 ++++++
>  6 files changed, 949 insertions(+), 0 deletions(-)
>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/Kconfig
>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/Makefile
>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h
> 
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index eaf78f4..08aae01 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -671,6 +671,8 @@ source "drivers/dma/sh/Kconfig"
>  
>  source "drivers/dma/ti/Kconfig"
>  
> +source "drivers/dma/fsl-dpaa2-qdma/Kconfig"
> +
>  # clients
>  comment "DMA Clients"
>  	depends on DMA_ENGINE
> diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
> index 6126e1c..2499ed8 100644
> --- a/drivers/dma/Makefile
> +++ b/drivers/dma/Makefile
> @@ -75,6 +75,7 @@ obj-$(CONFIG_UNIPHIER_MDMAC) += uniphier-mdmac.o
>  obj-$(CONFIG_XGENE_DMA) += xgene-dma.o
>  obj-$(CONFIG_ZX_DMA) += zx_dma.o
>  obj-$(CONFIG_ST_FDMA) += st_fdma.o
> +obj-$(CONFIG_FSL_DPAA2_QDMA) += fsl-dpaa2-qdma/
>  
>  obj-y += mediatek/
>  obj-y += qcom/
> diff --git a/drivers/dma/fsl-dpaa2-qdma/Kconfig b/drivers/dma/fsl-dpaa2-qdma/Kconfig
> new file mode 100644
> index 0000000..258ed6b
> --- /dev/null
> +++ b/drivers/dma/fsl-dpaa2-qdma/Kconfig
> @@ -0,0 +1,9 @@
> +menuconfig FSL_DPAA2_QDMA
> +	tristate "NXP DPAA2 QDMA"
> +	depends on ARM64
> +	depends on FSL_MC_BUS && FSL_MC_DPIO
> +	select DMA_ENGINE
> +	select DMA_VIRTUAL_CHANNELS
> +	help
> +	  NXP Data Path Acceleration Architecture 2 QDMA driver,
> +	  using the NXP MC bus driver.
> diff --git a/drivers/dma/fsl-dpaa2-qdma/Makefile b/drivers/dma/fsl-dpaa2-qdma/Makefile
> new file mode 100644
> index 0000000..c1d0226
> --- /dev/null
> +++ b/drivers/dma/fsl-dpaa2-qdma/Makefile
> @@ -0,0 +1,3 @@
> +# SPDX-License-Identifier: GPL-2.0
> +# Makefile for the NXP DPAA2 qDMA controllers
> +obj-$(CONFIG_FSL_DPAA2_QDMA) += dpaa2-qdma.o dpdmai.o
> diff --git a/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
> new file mode 100644
> index 0000000..0cdde0f
> --- /dev/null
> +++ b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
> @@ -0,0 +1,782 @@
> +// SPDX-License-Identifier: GPL-2.0
> +// Copyright 2014-2018 NXP
> +
> +/*
> + * Author: Changming Huang <jerry.huang@nxp.com>
> + *
> + * Driver for the NXP QDMA engine with QMan mode.
> + * Channel virtualization is supported through enqueuing of DMA jobs to,
> + * or dequeuing DMA jobs from different work queues with QMan portal.
> + * This module can be found on NXP LS2 SoCs.
> + *
> + */
> +
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/dmapool.h>
> +#include <linux/of_irq.h>
> +#include <linux/iommu.h>
> +#include <linux/sys_soc.h>
> +#include <linux/fsl/mc.h>
> +#include <soc/fsl/dpaa2-io.h>
> +
> +#include "../virt-dma.h"
> +#include "dpdmai_cmd.h"
> +#include "dpdmai.h"
> +#include "dpaa2-qdma.h"
> +
> +static bool smmu_disable = true;
> +
> +static struct dpaa2_qdma_chan *to_dpaa2_qdma_chan(struct dma_chan *chan)
> +{
> +	return container_of(chan, struct dpaa2_qdma_chan, vchan.chan);
> +}
> +
> +static struct dpaa2_qdma_comp *to_fsl_qdma_comp(struct virt_dma_desc *vd)
> +{
> +	return container_of(vd, struct dpaa2_qdma_comp, vdesc);
> +}
> +
> +static int dpaa2_qdma_alloc_chan_resources(struct dma_chan *chan)
> +{
> +	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
> +	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
> +	struct device *dev = &dpaa2_qdma->priv->dpdmai_dev->dev;
> +
> +	dpaa2_chan->fd_pool = dma_pool_create("fd_pool", dev,
> +					      FD_POOL_SIZE, 32, 0);
> +	if (!dpaa2_chan->fd_pool)
> +		return -ENOMEM;
> +
> +	return dpaa2_qdma->desc_allocated++;
> +}
> +
> +static void dpaa2_qdma_free_chan_resources(struct dma_chan *chan)
> +{
> +	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
> +	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
> +	unsigned long flags;
> +
> +	LIST_HEAD(head);
> +
> +	spin_lock_irqsave(&dpaa2_chan->vchan.lock, flags);
> +	vchan_get_all_descriptors(&dpaa2_chan->vchan, &head);
> +	spin_unlock_irqrestore(&dpaa2_chan->vchan.lock, flags);
> +
> +	vchan_dma_desc_free_list(&dpaa2_chan->vchan, &head);
> +
> +	dpaa2_dpdmai_free_comp(dpaa2_chan, &dpaa2_chan->comp_used);
> +	dpaa2_dpdmai_free_comp(dpaa2_chan, &dpaa2_chan->comp_free);
> +
> +	dma_pool_destroy(dpaa2_chan->fd_pool);
> +	dpaa2_qdma->desc_allocated--;
> +}
> +
> +/*
> + * Request a command descriptor for enqueue.
> + */
> +static struct dpaa2_qdma_comp *
> +dpaa2_qdma_request_desc(struct dpaa2_qdma_chan *dpaa2_chan)
> +{
> +	struct dpaa2_qdma_comp *comp_temp = NULL;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
> +	if (list_empty(&dpaa2_chan->comp_free)) {
> +		spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
> +		comp_temp = kzalloc(sizeof(*comp_temp), GFP_KERNEL);

GFP_NOWAIT?

> +		if (!comp_temp)
> +			goto err;
> +		comp_temp->fd_virt_addr =
> +			dma_pool_alloc(dpaa2_chan->fd_pool, GFP_NOWAIT,
> +				       &comp_temp->fd_bus_addr);
> +		if (!comp_temp->fd_virt_addr)

err handling seems incorrect, you dont clean up, caller doesnt check
return!

> +			goto err;
> +
> +		comp_temp->fl_virt_addr =
> +			(void *)((struct dpaa2_fd *)
> +				comp_temp->fd_virt_addr + 1);

casts and pointer math, what could go wrong!! This doesnt smell right!

> +		comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
> +					sizeof(struct dpaa2_fd);

why not use fl_virt_addr and get the bus_address?

> +		comp_temp->desc_virt_addr =
> +			(void *)((struct dpaa2_fl_entry *)
> +				comp_temp->fl_virt_addr + 3);
> +		comp_temp->desc_bus_addr = comp_temp->fl_bus_addr +
> +				sizeof(struct dpaa2_fl_entry) * 3;

pointer math in the two calls doesnt match and as I said doesnt look
good...

> +
> +		comp_temp->qchan = dpaa2_chan;
> +		return comp_temp;
> +	}
> +	comp_temp = list_first_entry(&dpaa2_chan->comp_free,
> +				     struct dpaa2_qdma_comp, list);
> +	list_del(&comp_temp->list);
> +	spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
> +
> +	comp_temp->qchan = dpaa2_chan;
> +err:
> +	return comp_temp;
> +}
> +
> +static void
> +dpaa2_qdma_populate_fd(u32 format, struct dpaa2_qdma_comp *dpaa2_comp)
> +{
> +	struct dpaa2_fd *fd;
> +
> +	fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;

whats with the casts! you seem to like them! You are casting away from
void!

> +	memset(fd, 0, sizeof(struct dpaa2_fd));
> +
> +	/* fd populated */
> +	dpaa2_fd_set_addr(fd, dpaa2_comp->fl_bus_addr);
> +	/* Bypass memory translation, Frame list format, short length disable */
> +	/* we need to disable BMT if fsl-mc use iova addr */
> +	if (smmu_disable)
> +		dpaa2_fd_set_bpid(fd, QMAN_FD_BMT_ENABLE);
> +	dpaa2_fd_set_format(fd, QMAN_FD_FMT_ENABLE | QMAN_FD_SL_DISABLE);
> +
> +	dpaa2_fd_set_frc(fd, format | QDMA_SER_CTX);
> +}
> +
> +/* first frame list for descriptor buffer */
> +static void
> +dpaa2_qdma_populate_first_framel(struct dpaa2_fl_entry *f_list,
> +				 struct dpaa2_qdma_comp *dpaa2_comp,
> +				 bool wrt_changed)
> +{
> +	struct dpaa2_qdma_sd_d *sdd;
> +
> +	sdd = (struct dpaa2_qdma_sd_d *)dpaa2_comp->desc_virt_addr;

again

> +	memset(sdd, 0, 2 * (sizeof(*sdd)));
> +
> +	/* source descriptor CMD */
> +	sdd->cmd = cpu_to_le32(QDMA_SD_CMD_RDTTYPE_COHERENT);
> +	sdd++;
> +
> +	/* dest descriptor CMD */
> +	if (wrt_changed)
> +		sdd->cmd = cpu_to_le32(LX2160_QDMA_DD_CMD_WRTTYPE_COHERENT);
> +	else
> +		sdd->cmd = cpu_to_le32(QDMA_DD_CMD_WRTTYPE_COHERENT);
> +
> +	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
> +
> +	/* first frame list to source descriptor */
> +	dpaa2_fl_set_addr(f_list, dpaa2_comp->desc_bus_addr);
> +	dpaa2_fl_set_len(f_list, 0x20);
> +	dpaa2_fl_set_format(f_list, QDMA_FL_FMT_SBF | QDMA_FL_SL_LONG);
> +
> +	/* bypass memory translation */
> +	if (smmu_disable)
> +		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
> +}
> +
> +/* source and destination frame list */
> +static void
> +dpaa2_qdma_populate_frames(struct dpaa2_fl_entry *f_list,
> +			   dma_addr_t dst, dma_addr_t src,
> +			   size_t len, uint8_t fmt)
> +{
> +	/* source frame list to source buffer */
> +	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
> +
> +	dpaa2_fl_set_addr(f_list, src);
> +	dpaa2_fl_set_len(f_list, len);
> +
> +	/* single buffer frame or scatter gather frame */
> +	dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
> +
> +	/* bypass memory translation */
> +	if (smmu_disable)
> +		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
> +
> +	f_list++;
> +
> +	/* destination frame list to destination buffer */
> +	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
> +
> +	dpaa2_fl_set_addr(f_list, dst);
> +	dpaa2_fl_set_len(f_list, len);
> +	dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
> +	/* single buffer frame or scatter gather frame */
> +	dpaa2_fl_set_final(f_list, QDMA_FL_F);
> +	/* bypass memory translation */
> +	if (smmu_disable)
> +		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
> +}
> +
> +static struct dma_async_tx_descriptor
> +*dpaa2_qdma_prep_memcpy(struct dma_chan *chan, dma_addr_t dst,
> +			dma_addr_t src, size_t len, ulong flags)
> +{
> +	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
> +	struct dpaa2_qdma_engine *dpaa2_qdma;
> +	struct dpaa2_qdma_comp *dpaa2_comp;
> +	struct dpaa2_fl_entry *f_list;
> +	bool wrt_changed;
> +	u32 format;
> +
> +	dpaa2_qdma = dpaa2_chan->qdma;
> +	dpaa2_comp = dpaa2_qdma_request_desc(dpaa2_chan);
> +	wrt_changed = (bool)dpaa2_qdma->qdma_wrtype_fixup;
> +
> +#ifdef LONG_FORMAT

 compile flag and define, so else part is dead code??

> +	format = QDMA_FD_LONG_FORMAT;
> +#else
> +	format = QDMA_FD_SHORT_FORMAT;
> +#endif
> +	/* populate Frame descriptor */
> +	dpaa2_qdma_populate_fd(format, dpaa2_comp);
> +
> +	f_list = (struct dpaa2_fl_entry *)dpaa2_comp->fl_virt_addr;
> +
> +#ifdef LONG_FORMAT
> +	/* first frame list for descriptor buffer (logn format) */
> +	dpaa2_qdma_populate_first_framel(f_list, dpaa2_comp, wrt_changed);
> +
> +	f_list++;
> +#endif
> +
> +	dpaa2_qdma_populate_frames(f_list, dst, src, len, QDMA_FL_FMT_SBF);
> +
> +	return vchan_tx_prep(&dpaa2_chan->vchan, &dpaa2_comp->vdesc, flags);
> +}
> +
> +static enum
> +dma_status dpaa2_qdma_tx_status(struct dma_chan *chan,
> +				dma_cookie_t cookie,
> +				struct dma_tx_state *txstate)
> +{
> +	return dma_cookie_status(chan, cookie, txstate);
> +}
> +
> +static void dpaa2_qdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
> +	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
> +	struct dpaa2_qdma_priv *priv = dpaa2_qdma->priv;
> +	struct dpaa2_qdma_comp *dpaa2_comp;
> +	struct virt_dma_desc *vdesc;
> +	struct dpaa2_fd *fd;
> +	unsigned long flags;
> +	int err;
> +
> +	spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
> +	spin_lock(&dpaa2_chan->vchan.lock);
> +	if (vchan_issue_pending(&dpaa2_chan->vchan)) {
> +		vdesc = vchan_next_desc(&dpaa2_chan->vchan);
> +		if (!vdesc)
> +			goto err_enqueue;
> +		dpaa2_comp = to_fsl_qdma_comp(vdesc);
> +
> +		fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;
> +
> +		list_del(&vdesc->node);
> +		list_add_tail(&dpaa2_comp->list, &dpaa2_chan->comp_used);

what does this list do?

> +
> +		/* TOBO: priority hard-coded to zero */

You mean TODO?

> +		err = dpaa2_io_service_enqueue_fq(NULL,
> +				priv->tx_queue_attr[0].fqid, fd);
> +		if (err) {
> +			list_del(&dpaa2_comp->list);
> +			list_add_tail(&dpaa2_comp->list,
> +				      &dpaa2_chan->comp_free);
> +		}
> +	}
> +err_enqueue:
> +	spin_unlock(&dpaa2_chan->vchan.lock);
> +	spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
> +}
> +
> +static int __cold dpaa2_qdma_setup(struct fsl_mc_device *ls_dev)
> +{
> +	struct dpaa2_qdma_priv_per_prio *ppriv;
> +	struct device *dev = &ls_dev->dev;
> +	struct dpaa2_qdma_priv *priv;
> +	u8 prio_def = DPDMAI_PRIO_NUM;
> +	int err;
> +	int i;
> +
> +	priv = dev_get_drvdata(dev);
> +
> +	priv->dev = dev;
> +	priv->dpqdma_id = ls_dev->obj_desc.id;
> +
> +	/*Get the handle for the DPDMAI this interface is associate with */

Please run checkpatch, it should have told you that you need space after
comment marker /* foo...

> +	err = dpdmai_open(priv->mc_io, 0, priv->dpqdma_id, &ls_dev->mc_handle);
> +	if (err) {
> +		dev_err(dev, "dpdmai_open() failed\n");
> +		return err;
> +	}
> +	dev_info(dev, "Opened dpdmai object successfully\n");
> +
> +	err = dpdmai_get_attributes(priv->mc_io, 0, ls_dev->mc_handle,
> +				    &priv->dpdmai_attr);
> +	if (err) {
> +		dev_err(dev, "dpdmai_get_attributes() failed\n");
> +		return err;

so you dont close what you opened in dpdmai_open() Please give a serious
thought and testing to this driver

> +	}
> +
> +	if (priv->dpdmai_attr.version.major > DPDMAI_VER_MAJOR) {
> +		dev_err(dev, "DPDMAI major version mismatch\n"
> +			     "Found %u.%u, supported version is %u.%u\n",
> +				priv->dpdmai_attr.version.major,
> +				priv->dpdmai_attr.version.minor,
> +				DPDMAI_VER_MAJOR, DPDMAI_VER_MINOR);
> +	}
> +
> +	if (priv->dpdmai_attr.version.minor > DPDMAI_VER_MINOR) {
> +		dev_err(dev, "DPDMAI minor version mismatch\n"
> +			     "Found %u.%u, supported version is %u.%u\n",
> +				priv->dpdmai_attr.version.major,
> +				priv->dpdmai_attr.version.minor,
> +				DPDMAI_VER_MAJOR, DPDMAI_VER_MINOR);

what is the implication of these error, why not bail out on these?

> +	}
> +
> +	priv->num_pairs = min(priv->dpdmai_attr.num_of_priorities, prio_def);
> +	ppriv = kcalloc(priv->num_pairs, sizeof(*ppriv), GFP_KERNEL);

what is the context of the fn, sleepy, atomic?

> +	if (!ppriv) {
> +		dev_err(dev, "kzalloc for ppriv failed\n");

this need not be logged, core will do so

> +		return -1;

really -1??

I think this driver needs more work, please fix these issues in the
comments above and also see in rest of the code
Peng Ma June 10, 2019, 9:51 a.m. UTC | #2 
Hi, Vinod,

Please see my comments inline, thanks very much.

Best Regards,
Peng

>-----Original Message-----
>From: Vinod Koul <vkoul@kernel.org>
>Sent: 2019年4月29日 13:32
>To: Peng Ma <peng.ma@nxp.com>
>Cc: dan.j.williams@intel.com; Leo Li <leoyang.li@nxp.com>;
>linux-kernel@vger.kernel.org; dmaengine@vger.kernel.org
>Subject: [EXT] Re: [V3 2/2] dmaengine: fsl-dpaa2-qdma: Add NXP dpaa2 qDMA
>controller driver for Layerscape SoCs
>
>Caution: EXT Email
>
>On 09-04-19, 15:22, Peng Ma wrote:
>> DPPA2(Data Path Acceleration Architecture 2) qDMA The qDMA supports
>> channel virtualization by allowing DMA jobs to be enqueued into
>> different frame queues. Core can initiate a DMA transaction by
>> preparing a frame descriptor(FD) for each DMA job and enqueuing this job to
>a frame queue.
>> through a hardware portal. The qDMA prefetches DMA jobs from the frame
>queues.
>> It then schedules and dispatches to internal DMA hardware engines,
>> which generate read and write requests. Both qDMA source data and
>> destination data can be either contiguous or non-contiguous using one or
>more scatter/gather tables.
>> The qDMA supports global bandwidth flow control where all DMA
>> transactions are stalled if the bandwidth threshold has been reached.
>> Also supported are transaction based read throttling.
>>
>> Add NXP dppa2 qDMA to support some of Layerscape SoCs.
>> such as: LS1088A, LS208xA, LX2, etc.
>>
>> Signed-off-by: Peng Ma <peng.ma@nxp.com>
>> ---
>> changed for v3:
>>       - Add depends on arm64 for dpaa2 qdma driver
>>       - The dpaa2_io_service_[de]register functions have a new
>parameter
>>       So update all calls to some functions
>>
>>  drivers/dma/Kconfig                     |    2 +
>>  drivers/dma/Makefile                    |    1 +
>>  drivers/dma/fsl-dpaa2-qdma/Kconfig      |    9 +
>>  drivers/dma/fsl-dpaa2-qdma/Makefile     |    3 +
>>  drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c |  782
>> +++++++++++++++++++++++++++++++
>> drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h |  152 ++++++
>>  6 files changed, 949 insertions(+), 0 deletions(-)  create mode
>> 100644 drivers/dma/fsl-dpaa2-qdma/Kconfig
>>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/Makefile
>>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
>>  create mode 100644 drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h
>>
>> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig index
>> eaf78f4..08aae01 100644
>> --- a/drivers/dma/Kconfig
>> +++ b/drivers/dma/Kconfig
>> @@ -671,6 +671,8 @@ source "drivers/dma/sh/Kconfig"
>>
>>  source "drivers/dma/ti/Kconfig"
>>
>> +source "drivers/dma/fsl-dpaa2-qdma/Kconfig"
>> +
>>  # clients
>>  comment "DMA Clients"
>>       depends on DMA_ENGINE
>> diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile index
>> 6126e1c..2499ed8 100644
>> --- a/drivers/dma/Makefile
>> +++ b/drivers/dma/Makefile
>> @@ -75,6 +75,7 @@ obj-$(CONFIG_UNIPHIER_MDMAC) +=
>uniphier-mdmac.o
>>  obj-$(CONFIG_XGENE_DMA) += xgene-dma.o
>>  obj-$(CONFIG_ZX_DMA) += zx_dma.o
>>  obj-$(CONFIG_ST_FDMA) += st_fdma.o
>> +obj-$(CONFIG_FSL_DPAA2_QDMA) += fsl-dpaa2-qdma/
>>
>>  obj-y += mediatek/
>>  obj-y += qcom/
>> diff --git a/drivers/dma/fsl-dpaa2-qdma/Kconfig
>> b/drivers/dma/fsl-dpaa2-qdma/Kconfig
>> new file mode 100644
>> index 0000000..258ed6b
>> --- /dev/null
>> +++ b/drivers/dma/fsl-dpaa2-qdma/Kconfig
>> @@ -0,0 +1,9 @@
>> +menuconfig FSL_DPAA2_QDMA
>> +     tristate "NXP DPAA2 QDMA"
>> +     depends on ARM64
>> +     depends on FSL_MC_BUS && FSL_MC_DPIO
>> +     select DMA_ENGINE
>> +     select DMA_VIRTUAL_CHANNELS
>> +     help
>> +       NXP Data Path Acceleration Architecture 2 QDMA driver,
>> +       using the NXP MC bus driver.
>> diff --git a/drivers/dma/fsl-dpaa2-qdma/Makefile
>> b/drivers/dma/fsl-dpaa2-qdma/Makefile
>> new file mode 100644
>> index 0000000..c1d0226
>> --- /dev/null
>> +++ b/drivers/dma/fsl-dpaa2-qdma/Makefile
>> @@ -0,0 +1,3 @@
>> +# SPDX-License-Identifier: GPL-2.0
>> +# Makefile for the NXP DPAA2 qDMA controllers
>> +obj-$(CONFIG_FSL_DPAA2_QDMA) += dpaa2-qdma.o dpdmai.o
>> diff --git a/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
>> b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
>> new file mode 100644
>> index 0000000..0cdde0f
>> --- /dev/null
>> +++ b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
>> @@ -0,0 +1,782 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +// Copyright 2014-2018 NXP
>> +
>> +/*
>> + * Author: Changming Huang <jerry.huang@nxp.com>
>> + *
>> + * Driver for the NXP QDMA engine with QMan mode.
>> + * Channel virtualization is supported through enqueuing of DMA jobs
>> +to,
>> + * or dequeuing DMA jobs from different work queues with QMan portal.
>> + * This module can be found on NXP LS2 SoCs.
>> + *
>> + */
>> +
>> +#include <linux/init.h>
>> +#include <linux/module.h>
>> +#include <linux/dmapool.h>
>> +#include <linux/of_irq.h>
>> +#include <linux/iommu.h>
>> +#include <linux/sys_soc.h>
>> +#include <linux/fsl/mc.h>
>> +#include <soc/fsl/dpaa2-io.h>
>> +
>> +#include "../virt-dma.h"
>> +#include "dpdmai_cmd.h"
>> +#include "dpdmai.h"
>> +#include "dpaa2-qdma.h"
>> +
>> +static bool smmu_disable = true;
>> +
>> +static struct dpaa2_qdma_chan *to_dpaa2_qdma_chan(struct dma_chan
>> +*chan) {
>> +     return container_of(chan, struct dpaa2_qdma_chan, vchan.chan); }
>> +
>> +static struct dpaa2_qdma_comp *to_fsl_qdma_comp(struct virt_dma_desc
>> +*vd) {
>> +     return container_of(vd, struct dpaa2_qdma_comp, vdesc); }
>> +
>> +static int dpaa2_qdma_alloc_chan_resources(struct dma_chan *chan) {
>> +     struct dpaa2_qdma_chan *dpaa2_chan =
>to_dpaa2_qdma_chan(chan);
>> +     struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
>> +     struct device *dev = &dpaa2_qdma->priv->dpdmai_dev->dev;
>> +
>> +     dpaa2_chan->fd_pool = dma_pool_create("fd_pool", dev,
>> +                                           FD_POOL_SIZE, 32,
>0);
>> +     if (!dpaa2_chan->fd_pool)
>> +             return -ENOMEM;
>> +
>> +     return dpaa2_qdma->desc_allocated++; }
>> +
>> +static void dpaa2_qdma_free_chan_resources(struct dma_chan *chan) {
>> +     struct dpaa2_qdma_chan *dpaa2_chan =
>to_dpaa2_qdma_chan(chan);
>> +     struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
>> +     unsigned long flags;
>> +
>> +     LIST_HEAD(head);
>> +
>> +     spin_lock_irqsave(&dpaa2_chan->vchan.lock, flags);
>> +     vchan_get_all_descriptors(&dpaa2_chan->vchan, &head);
>> +     spin_unlock_irqrestore(&dpaa2_chan->vchan.lock, flags);
>> +
>> +     vchan_dma_desc_free_list(&dpaa2_chan->vchan, &head);
>> +
>> +     dpaa2_dpdmai_free_comp(dpaa2_chan,
>&dpaa2_chan->comp_used);
>> +     dpaa2_dpdmai_free_comp(dpaa2_chan, &dpaa2_chan->comp_free);
>> +
>> +     dma_pool_destroy(dpaa2_chan->fd_pool);
>> +     dpaa2_qdma->desc_allocated--;
>> +}
>> +
>> +/*
>> + * Request a command descriptor for enqueue.
>> + */
>> +static struct dpaa2_qdma_comp *
>> +dpaa2_qdma_request_desc(struct dpaa2_qdma_chan *dpaa2_chan) {
>> +     struct dpaa2_qdma_comp *comp_temp = NULL;
>> +     unsigned long flags;
>> +
>> +     spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
>> +     if (list_empty(&dpaa2_chan->comp_free)) {
>> +             spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
>> +             comp_temp = kzalloc(sizeof(*comp_temp), GFP_KERNEL);
>
>GFP_NOWAIT?
[Peng Ma] Got it.
>
>> +             if (!comp_temp)
>> +                     goto err;
>> +             comp_temp->fd_virt_addr =
>> +                     dma_pool_alloc(dpaa2_chan->fd_pool,
>GFP_NOWAIT,
>> +                                    &comp_temp->fd_bus_addr);
>> +             if (!comp_temp->fd_virt_addr)
>
>err handling seems incorrect, you dont clean up, caller doesnt check return!
>
[Peng Ma] Yes, It's my fault.
>> +                     goto err;
>> +
>> +             comp_temp->fl_virt_addr =
>> +                     (void *)((struct dpaa2_fd *)
>> +                             comp_temp->fd_virt_addr + 1);
>
>casts and pointer math, what could go wrong!! This doesnt smell right!
>
>> +             comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
>> +                                     sizeof(struct dpaa2_fd);
>
>why not use fl_virt_addr and get the bus_address?
What you mean is I should use virt_to_phys to get the bus_address?
>
>> +             comp_temp->desc_virt_addr =
>> +                     (void *)((struct dpaa2_fl_entry *)
>> +                             comp_temp->fl_virt_addr + 3);
>> +             comp_temp->desc_bus_addr = comp_temp->fl_bus_addr +
>> +                             sizeof(struct dpaa2_fl_entry) * 3;
>
>pointer math in the two calls doesnt match and as I said doesnt look good...
Should I do this as follows:
-               comp_temp->fl_virt_addr =
-                       (void *)((struct dpaa2_fd *)
-                               comp_temp->fd_virt_addr + 1);
-               comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
-                                       sizeof(struct dpaa2_fd);
-               comp_temp->desc_virt_addr =
-                       (void *)((struct dpaa2_fl_entry *)
-                               comp_temp->fl_virt_addr + 3);
-               comp_temp->desc_bus_addr = comp_temp->fl_bus_addr +
-                               sizeof(struct dpaa2_fl_entry) * 3;
+               comp_temp->fd_virt_addr = (struct dpaa2_fd *)virt_addr_head++;
+               comp_temp->fl_virt_addr = (struct dpaa2_fl_entry *)virt_addr_head;
+               comp_temp->fl_bus_addr = virt_to_phys(comp_temp->fl_virt_addr);
+               virt_addr_head = (struct dpaa2_fl_entry *)virt_addr_head + 3;
+               comp_temp->desc_virt_addr = (struct dpaa2_qdma_sd_d *)virt_addr_head;
+               comp_temp->desc_bus_addr = virt_to_phys(comp_temp->desc_virt_addr);
>
>> +
>> +             comp_temp->qchan = dpaa2_chan;
>> +             return comp_temp;
>> +     }
>> +     comp_temp = list_first_entry(&dpaa2_chan->comp_free,
>> +                                  struct dpaa2_qdma_comp, list);
>> +     list_del(&comp_temp->list);
>> +     spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
>> +
>> +     comp_temp->qchan = dpaa2_chan;
>> +err:
>> +     return comp_temp;
>> +}
>> +
>> +static void
>> +dpaa2_qdma_populate_fd(u32 format, struct dpaa2_qdma_comp
>> +*dpaa2_comp) {
>> +     struct dpaa2_fd *fd;
>> +
>> +     fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;
>
>whats with the casts! you seem to like them! You are casting away from void!
This will avoid after change fd_virt_addr type.
>
>> +     memset(fd, 0, sizeof(struct dpaa2_fd));
>> +
>> +     /* fd populated */
>> +     dpaa2_fd_set_addr(fd, dpaa2_comp->fl_bus_addr);
>> +     /* Bypass memory translation, Frame list format, short length disable
>*/
>> +     /* we need to disable BMT if fsl-mc use iova addr */
>> +     if (smmu_disable)
>> +             dpaa2_fd_set_bpid(fd, QMAN_FD_BMT_ENABLE);
>> +     dpaa2_fd_set_format(fd, QMAN_FD_FMT_ENABLE |
>> + QMAN_FD_SL_DISABLE);
>> +
>> +     dpaa2_fd_set_frc(fd, format | QDMA_SER_CTX); }
>> +
>> +/* first frame list for descriptor buffer */ static void
>> +dpaa2_qdma_populate_first_framel(struct dpaa2_fl_entry *f_list,
>> +                              struct dpaa2_qdma_comp
>*dpaa2_comp,
>> +                              bool wrt_changed) {
>> +     struct dpaa2_qdma_sd_d *sdd;
>> +
>> +     sdd = (struct dpaa2_qdma_sd_d *)dpaa2_comp->desc_virt_addr;
>
>again
Same to before.
>
>> +     memset(sdd, 0, 2 * (sizeof(*sdd)));
>> +
>> +     /* source descriptor CMD */
>> +     sdd->cmd = cpu_to_le32(QDMA_SD_CMD_RDTTYPE_COHERENT);
>> +     sdd++;
>> +
>> +     /* dest descriptor CMD */
>> +     if (wrt_changed)
>> +             sdd->cmd =
>cpu_to_le32(LX2160_QDMA_DD_CMD_WRTTYPE_COHERENT);
>> +     else
>> +             sdd->cmd =
>cpu_to_le32(QDMA_DD_CMD_WRTTYPE_COHERENT);
>> +
>> +     memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
>> +
>> +     /* first frame list to source descriptor */
>> +     dpaa2_fl_set_addr(f_list, dpaa2_comp->desc_bus_addr);
>> +     dpaa2_fl_set_len(f_list, 0x20);
>> +     dpaa2_fl_set_format(f_list, QDMA_FL_FMT_SBF |
>QDMA_FL_SL_LONG);
>> +
>> +     /* bypass memory translation */
>> +     if (smmu_disable)
>> +             f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE); }
>> +
>> +/* source and destination frame list */ static void
>> +dpaa2_qdma_populate_frames(struct dpaa2_fl_entry *f_list,
>> +                        dma_addr_t dst, dma_addr_t src,
>> +                        size_t len, uint8_t fmt) {
>> +     /* source frame list to source buffer */
>> +     memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
>> +
>> +     dpaa2_fl_set_addr(f_list, src);
>> +     dpaa2_fl_set_len(f_list, len);
>> +
>> +     /* single buffer frame or scatter gather frame */
>> +     dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
>> +
>> +     /* bypass memory translation */
>> +     if (smmu_disable)
>> +             f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
>> +
>> +     f_list++;
>> +
>> +     /* destination frame list to destination buffer */
>> +     memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
>> +
>> +     dpaa2_fl_set_addr(f_list, dst);
>> +     dpaa2_fl_set_len(f_list, len);
>> +     dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
>> +     /* single buffer frame or scatter gather frame */
>> +     dpaa2_fl_set_final(f_list, QDMA_FL_F);
>> +     /* bypass memory translation */
>> +     if (smmu_disable)
>> +             f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE); }
>> +
>> +static struct dma_async_tx_descriptor *dpaa2_qdma_prep_memcpy(struct
>> +dma_chan *chan, dma_addr_t dst,
>> +                     dma_addr_t src, size_t len, ulong flags) {
>> +     struct dpaa2_qdma_chan *dpaa2_chan =
>to_dpaa2_qdma_chan(chan);
>> +     struct dpaa2_qdma_engine *dpaa2_qdma;
>> +     struct dpaa2_qdma_comp *dpaa2_comp;
>> +     struct dpaa2_fl_entry *f_list;
>> +     bool wrt_changed;
>> +     u32 format;
>> +
>> +     dpaa2_qdma = dpaa2_chan->qdma;
>> +     dpaa2_comp = dpaa2_qdma_request_desc(dpaa2_chan);
>> +     wrt_changed = (bool)dpaa2_qdma->qdma_wrtype_fixup;
>> +
>> +#ifdef LONG_FORMAT
>
> compile flag and define, so else part is dead code??
Ok, I will add it to Kconfig.
>
>> +     format = QDMA_FD_LONG_FORMAT;
>> +#else
>> +     format = QDMA_FD_SHORT_FORMAT;
>> +#endif
>> +     /* populate Frame descriptor */
>> +     dpaa2_qdma_populate_fd(format, dpaa2_comp);
>> +
>> +     f_list = (struct dpaa2_fl_entry *)dpaa2_comp->fl_virt_addr;
>> +
>> +#ifdef LONG_FORMAT
>> +     /* first frame list for descriptor buffer (logn format) */
>> +     dpaa2_qdma_populate_first_framel(f_list, dpaa2_comp,
>> +wrt_changed);
>> +
>> +     f_list++;
>> +#endif
>> +
>> +     dpaa2_qdma_populate_frames(f_list, dst, src, len,
>> + QDMA_FL_FMT_SBF);
>> +
>> +     return vchan_tx_prep(&dpaa2_chan->vchan, &dpaa2_comp->vdesc,
>> +flags); }
>> +
>> +static enum
>> +dma_status dpaa2_qdma_tx_status(struct dma_chan *chan,
>> +                             dma_cookie_t cookie,
>> +                             struct dma_tx_state *txstate) {
>> +     return dma_cookie_status(chan, cookie, txstate); }
>> +
>> +static void dpaa2_qdma_issue_pending(struct dma_chan *chan) {
>> +     struct dpaa2_qdma_chan *dpaa2_chan =
>to_dpaa2_qdma_chan(chan);
>> +     struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
>> +     struct dpaa2_qdma_priv *priv = dpaa2_qdma->priv;
>> +     struct dpaa2_qdma_comp *dpaa2_comp;
>> +     struct virt_dma_desc *vdesc;
>> +     struct dpaa2_fd *fd;
>> +     unsigned long flags;
>> +     int err;
>> +
>> +     spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
>> +     spin_lock(&dpaa2_chan->vchan.lock);
>> +     if (vchan_issue_pending(&dpaa2_chan->vchan)) {
>> +             vdesc = vchan_next_desc(&dpaa2_chan->vchan);
>> +             if (!vdesc)
>> +                     goto err_enqueue;
>> +             dpaa2_comp = to_fsl_qdma_comp(vdesc);
>> +
>> +             fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;
>> +
>> +             list_del(&vdesc->node);
>> +             list_add_tail(&dpaa2_comp->list,
>> + &dpaa2_chan->comp_used);
>
>what does this list do?
>
It is will used in interrupt to deal dma jobs.
>> +
>> +             /* TOBO: priority hard-coded to zero */
>
>You mean TODO?
Yes.
>
>> +             err = dpaa2_io_service_enqueue_fq(NULL,
>> +                             priv->tx_queue_attr[0].fqid, fd);
>> +             if (err) {
>> +                     list_del(&dpaa2_comp->list);
>> +                     list_add_tail(&dpaa2_comp->list,
>> +                                   &dpaa2_chan->comp_free);
>> +             }
>> +     }
>> +err_enqueue:
>> +     spin_unlock(&dpaa2_chan->vchan.lock);
>> +     spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags); }
>> +
>> +static int __cold dpaa2_qdma_setup(struct fsl_mc_device *ls_dev) {
>> +     struct dpaa2_qdma_priv_per_prio *ppriv;
>> +     struct device *dev = &ls_dev->dev;
>> +     struct dpaa2_qdma_priv *priv;
>> +     u8 prio_def = DPDMAI_PRIO_NUM;
>> +     int err;
>> +     int i;
>> +
>> +     priv = dev_get_drvdata(dev);
>> +
>> +     priv->dev = dev;
>> +     priv->dpqdma_id = ls_dev->obj_desc.id;
>> +
>> +     /*Get the handle for the DPDMAI this interface is associate with
>> + */
>
>Please run checkpatch, it should have told you that you need space after
>comment marker /* foo...
>
Ok, I will check it with --strict.
>> +     err = dpdmai_open(priv->mc_io, 0, priv->dpqdma_id,
>&ls_dev->mc_handle);
>> +     if (err) {
>> +             dev_err(dev, "dpdmai_open() failed\n");
>> +             return err;
>> +     }
>> +     dev_info(dev, "Opened dpdmai object successfully\n");
>> +
>> +     err = dpdmai_get_attributes(priv->mc_io, 0, ls_dev->mc_handle,
>> +                                 &priv->dpdmai_attr);
>> +     if (err) {
>> +             dev_err(dev, "dpdmai_get_attributes() failed\n");
>> +             return err;
>
>so you dont close what you opened in dpdmai_open() Please give a serious
>thought and testing to this driver
OK, got it.
>
>> +     }
>> +
>> +     if (priv->dpdmai_attr.version.major > DPDMAI_VER_MAJOR) {
>> +             dev_err(dev, "DPDMAI major version mismatch\n"
>> +                          "Found %u.%u, supported version
>is %u.%u\n",
>> +                             priv->dpdmai_attr.version.major,
>> +                             priv->dpdmai_attr.version.minor,
>> +                             DPDMAI_VER_MAJOR,
>DPDMAI_VER_MINOR);
>> +     }
>> +
>> +     if (priv->dpdmai_attr.version.minor > DPDMAI_VER_MINOR) {
>> +             dev_err(dev, "DPDMAI minor version mismatch\n"
>> +                          "Found %u.%u, supported version
>is %u.%u\n",
>> +                             priv->dpdmai_attr.version.major,
>> +                             priv->dpdmai_attr.version.minor,
>> +                             DPDMAI_VER_MAJOR,
>DPDMAI_VER_MINOR);
>
>what is the implication of these error, why not bail out on these?
There should be return.
>
>> +     }
>> +
>> +     priv->num_pairs = min(priv->dpdmai_attr.num_of_priorities,
>prio_def);
>> +     ppriv = kcalloc(priv->num_pairs, sizeof(*ppriv), GFP_KERNEL);
>
>what is the context of the fn, sleepy, atomic?
This function will be called on qdma probe,Here is not a critical area,
What's the problem about to use GFP_KERNEL, please let me know.
>
>> +     if (!ppriv) {
>> +             dev_err(dev, "kzalloc for ppriv failed\n");
>
>this need not be logged, core will do so
OK.
>
>> +             return -1;
>
>really -1??
I will update.
>
>I think this driver needs more work, please fix these issues in the comments
>above and also see in rest of the code
OK, I will check all of those patch about these issues in rest of the code. Thanks.
>
Best Regards,
Peng
>--
>~Vinod
Vinod Koul June 13, 2019, 11:03 a.m. UTC | #3 
On 10-06-19, 09:51, Peng Ma wrote:

> >> +                     goto err;
> >> +
> >> +             comp_temp->fl_virt_addr =
> >> +                     (void *)((struct dpaa2_fd *)
> >> +                             comp_temp->fd_virt_addr + 1);
> >
> >casts and pointer math, what could go wrong!! This doesnt smell right!
> >
> >> +             comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
> >> +                                     sizeof(struct dpaa2_fd);
> >
> >why not use fl_virt_addr and get the bus_address?
> What you mean is I should use virt_to_phys to get the bus_address?

Yes instead of maintaining both pointers, just use one and then when
required use one to get other. For bus address I would prefer
dma_map_single rather than virt_to_phys()
Peng Ma June 14, 2019, 1:41 a.m. UTC | #4 
Hi Vinod,

>-----Original Message-----
>From: Vinod Koul <vkoul@kernel.org>
>Sent: 2019年6月13日 19:03
>To: Peng Ma <peng.ma@nxp.com>
>Cc: dan.j.williams@intel.com; Leo Li <leoyang.li@nxp.com>;
>linux-kernel@vger.kernel.org; dmaengine@vger.kernel.org
>Subject: Re: [EXT] Re: [V3 2/2] dmaengine: fsl-dpaa2-qdma: Add NXP dpaa2
>qDMA controller driver for Layerscape SoCs
>
>Caution: EXT Email
>
>On 10-06-19, 09:51, Peng Ma wrote:
>
>> >> +                     goto err;
>> >> +
>> >> +             comp_temp->fl_virt_addr =
>> >> +                     (void *)((struct dpaa2_fd *)
>> >> +                             comp_temp->fd_virt_addr + 1);
>> >
>> >casts and pointer math, what could go wrong!! This doesnt smell right!
>> >
>> >> +             comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
>> >> +                                     sizeof(struct dpaa2_fd);
>> >
>> >why not use fl_virt_addr and get the bus_address?
>> What you mean is I should use virt_to_phys to get the bus_address?
>
>Yes instead of maintaining both pointers, just use one and then when required
>use one to get other. For bus address I would prefer dma_map_single rather
>than virt_to_phys()
[Peng Ma] ok, thanks, I have already send V4 for this series patch, Please review it in your spare time.
Patchwork link: https://patchwork.kernel.org/project/linux-dmaengine/list/?series=131767
Best Regars,
Peng
>--
>~Vinod
diff mbox series

Patch

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index eaf78f4..08aae01 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -671,6 +671,8 @@  source "drivers/dma/sh/Kconfig"
 
 source "drivers/dma/ti/Kconfig"
 
+source "drivers/dma/fsl-dpaa2-qdma/Kconfig"
+
 # clients
 comment "DMA Clients"
 	depends on DMA_ENGINE
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 6126e1c..2499ed8 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -75,6 +75,7 @@  obj-$(CONFIG_UNIPHIER_MDMAC) += uniphier-mdmac.o
 obj-$(CONFIG_XGENE_DMA) += xgene-dma.o
 obj-$(CONFIG_ZX_DMA) += zx_dma.o
 obj-$(CONFIG_ST_FDMA) += st_fdma.o
+obj-$(CONFIG_FSL_DPAA2_QDMA) += fsl-dpaa2-qdma/
 
 obj-y += mediatek/
 obj-y += qcom/
diff --git a/drivers/dma/fsl-dpaa2-qdma/Kconfig b/drivers/dma/fsl-dpaa2-qdma/Kconfig
new file mode 100644
index 0000000..258ed6b
--- /dev/null
+++ b/drivers/dma/fsl-dpaa2-qdma/Kconfig
@@ -0,0 +1,9 @@ 
+menuconfig FSL_DPAA2_QDMA
+	tristate "NXP DPAA2 QDMA"
+	depends on ARM64
+	depends on FSL_MC_BUS && FSL_MC_DPIO
+	select DMA_ENGINE
+	select DMA_VIRTUAL_CHANNELS
+	help
+	  NXP Data Path Acceleration Architecture 2 QDMA driver,
+	  using the NXP MC bus driver.
diff --git a/drivers/dma/fsl-dpaa2-qdma/Makefile b/drivers/dma/fsl-dpaa2-qdma/Makefile
new file mode 100644
index 0000000..c1d0226
--- /dev/null
+++ b/drivers/dma/fsl-dpaa2-qdma/Makefile
@@ -0,0 +1,3 @@ 
+# SPDX-License-Identifier: GPL-2.0
+# Makefile for the NXP DPAA2 qDMA controllers
+obj-$(CONFIG_FSL_DPAA2_QDMA) += dpaa2-qdma.o dpdmai.o
diff --git a/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
new file mode 100644
index 0000000..0cdde0f
--- /dev/null
+++ b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.c
@@ -0,0 +1,782 @@ 
+// SPDX-License-Identifier: GPL-2.0
+// Copyright 2014-2018 NXP
+
+/*
+ * Author: Changming Huang <jerry.huang@nxp.com>
+ *
+ * Driver for the NXP QDMA engine with QMan mode.
+ * Channel virtualization is supported through enqueuing of DMA jobs to,
+ * or dequeuing DMA jobs from different work queues with QMan portal.
+ * This module can be found on NXP LS2 SoCs.
+ *
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/dmapool.h>
+#include <linux/of_irq.h>
+#include <linux/iommu.h>
+#include <linux/sys_soc.h>
+#include <linux/fsl/mc.h>
+#include <soc/fsl/dpaa2-io.h>
+
+#include "../virt-dma.h"
+#include "dpdmai_cmd.h"
+#include "dpdmai.h"
+#include "dpaa2-qdma.h"
+
+static bool smmu_disable = true;
+
+static struct dpaa2_qdma_chan *to_dpaa2_qdma_chan(struct dma_chan *chan)
+{
+	return container_of(chan, struct dpaa2_qdma_chan, vchan.chan);
+}
+
+static struct dpaa2_qdma_comp *to_fsl_qdma_comp(struct virt_dma_desc *vd)
+{
+	return container_of(vd, struct dpaa2_qdma_comp, vdesc);
+}
+
+static int dpaa2_qdma_alloc_chan_resources(struct dma_chan *chan)
+{
+	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
+	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
+	struct device *dev = &dpaa2_qdma->priv->dpdmai_dev->dev;
+
+	dpaa2_chan->fd_pool = dma_pool_create("fd_pool", dev,
+					      FD_POOL_SIZE, 32, 0);
+	if (!dpaa2_chan->fd_pool)
+		return -ENOMEM;
+
+	return dpaa2_qdma->desc_allocated++;
+}
+
+static void dpaa2_qdma_free_chan_resources(struct dma_chan *chan)
+{
+	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
+	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
+	unsigned long flags;
+
+	LIST_HEAD(head);
+
+	spin_lock_irqsave(&dpaa2_chan->vchan.lock, flags);
+	vchan_get_all_descriptors(&dpaa2_chan->vchan, &head);
+	spin_unlock_irqrestore(&dpaa2_chan->vchan.lock, flags);
+
+	vchan_dma_desc_free_list(&dpaa2_chan->vchan, &head);
+
+	dpaa2_dpdmai_free_comp(dpaa2_chan, &dpaa2_chan->comp_used);
+	dpaa2_dpdmai_free_comp(dpaa2_chan, &dpaa2_chan->comp_free);
+
+	dma_pool_destroy(dpaa2_chan->fd_pool);
+	dpaa2_qdma->desc_allocated--;
+}
+
+/*
+ * Request a command descriptor for enqueue.
+ */
+static struct dpaa2_qdma_comp *
+dpaa2_qdma_request_desc(struct dpaa2_qdma_chan *dpaa2_chan)
+{
+	struct dpaa2_qdma_comp *comp_temp = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
+	if (list_empty(&dpaa2_chan->comp_free)) {
+		spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
+		comp_temp = kzalloc(sizeof(*comp_temp), GFP_KERNEL);
+		if (!comp_temp)
+			goto err;
+		comp_temp->fd_virt_addr =
+			dma_pool_alloc(dpaa2_chan->fd_pool, GFP_NOWAIT,
+				       &comp_temp->fd_bus_addr);
+		if (!comp_temp->fd_virt_addr)
+			goto err;
+
+		comp_temp->fl_virt_addr =
+			(void *)((struct dpaa2_fd *)
+				comp_temp->fd_virt_addr + 1);
+		comp_temp->fl_bus_addr = comp_temp->fd_bus_addr +
+					sizeof(struct dpaa2_fd);
+		comp_temp->desc_virt_addr =
+			(void *)((struct dpaa2_fl_entry *)
+				comp_temp->fl_virt_addr + 3);
+		comp_temp->desc_bus_addr = comp_temp->fl_bus_addr +
+				sizeof(struct dpaa2_fl_entry) * 3;
+
+		comp_temp->qchan = dpaa2_chan;
+		return comp_temp;
+	}
+	comp_temp = list_first_entry(&dpaa2_chan->comp_free,
+				     struct dpaa2_qdma_comp, list);
+	list_del(&comp_temp->list);
+	spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
+
+	comp_temp->qchan = dpaa2_chan;
+err:
+	return comp_temp;
+}
+
+static void
+dpaa2_qdma_populate_fd(u32 format, struct dpaa2_qdma_comp *dpaa2_comp)
+{
+	struct dpaa2_fd *fd;
+
+	fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;
+	memset(fd, 0, sizeof(struct dpaa2_fd));
+
+	/* fd populated */
+	dpaa2_fd_set_addr(fd, dpaa2_comp->fl_bus_addr);
+	/* Bypass memory translation, Frame list format, short length disable */
+	/* we need to disable BMT if fsl-mc use iova addr */
+	if (smmu_disable)
+		dpaa2_fd_set_bpid(fd, QMAN_FD_BMT_ENABLE);
+	dpaa2_fd_set_format(fd, QMAN_FD_FMT_ENABLE | QMAN_FD_SL_DISABLE);
+
+	dpaa2_fd_set_frc(fd, format | QDMA_SER_CTX);
+}
+
+/* first frame list for descriptor buffer */
+static void
+dpaa2_qdma_populate_first_framel(struct dpaa2_fl_entry *f_list,
+				 struct dpaa2_qdma_comp *dpaa2_comp,
+				 bool wrt_changed)
+{
+	struct dpaa2_qdma_sd_d *sdd;
+
+	sdd = (struct dpaa2_qdma_sd_d *)dpaa2_comp->desc_virt_addr;
+	memset(sdd, 0, 2 * (sizeof(*sdd)));
+
+	/* source descriptor CMD */
+	sdd->cmd = cpu_to_le32(QDMA_SD_CMD_RDTTYPE_COHERENT);
+	sdd++;
+
+	/* dest descriptor CMD */
+	if (wrt_changed)
+		sdd->cmd = cpu_to_le32(LX2160_QDMA_DD_CMD_WRTTYPE_COHERENT);
+	else
+		sdd->cmd = cpu_to_le32(QDMA_DD_CMD_WRTTYPE_COHERENT);
+
+	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
+
+	/* first frame list to source descriptor */
+	dpaa2_fl_set_addr(f_list, dpaa2_comp->desc_bus_addr);
+	dpaa2_fl_set_len(f_list, 0x20);
+	dpaa2_fl_set_format(f_list, QDMA_FL_FMT_SBF | QDMA_FL_SL_LONG);
+
+	/* bypass memory translation */
+	if (smmu_disable)
+		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
+}
+
+/* source and destination frame list */
+static void
+dpaa2_qdma_populate_frames(struct dpaa2_fl_entry *f_list,
+			   dma_addr_t dst, dma_addr_t src,
+			   size_t len, uint8_t fmt)
+{
+	/* source frame list to source buffer */
+	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
+
+	dpaa2_fl_set_addr(f_list, src);
+	dpaa2_fl_set_len(f_list, len);
+
+	/* single buffer frame or scatter gather frame */
+	dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
+
+	/* bypass memory translation */
+	if (smmu_disable)
+		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
+
+	f_list++;
+
+	/* destination frame list to destination buffer */
+	memset(f_list, 0, sizeof(struct dpaa2_fl_entry));
+
+	dpaa2_fl_set_addr(f_list, dst);
+	dpaa2_fl_set_len(f_list, len);
+	dpaa2_fl_set_format(f_list, (fmt | QDMA_FL_SL_LONG));
+	/* single buffer frame or scatter gather frame */
+	dpaa2_fl_set_final(f_list, QDMA_FL_F);
+	/* bypass memory translation */
+	if (smmu_disable)
+		f_list->bpid = cpu_to_le16(QDMA_FL_BMT_ENABLE);
+}
+
+static struct dma_async_tx_descriptor
+*dpaa2_qdma_prep_memcpy(struct dma_chan *chan, dma_addr_t dst,
+			dma_addr_t src, size_t len, ulong flags)
+{
+	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
+	struct dpaa2_qdma_engine *dpaa2_qdma;
+	struct dpaa2_qdma_comp *dpaa2_comp;
+	struct dpaa2_fl_entry *f_list;
+	bool wrt_changed;
+	u32 format;
+
+	dpaa2_qdma = dpaa2_chan->qdma;
+	dpaa2_comp = dpaa2_qdma_request_desc(dpaa2_chan);
+	wrt_changed = (bool)dpaa2_qdma->qdma_wrtype_fixup;
+
+#ifdef LONG_FORMAT
+	format = QDMA_FD_LONG_FORMAT;
+#else
+	format = QDMA_FD_SHORT_FORMAT;
+#endif
+	/* populate Frame descriptor */
+	dpaa2_qdma_populate_fd(format, dpaa2_comp);
+
+	f_list = (struct dpaa2_fl_entry *)dpaa2_comp->fl_virt_addr;
+
+#ifdef LONG_FORMAT
+	/* first frame list for descriptor buffer (logn format) */
+	dpaa2_qdma_populate_first_framel(f_list, dpaa2_comp, wrt_changed);
+
+	f_list++;
+#endif
+
+	dpaa2_qdma_populate_frames(f_list, dst, src, len, QDMA_FL_FMT_SBF);
+
+	return vchan_tx_prep(&dpaa2_chan->vchan, &dpaa2_comp->vdesc, flags);
+}
+
+static enum
+dma_status dpaa2_qdma_tx_status(struct dma_chan *chan,
+				dma_cookie_t cookie,
+				struct dma_tx_state *txstate)
+{
+	return dma_cookie_status(chan, cookie, txstate);
+}
+
+static void dpaa2_qdma_issue_pending(struct dma_chan *chan)
+{
+	struct dpaa2_qdma_chan *dpaa2_chan = to_dpaa2_qdma_chan(chan);
+	struct dpaa2_qdma_engine *dpaa2_qdma = dpaa2_chan->qdma;
+	struct dpaa2_qdma_priv *priv = dpaa2_qdma->priv;
+	struct dpaa2_qdma_comp *dpaa2_comp;
+	struct virt_dma_desc *vdesc;
+	struct dpaa2_fd *fd;
+	unsigned long flags;
+	int err;
+
+	spin_lock_irqsave(&dpaa2_chan->queue_lock, flags);
+	spin_lock(&dpaa2_chan->vchan.lock);
+	if (vchan_issue_pending(&dpaa2_chan->vchan)) {
+		vdesc = vchan_next_desc(&dpaa2_chan->vchan);
+		if (!vdesc)
+			goto err_enqueue;
+		dpaa2_comp = to_fsl_qdma_comp(vdesc);
+
+		fd = (struct dpaa2_fd *)dpaa2_comp->fd_virt_addr;
+
+		list_del(&vdesc->node);
+		list_add_tail(&dpaa2_comp->list, &dpaa2_chan->comp_used);
+
+		/* TOBO: priority hard-coded to zero */
+		err = dpaa2_io_service_enqueue_fq(NULL,
+				priv->tx_queue_attr[0].fqid, fd);
+		if (err) {
+			list_del(&dpaa2_comp->list);
+			list_add_tail(&dpaa2_comp->list,
+				      &dpaa2_chan->comp_free);
+		}
+	}
+err_enqueue:
+	spin_unlock(&dpaa2_chan->vchan.lock);
+	spin_unlock_irqrestore(&dpaa2_chan->queue_lock, flags);
+}
+
+static int __cold dpaa2_qdma_setup(struct fsl_mc_device *ls_dev)
+{
+	struct dpaa2_qdma_priv_per_prio *ppriv;
+	struct device *dev = &ls_dev->dev;
+	struct dpaa2_qdma_priv *priv;
+	u8 prio_def = DPDMAI_PRIO_NUM;
+	int err;
+	int i;
+
+	priv = dev_get_drvdata(dev);
+
+	priv->dev = dev;
+	priv->dpqdma_id = ls_dev->obj_desc.id;
+
+	/*Get the handle for the DPDMAI this interface is associate with */
+	err = dpdmai_open(priv->mc_io, 0, priv->dpqdma_id, &ls_dev->mc_handle);
+	if (err) {
+		dev_err(dev, "dpdmai_open() failed\n");
+		return err;
+	}
+	dev_info(dev, "Opened dpdmai object successfully\n");
+
+	err = dpdmai_get_attributes(priv->mc_io, 0, ls_dev->mc_handle,
+				    &priv->dpdmai_attr);
+	if (err) {
+		dev_err(dev, "dpdmai_get_attributes() failed\n");
+		return err;
+	}
+
+	if (priv->dpdmai_attr.version.major > DPDMAI_VER_MAJOR) {
+		dev_err(dev, "DPDMAI major version mismatch\n"
+			     "Found %u.%u, supported version is %u.%u\n",
+				priv->dpdmai_attr.version.major,
+				priv->dpdmai_attr.version.minor,
+				DPDMAI_VER_MAJOR, DPDMAI_VER_MINOR);
+	}
+
+	if (priv->dpdmai_attr.version.minor > DPDMAI_VER_MINOR) {
+		dev_err(dev, "DPDMAI minor version mismatch\n"
+			     "Found %u.%u, supported version is %u.%u\n",
+				priv->dpdmai_attr.version.major,
+				priv->dpdmai_attr.version.minor,
+				DPDMAI_VER_MAJOR, DPDMAI_VER_MINOR);
+	}
+
+	priv->num_pairs = min(priv->dpdmai_attr.num_of_priorities, prio_def);
+	ppriv = kcalloc(priv->num_pairs, sizeof(*ppriv), GFP_KERNEL);
+	if (!ppriv) {
+		dev_err(dev, "kzalloc for ppriv failed\n");
+		return -1;
+	}
+	priv->ppriv = ppriv;
+
+	for (i = 0; i < priv->num_pairs; i++) {
+		err = dpdmai_get_rx_queue(priv->mc_io, 0, ls_dev->mc_handle,
+					  i, &priv->rx_queue_attr[i]);
+		if (err) {
+			dev_err(dev, "dpdmai_get_rx_queue() failed\n");
+			return err;
+		}
+		ppriv->rsp_fqid = priv->rx_queue_attr[i].fqid;
+
+		err = dpdmai_get_tx_queue(priv->mc_io, 0, ls_dev->mc_handle,
+					  i, &priv->tx_queue_attr[i]);
+		if (err) {
+			dev_err(dev, "dpdmai_get_tx_queue() failed\n");
+			return err;
+		}
+		ppriv->req_fqid = priv->tx_queue_attr[i].fqid;
+		ppriv->prio = i;
+		ppriv->priv = priv;
+		ppriv++;
+	}
+
+	return 0;
+}
+
+static void dpaa2_qdma_fqdan_cb(struct dpaa2_io_notification_ctx *ctx)
+{
+	struct dpaa2_qdma_priv_per_prio *ppriv = container_of(ctx,
+			struct dpaa2_qdma_priv_per_prio, nctx);
+	struct dpaa2_qdma_comp *dpaa2_comp, *_comp_tmp;
+	struct dpaa2_qdma_priv *priv = ppriv->priv;
+	u32 n_chans = priv->dpaa2_qdma->n_chans;
+	struct dpaa2_qdma_chan *qchan;
+	const struct dpaa2_fd *fd_eq;
+	const struct dpaa2_fd *fd;
+	struct dpaa2_dq *dq;
+	int is_last = 0;
+	int found;
+	u8 status;
+	int err;
+	int i;
+
+	do {
+		err = dpaa2_io_service_pull_fq(NULL, ppriv->rsp_fqid,
+					       ppriv->store);
+	} while (err);
+
+	while (!is_last) {
+		do {
+			dq = dpaa2_io_store_next(ppriv->store, &is_last);
+		} while (!is_last && !dq);
+		if (!dq) {
+			dev_err(priv->dev, "FQID returned no valid frames!\n");
+			continue;
+		}
+
+		/* obtain FD and process the error */
+		fd = dpaa2_dq_fd(dq);
+
+		status = dpaa2_fd_get_ctrl(fd) & 0xff;
+		if (status)
+			dev_err(priv->dev, "FD error occurred\n");
+		found = 0;
+		for (i = 0; i < n_chans; i++) {
+			qchan = &priv->dpaa2_qdma->chans[i];
+			spin_lock(&qchan->queue_lock);
+			if (list_empty(&qchan->comp_used)) {
+				spin_unlock(&qchan->queue_lock);
+				continue;
+			}
+			list_for_each_entry_safe(dpaa2_comp, _comp_tmp,
+						 &qchan->comp_used, list) {
+				fd_eq = (struct dpaa2_fd *)
+					dpaa2_comp->fd_virt_addr;
+
+				if (le64_to_cpu(fd_eq->simple.addr) ==
+				    le64_to_cpu(fd->simple.addr)) {
+					spin_lock(&qchan->vchan.lock);
+					vchan_cookie_complete(&
+							dpaa2_comp->vdesc);
+					spin_unlock(&qchan->vchan.lock);
+					found = 1;
+					break;
+				}
+			}
+			spin_unlock(&qchan->queue_lock);
+			if (found)
+				break;
+		}
+	}
+
+	dpaa2_io_service_rearm(NULL, ctx);
+}
+
+static int __cold dpaa2_qdma_dpio_setup(struct dpaa2_qdma_priv *priv)
+{
+	struct dpaa2_qdma_priv_per_prio *ppriv;
+	struct device *dev = priv->dev;
+	int err, i, num;
+
+	num = priv->num_pairs;
+	ppriv = priv->ppriv;
+	for (i = 0; i < num; i++) {
+		ppriv->nctx.is_cdan = 0;
+		ppriv->nctx.desired_cpu = 1;
+		ppriv->nctx.id = ppriv->rsp_fqid;
+		ppriv->nctx.cb = dpaa2_qdma_fqdan_cb;
+		err = dpaa2_io_service_register(NULL, &ppriv->nctx, dev);
+		if (err) {
+			dev_err(dev, "Notification register failed\n");
+			goto err_service;
+		}
+
+		ppriv->store =
+			dpaa2_io_store_create(DPAA2_QDMA_STORE_SIZE, dev);
+		if (!ppriv->store) {
+			dev_err(dev, "dpaa2_io_store_create() failed\n");
+			goto err_store;
+		}
+
+		ppriv++;
+	}
+	return 0;
+
+err_store:
+	dpaa2_io_service_deregister(NULL, &ppriv->nctx, dev);
+err_service:
+	ppriv--;
+	while (ppriv >= priv->ppriv) {
+		dpaa2_io_service_deregister(NULL, &ppriv->nctx, dev);
+		dpaa2_io_store_destroy(ppriv->store);
+		ppriv--;
+	}
+	return -1;
+}
+
+static void dpaa2_dpmai_store_free(struct dpaa2_qdma_priv *priv)
+{
+	struct dpaa2_qdma_priv_per_prio *ppriv = priv->ppriv;
+	int i;
+
+	for (i = 0; i < priv->num_pairs; i++) {
+		dpaa2_io_store_destroy(ppriv->store);
+		ppriv++;
+	}
+}
+
+static void dpaa2_dpdmai_dpio_free(struct dpaa2_qdma_priv *priv)
+{
+	struct dpaa2_qdma_priv_per_prio *ppriv = priv->ppriv;
+	struct device *dev = priv->dev;
+	int i;
+
+	for (i = 0; i < priv->num_pairs; i++) {
+		dpaa2_io_service_deregister(NULL, &ppriv->nctx, dev);
+		ppriv++;
+	}
+}
+
+static int __cold dpaa2_dpdmai_bind(struct dpaa2_qdma_priv *priv)
+{
+	int err;
+	int i, num;
+	struct device *dev = priv->dev;
+	struct dpaa2_qdma_priv_per_prio *ppriv;
+	struct dpdmai_rx_queue_cfg rx_queue_cfg;
+	struct fsl_mc_device *ls_dev = to_fsl_mc_device(dev);
+
+	num = priv->num_pairs;
+	ppriv = priv->ppriv;
+	for (i = 0; i < num; i++) {
+		rx_queue_cfg.options = DPDMAI_QUEUE_OPT_USER_CTX |
+					DPDMAI_QUEUE_OPT_DEST;
+		rx_queue_cfg.user_ctx = ppriv->nctx.qman64;
+		rx_queue_cfg.dest_cfg.dest_type = DPDMAI_DEST_DPIO;
+		rx_queue_cfg.dest_cfg.dest_id = ppriv->nctx.dpio_id;
+		rx_queue_cfg.dest_cfg.priority = ppriv->prio;
+		err = dpdmai_set_rx_queue(priv->mc_io, 0, ls_dev->mc_handle,
+					  rx_queue_cfg.dest_cfg.priority,
+					  &rx_queue_cfg);
+		if (err) {
+			dev_err(dev, "dpdmai_set_rx_queue() failed\n");
+			return err;
+		}
+
+		ppriv++;
+	}
+
+	return 0;
+}
+
+static int __cold dpaa2_dpdmai_dpio_unbind(struct dpaa2_qdma_priv *priv)
+{
+	int i;
+	int err = 0;
+	struct device *dev = priv->dev;
+	struct fsl_mc_device *ls_dev = to_fsl_mc_device(dev);
+	struct dpaa2_qdma_priv_per_prio *ppriv = priv->ppriv;
+
+	for (i = 0; i < priv->num_pairs; i++) {
+		ppriv->nctx.qman64 = 0;
+		ppriv->nctx.dpio_id = 0;
+		ppriv++;
+	}
+
+	err = dpdmai_reset(priv->mc_io, 0, ls_dev->mc_handle);
+	if (err)
+		dev_err(dev, "dpdmai_reset() failed\n");
+
+	return err;
+}
+
+static void dpaa2_dpdmai_free_comp(struct dpaa2_qdma_chan *qchan,
+				   struct list_head *head)
+{
+	struct dpaa2_qdma_comp *comp_tmp, *_comp_tmp;
+
+	list_for_each_entry_safe(comp_tmp, _comp_tmp,
+				 head, list) {
+		dma_pool_free(qchan->fd_pool,
+			      comp_tmp->fd_virt_addr,
+			      comp_tmp->fd_bus_addr);
+		list_del(&comp_tmp->list);
+		kfree(comp_tmp);
+	}
+}
+
+static void dpaa2_dpdmai_free_channels(struct dpaa2_qdma_engine *dpaa2_qdma)
+{
+	struct dpaa2_qdma_chan *qchan;
+	int num, i;
+
+	num = dpaa2_qdma->n_chans;
+	for (i = 0; i < num; i++) {
+		qchan = &dpaa2_qdma->chans[i];
+		dpaa2_dpdmai_free_comp(qchan, &qchan->comp_used);
+		dpaa2_dpdmai_free_comp(qchan, &qchan->comp_free);
+		dma_pool_destroy(qchan->fd_pool);
+	}
+}
+
+static void dpaa2_qdma_free_desc(struct virt_dma_desc *vdesc)
+{
+	struct dpaa2_qdma_comp *dpaa2_comp;
+	struct dpaa2_qdma_chan *qchan;
+
+	dpaa2_comp = to_fsl_qdma_comp(vdesc);
+	qchan = dpaa2_comp->qchan;
+	list_del(&dpaa2_comp->list);
+	list_add_tail(&dpaa2_comp->list, &qchan->comp_free);
+}
+
+static int dpaa2_dpdmai_init_channels(struct dpaa2_qdma_engine *dpaa2_qdma)
+{
+	struct dpaa2_qdma_chan *dpaa2_chan;
+	int i;
+
+	INIT_LIST_HEAD(&dpaa2_qdma->dma_dev.channels);
+	for (i = 0; i < dpaa2_qdma->n_chans; i++) {
+		dpaa2_chan = &dpaa2_qdma->chans[i];
+		dpaa2_chan->qdma = dpaa2_qdma;
+		dpaa2_chan->vchan.desc_free = dpaa2_qdma_free_desc;
+		vchan_init(&dpaa2_chan->vchan, &dpaa2_qdma->dma_dev);
+		spin_lock_init(&dpaa2_chan->queue_lock);
+		INIT_LIST_HEAD(&dpaa2_chan->comp_used);
+		INIT_LIST_HEAD(&dpaa2_chan->comp_free);
+	}
+	return 0;
+}
+
+static int dpaa2_qdma_probe(struct fsl_mc_device *dpdmai_dev)
+{
+	struct dpaa2_qdma_priv *priv;
+	struct device *dev = &dpdmai_dev->dev;
+	struct dpaa2_qdma_engine *dpaa2_qdma;
+	int err;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+	dev_set_drvdata(dev, priv);
+	priv->dpdmai_dev = dpdmai_dev;
+
+	priv->iommu_domain = iommu_get_domain_for_dev(dev);
+	if (priv->iommu_domain)
+		smmu_disable = false;
+
+	/* obtain a MC portal */
+	err = fsl_mc_portal_allocate(dpdmai_dev, 0, &priv->mc_io);
+	if (err) {
+		if (err == -ENXIO)
+			err = -EPROBE_DEFER;
+		else
+			dev_err(dev, "MC portal allocation failed\n");
+		goto err_mcportal;
+	}
+
+	/* DPDMAI initialization */
+	err = dpaa2_qdma_setup(dpdmai_dev);
+	if (err) {
+		dev_err(dev, "dpaa2_dpdmai_setup() failed\n");
+		goto err_dpdmai_setup;
+	}
+
+	/* DPIO */
+	err = dpaa2_qdma_dpio_setup(priv);
+	if (err) {
+		dev_err(dev, "dpaa2_dpdmai_dpio_setup() failed\n");
+		goto err_dpio_setup;
+	}
+
+	/* DPDMAI binding to DPIO */
+	err = dpaa2_dpdmai_bind(priv);
+	if (err) {
+		dev_err(dev, "dpaa2_dpdmai_bind() failed\n");
+		goto err_bind;
+	}
+
+	/* DPDMAI enable */
+	err = dpdmai_enable(priv->mc_io, 0, dpdmai_dev->mc_handle);
+	if (err) {
+		dev_err(dev, "dpdmai_enable() faile\n");
+		goto err_enable;
+	}
+
+	dpaa2_qdma = kzalloc(sizeof(*dpaa2_qdma), GFP_KERNEL);
+	if (!dpaa2_qdma) {
+		err = -ENOMEM;
+		goto err_eng;
+	}
+
+	priv->dpaa2_qdma = dpaa2_qdma;
+	dpaa2_qdma->priv = priv;
+
+	dpaa2_qdma->desc_allocated = 0;
+	dpaa2_qdma->n_chans = NUM_CH;
+
+	dpaa2_dpdmai_init_channels(dpaa2_qdma);
+
+	if (soc_device_match(soc_fixup_tuning))
+		dpaa2_qdma->qdma_wrtype_fixup = true;
+	else
+		dpaa2_qdma->qdma_wrtype_fixup = false;
+
+	dma_cap_set(DMA_PRIVATE, dpaa2_qdma->dma_dev.cap_mask);
+	dma_cap_set(DMA_SLAVE, dpaa2_qdma->dma_dev.cap_mask);
+	dma_cap_set(DMA_MEMCPY, dpaa2_qdma->dma_dev.cap_mask);
+
+	dpaa2_qdma->dma_dev.dev = dev;
+	dpaa2_qdma->dma_dev.device_alloc_chan_resources =
+		dpaa2_qdma_alloc_chan_resources;
+	dpaa2_qdma->dma_dev.device_free_chan_resources =
+		dpaa2_qdma_free_chan_resources;
+	dpaa2_qdma->dma_dev.device_tx_status = dpaa2_qdma_tx_status;
+	dpaa2_qdma->dma_dev.device_prep_dma_memcpy = dpaa2_qdma_prep_memcpy;
+	dpaa2_qdma->dma_dev.device_issue_pending = dpaa2_qdma_issue_pending;
+
+	err = dma_async_device_register(&dpaa2_qdma->dma_dev);
+	if (err) {
+		dev_err(dev, "Can't register NXP QDMA engine.\n");
+		goto err_eng;
+	}
+
+	return 0;
+
+err_eng:
+	dpdmai_disable(priv->mc_io, 0, dpdmai_dev->mc_handle);
+err_enable:
+	dpaa2_dpdmai_dpio_unbind(priv);
+err_bind:
+	dpaa2_dpmai_store_free(priv);
+	dpaa2_dpdmai_dpio_free(priv);
+err_dpio_setup:
+	dpdmai_close(priv->mc_io, 0, dpdmai_dev->mc_handle);
+err_dpdmai_setup:
+	fsl_mc_portal_free(priv->mc_io);
+err_mcportal:
+	kfree(priv->ppriv);
+	kfree(priv);
+	dev_set_drvdata(dev, NULL);
+	return err;
+}
+
+static int dpaa2_qdma_remove(struct fsl_mc_device *ls_dev)
+{
+	struct device *dev;
+	struct dpaa2_qdma_priv *priv;
+	struct dpaa2_qdma_engine *dpaa2_qdma;
+
+	dev = &ls_dev->dev;
+	priv = dev_get_drvdata(dev);
+	dpaa2_qdma = priv->dpaa2_qdma;
+
+	dpdmai_disable(priv->mc_io, 0, ls_dev->mc_handle);
+	dpaa2_dpdmai_dpio_unbind(priv);
+	dpaa2_dpmai_store_free(priv);
+	dpaa2_dpdmai_dpio_free(priv);
+	dpdmai_close(priv->mc_io, 0, ls_dev->mc_handle);
+	fsl_mc_portal_free(priv->mc_io);
+	dev_set_drvdata(dev, NULL);
+	dpaa2_dpdmai_free_channels(dpaa2_qdma);
+
+	dma_async_device_unregister(&dpaa2_qdma->dma_dev);
+	kfree(priv);
+	kfree(dpaa2_qdma);
+
+	return 0;
+}
+
+static const struct fsl_mc_device_id dpaa2_qdma_id_table[] = {
+	{
+		.vendor = FSL_MC_VENDOR_FREESCALE,
+		.obj_type = "dpdmai",
+	},
+	{ .vendor = 0x0 }
+};
+
+static struct fsl_mc_driver dpaa2_qdma_driver = {
+	.driver		= {
+		.name	= "dpaa2-qdma",
+		.owner  = THIS_MODULE,
+	},
+	.probe          = dpaa2_qdma_probe,
+	.remove		= dpaa2_qdma_remove,
+	.match_id_table	= dpaa2_qdma_id_table
+};
+
+static int __init dpaa2_qdma_driver_init(void)
+{
+	return fsl_mc_driver_register(&(dpaa2_qdma_driver));
+}
+late_initcall(dpaa2_qdma_driver_init);
+
+static void __exit fsl_qdma_exit(void)
+{
+	fsl_mc_driver_unregister(&(dpaa2_qdma_driver));
+}
+module_exit(fsl_qdma_exit);
+
+MODULE_ALIAS("platform:fsl-dpaa2-qdma");
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("NXP Layerscape DPAA2 qDMA engine driver");
diff --git a/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h
new file mode 100644
index 0000000..3fcade8
--- /dev/null
+++ b/drivers/dma/fsl-dpaa2-qdma/dpaa2-qdma.h
@@ -0,0 +1,152 @@ 
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright 2014-2018 NXP */
+
+#ifndef __DPAA2_QDMA_H
+#define __DPAA2_QDMA_H
+
+#define LONG_FORMAT 1
+
+#define DPAA2_QDMA_STORE_SIZE 16
+#define NUM_CH 8
+
+struct dpaa2_qdma_sd_d {
+	u32 rsv:32;
+	union {
+		struct {
+			u32 ssd:12; /* souce stride distance */
+			u32 sss:12; /* souce stride size */
+			u32 rsv1:8;
+		} sdf;
+		struct {
+			u32 dsd:12; /* Destination stride distance */
+			u32 dss:12; /* Destination stride size */
+			u32 rsv2:8;
+		} ddf;
+	} df;
+	u32 rbpcmd;	/* Route-by-port command */
+	u32 cmd;
+} __attribute__((__packed__));
+
+/* Source descriptor command read transaction type for RBP=0: */
+/* coherent copy of cacheable memory */
+#define QDMA_SD_CMD_RDTTYPE_COHERENT (0xb << 28)
+/* Destination descriptor command write transaction type for RBP=0: */
+/* coherent copy of cacheable memory */
+#define QDMA_DD_CMD_WRTTYPE_COHERENT (0x6 << 28)
+#define LX2160_QDMA_DD_CMD_WRTTYPE_COHERENT (0xb << 28)
+
+#define QMAN_FD_FMT_ENABLE	BIT(0) /* frame list table enable */
+#define QMAN_FD_BMT_ENABLE	BIT(15) /* bypass memory translation */
+#define QMAN_FD_BMT_DISABLE	(0) /* bypass memory translation */
+#define QMAN_FD_SL_DISABLE	(0) /* short lengthe disabled */
+#define QMAN_FD_SL_ENABLE	BIT(14) /* short lengthe enabled */
+
+#define QDMA_FINAL_BIT_DISABLE	(0) /* final bit disable */
+#define QDMA_FINAL_BIT_ENABLE	BIT(31) /* final bit enable */
+
+#define QDMA_FD_SHORT_FORMAT	BIT(11) /* short format */
+#define QDMA_FD_LONG_FORMAT	(0) /* long format */
+#define QDMA_SER_DISABLE	(8) /* no notification */
+#define QDMA_SER_CTX		BIT(8) /* notification by FQD_CTX[fqid] */
+#define QDMA_SER_DEST		(2 << 8) /* notification by destination desc */
+#define QDMA_SER_BOTH		(3 << 8) /* soruce and dest notification */
+#define QDMA_FD_SPF_ENALBE	BIT(30) /* source prefetch enable */
+
+#define QMAN_FD_VA_ENABLE	BIT(14) /* Address used is virtual address */
+#define QMAN_FD_VA_DISABLE	(0)/* Address used is a real address */
+/* Flow Context: 49bit physical address */
+#define QMAN_FD_CBMT_ENABLE	BIT(15)
+#define QMAN_FD_CBMT_DISABLE	(0) /* Flow Context: 64bit virtual address */
+#define QMAN_FD_SC_DISABLE	(0) /* stashing control */
+
+#define QDMA_FL_FMT_SBF		(0x0) /* Single buffer frame */
+#define QDMA_FL_FMT_SGE		(0x2) /* Scatter gather frame */
+#define QDMA_FL_BMT_ENABLE	BIT(15) /* enable bypass memory translation */
+#define QDMA_FL_BMT_DISABLE	(0x0) /* enable bypass memory translation */
+#define QDMA_FL_SL_LONG		(0x0)/* long length */
+#define QDMA_FL_SL_SHORT	(0x1) /* short length */
+#define QDMA_FL_F		(0x1)/* last frame list bit */
+
+/*Description of Frame list table structure*/
+struct dpaa2_qdma_chan {
+	struct virt_dma_chan		vchan;
+	struct virt_dma_desc		vdesc;
+	enum dma_status			status;
+	struct dpaa2_qdma_engine	*qdma;
+
+	struct mutex			dpaa2_queue_mutex;
+	spinlock_t			queue_lock;
+	struct dma_pool			*fd_pool;
+
+	struct list_head		comp_used;
+	struct list_head		comp_free;
+
+};
+
+struct dpaa2_qdma_comp {
+	dma_addr_t		fd_bus_addr;
+	dma_addr_t		fl_bus_addr;
+	dma_addr_t		desc_bus_addr;
+	void			*fd_virt_addr;
+	void			*fl_virt_addr;
+	void			*desc_virt_addr;
+	struct dpaa2_qdma_chan	*qchan;
+	struct virt_dma_desc	vdesc;
+	struct list_head	list;
+};
+
+struct dpaa2_qdma_engine {
+	struct dma_device	dma_dev;
+	u32			n_chans;
+	struct dpaa2_qdma_chan	chans[NUM_CH];
+	int			qdma_wrtype_fixup;
+	int			desc_allocated;
+
+	struct dpaa2_qdma_priv *priv;
+};
+
+/*
+ * dpaa2_qdma_priv - driver private data
+ */
+struct dpaa2_qdma_priv {
+	int dpqdma_id;
+
+	struct iommu_domain	*iommu_domain;
+	struct dpdmai_attr	dpdmai_attr;
+	struct device		*dev;
+	struct fsl_mc_io	*mc_io;
+	struct fsl_mc_device	*dpdmai_dev;
+	u8			num_pairs;
+
+	struct dpaa2_qdma_engine	*dpaa2_qdma;
+	struct dpaa2_qdma_priv_per_prio	*ppriv;
+
+	struct dpdmai_rx_queue_attr rx_queue_attr[DPDMAI_PRIO_NUM];
+	struct dpdmai_tx_queue_attr tx_queue_attr[DPDMAI_PRIO_NUM];
+};
+
+struct dpaa2_qdma_priv_per_prio {
+	int req_fqid;
+	int rsp_fqid;
+	int prio;
+
+	struct dpaa2_io_store *store;
+	struct dpaa2_io_notification_ctx nctx;
+
+	struct dpaa2_qdma_priv *priv;
+};
+
+static struct soc_device_attribute soc_fixup_tuning[] = {
+	{ .family = "QorIQ LX2160A"},
+	{ },
+};
+
+/* FD pool size: one FD + 3 Frame list + 2 source/destination descriptor */
+#define FD_POOL_SIZE (sizeof(struct dpaa2_fd) + \
+		sizeof(struct dpaa2_fl_entry) * 3 + \
+		sizeof(struct dpaa2_qdma_sd_d) * 2)
+
+static void dpaa2_dpdmai_free_channels(struct dpaa2_qdma_engine *dpaa2_qdma);
+static void dpaa2_dpdmai_free_comp(struct dpaa2_qdma_chan *qchan,
+				   struct list_head *head);
+#endif /* __DPAA2_QDMA_H */