On 16-10-20, 02:39, Sanjay R Mehta wrote:
> From: Sanjay R Mehta <sanju.me...@amd.com>
>
> Adding support for AMD PTDMA controller. It performs high-bandwidth
Add support or Adds support.. would be apt!
> memory to memory and IO copy operation. Device commands are managed
> via a circular queue of 'descriptors', each of which specifies source
> and destination addresses for copying a single buffer of data.
>
> Signed-off-by: Sanjay R Mehta <sanju.me...@amd.com>
> ---
> MAINTAINERS | 6 +
> drivers/dma/Kconfig | 2 +
> drivers/dma/Makefile | 1 +
> drivers/dma/ptdma/Kconfig | 11 ++
> drivers/dma/ptdma/Makefile | 10 ++
> drivers/dma/ptdma/ptdma-dev.c | 296 +++++++++++++++++++++++++++++++++++++++
> drivers/dma/ptdma/ptdma-pci.c | 252 +++++++++++++++++++++++++++++++++
> drivers/dma/ptdma/ptdma.h | 316
> ++++++++++++++++++++++++++++++++++++++++++
> 8 files changed, 894 insertions(+)
> create mode 100644 drivers/dma/ptdma/Kconfig
> create mode 100644 drivers/dma/ptdma/Makefile
> create mode 100644 drivers/dma/ptdma/ptdma-dev.c
> create mode 100644 drivers/dma/ptdma/ptdma-pci.c
> create mode 100644 drivers/dma/ptdma/ptdma.h
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 33b27e6..f6ae0d1 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -943,6 +943,12 @@ S: Supported
> F: arch/arm64/boot/dts/amd/amd-seattle-xgbe*.dtsi
> F: drivers/net/ethernet/amd/xgbe/
>
> +AMD PTDMA DRIVER
> +M: Sanjay R Mehta <sanju.me...@amd.com>
> +L: dmaeng...@vger.kernel.org
> +S: Maintained
> +F: drivers/dma/ptdma/
> +
> ANALOG DEVICES INC AD5686 DRIVER
> M: Michael Hennerich <michael.henner...@analog.com>
> L: linux...@vger.kernel.org
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index 518a143..a21c983 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -748,6 +748,8 @@ source "drivers/dma/ti/Kconfig"
>
> source "drivers/dma/fsl-dpaa2-qdma/Kconfig"
>
> +source "drivers/dma/ptdma/Kconfig"
> +
> # clients
> comment "DMA Clients"
> depends on DMA_ENGINE
> diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
> index e60f813..2785756 100644
> --- a/drivers/dma/Makefile
> +++ b/drivers/dma/Makefile
> @@ -83,6 +83,7 @@ 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-$(CONFIG_AMD_PTDMA) += ptdma/
Please keep this sorted :(
> obj-y += mediatek/
> obj-y += qcom/
> diff --git a/drivers/dma/ptdma/Kconfig b/drivers/dma/ptdma/Kconfig
> new file mode 100644
> index 0000000..f93f9c2
> --- /dev/null
> +++ b/drivers/dma/ptdma/Kconfig
> @@ -0,0 +1,11 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +config AMD_PTDMA
> + tristate "AMD PassThru DMA Engine"
> + depends on X86_64 && PCI
> + help
> + Enable support for the AMD PTDMA controller. This controller
> + provides DMA capabilities & performs high bandwidth memory to
> + memory and IO copy operation and performs DMA transfer through
> + queue based descriptor management. This DMA controller is intended
> + to use with AMD Non-Transparent Bridge devices and not for general
> + purpose slave DMA.
s/slave/peripheral
> +static int pt_core_execute_cmd(struct ptdma_desc *desc,
> + struct pt_cmd_queue *cmd_q)
> +{
> + __le32 *mp;
> + u32 *dp;
> + u32 tail;
> + int i;
> +
> + if (desc->dw0.soc) {
> + desc->dw0.ioc = 1;
> + desc->dw0.soc = 0;
> + }
> + mutex_lock(&cmd_q->q_mutex);
> +
> + mp = (__le32 *)&cmd_q->qbase[cmd_q->qidx];
why not make the qidx __le32 instead of int ?
> + dp = (u32 *)desc;
> + for (i = 0; i < 8; i++)
why 8..?
> + mp[i] = cpu_to_le32(dp[i]); /* handle endianness */
Where is this used..? Why not drop this..
> +static irqreturn_t pt_core_irq_handler(int irq, void *data)
> +{
> + struct pt_device *pt = (struct pt_device *)data;
why do you need cast away from void *
> +int pt_core_init(struct pt_device *pt)
> +{
> + char dma_pool_name[MAX_DMAPOOL_NAME_LEN];
> + struct pt_cmd_queue *cmd_q = &pt->cmd_q;
> + u32 dma_addr_lo, dma_addr_hi;
> + struct device *dev = pt->dev;
> + struct dma_pool *dma_pool;
> + int ret;
> +
> + /* Allocate a dma pool for the queue */
> + snprintf(dma_pool_name, sizeof(dma_pool_name), "%s_q", pt->name);
> +
> + dma_pool = dma_pool_create(dma_pool_name, dev,
> + PT_DMAPOOL_MAX_SIZE,
> + PT_DMAPOOL_ALIGN, 0);
> + if (!dma_pool) {
> + dev_err(dev, "unable to allocate dma pool\n");
> + ret = -ENOMEM;
> + return ret;
return -ENOMEM?
> + }
> +
> + /* ptdma core initialisation */
> + iowrite32(CMD_CONFIG_VHB_EN, pt->io_regs + CMD_CONFIG_OFFSET);
> + iowrite32(CMD_QUEUE_PRIO, pt->io_regs + CMD_QUEUE_PRIO_OFFSET);
> + iowrite32(CMD_TIMEOUT_DISABLE, pt->io_regs + CMD_TIMEOUT_OFFSET);
> + iowrite32(CMD_CLK_GATE_CONFIG, pt->io_regs + CMD_CLK_GATE_CTL_OFFSET);
> + iowrite32(CMD_CONFIG_REQID, pt->io_regs + CMD_REQID_CONFIG_OFFSET);
> +
> + cmd_q->pt = pt;
> + cmd_q->dma_pool = dma_pool;
> + mutex_init(&cmd_q->q_mutex);
> +
> + /* Page alignment satisfies our needs for N <= 128 */
> + cmd_q->qsize = Q_SIZE(Q_DESC_SIZE);
> + cmd_q->qbase = dma_alloc_coherent(dev, cmd_q->qsize,
> + &cmd_q->qbase_dma,
> + GFP_KERNEL);
> + if (!cmd_q->qbase) {
> + dev_err(dev, "unable to allocate command queue\n");
> + ret = -ENOMEM;
> + goto e_dma_alloc;
> + }
> +
> + cmd_q->qidx = 0;
> +
> + /* Preset some register values */
> + cmd_q->reg_control = pt->io_regs + CMD_Q_STATUS_INCR;
> + pt_init_cmdq_regs(cmd_q);
> +
> + /* Turn off the queues and disable interrupts until ready */
> + pt_core_disable_queue_interrupts(pt);
> +
> + cmd_q->qcontrol = 0; /* Start with nothing */
> + iowrite32(cmd_q->qcontrol, cmd_q->reg_control);
> +
> + ioread32(cmd_q->reg_int_status);
> + ioread32(cmd_q->reg_status);
> +
> + /* Clear the interrupt status */
> + iowrite32(SUPPORTED_INTERRUPTS, cmd_q->reg_interrupt_status);
> +
> + /* Request an irq */
> + ret = request_irq(pt->pt_irq, pt_core_irq_handler, 0, pt->name, pt);
> + if (ret) {
> + dev_err(dev, "unable to allocate an IRQ\n");
Do log the error while you are it!
> + goto e_pool;
> + }
> +
> + /* Update the device registers with queue information. */
> +
> + cmd_q->qcontrol &= ~(CMD_Q_SIZE << CMD_Q_SHIFT);
> + cmd_q->qcontrol |= QUEUE_SIZE_VAL << CMD_Q_SHIFT;
Please use helpers in bitfield.h to do the shifts for you, no need to
define the shift values
> +#include <linux/device.h>
> +#include <linux/dma-mapping.h>
> +#include <linux/delay.h>
> +#include <linux/interrupt.h>
> +#include <linux/kernel.h>
> +#include <linux/kthread.h>
> +#include <linux/module.h>
> +#include <linux/pci_ids.h>
> +#include <linux/pci.h>
> +#include <linux/spinlock.h>
> +#include <linux/sched.h>
why do you need sched.h here?
> +
> +#include "ptdma.h"
> +
> +/* Ever-increasing value to produce unique unit numbers */
> +static atomic_t pt_ordinal;
What is the need of that?
> +static int pt_get_irqs(struct pt_device *pt)
> +{
> + struct device *dev = pt->dev;
> + int ret;
> +
> + ret = pt_get_msix_irqs(pt);
> + if (!ret)
> + return 0;
> +
> + /* Couldn't get MSI-X vectors, try MSI */
> + dev_dbg(dev, "could not enable MSI-X (%d), trying MSI\n", ret);
> + ret = pt_get_msi_irq(pt);
> + if (!ret)
> + return 0;
> +
> + /* Couldn't get MSI interrupt */
> + dev_dbg(dev, "could not enable MSI (%d)\n", ret);
Should this not be an error?
> +/*
> + * descriptor for PTDMA commands
> + * 8 32-bit words:
> + * word 0: function; engine; control bits
> + * word 1: length of source data
> + * word 2: low 32 bits of source pointer
> + * word 3: upper 16 bits of source pointer; source memory type
> + * word 4: low 32 bits of destination pointer
> + * word 5: upper 16 bits of destination pointer; destination memory type
> + * word 6: reserved 32 bits
> + * word 7: reserved 32 bits
> + */
> +
> +union dword0 {
> + struct {
> + unsigned int soc:1;
> + unsigned int ioc:1;
> + unsigned int rsvd1:1;
> + unsigned int init:1;
> + unsigned int eom:1;
> + unsigned int function:15;
> + unsigned int engine:4;
> + unsigned int prot:1;
> + unsigned int rsvd2:7;
> + };
Do you really want to use union and bitfields here, this seem like HW
description, best to use FIELD_PREP or FIELD_GET for these
Been there, done that would advise to avoid this approach.
--
~Vinod
