dmaengine: add Qualcomm Technologies HIDMA channel driver
authorSinan Kaya <okaya@codeaurora.org>
Fri, 5 Feb 2016 04:34:35 +0000 (23:34 -0500)
committerVinod Koul <vinod.koul@intel.com>
Fri, 11 Mar 2016 02:12:30 +0000 (07:42 +0530)
This patch adds support for hidma engine. The driver consists of two
logical blocks. The DMA engine interface and the low-level interface.
The hardware only supports memcpy/memset and this driver only support
memcpy interface. HW and driver doesn't support slave interface.

Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
Reviewed-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
drivers/dma/qcom/Kconfig
drivers/dma/qcom/hidma.c [new file with mode: 0644]
drivers/dma/qcom/hidma.h [new file with mode: 0644]

index c975b11..a7761c4 100644 (file)
@@ -17,3 +17,13 @@ config QCOM_HIDMA_MGMT
          start managing the channels. In a virtualized environment,
          the guest OS would run QCOM_HIDMA channel driver and the
          host would run the QCOM_HIDMA_MGMT management driver.
+
+config QCOM_HIDMA
+       tristate "Qualcomm Technologies HIDMA Channel support"
+       select DMA_ENGINE
+       help
+         Enable support for the Qualcomm Technologies HIDMA controller.
+         The HIDMA controller supports optimized buffer copies
+         (user to kernel, kernel to kernel, etc.).  It only supports
+         memcpy interface. The core is not intended for general
+         purpose slave DMA.
diff --git a/drivers/dma/qcom/hidma.c b/drivers/dma/qcom/hidma.c
new file mode 100644 (file)
index 0000000..cccc78e
--- /dev/null
@@ -0,0 +1,706 @@
+/*
+ * Qualcomm Technologies HIDMA DMA engine interface
+ *
+ * Copyright (c) 2015, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+/*
+ * Copyright (C) Freescale Semicondutor, Inc. 2007, 2008.
+ * Copyright (C) Semihalf 2009
+ * Copyright (C) Ilya Yanok, Emcraft Systems 2010
+ * Copyright (C) Alexander Popov, Promcontroller 2014
+ *
+ * Written by Piotr Ziecik <kosmo@semihalf.com>. Hardware description
+ * (defines, structures and comments) was taken from MPC5121 DMA driver
+ * written by Hongjun Chen <hong-jun.chen@freescale.com>.
+ *
+ * Approved as OSADL project by a majority of OSADL members and funded
+ * by OSADL membership fees in 2009;  for details see www.osadl.org.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * The full GNU General Public License is included in this distribution in the
+ * file called COPYING.
+ */
+
+/* Linux Foundation elects GPLv2 license only. */
+
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/of_dma.h>
+#include <linux/property.h>
+#include <linux/delay.h>
+#include <linux/acpi.h>
+#include <linux/irq.h>
+#include <linux/atomic.h>
+#include <linux/pm_runtime.h>
+
+#include "../dmaengine.h"
+#include "hidma.h"
+
+/*
+ * Default idle time is 2 seconds. This parameter can
+ * be overridden by changing the following
+ * /sys/bus/platform/devices/QCOM8061:<xy>/power/autosuspend_delay_ms
+ * during kernel boot.
+ */
+#define HIDMA_AUTOSUSPEND_TIMEOUT              2000
+#define HIDMA_ERR_INFO_SW                      0xFF
+#define HIDMA_ERR_CODE_UNEXPECTED_TERMINATE    0x0
+#define HIDMA_NR_DEFAULT_DESC                  10
+
+static inline struct hidma_dev *to_hidma_dev(struct dma_device *dmadev)
+{
+       return container_of(dmadev, struct hidma_dev, ddev);
+}
+
+static inline
+struct hidma_dev *to_hidma_dev_from_lldev(struct hidma_lldev **_lldevp)
+{
+       return container_of(_lldevp, struct hidma_dev, lldev);
+}
+
+static inline struct hidma_chan *to_hidma_chan(struct dma_chan *dmach)
+{
+       return container_of(dmach, struct hidma_chan, chan);
+}
+
+static inline
+struct hidma_desc *to_hidma_desc(struct dma_async_tx_descriptor *t)
+{
+       return container_of(t, struct hidma_desc, desc);
+}
+
+static void hidma_free(struct hidma_dev *dmadev)
+{
+       INIT_LIST_HEAD(&dmadev->ddev.channels);
+}
+
+static unsigned int nr_desc_prm;
+module_param(nr_desc_prm, uint, 0644);
+MODULE_PARM_DESC(nr_desc_prm, "number of descriptors (default: 0)");
+
+
+/* process completed descriptors */
+static void hidma_process_completed(struct hidma_chan *mchan)
+{
+       struct dma_device *ddev = mchan->chan.device;
+       struct hidma_dev *mdma = to_hidma_dev(ddev);
+       struct dma_async_tx_descriptor *desc;
+       dma_cookie_t last_cookie;
+       struct hidma_desc *mdesc;
+       unsigned long irqflags;
+       struct list_head list;
+
+       INIT_LIST_HEAD(&list);
+
+       /* Get all completed descriptors */
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       list_splice_tail_init(&mchan->completed, &list);
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       /* Execute callbacks and run dependencies */
+       list_for_each_entry(mdesc, &list, node) {
+               enum dma_status llstat;
+
+               desc = &mdesc->desc;
+
+               spin_lock_irqsave(&mchan->lock, irqflags);
+               dma_cookie_complete(desc);
+               spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+               llstat = hidma_ll_status(mdma->lldev, mdesc->tre_ch);
+               if (desc->callback && (llstat == DMA_COMPLETE))
+                       desc->callback(desc->callback_param);
+
+               last_cookie = desc->cookie;
+               dma_run_dependencies(desc);
+       }
+
+       /* Free descriptors */
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       list_splice_tail_init(&list, &mchan->free);
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+}
+
+/*
+ * Called once for each submitted descriptor.
+ * PM is locked once for each descriptor that is currently
+ * in execution.
+ */
+static void hidma_callback(void *data)
+{
+       struct hidma_desc *mdesc = data;
+       struct hidma_chan *mchan = to_hidma_chan(mdesc->desc.chan);
+       struct dma_device *ddev = mchan->chan.device;
+       struct hidma_dev *dmadev = to_hidma_dev(ddev);
+       unsigned long irqflags;
+       bool queued = false;
+
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       if (mdesc->node.next) {
+               /* Delete from the active list, add to completed list */
+               list_move_tail(&mdesc->node, &mchan->completed);
+               queued = true;
+
+               /* calculate the next running descriptor */
+               mchan->running = list_first_entry(&mchan->active,
+                                                 struct hidma_desc, node);
+       }
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       hidma_process_completed(mchan);
+
+       if (queued) {
+               pm_runtime_mark_last_busy(dmadev->ddev.dev);
+               pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       }
+}
+
+static int hidma_chan_init(struct hidma_dev *dmadev, u32 dma_sig)
+{
+       struct hidma_chan *mchan;
+       struct dma_device *ddev;
+
+       mchan = devm_kzalloc(dmadev->ddev.dev, sizeof(*mchan), GFP_KERNEL);
+       if (!mchan)
+               return -ENOMEM;
+
+       ddev = &dmadev->ddev;
+       mchan->dma_sig = dma_sig;
+       mchan->dmadev = dmadev;
+       mchan->chan.device = ddev;
+       dma_cookie_init(&mchan->chan);
+
+       INIT_LIST_HEAD(&mchan->free);
+       INIT_LIST_HEAD(&mchan->prepared);
+       INIT_LIST_HEAD(&mchan->active);
+       INIT_LIST_HEAD(&mchan->completed);
+
+       spin_lock_init(&mchan->lock);
+       list_add_tail(&mchan->chan.device_node, &ddev->channels);
+       dmadev->ddev.chancnt++;
+       return 0;
+}
+
+static void hidma_issue_task(unsigned long arg)
+{
+       struct hidma_dev *dmadev = (struct hidma_dev *)arg;
+
+       pm_runtime_get_sync(dmadev->ddev.dev);
+       hidma_ll_start(dmadev->lldev);
+}
+
+static void hidma_issue_pending(struct dma_chan *dmach)
+{
+       struct hidma_chan *mchan = to_hidma_chan(dmach);
+       struct hidma_dev *dmadev = mchan->dmadev;
+       unsigned long flags;
+       int status;
+
+       spin_lock_irqsave(&mchan->lock, flags);
+       if (!mchan->running) {
+               struct hidma_desc *desc = list_first_entry(&mchan->active,
+                                                          struct hidma_desc,
+                                                          node);
+               mchan->running = desc;
+       }
+       spin_unlock_irqrestore(&mchan->lock, flags);
+
+       /* PM will be released in hidma_callback function. */
+       status = pm_runtime_get(dmadev->ddev.dev);
+       if (status < 0)
+               tasklet_schedule(&dmadev->task);
+       else
+               hidma_ll_start(dmadev->lldev);
+}
+
+static enum dma_status hidma_tx_status(struct dma_chan *dmach,
+                                      dma_cookie_t cookie,
+                                      struct dma_tx_state *txstate)
+{
+       struct hidma_chan *mchan = to_hidma_chan(dmach);
+       enum dma_status ret;
+
+       ret = dma_cookie_status(dmach, cookie, txstate);
+       if (ret == DMA_COMPLETE)
+               return ret;
+
+       if (mchan->paused && (ret == DMA_IN_PROGRESS)) {
+               unsigned long flags;
+               dma_cookie_t runcookie;
+
+               spin_lock_irqsave(&mchan->lock, flags);
+               if (mchan->running)
+                       runcookie = mchan->running->desc.cookie;
+               else
+                       runcookie = -EINVAL;
+
+               if (runcookie == cookie)
+                       ret = DMA_PAUSED;
+
+               spin_unlock_irqrestore(&mchan->lock, flags);
+       }
+
+       return ret;
+}
+
+/*
+ * Submit descriptor to hardware.
+ * Lock the PM for each descriptor we are sending.
+ */
+static dma_cookie_t hidma_tx_submit(struct dma_async_tx_descriptor *txd)
+{
+       struct hidma_chan *mchan = to_hidma_chan(txd->chan);
+       struct hidma_dev *dmadev = mchan->dmadev;
+       struct hidma_desc *mdesc;
+       unsigned long irqflags;
+       dma_cookie_t cookie;
+
+       pm_runtime_get_sync(dmadev->ddev.dev);
+       if (!hidma_ll_isenabled(dmadev->lldev)) {
+               pm_runtime_mark_last_busy(dmadev->ddev.dev);
+               pm_runtime_put_autosuspend(dmadev->ddev.dev);
+               return -ENODEV;
+       }
+
+       mdesc = container_of(txd, struct hidma_desc, desc);
+       spin_lock_irqsave(&mchan->lock, irqflags);
+
+       /* Move descriptor to active */
+       list_move_tail(&mdesc->node, &mchan->active);
+
+       /* Update cookie */
+       cookie = dma_cookie_assign(txd);
+
+       hidma_ll_queue_request(dmadev->lldev, mdesc->tre_ch);
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       return cookie;
+}
+
+static int hidma_alloc_chan_resources(struct dma_chan *dmach)
+{
+       struct hidma_chan *mchan = to_hidma_chan(dmach);
+       struct hidma_dev *dmadev = mchan->dmadev;
+       struct hidma_desc *mdesc, *tmp;
+       unsigned long irqflags;
+       LIST_HEAD(descs);
+       unsigned int i;
+       int rc = 0;
+
+       if (mchan->allocated)
+               return 0;
+
+       /* Alloc descriptors for this channel */
+       for (i = 0; i < dmadev->nr_descriptors; i++) {
+               mdesc = kzalloc(sizeof(struct hidma_desc), GFP_NOWAIT);
+               if (!mdesc) {
+                       rc = -ENOMEM;
+                       break;
+               }
+               dma_async_tx_descriptor_init(&mdesc->desc, dmach);
+               mdesc->desc.tx_submit = hidma_tx_submit;
+
+               rc = hidma_ll_request(dmadev->lldev, mchan->dma_sig,
+                                     "DMA engine", hidma_callback, mdesc,
+                                     &mdesc->tre_ch);
+               if (rc) {
+                       dev_err(dmach->device->dev,
+                               "channel alloc failed at %u\n", i);
+                       kfree(mdesc);
+                       break;
+               }
+               list_add_tail(&mdesc->node, &descs);
+       }
+
+       if (rc) {
+               /* return the allocated descriptors */
+               list_for_each_entry_safe(mdesc, tmp, &descs, node) {
+                       hidma_ll_free(dmadev->lldev, mdesc->tre_ch);
+                       kfree(mdesc);
+               }
+               return rc;
+       }
+
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       list_splice_tail_init(&descs, &mchan->free);
+       mchan->allocated = true;
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+       return 1;
+}
+
+static struct dma_async_tx_descriptor *
+hidma_prep_dma_memcpy(struct dma_chan *dmach, dma_addr_t dest, dma_addr_t src,
+               size_t len, unsigned long flags)
+{
+       struct hidma_chan *mchan = to_hidma_chan(dmach);
+       struct hidma_desc *mdesc = NULL;
+       struct hidma_dev *mdma = mchan->dmadev;
+       unsigned long irqflags;
+
+       /* Get free descriptor */
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       if (!list_empty(&mchan->free)) {
+               mdesc = list_first_entry(&mchan->free, struct hidma_desc, node);
+               list_del(&mdesc->node);
+       }
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       if (!mdesc)
+               return NULL;
+
+       hidma_ll_set_transfer_params(mdma->lldev, mdesc->tre_ch,
+                                    src, dest, len, flags);
+
+       /* Place descriptor in prepared list */
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       list_add_tail(&mdesc->node, &mchan->prepared);
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       return &mdesc->desc;
+}
+
+static int hidma_terminate_channel(struct dma_chan *chan)
+{
+       struct hidma_chan *mchan = to_hidma_chan(chan);
+       struct hidma_dev *dmadev = to_hidma_dev(mchan->chan.device);
+       struct hidma_desc *tmp, *mdesc;
+       unsigned long irqflags;
+       LIST_HEAD(list);
+       int rc;
+
+       pm_runtime_get_sync(dmadev->ddev.dev);
+       /* give completed requests a chance to finish */
+       hidma_process_completed(mchan);
+
+       spin_lock_irqsave(&mchan->lock, irqflags);
+       list_splice_init(&mchan->active, &list);
+       list_splice_init(&mchan->prepared, &list);
+       list_splice_init(&mchan->completed, &list);
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+
+       /* this suspends the existing transfer */
+       rc = hidma_ll_pause(dmadev->lldev);
+       if (rc) {
+               dev_err(dmadev->ddev.dev, "channel did not pause\n");
+               goto out;
+       }
+
+       /* return all user requests */
+       list_for_each_entry_safe(mdesc, tmp, &list, node) {
+               struct dma_async_tx_descriptor *txd = &mdesc->desc;
+               dma_async_tx_callback callback = mdesc->desc.callback;
+               void *param = mdesc->desc.callback_param;
+
+               dma_descriptor_unmap(txd);
+
+               if (callback)
+                       callback(param);
+
+               dma_run_dependencies(txd);
+
+               /* move myself to free_list */
+               list_move(&mdesc->node, &mchan->free);
+       }
+
+       rc = hidma_ll_resume(dmadev->lldev);
+out:
+       pm_runtime_mark_last_busy(dmadev->ddev.dev);
+       pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       return rc;
+}
+
+static int hidma_terminate_all(struct dma_chan *chan)
+{
+       struct hidma_chan *mchan = to_hidma_chan(chan);
+       struct hidma_dev *dmadev = to_hidma_dev(mchan->chan.device);
+       int rc;
+
+       rc = hidma_terminate_channel(chan);
+       if (rc)
+               return rc;
+
+       /* reinitialize the hardware */
+       pm_runtime_get_sync(dmadev->ddev.dev);
+       rc = hidma_ll_setup(dmadev->lldev);
+       pm_runtime_mark_last_busy(dmadev->ddev.dev);
+       pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       return rc;
+}
+
+static void hidma_free_chan_resources(struct dma_chan *dmach)
+{
+       struct hidma_chan *mchan = to_hidma_chan(dmach);
+       struct hidma_dev *mdma = mchan->dmadev;
+       struct hidma_desc *mdesc, *tmp;
+       unsigned long irqflags;
+       LIST_HEAD(descs);
+
+       /* terminate running transactions and free descriptors */
+       hidma_terminate_channel(dmach);
+
+       spin_lock_irqsave(&mchan->lock, irqflags);
+
+       /* Move data */
+       list_splice_tail_init(&mchan->free, &descs);
+
+       /* Free descriptors */
+       list_for_each_entry_safe(mdesc, tmp, &descs, node) {
+               hidma_ll_free(mdma->lldev, mdesc->tre_ch);
+               list_del(&mdesc->node);
+               kfree(mdesc);
+       }
+
+       mchan->allocated = 0;
+       spin_unlock_irqrestore(&mchan->lock, irqflags);
+}
+
+static int hidma_pause(struct dma_chan *chan)
+{
+       struct hidma_chan *mchan;
+       struct hidma_dev *dmadev;
+
+       mchan = to_hidma_chan(chan);
+       dmadev = to_hidma_dev(mchan->chan.device);
+       if (!mchan->paused) {
+               pm_runtime_get_sync(dmadev->ddev.dev);
+               if (hidma_ll_pause(dmadev->lldev))
+                       dev_warn(dmadev->ddev.dev, "channel did not stop\n");
+               mchan->paused = true;
+               pm_runtime_mark_last_busy(dmadev->ddev.dev);
+               pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       }
+       return 0;
+}
+
+static int hidma_resume(struct dma_chan *chan)
+{
+       struct hidma_chan *mchan;
+       struct hidma_dev *dmadev;
+       int rc = 0;
+
+       mchan = to_hidma_chan(chan);
+       dmadev = to_hidma_dev(mchan->chan.device);
+       if (mchan->paused) {
+               pm_runtime_get_sync(dmadev->ddev.dev);
+               rc = hidma_ll_resume(dmadev->lldev);
+               if (!rc)
+                       mchan->paused = false;
+               else
+                       dev_err(dmadev->ddev.dev,
+                               "failed to resume the channel");
+               pm_runtime_mark_last_busy(dmadev->ddev.dev);
+               pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       }
+       return rc;
+}
+
+static irqreturn_t hidma_chirq_handler(int chirq, void *arg)
+{
+       struct hidma_lldev *lldev = arg;
+
+       /*
+        * All interrupts are request driven.
+        * HW doesn't send an interrupt by itself.
+        */
+       return hidma_ll_inthandler(chirq, lldev);
+}
+
+static int hidma_probe(struct platform_device *pdev)
+{
+       struct hidma_dev *dmadev;
+       struct resource *trca_resource;
+       struct resource *evca_resource;
+       int chirq;
+       void __iomem *evca;
+       void __iomem *trca;
+       int rc;
+
+       pm_runtime_set_autosuspend_delay(&pdev->dev, HIDMA_AUTOSUSPEND_TIMEOUT);
+       pm_runtime_use_autosuspend(&pdev->dev);
+       pm_runtime_set_active(&pdev->dev);
+       pm_runtime_enable(&pdev->dev);
+
+       trca_resource = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       trca = devm_ioremap_resource(&pdev->dev, trca_resource);
+       if (IS_ERR(trca)) {
+               rc = -ENOMEM;
+               goto bailout;
+       }
+
+       evca_resource = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+       evca = devm_ioremap_resource(&pdev->dev, evca_resource);
+       if (IS_ERR(evca)) {
+               rc = -ENOMEM;
+               goto bailout;
+       }
+
+       /*
+        * This driver only handles the channel IRQs.
+        * Common IRQ is handled by the management driver.
+        */
+       chirq = platform_get_irq(pdev, 0);
+       if (chirq < 0) {
+               rc = -ENODEV;
+               goto bailout;
+       }
+
+       dmadev = devm_kzalloc(&pdev->dev, sizeof(*dmadev), GFP_KERNEL);
+       if (!dmadev) {
+               rc = -ENOMEM;
+               goto bailout;
+       }
+
+       INIT_LIST_HEAD(&dmadev->ddev.channels);
+       spin_lock_init(&dmadev->lock);
+       dmadev->ddev.dev = &pdev->dev;
+       pm_runtime_get_sync(dmadev->ddev.dev);
+
+       dma_cap_set(DMA_MEMCPY, dmadev->ddev.cap_mask);
+       if (WARN_ON(!pdev->dev.dma_mask)) {
+               rc = -ENXIO;
+               goto dmafree;
+       }
+
+       dmadev->dev_evca = evca;
+       dmadev->evca_resource = evca_resource;
+       dmadev->dev_trca = trca;
+       dmadev->trca_resource = trca_resource;
+       dmadev->ddev.device_prep_dma_memcpy = hidma_prep_dma_memcpy;
+       dmadev->ddev.device_alloc_chan_resources = hidma_alloc_chan_resources;
+       dmadev->ddev.device_free_chan_resources = hidma_free_chan_resources;
+       dmadev->ddev.device_tx_status = hidma_tx_status;
+       dmadev->ddev.device_issue_pending = hidma_issue_pending;
+       dmadev->ddev.device_pause = hidma_pause;
+       dmadev->ddev.device_resume = hidma_resume;
+       dmadev->ddev.device_terminate_all = hidma_terminate_all;
+       dmadev->ddev.copy_align = 8;
+
+       device_property_read_u32(&pdev->dev, "desc-count",
+                                &dmadev->nr_descriptors);
+
+       if (!dmadev->nr_descriptors && nr_desc_prm)
+               dmadev->nr_descriptors = nr_desc_prm;
+
+       if (!dmadev->nr_descriptors)
+               dmadev->nr_descriptors = HIDMA_NR_DEFAULT_DESC;
+
+       dmadev->chidx = readl(dmadev->dev_trca + 0x28);
+
+       /* Set DMA mask to 64 bits. */
+       rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
+       if (rc) {
+               dev_warn(&pdev->dev, "unable to set coherent mask to 64");
+               rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
+               if (rc)
+                       goto dmafree;
+       }
+
+       dmadev->lldev = hidma_ll_init(dmadev->ddev.dev,
+                                     dmadev->nr_descriptors, dmadev->dev_trca,
+                                     dmadev->dev_evca, dmadev->chidx);
+       if (!dmadev->lldev) {
+               rc = -EPROBE_DEFER;
+               goto dmafree;
+       }
+
+       rc = devm_request_irq(&pdev->dev, chirq, hidma_chirq_handler, 0,
+                             "qcom-hidma", dmadev->lldev);
+       if (rc)
+               goto uninit;
+
+       INIT_LIST_HEAD(&dmadev->ddev.channels);
+       rc = hidma_chan_init(dmadev, 0);
+       if (rc)
+               goto uninit;
+
+       rc = dma_async_device_register(&dmadev->ddev);
+       if (rc)
+               goto uninit;
+
+       dmadev->irq = chirq;
+       tasklet_init(&dmadev->task, hidma_issue_task, (unsigned long)dmadev);
+       dev_info(&pdev->dev, "HI-DMA engine driver registration complete\n");
+       platform_set_drvdata(pdev, dmadev);
+       pm_runtime_mark_last_busy(dmadev->ddev.dev);
+       pm_runtime_put_autosuspend(dmadev->ddev.dev);
+       return 0;
+
+uninit:
+       hidma_ll_uninit(dmadev->lldev);
+dmafree:
+       if (dmadev)
+               hidma_free(dmadev);
+bailout:
+       pm_runtime_put_sync(&pdev->dev);
+       pm_runtime_disable(&pdev->dev);
+       return rc;
+}
+
+static int hidma_remove(struct platform_device *pdev)
+{
+       struct hidma_dev *dmadev = platform_get_drvdata(pdev);
+
+       pm_runtime_get_sync(dmadev->ddev.dev);
+       dma_async_device_unregister(&dmadev->ddev);
+       devm_free_irq(dmadev->ddev.dev, dmadev->irq, dmadev->lldev);
+       hidma_ll_uninit(dmadev->lldev);
+       hidma_free(dmadev);
+
+       dev_info(&pdev->dev, "HI-DMA engine removed\n");
+       pm_runtime_put_sync_suspend(&pdev->dev);
+       pm_runtime_disable(&pdev->dev);
+
+       return 0;
+}
+
+#if IS_ENABLED(CONFIG_ACPI)
+static const struct acpi_device_id hidma_acpi_ids[] = {
+       {"QCOM8061"},
+       {},
+};
+#endif
+
+static const struct of_device_id hidma_match[] = {
+       {.compatible = "qcom,hidma-1.0",},
+       {},
+};
+
+MODULE_DEVICE_TABLE(of, hidma_match);
+
+static struct platform_driver hidma_driver = {
+       .probe = hidma_probe,
+       .remove = hidma_remove,
+       .driver = {
+                  .name = "hidma",
+                  .of_match_table = hidma_match,
+                  .acpi_match_table = ACPI_PTR(hidma_acpi_ids),
+       },
+};
+
+module_platform_driver(hidma_driver);
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/dma/qcom/hidma.h b/drivers/dma/qcom/hidma.h
new file mode 100644 (file)
index 0000000..231e306
--- /dev/null
@@ -0,0 +1,160 @@
+/*
+ * Qualcomm Technologies HIDMA data structures
+ *
+ * Copyright (c) 2014, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#ifndef QCOM_HIDMA_H
+#define QCOM_HIDMA_H
+
+#include <linux/kfifo.h>
+#include <linux/interrupt.h>
+#include <linux/dmaengine.h>
+
+#define TRE_SIZE                       32 /* each TRE is 32 bytes  */
+#define TRE_CFG_IDX                    0
+#define TRE_LEN_IDX                    1
+#define TRE_SRC_LOW_IDX                2
+#define TRE_SRC_HI_IDX                 3
+#define TRE_DEST_LOW_IDX               4
+#define TRE_DEST_HI_IDX                5
+
+struct hidma_tx_status {
+       u8 err_info;                    /* error record in this transfer    */
+       u8 err_code;                    /* completion code                  */
+};
+
+struct hidma_tre {
+       atomic_t allocated;             /* if this channel is allocated     */
+       bool queued;                    /* flag whether this is pending     */
+       u16 status;                     /* status                           */
+       u32 chidx;                      /* index of the tre                 */
+       u32 dma_sig;                    /* signature of the tre             */
+       const char *dev_name;           /* name of the device               */
+       void (*callback)(void *data);   /* requester callback               */
+       void *data;                     /* Data associated with this channel*/
+       struct hidma_lldev *lldev;      /* lldma device pointer             */
+       u32 tre_local[TRE_SIZE / sizeof(u32) + 1]; /* TRE local copy        */
+       u32 tre_index;                  /* the offset where this was written*/
+       u32 int_flags;                  /* interrupt flags                  */
+};
+
+struct hidma_lldev {
+       bool initialized;               /* initialized flag               */
+       u8 trch_state;                  /* trch_state of the device       */
+       u8 evch_state;                  /* evch_state of the device       */
+       u8 chidx;                       /* channel index in the core      */
+       u32 nr_tres;                    /* max number of configs          */
+       spinlock_t lock;                /* reentrancy                     */
+       struct hidma_tre *trepool;      /* trepool of user configs */
+       struct device *dev;             /* device                         */
+       void __iomem *trca;             /* Transfer Channel address       */
+       void __iomem *evca;             /* Event Channel address          */
+       struct hidma_tre
+               **pending_tre_list;     /* Pointers to pending TREs       */
+       struct hidma_tx_status
+               *tx_status_list;        /* Pointers to pending TREs status*/
+       s32 pending_tre_count;          /* Number of TREs pending         */
+
+       void *tre_ring;                 /* TRE ring                       */
+       dma_addr_t tre_ring_handle;     /* TRE ring to be shared with HW  */
+       u32 tre_ring_size;              /* Byte size of the ring          */
+       u32 tre_processed_off;          /* last processed TRE             */
+
+       void *evre_ring;                /* EVRE ring                       */
+       dma_addr_t evre_ring_handle;    /* EVRE ring to be shared with HW  */
+       u32 evre_ring_size;             /* Byte size of the ring           */
+       u32 evre_processed_off;         /* last processed EVRE             */
+
+       u32 tre_write_offset;           /* TRE write location              */
+       struct tasklet_struct task;     /* task delivering notifications   */
+       DECLARE_KFIFO_PTR(handoff_fifo,
+               struct hidma_tre *);    /* pending TREs FIFO               */
+};
+
+struct hidma_desc {
+       struct dma_async_tx_descriptor  desc;
+       /* link list node for this channel*/
+       struct list_head                node;
+       u32                             tre_ch;
+};
+
+struct hidma_chan {
+       bool                            paused;
+       bool                            allocated;
+       char                            dbg_name[16];
+       u32                             dma_sig;
+
+       /*
+        * active descriptor on this channel
+        * It is used by the DMA complete notification to
+        * locate the descriptor that initiated the transfer.
+        */
+       struct dentry                   *debugfs;
+       struct dentry                   *stats;
+       struct hidma_dev                *dmadev;
+       struct hidma_desc               *running;
+
+       struct dma_chan                 chan;
+       struct list_head                free;
+       struct list_head                prepared;
+       struct list_head                active;
+       struct list_head                completed;
+
+       /* Lock for this structure */
+       spinlock_t                      lock;
+};
+
+struct hidma_dev {
+       int                             irq;
+       int                             chidx;
+       u32                             nr_descriptors;
+
+       struct hidma_lldev              *lldev;
+       void                            __iomem *dev_trca;
+       struct resource                 *trca_resource;
+       void                            __iomem *dev_evca;
+       struct resource                 *evca_resource;
+
+       /* used to protect the pending channel list*/
+       spinlock_t                      lock;
+       struct dma_device               ddev;
+
+       struct dentry                   *debugfs;
+       struct dentry                   *stats;
+
+       /* Task delivering issue_pending */
+       struct tasklet_struct           task;
+};
+
+int hidma_ll_request(struct hidma_lldev *llhndl, u32 dev_id,
+                       const char *dev_name,
+                       void (*callback)(void *data), void *data, u32 *tre_ch);
+
+void hidma_ll_free(struct hidma_lldev *llhndl, u32 tre_ch);
+enum dma_status hidma_ll_status(struct hidma_lldev *llhndl, u32 tre_ch);
+bool hidma_ll_isenabled(struct hidma_lldev *llhndl);
+void hidma_ll_queue_request(struct hidma_lldev *llhndl, u32 tre_ch);
+void hidma_ll_start(struct hidma_lldev *llhndl);
+int hidma_ll_pause(struct hidma_lldev *llhndl);
+int hidma_ll_resume(struct hidma_lldev *llhndl);
+void hidma_ll_set_transfer_params(struct hidma_lldev *llhndl, u32 tre_ch,
+       dma_addr_t src, dma_addr_t dest, u32 len, u32 flags);
+int hidma_ll_setup(struct hidma_lldev *lldev);
+struct hidma_lldev *hidma_ll_init(struct device *dev, u32 max_channels,
+                       void __iomem *trca, void __iomem *evca,
+                       u8 chidx);
+int hidma_ll_uninit(struct hidma_lldev *llhndl);
+irqreturn_t hidma_ll_inthandler(int irq, void *arg);
+void hidma_cleanup_pending_tre(struct hidma_lldev *llhndl, u8 err_info,
+                               u8 err_code);
+#endif