[PATCH RFC v2 11/14] dmaengine: idxd: connect idxd to dmaengine subsystem

From: Dave Jiang
Date: Thu Dec 12 2019 - 13:25:20 EST

Add plumbing for dmaengine subsystem connection. The driver register a DMA
device per DSA device. The channels are dynamically registered when a
workqueue is configured to be "kernel:dmanegine" type. The driver will
utilize the newly introduced DMA request API calls to provide a lockless
descriptor submission path.

Signed-off-by: Dave Jiang <dave.jiang@xxxxxxxxx>
---
drivers/dma/Kconfig | 1
drivers/dma/idxd/Makefile | 2 -
drivers/dma/idxd/device.c | 2 +
drivers/dma/idxd/dma.c | 119 +++++++++++++++++++++++++++++++++++++++++++++
drivers/dma/idxd/idxd.h | 14 +++++
drivers/dma/idxd/init.c | 48 ++++++++++++++++++
drivers/dma/idxd/irq.c | 101 ++++++++++++++++++++++++++++++++++++++
drivers/dma/idxd/submit.c | 51 +++++++++++++++++++
drivers/dma/idxd/sysfs.c | 28 +++++++++++
9 files changed, 364 insertions(+), 2 deletions(-)
create mode 100644 drivers/dma/idxd/dma.c

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index d6df373d2607..d1474c7c5151 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -282,6 +282,7 @@ config INTEL_IDXD
tristate "Intel Data Accelerators support"
depends on PCI && X86_64
select DMA_ENGINE
+ select DMA_ENGINE_REQUEST
select SBITMAP
help
Enable support for the Intel(R) data accelerators present
diff --git a/drivers/dma/idxd/Makefile b/drivers/dma/idxd/Makefile
index 50eca12015e2..a036ba0e77d2 100644
--- a/drivers/dma/idxd/Makefile
+++ b/drivers/dma/idxd/Makefile
@@ -1,2 +1,2 @@
obj-$(CONFIG_INTEL_IDXD) += idxd.o
-idxd-y := init.o irq.o device.o sysfs.o submit.o
+idxd-y := init.o irq.o device.o sysfs.o submit.o dma.o
diff --git a/drivers/dma/idxd/device.c b/drivers/dma/idxd/device.c
index 74a60a8bef76..49638d3a2151 100644
--- a/drivers/dma/idxd/device.c
+++ b/drivers/dma/idxd/device.c
@@ -5,7 +5,9 @@
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/io-64-nonatomic-lo-hi.h>
+#include <linux/dmaengine.h>
#include <uapi/linux/idxd.h>
+#include "../dmaengine.h"
#include "idxd.h"
#include "registers.h"

diff --git a/drivers/dma/idxd/dma.c b/drivers/dma/idxd/dma.c
new file mode 100644
index 000000000000..07fbc98668ae
--- /dev/null
+++ b/drivers/dma/idxd/dma.c
@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2019 Intel Corporation. All rights rsvd. */
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/device.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
+#include <linux/dmaengine.h>
+#include <uapi/linux/idxd.h>
+#include "../dmaengine.h"
+#include "registers.h"
+#include "idxd.h"
+
+void idxd_parse_completion_status(u8 status, enum dmaengine_tx_result *res)
+{
+ u8 code = status & DSA_COMP_STATUS_MASK;
+
+ switch (code) {
+ case DSA_COMP_SUCCESS:
+ *res = DMA_TRANS_NOERROR;
+ break;
+ case DSA_COMP_HW_ERR1:
+ *res = DMA_TRANS_READ_FAILED;
+ break;
+ default:
+ *res = DMA_TRANS_ERROR;
+ break;
+ }
+}
+
+static int idxd_dma_submit_request(struct dma_chan *chan,
+ struct dma_request *req)
+{
+ struct idxd_wq *wq = container_of(chan, struct idxd_wq, dma_chan);
+
+ if (req->cmd == DMA_MEMCPY)
+ return idxd_submit_memcpy(wq, req);
+
+ return -EINVAL;
+}
+
+static int idxd_dma_alloc_chan_resources(struct dma_chan *chan)
+{
+ struct idxd_wq *wq = container_of(chan, struct idxd_wq, dma_chan);
+ struct device *dev = &wq->idxd->pdev->dev;
+
+ idxd_wq_get(wq);
+ dev_dbg(dev, "%s: client_count: %d\n", __func__, idxd_wq_refcount(wq));
+ return 0;
+}
+
+static void idxd_dma_free_chan_resources(struct dma_chan *chan)
+{
+ struct idxd_wq *wq = container_of(chan, struct idxd_wq, dma_chan);
+ struct device *dev = &wq->idxd->pdev->dev;
+
+ idxd_wq_put(wq);
+ dev_dbg(dev, "%s: client_count: %d\n", __func__, idxd_wq_refcount(wq));
+}
+
+int idxd_register_dma_device(struct idxd_device *idxd)
+{
+ struct dma_device *dma = &idxd->dma_dev;
+
+ INIT_LIST_HEAD(&dma->channels);
+ dma->dev = &idxd->pdev->dev;
+
+ if (idxd->hw.opcap.bits[0] & IDXD_OPCAP_MEMMOVE)
+ dma_cap_set(DMA_MEMCPY, dma->cap_mask);
+
+ if (idxd->hw.opcap.bits[0] & IDXD_OPCAP_NOOP)
+ dma_cap_set(DMA_INTERRUPT, dma->cap_mask);
+
+ dma->device_submit_request = idxd_dma_submit_request;
+ dma->device_alloc_chan_resources = idxd_dma_alloc_chan_resources;
+ dma->device_free_chan_resources = idxd_dma_free_chan_resources;
+
+ return dma_async_request_device_register(&idxd->dma_dev);
+}
+
+void idxd_unregister_dma_device(struct idxd_device *idxd)
+{
+ dma_async_device_unregister(&idxd->dma_dev);
+}
+
+int idxd_register_dma_channel(struct idxd_wq *wq)
+{
+ struct idxd_device *idxd = wq->idxd;
+ struct dma_device *dma = &idxd->dma_dev;
+ struct dma_chan *chan = &wq->dma_chan;
+ struct idxd_group *group = wq->group;
+ int rc;
+
+ memset(&wq->dma_chan, 0, sizeof(struct dma_chan));
+ chan->device = dma;
+ list_add_tail(&chan->device_node, &dma->channels);
+ chan->max_sgs = wq->batch_size;
+ chan->depth = wq->size +
+ idxd->hw.gen_cap.max_descs_per_engine * group->num_engines;
+
+ rc = dma_async_device_channel_register(dma, chan);
+ if (rc < 0)
+ return rc;
+
+ rc = dma_chan_alloc_request_resources(chan);
+ if (rc < 0) {
+ dma_async_device_channel_unregister(dma, chan);
+ return rc;
+ }
+
+ return 0;
+}
+
+void idxd_unregister_dma_channel(struct idxd_wq *wq)
+{
+ dma_chan_free_request_resources(&wq->dma_chan);
+ dma_async_device_channel_unregister(&wq->idxd->dma_dev, &wq->dma_chan);
+}
diff --git a/drivers/dma/idxd/idxd.h b/drivers/dma/idxd/idxd.h
index 654351105d85..7f94e50267cc 100644
--- a/drivers/dma/idxd/idxd.h
+++ b/drivers/dma/idxd/idxd.h
@@ -4,6 +4,7 @@
#define _IDXD_H_

#include <linux/sbitmap.h>
+#include <linux/dmaengine.h>
#include <linux/percpu-rwsem.h>
#include <linux/wait.h>
#include "registers.h"
@@ -96,6 +97,7 @@ struct idxd_wq {
int batches_size;
int batch_size;
struct sbitmap sbmap;
+ struct dma_chan dma_chan;
struct percpu_rw_semaphore submit_lock;
wait_queue_head_t submit_waitq;
char name[WQ_NAME_SIZE + 1];
@@ -167,6 +169,8 @@ struct idxd_device {
struct msix_entry *msix_entries;
int num_wq_irqs;
struct idxd_irq_entry *irq_entries;
+
+ struct dma_device dma_dev;
};

/* IDXD software descriptor */
@@ -181,6 +185,7 @@ struct idxd_desc {
struct list_head list;
int id;
struct idxd_wq *wq;
+ struct dma_request *req;
};

#define confdev_to_idxd(dev) container_of(dev, struct idxd_device, conf_dev)
@@ -252,4 +257,13 @@ int idxd_wq_disable(struct idxd_wq *wq);
int idxd_wq_map_portal(struct idxd_wq *wq);
void idxd_wq_unmap_portal(struct idxd_wq *wq);

+/* submission */
+int idxd_submit_memcpy(struct idxd_wq *wq, struct dma_request *req);
+
+/* dmaengine */
+int idxd_register_dma_device(struct idxd_device *idxd);
+void idxd_unregister_dma_device(struct idxd_device *idxd);
+int idxd_register_dma_channel(struct idxd_wq *wq);
+void idxd_unregister_dma_channel(struct idxd_wq *wq);
+void idxd_parse_completion_status(u8 status, enum dmaengine_tx_result *res);
#endif
diff --git a/drivers/dma/idxd/init.c b/drivers/dma/idxd/init.c
index bec686cec27b..3ec1ef0f733e 100644
--- a/drivers/dma/idxd/init.c
+++ b/drivers/dma/idxd/init.c
@@ -15,6 +15,8 @@
#include <linux/device.h>
#include <linux/idr.h>
#include <uapi/linux/idxd.h>
+#include <linux/dmaengine.h>
+#include "../dmaengine.h"
#include "registers.h"
#include "idxd.h"

@@ -395,6 +397,50 @@ static int idxd_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
return 0;
}

+static void idxd_flush_pending_llist(struct idxd_irq_entry *ie)
+{
+ struct idxd_desc *desc, *itr;
+ struct llist_node *head;
+ struct dma_request *req;
+
+ head = llist_del_all(&ie->pending_llist);
+ if (!head)
+ return;
+
+ llist_for_each_entry_safe(desc, itr, head, llnode) {
+ req = desc->req;
+ if (!desc->completion->status)
+ req->result.result = DMA_TRANS_ABORTED;
+ else if (desc->completion->status == DSA_COMP_SUCCESS)
+ req->result.result = DMA_TRANS_NOERROR;
+ else
+ req->result.result = DMA_TRANS_ERROR;
+
+ dmaengine_request_complete(req);
+ idxd_free_desc(desc->wq, desc);
+ }
+}
+
+static void idxd_flush_work_list(struct idxd_irq_entry *ie)
+{
+ struct idxd_desc *desc, *iter;
+ struct dma_request *req;
+
+ list_for_each_entry_safe(desc, iter, &ie->work_list, list) {
+ req = desc->req;
+ list_del(&desc->list);
+ if (!desc->completion->status)
+ req->result.result = DMA_TRANS_ABORTED;
+ else if (desc->completion->status == DSA_COMP_SUCCESS)
+ req->result.result = DMA_TRANS_NOERROR;
+ else
+ req->result.result = DMA_TRANS_ERROR;
+
+ dmaengine_request_complete(req);
+ idxd_free_desc(desc->wq, desc);
+ }
+}
+
static void idxd_shutdown(struct pci_dev *pdev)
{
struct idxd_device *idxd = pci_get_drvdata(pdev);
@@ -418,6 +464,8 @@ static void idxd_shutdown(struct pci_dev *pdev)
synchronize_irq(idxd->msix_entries[i].vector);
if (i == 0)
continue;
+ idxd_flush_pending_llist(irq_entry);
+ idxd_flush_work_list(irq_entry);
}
}

diff --git a/drivers/dma/idxd/irq.c b/drivers/dma/idxd/irq.c
index de4b80973c2f..b4adeb2817d1 100644
--- a/drivers/dma/idxd/irq.c
+++ b/drivers/dma/idxd/irq.c
@@ -5,7 +5,9 @@
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/io-64-nonatomic-lo-hi.h>
+#include <linux/dmaengine.h>
#include <uapi/linux/idxd.h>
+#include "../dmaengine.h"
#include "idxd.h"
#include "registers.h"

@@ -146,11 +148,110 @@ irqreturn_t idxd_misc_thread(int vec, void *data)
return IRQ_HANDLED;
}

+static int irq_process_pending_llist(struct idxd_irq_entry *irq_entry,
+ int *processed)
+{
+ struct idxd_desc *desc, *t;
+ struct llist_node *head;
+ int queued = 0;
+ struct dma_request *req;
+
+ head = llist_del_all(&irq_entry->pending_llist);
+ if (!head)
+ return 0;
+
+ llist_for_each_entry_safe(desc, t, head, llnode) {
+ req = desc->req;
+ if (desc->completion->status) {
+ if ((desc->completion->status & DSA_COMP_STATUS_MASK) !=
+ DSA_COMP_SUCCESS)
+ idxd_parse_completion_status(desc->completion->status,
+ &req->result.result);
+
+ dmaengine_request_complete(req);
+ idxd_free_desc(desc->wq, desc);
+ (*processed)++;
+ } else {
+ list_add_tail(&desc->list, &irq_entry->work_list);
+ queued++;
+ }
+ }
+
+ return queued;
+}
+
+static int irq_process_work_list(struct idxd_irq_entry *irq_entry,
+ int *processed)
+{
+ struct list_head *node, *next;
+ int queued = 0;
+ struct dma_request *req;
+
+ if (list_empty(&irq_entry->work_list))
+ return 0;
+
+ list_for_each_safe(node, next, &irq_entry->work_list) {
+ struct idxd_desc *desc =
+ container_of(node, struct idxd_desc, list);
+
+ req = desc->req;
+ if (desc->completion->status) {
+ list_del(&desc->list);
+ /* process and callback */
+ if ((desc->completion->status & DSA_COMP_STATUS_MASK) !=
+ DSA_COMP_SUCCESS)
+ idxd_parse_completion_status(desc->completion->status,
+ &req->result.result);
+
+ dmaengine_request_complete(req);
+ idxd_free_desc(desc->wq, desc);
+ (*processed)++;
+ } else {
+ queued++;
+ }
+ }
+
+ return queued;
+}
+
irqreturn_t idxd_wq_thread(int irq, void *data)
{
struct idxd_irq_entry *irq_entry = data;
+ int rc, processed = 0, retry = 0;
+
+ /*
+ * There are two lists we are processing. The pending_llist is where
+ * submmiter adds all the submitted descriptor after sending it to
+ * the workqueue. It's a lockless singly linked list. The work_list
+ * is the common linux double linked list. We are in a scenario of
+ * multiple producers and a single consumer. The producers are all
+ * the kernel submitters of descriptors, and the consumer is the
+ * kernel irq handler thread for the msix vector when using threaded
+ * irq. To work with the restrictions of llist to remain lockless,
+ * we are doing the following steps:
+ * 1. Iterate through the work_list and process any completed
+ * descriptor. Delete the completed entries during iteration.
+ * 2. llist_del_all() from the pending list.
+ * 3. Iterate through the llist that was deleted from the pending list
+ * and process the completed entries.
+ * 4. If the entry is still waiting on hardware, list_add_tail() to
+ * the work_list.
+ * 5. Repeat until no more descriptors.
+ */
+ do {
+ rc = irq_process_work_list(irq_entry, &processed);
+ if (rc != 0) {
+ retry++;
+ continue;
+ }
+
+ rc = irq_process_pending_llist(irq_entry, &processed);
+ } while (rc != 0 && retry != 10);

idxd_unmask_msix_vector(irq_entry->idxd, irq_entry->id);

+ if (processed == 0)
+ return IRQ_NONE;
+
return IRQ_HANDLED;
}
diff --git a/drivers/dma/idxd/submit.c b/drivers/dma/idxd/submit.c
index 2dcd13f9f654..f7baa1bbb0c7 100644
--- a/drivers/dma/idxd/submit.c
+++ b/drivers/dma/idxd/submit.c
@@ -87,7 +87,9 @@ static int idxd_submit_desc(struct idxd_wq *wq, struct idxd_desc *desc,
* Pending the descriptor to the lockless list for the irq_entry
* that we designated the descriptor to.
*/
- llist_add(&desc->llnode, &idxd->irq_entries[vec].pending_llist);
+ if (desc->req->flags & DMA_PREP_INTERRUPT)
+ llist_add(&desc->llnode,
+ &idxd->irq_entries[vec].pending_llist);

return 0;
}
@@ -125,3 +127,50 @@ static inline void set_completion_address(struct idxd_desc *desc,
{
*compl_addr = desc->compl_dma;
}
+
+static void op_flag_setup(struct idxd_wq *wq, struct dma_request *req,
+ u32 *desc_flags)
+{
+ *desc_flags = IDXD_OP_FLAG_CRAV | IDXD_OP_FLAG_RCR;
+ if (req->flags & DMA_PREP_INTERRUPT)
+ *desc_flags |= IDXD_OP_FLAG_RCI;
+ if (req->flags & DMA_PREP_FENCE)
+ *desc_flags |= IDXD_OP_FLAG_FENCE;
+}
+
+int idxd_submit_memcpy(struct idxd_wq *wq, struct dma_request *req)
+{
+ u32 desc_flags;
+ struct idxd_device *idxd = wq->idxd;
+ struct idxd_desc *desc;
+ int rc;
+ bool nonblock;
+ u64 compl_addr, src, dst;
+
+ if (wq->state != IDXD_WQ_ENABLED)
+ return -EPERM;
+
+ if (req->bvec.bv_len > idxd->max_xfer_bytes)
+ return -EINVAL;
+
+ op_flag_setup(wq, req, &desc_flags);
+ nonblock = !!(req->flags & DMA_SUBMIT_NONBLOCK);
+ desc = idxd_alloc_desc(wq, nonblock);
+ if (IS_ERR(desc))
+ return PTR_ERR(desc);
+
+ set_completion_address(desc, &compl_addr);
+ set_desc_addresses(req, &src, &dst);
+ idxd_prep_desc_common(wq, desc->hw, DSA_OPCODE_MEMMOVE,
+ src, dst, req->bvec.bv_len, compl_addr,
+ desc_flags);
+ desc->req = req;
+
+ rc = idxd_submit_desc(wq, desc, nonblock);
+ if (rc < 0) {
+ idxd_free_desc(wq, desc);
+ return rc;
+ }
+
+ return 0;
+}
diff --git a/drivers/dma/idxd/sysfs.c b/drivers/dma/idxd/sysfs.c
index fab916a555b2..a7468d22c287 100644
--- a/drivers/dma/idxd/sysfs.c
+++ b/drivers/dma/idxd/sysfs.c
@@ -55,6 +55,14 @@ static inline bool is_idxd_wq_dev(struct device *dev)
return dev ? dev->type == &idxd_wq_device_type : false;
}

+static inline bool is_idxd_wq_dmaengine(struct idxd_wq *wq)
+{
+ if (wq->type == IDXD_WQT_KERNEL &&
+ strcmp(wq->name, "dmaengine") == 0)
+ return true;
+ return false;
+}
+
static int idxd_config_bus_match(struct device *dev,
struct device_driver *drv)
{
@@ -122,6 +130,12 @@ static int idxd_config_bus_probe(struct device *dev)
spin_unlock_irqrestore(&idxd->dev_lock, flags);
dev_info(dev, "Device %s enabled\n", dev_name(dev));

+ rc = idxd_register_dma_device(idxd);
+ if (rc < 0) {
+ spin_unlock_irqrestore(&idxd->dev_lock, flags);
+ dev_dbg(dev, "Failed to register dmaengine device\n");
+ return rc;
+ }
return 0;
} else if (is_idxd_wq_dev(dev)) {
struct idxd_wq *wq = confdev_to_wq(dev);
@@ -194,6 +208,16 @@ static int idxd_config_bus_probe(struct device *dev)
wq->client_count = 0;

dev_info(dev, "wq %s enabled\n", dev_name(&wq->conf_dev));
+
+ if (is_idxd_wq_dmaengine(wq)) {
+ rc = idxd_register_dma_channel(wq);
+ if (rc < 0) {
+ dev_dbg(dev, "DMA channel register failed\n");
+ mutex_unlock(&wq->wq_lock);
+ return rc;
+ }
+ }
+
mutex_unlock(&wq->wq_lock);
return 0;
}
@@ -215,6 +239,9 @@ static void disable_wq(struct idxd_wq *wq)
return;
}

+ if (is_idxd_wq_dmaengine(wq))
+ idxd_unregister_dma_channel(wq);
+
if (idxd_wq_refcount(wq))
dev_warn(dev, "Clients has claim on wq %d: %d\n",
wq->id, idxd_wq_refcount(wq));
@@ -264,6 +291,7 @@ static int idxd_config_bus_remove(struct device *dev)
device_release_driver(&wq->conf_dev);
}

+ idxd_unregister_dma_device(idxd);
spin_lock_irqsave(&idxd->dev_lock, flags);
rc = idxd_device_disable(idxd);
spin_unlock_irqrestore(&idxd->dev_lock, flags);