Re: [PATCH v2 2/8] accel/qaic: Add uapi and core driver file

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 2/22/2023 8:52 AM, Dafna Hirschfeld wrote:
On 17.02.2023 11:15, Jeffrey Hugo wrote:
On 2/16/2023 7:13 AM, Jacek Lawrynowicz wrote:
Hi,

On 06.02.2023 16:41, Jeffrey Hugo wrote:
Add the QAIC driver uapi file and core driver file that binds to the PCIe
device.  The core driver file also creates the accel device and manages
all the interconnections between the different parts of the driver.

The driver can be built as a module.  If so, it will be called "qaic.ko".

Signed-off-by: Jeffrey Hugo <quic_jhugo@xxxxxxxxxxx>
Reviewed-by: Carl Vanderlip <quic_carlv@xxxxxxxxxxx>
---
 drivers/accel/qaic/qaic.h     | 321 ++++++++++++++++++
 drivers/accel/qaic/qaic_drv.c | 771 ++++++++++++++++++++++++++++++++++++++++++
 include/uapi/drm/qaic_accel.h | 283 ++++++++++++++++
 3 files changed, 1375 insertions(+)
 create mode 100644 drivers/accel/qaic/qaic.h
 create mode 100644 drivers/accel/qaic/qaic_drv.c
 create mode 100644 include/uapi/drm/qaic_accel.h

diff --git a/drivers/accel/qaic/qaic.h b/drivers/accel/qaic/qaic.h
new file mode 100644
index 0000000..3f7ea76
--- /dev/null
+++ b/drivers/accel/qaic/qaic.h
@@ -0,0 +1,321 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ *
+ * Copyright (c) 2019-2021, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef QAICINTERNAL_H_

Please use guard macro that matches the file name: _QAIC_H_

Before moving to DRM/ACCEL, this conflicted with the uapi file. However, that is no longer the case, so yes, this should be changed. Will do.


+#define QAICINTERNAL_H_
+
+#include <linux/interrupt.h>
+#include <linux/kref.h>
+#include <linux/mhi.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/spinlock.h>
+#include <linux/srcu.h>
+#include <linux/wait.h>
+#include <linux/workqueue.h>
+#include <drm/drm_device.h>
+#include <drm/drm_gem.h>
+
+#define QAIC_DBC_BASE        0x20000
+#define QAIC_DBC_SIZE        0x1000
+
+#define QAIC_NO_PARTITION    -1
+
+#define QAIC_DBC_OFF(i)        ((i) * QAIC_DBC_SIZE + QAIC_DBC_BASE)
+
+#define to_qaic_bo(obj) container_of(obj, struct qaic_bo, base)
+
+extern bool poll_datapath;
+
+struct qaic_user {
+    /* Uniquely identifies this user for the device */
+    int            handle;
+    struct kref        ref_count;
+    /* Char device opened by this user */
+    struct qaic_drm_device    *qddev;
+    /* Node in list of users that opened this drm device */
+    struct list_head    node;
+    /* SRCU used to synchronize this user during cleanup */
+    struct srcu_struct    qddev_lock;
+    atomic_t        chunk_id;
+};
+
+struct dma_bridge_chan {
+    /* Pointer to device strcut maintained by driver */
+    struct qaic_device    *qdev;
+    /* ID of this DMA bridge channel(DBC) */
+    unsigned int        id;
+    /* Synchronizes access to xfer_list */
+    spinlock_t        xfer_lock;
+    /* Base address of request queue */
+    void            *req_q_base;
+    /* Base address of response queue */
+    void            *rsp_q_base;
+    /*
+     * Base bus address of request queue. Response queue bus address can be
+     * calculated by adding request queue size to this variable
+     */
+    dma_addr_t        dma_addr;
+    /* Total size of request and response queue in byte */
+    u32            total_size;
+    /* Capacity of request/response queue */
+    u32            nelem;
+    /* The user that opened this DBC */
+    struct qaic_user    *usr;
+    /*
+     * Request ID of next memory handle that goes in request queue. One
+     * memory handle can enqueue more than one request elements, all
+     * this requests that belong to same memory handle have same request ID
+     */
+    u16            next_req_id;
+    /* TRUE: DBC is in use; FALSE: DBC not in use */

Use standard "true"/"false" instead of custom "TRUE"/"FALSE" macros.
This applies here and in multiple other places in the driver.

I think you are getting at that the documentation could be confusing. I don't appear to see custom macro use in the code.  Will try to clarify that here.

+    bool            in_use;
+    /*
+     * Base address of device registers. Used to read/write request and
+     * response queue's head and tail pointer of this DBC.
+     */
+    void __iomem        *dbc_base;
+    /* Head of list where each node is a memory handle queued in request queue */
+    struct list_head    xfer_list;
+    /* Synchronizes DBC readers during cleanup */
+    struct srcu_struct    ch_lock;
+    /*
+     * When this DBC is released, any thread waiting on this wait queue is
+     * woken up
+     */
+    wait_queue_head_t    dbc_release;
+    /* Head of list where each node is a bo associated with this DBC */
+    struct list_head    bo_lists;
+    /* The irq line for this DBC.  Used for polling */
+    unsigned int        irq;
+    /* Polling work item to simulate interrupts */
+    struct work_struct    poll_work;
+};
+
+struct qaic_device {
+    /* Pointer to base PCI device struct of our physical device */
+    struct pci_dev        *pdev;
+    /* Mask of all bars of this device */
+    int            bars;
+    /* Req. ID of request that will be queued next in MHI control device */
+    u32            next_seq_num;
+    /* Base address of bar 0 */
+    void __iomem        *bar_0;
+    /* Base address of bar 2 */
+    void __iomem        *bar_2;
+    /* Controller structure for MHI devices */
+    struct mhi_controller    *mhi_cntl;
+    /* MHI control channel device */
+    struct mhi_device    *cntl_ch;
+    /* List of requests queued in MHI control device */
+    struct list_head    cntl_xfer_list;
+    /* Synchronizes MHI control device transactions and its xfer list */
+    struct mutex        cntl_mutex;
+    /* Base actual physical representation of drm device */
+    struct qaic_drm_device    *base_dev;
+    /* Array of DBC struct of this device */
+    struct dma_bridge_chan    *dbc;
+    /* Work queue for tasks related to MHI control device */
+    struct workqueue_struct    *cntl_wq;
+    /* Synchronizes all the users of device during cleanup */
+    struct srcu_struct    dev_lock;
+    /* TRUE: Device under reset; FALSE: Device not under reset */
+    bool            in_reset;
+    /*
+     * TRUE: A tx MHI transaction has failed and a rx buffer is still queued
+     * in control device. Such a buffer is considered lost rx buffer
+     * FALSE: No rx buffer is lost in control device
+     */
+    bool            cntl_lost_buf;
+    /* Maximum number of DBC supported by this device */
+    u32            num_dbc;
+    /* Head in list of drm device created on top of this device */
+    struct list_head    qaic_drm_devices;
+    /* Synchronizes access of qaic_drm_devices list */
+    struct mutex        qaic_drm_devices_mutex;
+    /* Generate the CRC of a control message */
+    u32 (*gen_crc)(void *msg);
+    /* Validate the CRC of a control message */
+    bool (*valid_crc)(void *msg);
+};
+
+struct qaic_drm_device {
+    /* Pointer to the root device struct driven by this driver */
+    struct qaic_device    *qdev;
+    /* Node in list of drm devices maintained by root device */
+    struct list_head    node;
+    /*
+     * The physical device can be partition in number of logical devices. +     * And each logical device is given a partition id. This member stores +     * that id. QAIC_NO_PARTITION is a sentinel used to mark that this drm
+     * device is the actual physical device
+     */
+    s32            partition_id;
+    /*
+     * It points to the user that created this drm device. It is NULL
+     * when this drm device represents the physical device i.e.
+     * partition_id is QAIC_NO_PARTITION
+     */
+    struct qaic_user    *owner;
+    /* Pointer to the drm device struct of this drm device */
+    struct drm_device    *ddev;
+    /* Head in list of users who have opened this drm device */
+    struct list_head    users;
+    /* Synchronizes access to users list */
+    struct mutex        users_mutex;
+};
+
+struct qaic_bo {
+    struct drm_gem_object    base;

Any reason why drm_gem_shmem_object cannot be used as a base?

I beleive that is incompatible with our memory allocator in patch 5.

+    /* Scatter/gather table for allocate/imported BO */
+    struct sg_table        *sgt;
+    /* BO size requested by user. GEM object might be bigger in size. */
+    u64            size;
+    /* Head in list of slices of this BO */
+    struct list_head    slices;
+    /* Total nents, for all slices of this BO */
+    int            total_slice_nents;
+    /*
+     * Direction of transfer. It can assume only two value DMA_TO_DEVICE and
+     * DMA_FROM_DEVICE.
+     */
+    int            dir;
+    /* The pointer of the DBC which operates on this BO */
+    struct dma_bridge_chan    *dbc;
+    /* Number of slice that belongs to this buffer */
+    u32            nr_slice;
+    /* Number of slice that have been transferred by DMA engine */
+    u32            nr_slice_xfer_done;
+    /* TRUE = BO is queued for execution, FALSE = BO is not queued */
+    bool            queued;
+    /*
+     * If TRUE then user has attached slicing information to this BO by
+     * calling DRM_IOCTL_QAIC_ATTACH_SLICE_BO ioctl.
+     */
+    bool            sliced;
+    /* Request ID of this BO if it is queued for execution */
+    u16            req_id;
+    /* Handle assigned to this BO */
+    u32            handle;
+    /* Wait on this for completion of DMA transfer of this BO */
+    struct completion    xfer_done;
+    /*
+     * Node in linked list where head is dbc->xfer_list.
+     * This link list contain BO's that are queued for DMA transfer.
+     */
+    struct list_head    xfer_list;
+    /*
+     * Node in linked list where head is dbc->bo_lists.
+     * This link list contain BO's that are associated with the DBC it is
+     * linked to.
+     */
+    struct list_head    bo_list;
+    struct {
+        /*
+         * Latest timestamp(ns) at which kernel received a request to
+         * execute this BO
+         */
+        u64        req_received_ts;
+        /*
+         * Latest timestamp(ns) at which kernel enqueued requests of
+         * this BO for execution in DMA queue
+         */
+        u64        req_submit_ts;
+        /*
+         * Latest timestamp(ns) at which kernel received a completion
+         * interrupt for requests of this BO
+         */
+        u64        req_processed_ts;
+        /*
+         * Number of elements already enqueued in DMA queue before
+         * enqueuing requests of this BO
+         */
+        u32        queue_level_before;
+    } perf_stats;
+
+};
+
+struct bo_slice {
+    /* Mapped pages */
+    struct sg_table        *sgt;
+    /* Number of requests required to queue in DMA queue */
+    int            nents;
+    /* See enum dma_data_direction */
+    int            dir;
+    /* Actual requests that will be copied in DMA queue */
+    struct dbc_req        *reqs;
+    struct kref        ref_count;
+    /* TRUE: No DMA transfer required */
+    bool            no_xfer;
+    /* Pointer to the parent BO handle */
+    struct qaic_bo        *bo;
+    /* Node in list of slices maintained by parent BO */
+    struct list_head    slice;
+    /* Size of this slice in bytes */
+    u64            size;
+    /* Offset of this slice in buffer */
+    u64            offset;
+};
+
+int get_dbc_req_elem_size(void);
+int get_dbc_rsp_elem_size(void);
+int get_cntl_version(struct qaic_device *qdev, struct qaic_user *usr,
+             u16 *major, u16 *minor);
+int qaic_manage_ioctl(struct drm_device *dev, void *data,
+              struct drm_file *file_priv);
+int qaic_execute_ioctl(struct qaic_device *qdev, struct qaic_user *usr,
+               unsigned long arg, bool is_partial);
+int qaic_wait_exec_ioctl(struct qaic_device *qdev, struct qaic_user *usr,
+             unsigned long arg);
+int qaic_query_ioctl(struct qaic_device *qdev, struct qaic_user *usr,
+             unsigned long arg);
+int qaic_data_mmap(struct qaic_device *qdev, struct qaic_user *usr,
+           struct vm_area_struct *vma);
+int qaic_data_get_reservation(struct qaic_device *qdev, struct qaic_user *usr,
+                  void *data, u32 *partition_id,
+                  u16 *remove);
+void qaic_mhi_ul_xfer_cb(struct mhi_device *mhi_dev,
+             struct mhi_result *mhi_result);
+
+void qaic_mhi_dl_xfer_cb(struct mhi_device *mhi_dev,
+             struct mhi_result *mhi_result);
+
+int qaic_control_open(struct qaic_device *qdev);
+void qaic_control_close(struct qaic_device *qdev);
+void qaic_release_usr(struct qaic_device *qdev, struct qaic_user *usr);
+
+irqreturn_t dbc_irq_threaded_fn(int irq, void *data);
+irqreturn_t dbc_irq_handler(int irq, void *data);
+int disable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr); +void enable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr);
+void wakeup_dbc(struct qaic_device *qdev, u32 dbc_id);
+void release_dbc(struct qaic_device *qdev, u32 dbc_id);
+
+void wake_all_cntl(struct qaic_device *qdev);
+void qaic_dev_reset_clean_local_state(struct qaic_device *qdev, bool exit_reset);
+
+struct drm_gem_object *qaic_gem_prime_import(struct drm_device *dev,
+                         struct dma_buf *dma_buf);
+
+int qaic_create_bo_ioctl(struct drm_device *dev, void *data,
+             struct drm_file *file_priv);
+int qaic_mmap_bo_ioctl(struct drm_device *dev, void *data,
+               struct drm_file *file_priv);
+int qaic_attach_slice_bo_ioctl(struct drm_device *dev, void *data,
+                   struct drm_file *file_priv);
+int qaic_execute_bo_ioctl(struct drm_device *dev, void *data,
+              struct drm_file *file_priv);
+int qaic_partial_execute_bo_ioctl(struct drm_device *dev, void *data,
+                  struct drm_file *file_priv);
+int qaic_wait_bo_ioctl(struct drm_device *dev, void *data,
+               struct drm_file *file_priv);
+int qaic_test_print_bo_ioctl(struct drm_device *dev, void *data,
+                 struct drm_file *file_priv);
+int qaic_perf_stats_bo_ioctl(struct drm_device *dev, void *data,
+                 struct drm_file *file_priv);

You don't need to break these lines. You can use up to 100 columns in the whole driver.
It will be more readable and checkpatch won't complain.

Some of this predates the 100 column change.  Checkpatch already complains about the uapi file.

Will try to fix here.

+void irq_polling_work(struct work_struct *work);
+
+#endif /* QAICINTERNAL_H_ */
diff --git a/drivers/accel/qaic/qaic_drv.c b/drivers/accel/qaic/qaic_drv.c
new file mode 100644
index 0000000..602a784
--- /dev/null
+++ b/drivers/accel/qaic/qaic_drv.c
@@ -0,0 +1,771 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */ +/* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/idr.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/kref.h>
+#include <linux/mhi.h>
+#include <linux/module.h>
+#include <linux/msi.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/sched.h>

Is <linux/sched.h> used here?
Feels like there are couple other unused includes in this file.

Actually I think that can be removed now.  Will check.
The other ones look sane to me.  Are there specific ones you question?


+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+#include <linux/wait.h>
+#include <drm/drm_accel.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+#include <drm/drm_gem.h>
+#include <drm/drm_ioctl.h>
+#include <uapi/drm/qaic_accel.h>
+
+#include "mhi_controller.h"
+#include "mhi_qaic_ctrl.h"
+#include "qaic.h"
+
+MODULE_IMPORT_NS(DMA_BUF);
+
+#define PCI_DEV_AIC100            0xa100
+#define QAIC_NAME            "qaic"
+#define QAIC_DESC            "Qualcomm Cloud AI Accelerators"
+
+static unsigned int datapath_polling;
+module_param(datapath_polling, uint, 0400);
+bool poll_datapath;
+
+static u16 cntl_major = 5;
+static u16 cntl_minor;/* 0 */

Missing space before the comment.
And also you could convert both vars to macros as they are constants.

Sure

+static bool link_up;
+static DEFINE_IDA(qaic_usrs);
+
+static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id,
+                  struct qaic_user *owner);
+static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id,
+                    struct qaic_user *owner);
+
+static void free_usr(struct kref *kref)
+{
+    struct qaic_user *usr = container_of(kref, struct qaic_user, ref_count);
+
+    cleanup_srcu_struct(&usr->qddev_lock);
+    ida_free(&qaic_usrs, usr->handle);
+    kfree(usr);
+}
+
+static int qaic_open(struct drm_device *dev, struct drm_file *file)
+{
+    struct qaic_drm_device *qddev = dev->dev_private;
+    struct qaic_device *qdev = qddev->qdev;
+    struct qaic_user *usr;
+    int rcu_id;
+    int ret;
+
+    rcu_id = srcu_read_lock(&qdev->dev_lock);
+    if (qdev->in_reset) {
+        srcu_read_unlock(&qdev->dev_lock, rcu_id);
+        return -ENODEV;
+    }
+
+    usr = kmalloc(sizeof(*usr), GFP_KERNEL);
+    if (!usr) {
+        srcu_read_unlock(&qdev->dev_lock, rcu_id);
+        return -ENOMEM;
+    }
+
+    usr->handle = ida_alloc(&qaic_usrs, GFP_KERNEL);
+    if (usr->handle < 0) {
+        srcu_read_unlock(&qdev->dev_lock, rcu_id);
+        kfree(usr);
+        return usr->handle;

hi, use after free here

Nice catch.  Thanks.  Also noting the other things you mentioned.


+    }
+    usr->qddev = qddev;
+    atomic_set(&usr->chunk_id, 0);
+    init_srcu_struct(&usr->qddev_lock);
+    kref_init(&usr->ref_count);
+
+    ret = mutex_lock_interruptible(&qddev->users_mutex);
+    if (ret) {
+        cleanup_srcu_struct(&usr->qddev_lock);
+        kfree(usr);
+        srcu_read_unlock(&qdev->dev_lock, rcu_id);
+        return ret;
+    }
+
+    list_add(&usr->node, &qddev->users);
+    mutex_unlock(&qddev->users_mutex);
+
+    file->driver_priv = usr;
+
+    srcu_read_unlock(&qdev->dev_lock, rcu_id);
+    return 0;
+}
+
+static void qaic_postclose(struct drm_device *dev, struct drm_file *file)
+{
+    struct qaic_user *usr = file->driver_priv;
+    struct qaic_drm_device *qddev;
+    struct qaic_device *qdev;
+    int qdev_rcu_id;
+    int usr_rcu_id;
+    int i;
+
+    qddev = usr->qddev;
+    usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+    if (qddev) {
+        qdev = qddev->qdev;
+        qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+        if (!qdev->in_reset) {
+            qaic_release_usr(qdev, usr);
+            for (i = 0; i < qdev->num_dbc; ++i)
+                if (qdev->dbc[i].usr &&
+                    qdev->dbc[i].usr->handle == usr->handle)
+                    release_dbc(qdev, i);
+
+            /* Remove child devices */
+            if (qddev->partition_id == QAIC_NO_PARTITION)
+                qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION, usr);
+        }
+        srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+
+        mutex_lock(&qddev->users_mutex);
+        if (!list_empty(&usr->node))
+            list_del_init(&usr->node);
+        mutex_unlock(&qddev->users_mutex);
+    }
+
+    srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+    kref_put(&usr->ref_count, free_usr);
+
+    file->driver_priv = NULL;
+}
+
+static int qaic_part_dev_ioctl(struct drm_device *dev, void *data,
+                   struct drm_file *file_priv)
+{
+    struct qaic_device *qdev;
+    struct qaic_user *usr;
+    u32 partition_id;
+    int qdev_rcu_id;
+    int usr_rcu_id;
+    int ret = 0;
+    u16 remove;
+
+    usr = file_priv->driver_priv;
+    if (!usr)
+        return -EINVAL;
+
+    usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+    if (!usr->qddev) {
+        srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+        return -ENODEV;
+    }
+
+    qdev = usr->qddev->qdev;
+    if (!qdev) {
+        srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+        return -ENODEV;
+    }
+
+    qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+    if (qdev->in_reset) {
+        ret = -ENODEV;
+        goto out;
+    }
+
+    /* This IOCTL is only supported for base devices. */
+    if (usr->qddev->partition_id != QAIC_NO_PARTITION) {
+        ret = -ENOTTY;
+        goto out;
+    }
+
+    ret = qaic_data_get_reservation(qdev, usr, data, &partition_id,
+                    &remove);
+    if (ret)
+        goto out;
+
+    if (remove == 1)
+        qaic_destroy_drm_device(qdev, partition_id, usr);
+    else
+        ret = qaic_create_drm_device(qdev, partition_id, usr);
+
+out:
+    srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+    srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+
+    return ret;
+}
+
+DEFINE_DRM_ACCEL_FOPS(qaic_accel_fops);
+
+static const struct drm_ioctl_desc qaic_drm_ioctls[] = {
+    DRM_IOCTL_DEF_DRV(QAIC_MANAGE, qaic_manage_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_CREATE_BO, qaic_create_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_MMAP_BO, qaic_mmap_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_ATTACH_SLICE_BO, qaic_attach_slice_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_EXECUTE_BO, qaic_execute_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_PARTIAL_EXECUTE_BO, qaic_partial_execute_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_WAIT_BO, qaic_wait_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_PERF_STATS_BO, qaic_perf_stats_bo_ioctl, DRM_RENDER_ALLOW), +    DRM_IOCTL_DEF_DRV(QAIC_PART_DEV, qaic_part_dev_ioctl, DRM_RENDER_ALLOW),
+};
+
+static const struct drm_driver qaic_accel_driver = {
+    .driver_features    = DRIVER_GEM | DRIVER_COMPUTE_ACCEL,
+
+    .name            = QAIC_NAME,
+    .desc            = QAIC_DESC,
+    .date            = "20190618",
+
+    .fops            = &qaic_accel_fops,
+    .open            = qaic_open,
+    .postclose        = qaic_postclose,
+
+    .ioctls            = qaic_drm_ioctls,
+    .num_ioctls        = ARRAY_SIZE(qaic_drm_ioctls),
+    .prime_fd_to_handle    = drm_gem_prime_fd_to_handle,
+    .gem_prime_import    = qaic_gem_prime_import,
+};
+
+static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id,
+                  struct qaic_user *owner)
+{
+    struct qaic_drm_device *qddev;
+    struct drm_device *ddev;
+    struct device *pdev;
+    int ret;
+
+    /*
+     * Partition id QAIC_NO_PARTITION indicates that the device was created
+     * on mhi_probe and id > QAIC_NO_PARTITION indicates a partition
+     * created using IOCTL. So, pdev for primary device is the pci dev and
+     * the parent for partition dev is the primary device.
+     */
+    if (partition_id == QAIC_NO_PARTITION)
+        pdev = &qdev->pdev->dev;
+    else
+        pdev = qdev->base_dev->ddev->dev;
+
+    qddev = kzalloc(sizeof(*qddev), GFP_KERNEL);
+    if (!qddev) {
+        ret = -ENOMEM;
+        goto qddev_fail;

qddev_fail label is just before returning so better return here directly

+    }
+
+    ddev = drm_dev_alloc(&qaic_accel_driver, pdev);
+    if (IS_ERR(ddev)) {
+        ret = PTR_ERR(ddev);
+        goto ddev_fail;
+    }
+
+    ddev->dev_private = qddev;
+    qddev->ddev = ddev;
+
+    if (partition_id == QAIC_NO_PARTITION)
+        qdev->base_dev = qddev;
+    qddev->qdev = qdev;
+    qddev->partition_id = partition_id;
+    qddev->owner = owner;
+    INIT_LIST_HEAD(&qddev->users);
+    mutex_init(&qddev->users_mutex);
+
+    mutex_lock(&qdev->qaic_drm_devices_mutex);
+    list_add(&qddev->node, &qdev->qaic_drm_devices);
+    mutex_unlock(&qdev->qaic_drm_devices_mutex);
+
+    ret = drm_dev_register(ddev, 0);
+    if (ret) {
+        pci_dbg(qdev->pdev, "%s: drm_dev_register failed %d\n", __func__, ret);
+        goto drm_reg_fail;
+    }
+
+    return 0;
+
+drm_reg_fail:
+    mutex_destroy(&qddev->users_mutex);
+    mutex_lock(&qdev->qaic_drm_devices_mutex);
+    list_del(&qddev->node);
+    mutex_unlock(&qdev->qaic_drm_devices_mutex);
+    if (partition_id == QAIC_NO_PARTITION)
+        qdev->base_dev = NULL;
+    drm_dev_put(ddev);
+ddev_fail:
+    kfree(qddev);
+qddev_fail:

no point for this label

+    return ret;
+}
+
+static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id,
+                    struct qaic_user *owner)
+{
+    struct qaic_drm_device *qddev;
+    struct qaic_drm_device *q;
+    struct qaic_user *usr;
+
+    list_for_each_entry_safe(qddev, q, &qdev->qaic_drm_devices, node) {
+        /*
+         * Skip devices in case we just want to remove devices
+         * specific to a owner or partition id.
+         *
+         * owner    partition_id    notes
+         * ----------------------------------
+         * NULL        NO_PARTITION    delete base + all derived (qdev
+         *                reset)
+         * !NULL    NO_PARTITION    delete derived devs created by
+         *                owner.
+         * !NULL    >NO_PARTITION    delete derived dev identified by
+         *                the partition id and created by
+         *                owner
+         * NULL        >NO_PARTITION    invalid (no-op)
+         *
+         * if partition_id is any value < QAIC_NO_PARTITION this will be
+         * a no-op.
+         */
+        if (owner && owner != qddev->owner)
+            continue;
+
+        if (partition_id != QAIC_NO_PARTITION &&
+            partition_id != qddev->partition_id && !owner)
+            continue;
+
+        /*
+         * Existing users get unresolvable errors till they close FDs.
+         * Need to sync carefully with users calling close().  The
+         * list of users can be modified elsewhere when the lock isn't
+         * held here, but the sync'ing the srcu with the mutex held
+         * could deadlock.  Grab the mutex so that the list will be
+         * unmodified.  The user we get will exist as long as the
+         * lock is held.  Signal that the qcdev is going away, and
+         * grab a reference to the user so they don't go away for
+         * synchronize_srcu().  Then release the mutex to avoid
+         * deadlock and make sure the user has observed the signal.
+         * With the lock released, we cannot maintain any state of the
+         * user list.
+         */
+        mutex_lock(&qddev->users_mutex);
+        while (!list_empty(&qddev->users)) {
+            usr = list_first_entry(&qddev->users, struct qaic_user,
+                           node);
+            list_del_init(&usr->node);
+            kref_get(&usr->ref_count);
+            usr->qddev = NULL;
+            mutex_unlock(&qddev->users_mutex);
+            synchronize_srcu(&usr->qddev_lock);
+            kref_put(&usr->ref_count, free_usr);
+            mutex_lock(&qddev->users_mutex);
+        }
+        mutex_unlock(&qddev->users_mutex);
+
+        if (qddev->ddev) {
+            drm_dev_unregister(qddev->ddev);
+            drm_dev_put(qddev->ddev);
+        }
+
+        list_del(&qddev->node);
+        kfree(qddev);
+    }
+}
+
+static int qaic_mhi_probe(struct mhi_device *mhi_dev,
+              const struct mhi_device_id *id)
+{
+    struct qaic_device *qdev;
+    u16 major, minor;
+    int ret;
+
+    /*
+     * Invoking this function indicates that the control channel to the
+     * device is available.  We use that as a signal to indicate that
+     * the device side firmware has booted.  The device side firmware
+     * manages the device resources, so we need to communicate with it
+     * via the control channel in order to utilize the device. Therefore +     * we wait until this signal to create the drm dev that userspace will +     * use to control the device, because without the device side firmware,
+     * userspace can't do anything useful.
+     */
+
+    qdev = pci_get_drvdata(to_pci_dev(mhi_dev->mhi_cntrl->cntrl_dev));
+
+    qdev->in_reset = false;
+
+    dev_set_drvdata(&mhi_dev->dev, qdev);
+    qdev->cntl_ch = mhi_dev;
+
+    ret = qaic_control_open(qdev);
+    if (ret) {
+        pci_dbg(qdev->pdev, "%s: control_open failed %d\n", __func__, ret);
+        goto err;
+    }
+
+    ret = get_cntl_version(qdev, NULL, &major, &minor);
+    if (ret || major != cntl_major || minor > cntl_minor) {
+        pci_err(qdev->pdev, "%s: Control protocol version (%d.%d) not supported.  Supported version is (%d.%d). Ret: %d\n",
+            __func__, major, minor, cntl_major, cntl_minor, ret);
+        ret = -EINVAL;
+        goto close_control;
+    }
+
+    ret = qaic_create_drm_device(qdev, QAIC_NO_PARTITION, NULL);
+
+    return ret;
+
+close_control:
+    qaic_control_close(qdev);
+err:

same here, not sure there is a point to this label

Thanks,
Dafna

+    return ret;
+}
+
+static void qaic_mhi_remove(struct mhi_device *mhi_dev)
+{

Add a comment here

Presumably why this is an empty function?
Will do.

+}
+
+static void qaic_notify_reset(struct qaic_device *qdev)
+{
+    int i;
+
+    qdev->in_reset = true;
+    /* wake up any waiters to avoid waiting for timeouts at sync */
+    wake_all_cntl(qdev);
+    for (i = 0; i < qdev->num_dbc; ++i)
+        wakeup_dbc(qdev, i);
+    synchronize_srcu(&qdev->dev_lock);
+}
+
+void qaic_dev_reset_clean_local_state(struct qaic_device *qdev, bool exit_reset)
+{
+    int i;
+
+    qaic_notify_reset(qdev);
+
+    /* remove drmdevs to prevent new users from coming in */
+    if (qdev->base_dev)
+        qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION, NULL);
+
+    /* start tearing things down */
+    for (i = 0; i < qdev->num_dbc; ++i)
+        release_dbc(qdev, i);
+
+    if (exit_reset)
+        qdev->in_reset = false;
+}
+
+static int qaic_pci_probe(struct pci_dev *pdev,
+              const struct pci_device_id *id)

Please try to simplify this function. Maybe move irq init to separate function. It will be more readable and there will less of a error handling spaghetti at the bottom.

I guess I tend towards not breaking something up if there is not reuse elsewhere, but I think I see your point.  Will have a look.


+{
+    int ret;
+    int i;
+    int mhi_irq;
+    struct qaic_device *qdev;
+
+    qdev = devm_kzalloc(&pdev->dev, sizeof(*qdev), GFP_KERNEL);
+    if (!qdev)
+        return -ENOMEM;
+
+    if (id->device == PCI_DEV_AIC100) {
+        qdev->num_dbc = 16;
+        qdev->dbc = devm_kcalloc(&pdev->dev, qdev->num_dbc, sizeof(*qdev->dbc),
+                     GFP_KERNEL);
+        if (!qdev->dbc)
+            return -ENOMEM;
+    }
+
+    qdev->cntl_wq = alloc_workqueue("qaic_cntl", WQ_UNBOUND, 0);
+    if (!qdev->cntl_wq) {
+        ret = -ENOMEM;
+        goto wq_fail;
+    }
+    pci_set_drvdata(pdev, qdev);
+    qdev->pdev = pdev;
+    mutex_init(&qdev->cntl_mutex);
+    INIT_LIST_HEAD(&qdev->cntl_xfer_list);
+    init_srcu_struct(&qdev->dev_lock);
+    INIT_LIST_HEAD(&qdev->qaic_drm_devices);
+    mutex_init(&qdev->qaic_drm_devices_mutex);
+    for (i = 0; i < qdev->num_dbc; ++i) {
+        spin_lock_init(&qdev->dbc[i].xfer_lock);
+        qdev->dbc[i].qdev = qdev;
+        qdev->dbc[i].id = i;
+        INIT_LIST_HEAD(&qdev->dbc[i].xfer_list);
+        init_srcu_struct(&qdev->dbc[i].ch_lock);
+        init_waitqueue_head(&qdev->dbc[i].dbc_release);
+        INIT_LIST_HEAD(&qdev->dbc[i].bo_lists);
+    }
+
+    qdev->bars = pci_select_bars(pdev, IORESOURCE_MEM);
+
+    /* make sure the device has the expected BARs */
+    if (qdev->bars != (BIT(0) | BIT(2) | BIT(4))) {
+        pci_dbg(pdev, "%s: expected BARs 0, 2, and 4 not found in device.  Found 0x%x\n",
+            __func__, qdev->bars);
+        ret = -EINVAL;
+        goto bar_fail;
+    }
+
+    ret = pci_enable_device(pdev);
+    if (ret)
+        goto enable_fail;
+
+    ret = pci_request_selected_regions(pdev, qdev->bars, "aic100");
+    if (ret)
+        goto request_regions_fail;
+
+    pci_set_master(pdev);
+
+    ret = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
+    if (ret)
+        goto dma_mask_fail;
+    ret = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
+    if (ret)
+        goto dma_mask_fail;
+    ret = dma_set_max_seg_size(&pdev->dev, UINT_MAX);
+    if (ret)
+        goto dma_mask_fail;
+
+    qdev->bar_0 = pci_ioremap_bar(pdev, 0);
+    if (!qdev->bar_0) {
+        ret = -ENOMEM;
+        goto ioremap_0_fail;
+    }
+
+    qdev->bar_2 = pci_ioremap_bar(pdev, 2);
+    if (!qdev->bar_2) {
+        ret = -ENOMEM;
+        goto ioremap_2_fail;
+    }
+
+    for (i = 0; i < qdev->num_dbc; ++i)
+        qdev->dbc[i].dbc_base = qdev->bar_2 + QAIC_DBC_OFF(i);
+
+    ret = pci_alloc_irq_vectors(pdev, 1, 32, PCI_IRQ_MSI);
+    if (ret < 0)
+        goto alloc_irq_fail;
+
+    if (ret < 32) {
+        pci_err(pdev, "%s: Requested 32 MSIs.  Obtained %d MSIs which is less than the 32 required.\n",
+            __func__, ret);
+        ret = -ENODEV;
+        goto invalid_msi_config;
+    }
+
+    mhi_irq = pci_irq_vector(pdev, 0);
+    if (mhi_irq < 0) {
+        ret = mhi_irq;
+        goto get_mhi_irq_fail;
+    }
+
+    for (i = 0; i < qdev->num_dbc; ++i) {
+        ret = devm_request_threaded_irq(&pdev->dev,
+                        pci_irq_vector(pdev, i + 1),
+                        dbc_irq_handler,
+                        dbc_irq_threaded_fn,
+                        IRQF_SHARED,
+                        "qaic_dbc",
+                        &qdev->dbc[i]);
+        if (ret)
+            goto get_dbc_irq_failed;
+
+        if (poll_datapath) {
+            qdev->dbc[i].irq = pci_irq_vector(pdev, i + 1);
+            disable_irq_nosync(qdev->dbc[i].irq);
+            INIT_WORK(&qdev->dbc[i].poll_work, irq_polling_work);
+        }
+    }
+
+    qdev->mhi_cntl = qaic_mhi_register_controller(pdev, qdev->bar_0, mhi_irq);
+    if (IS_ERR(qdev->mhi_cntl)) {
+        ret = PTR_ERR(qdev->mhi_cntl);
+        goto mhi_register_fail;
+    }
+
+    return 0;
+
+mhi_register_fail:
+get_dbc_irq_failed:

I don't think that duplicated goto statements are allowed by the coding style.
These should be rather named something like err_free_irq.
See https://www.kernel.org/doc/html/v4.10/process/coding-style.html#centralized-exiting-of-functions

I took another look at that link, and I do not beleive it prohibits duplicated goto labels, but they probably could be renamed and simplified as you indicate.  Will have a look at that.


+    for (i = 0; i < qdev->num_dbc; ++i)
+        devm_free_irq(&pdev->dev, pci_irq_vector(pdev, i + 1),
+                  &qdev->dbc[i]);
+get_mhi_irq_fail:
+invalid_msi_config:
+    pci_free_irq_vectors(pdev);
+alloc_irq_fail:
+    iounmap(qdev->bar_2);
+ioremap_2_fail:
+    iounmap(qdev->bar_0);
+ioremap_0_fail:
+dma_mask_fail:
+    pci_clear_master(pdev);
+    pci_release_selected_regions(pdev, qdev->bars);
+request_regions_fail:
+    pci_disable_device(pdev);
+enable_fail:
+    pci_set_drvdata(pdev, NULL);
+bar_fail:
+    for (i = 0; i < qdev->num_dbc; ++i)
+        cleanup_srcu_struct(&qdev->dbc[i].ch_lock);
+    cleanup_srcu_struct(&qdev->dev_lock);
+    destroy_workqueue(qdev->cntl_wq);
+wq_fail:
+    return ret;
+}
+
+static void qaic_pci_remove(struct pci_dev *pdev)
+{
+    struct qaic_device *qdev = pci_get_drvdata(pdev);
+    int i;
+
+    if (!qdev)
+        return;
+
+    qaic_dev_reset_clean_local_state(qdev, false);
+    qaic_mhi_free_controller(qdev->mhi_cntl, link_up);
+    for (i = 0; i < qdev->num_dbc; ++i) {
+        devm_free_irq(&pdev->dev, pci_irq_vector(pdev, i + 1),
+                  &qdev->dbc[i]);
+        cleanup_srcu_struct(&qdev->dbc[i].ch_lock);
+    }
+    destroy_workqueue(qdev->cntl_wq);
+    pci_free_irq_vectors(pdev);
+    iounmap(qdev->bar_0);
+    pci_clear_master(pdev);
+    pci_release_selected_regions(pdev, qdev->bars);
+    pci_disable_device(pdev);
+    pci_set_drvdata(pdev, NULL);
+}
+
+static void qaic_pci_shutdown(struct pci_dev *pdev)
+{
+    /* see qaic_exit for what link_up is doing */
+    link_up = true;
+    qaic_pci_remove(pdev);
+}
+
+static pci_ers_result_t qaic_pci_error_detected(struct pci_dev *pdev,
+                        pci_channel_state_t error)
+{
+    return PCI_ERS_RESULT_NEED_RESET;
+}
+
+static void qaic_pci_reset_prepare(struct pci_dev *pdev)
+{
+    struct qaic_device *qdev = pci_get_drvdata(pdev);
+
+    qaic_notify_reset(qdev);
+    qaic_mhi_start_reset(qdev->mhi_cntl);
+    qaic_dev_reset_clean_local_state(qdev, false);
+}
+
+static void qaic_pci_reset_done(struct pci_dev *pdev)
+{
+    struct qaic_device *qdev = pci_get_drvdata(pdev);
+
+    qdev->in_reset = false;
+    qaic_mhi_reset_done(qdev->mhi_cntl);
+}
+
+static const struct mhi_device_id qaic_mhi_match_table[] = {
+    { .chan = "QAIC_CONTROL", },
+    {},
+};
+
+static struct mhi_driver qaic_mhi_driver = {
+    .id_table = qaic_mhi_match_table,
+    .remove = qaic_mhi_remove,
+    .probe = qaic_mhi_probe,
+    .ul_xfer_cb = qaic_mhi_ul_xfer_cb,
+    .dl_xfer_cb = qaic_mhi_dl_xfer_cb,
+    .driver = {
+        .name = "qaic_mhi",
+    },
+};
+
+static const struct pci_device_id qaic_ids[] = {
+    { PCI_DEVICE(PCI_VENDOR_ID_QCOM, PCI_DEV_AIC100), },
+    { }
+};
+MODULE_DEVICE_TABLE(pci, qaic_ids);
+
+static const struct pci_error_handlers qaic_pci_err_handler = {
+    .error_detected = qaic_pci_error_detected,
+    .reset_prepare = qaic_pci_reset_prepare,
+    .reset_done = qaic_pci_reset_done,
+};
+
+static struct pci_driver qaic_pci_driver = {
+    .name = QAIC_NAME,
+    .id_table = qaic_ids,
+    .probe = qaic_pci_probe,
+    .remove = qaic_pci_remove,
+    .shutdown = qaic_pci_shutdown,
+    .err_handler = &qaic_pci_err_handler,
+};
+
+static int __init qaic_init(void)
+{
+    int ret;
+
+    if (datapath_polling)
+        poll_datapath = true;
+
+    ret = mhi_driver_register(&qaic_mhi_driver);
+    if (ret) {
+        pr_debug("qaic: mhi_driver_register failed %d\n", ret);
+        goto free_class;
+    }
+
+    ret = pci_register_driver(&qaic_pci_driver);
+    if (ret) {
+        pr_debug("qaic: pci_register_driver failed %d\n", ret);
+        goto free_mhi;
+    }
+
+    ret = mhi_qaic_ctrl_init();
+    if (ret) {
+        pr_debug("qaic: mhi_qaic_ctrl_init failed %d\n", ret);
+        goto free_pci;
+    }
+
+    return 0;
+
+free_pci:
+    pci_unregister_driver(&qaic_pci_driver);
+free_mhi:
+    mhi_driver_unregister(&qaic_mhi_driver);
+free_class:

This label doesn't free anything. It should be renamed.

Doh.  Holdover from a previous version.  It does need to be renamed.


+    return ret;
+}
+
+static void __exit qaic_exit(void)
+{
+    /*
+     * We assume that qaic_pci_remove() is called due to a hotplug event
+     * which would mean that the link is down, and thus
+     * qaic_mhi_free_controller() should not try to access the device during
+     * cleanup.
+     * We call pci_unregister_driver() below, which also triggers
+     * qaic_pci_remove(), but since this is module exit, we expect the link +     * to the device to be up, in which case qaic_mhi_free_controller() +     * should try to access the device during cleanup to put the device in
+     * a sane state.
+     * For that reason, we set link_up here to let qaic_mhi_free_controller
+     * know the expected link state.  Since the module is going to be
+     * removed at the end of this, we don't need to worry about
+     * reinitializing the link_up state after the cleanup is done.
+     */
+    link_up = true;

Maybe you could just use pdev->current_state instead of link_up?

Maybe.  Will have a look at that.


+    mhi_qaic_ctrl_deinit();
+    pci_unregister_driver(&qaic_pci_driver);
+    mhi_driver_unregister(&qaic_mhi_driver);
+}
+
+module_init(qaic_init);
+module_exit(qaic_exit);
+
+MODULE_AUTHOR(QAIC_DESC " Kernel Driver Team");
+MODULE_DESCRIPTION(QAIC_DESC " Accel Driver");
+MODULE_LICENSE("GPL");
diff --git a/include/uapi/drm/qaic_accel.h b/include/uapi/drm/qaic_accel.h
new file mode 100644
index 0000000..d5fa6f5
--- /dev/null
+++ b/include/uapi/drm/qaic_accel.h
@@ -0,0 +1,283 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
+ *
+ * Copyright (c) 2019-2020, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef QAIC_ACCEL_H_
+#define QAIC_ACCEL_H_
+
+#include <linux/ioctl.h>
+#include <linux/types.h>

These two headers should not be needed here.

Fair enough.  Listed out of paranoia since they define things used in this header.


+#include "drm.h"
+
+#if defined(__CPLUSPLUS)

Use lowercase here: __cplusplus

Doh.  Not sure why uppercase was used.  The referenced examples sure didn't have that.


+extern "C" {
+#endif
+
+#define QAIC_MANAGE_MAX_MSG_LENGTH SZ_4K    /**<
+                          * The length(4K) includes len and
+                          * count fields of qaic_manage_msg
+                          */

I guess these are doxygen style commands but you should be using kernel-doc here.
See https://docs.kernel.org/doc-guide/kernel-doc.html.
This can be used to verify the header:
$ scripts/kernel-doc -v -none include/uapi/drm/qaic_accel.h

include/uapi/drm/drm.h was used as the example for these, and thus it was assumed that was the DRM style.

I'm not convinced kernel-doc is applicable to all of these.  Will review.


+
+enum qaic_sem_flags {

Is there any specific reason for enums if all values are explicitly given?

It is a style pulled from elsewhere.  I can remove the enums.


+    SEM_INSYNCFENCE =    0x1,

All these enums/defines end up in a global user space namespace.
I would advise to prefix everything with QAIC_ or DRM_QAIC_ (e.g. QAIC_SEM_INSYNCFENCE)
to avoid conflicts with other drivers or user space libs.

Good point.  Will do.


+    SEM_OUTSYNCFENCE =    0x2,
+};
+
+enum qaic_sem_cmd {
+    SEM_NOP =        0,
+    SEM_INIT =        1,
+    SEM_INC =        2,
+    SEM_DEC =        3,
+    SEM_WAIT_EQUAL =    4,
+    SEM_WAIT_GT_EQ =    5, /**< Greater than or equal */
+    SEM_WAIT_GT_0 =        6, /**< Greater than 0 */
+};
+
+enum qaic_manage_transaction_type {
+    TRANS_UNDEFINED =            0,
+    TRANS_PASSTHROUGH_FROM_USR =        1,
+    TRANS_PASSTHROUGH_TO_USR =        2,
+    TRANS_PASSTHROUGH_FROM_DEV =        3,
+    TRANS_PASSTHROUGH_TO_DEV =        4,
+    TRANS_DMA_XFER_FROM_USR =        5,
+    TRANS_DMA_XFER_TO_DEV =            6,
+    TRANS_ACTIVATE_FROM_USR =        7,
+    TRANS_ACTIVATE_FROM_DEV =        8,
+    TRANS_ACTIVATE_TO_DEV =            9,
+    TRANS_DEACTIVATE_FROM_USR =        10,
+    TRANS_DEACTIVATE_FROM_DEV =        11,
+    TRANS_STATUS_FROM_USR =            12,
+    TRANS_STATUS_TO_USR =            13,
+    TRANS_STATUS_FROM_DEV =            14,
+    TRANS_STATUS_TO_DEV =            15,
+    TRANS_TERMINATE_FROM_DEV =        16,
+    TRANS_TERMINATE_TO_DEV =        17,
+    TRANS_DMA_XFER_CONT =            18,
+    TRANS_VALIDATE_PARTITION_FROM_DEV =    19,
+    TRANS_VALIDATE_PARTITION_TO_DEV =    20,
+    TRANS_MAX =                21
+};
+
+struct qaic_manage_trans_hdr {
+    __u32 type;    /**< in, value from enum manage_transaction_type */
+    __u32 len;    /**< in, length of this transaction, including the header */
+};
+
+struct qaic_manage_trans_passthrough {
+    struct qaic_manage_trans_hdr hdr;
+    __u8 data[];    /**< in, userspace must encode in little endian */
+};
+
+struct qaic_manage_trans_dma_xfer {
+    struct qaic_manage_trans_hdr hdr;
+    __u32 tag;    /**< in, device specific */
+    __u32 count;    /**< in */
+    __u64 addr;    /**< in, address of the data to transferred via DMA */ +    __u64 size;    /**< in, length of the data to transferred via DMA */
+};
+
+struct qaic_manage_trans_activate_to_dev {
+    struct qaic_manage_trans_hdr hdr;
+    __u32 queue_size;    /**<
+                  * in, number of elements in DBC request
+                  * and respose queue
+                  */
+    __u32 eventfd;        /**< in */
+    __u32 options;        /**< in, device specific */
+    __u32 pad;        /**< pad must be 0 */
+};
+
+struct qaic_manage_trans_activate_from_dev {
+    struct qaic_manage_trans_hdr hdr;
+    __u32 status;    /**< out, status of activate transaction */
+    __u32 dbc_id;    /**< out, Identifier of assigned DMA Bridge channel */
+    __u64 options;    /**< out */
+};
+
+struct qaic_manage_trans_deactivate {
+    struct qaic_manage_trans_hdr hdr;
+    __u32 dbc_id;    /**< in, Identifier of assigned DMA Bridge channel */
+    __u32 pad;    /**< pad must be 0 */
+};
+
+struct qaic_manage_trans_status_to_dev {
+    struct qaic_manage_trans_hdr hdr;
+};
+
+struct qaic_manage_trans_status_from_dev {
+    struct qaic_manage_trans_hdr hdr;
+    __u16 major;    /**< out, major vesrion of NNC protocol used by device */ +    __u16 minor;    /**< out, minor vesrion of NNC protocol used by device */

vesrion -> version

Will fix.


+    __u32 status;    /**< out, status of query transaction  */
+    __u64 status_flags;    /**<
+                  * out
+                  * 0    : If set then device has CRC check enabled
+                  * 1:63 : Unused
+                  */
+};
+
+struct qaic_manage_msg {
+    __u32 len;    /**< in, Length of valid data - ie sum of all transactions */
+    __u32 count;    /**< in, Number of transactions in message */
+    __u64 data;    /**< in, Pointer to array of transactions */
+};
+
+struct qaic_create_bo {
+    __u64 size;    /**< in, Size of BO in byte */
+    __u32 handle;    /**< out, Returned GEM handle for the BO */
+    __u32 pad;    /**< pad must be 0 */
+};
+
+struct qaic_mmap_bo {
+    __u32 handle;    /**< in, Handle for the BO being mapped. */

The comment is missleading. BO is not mapped by this ioctl().

Its mapping the handle to the offset, which is a type of mapping.  I think I get your point though.  Will try to wordsmith this better.


+    __u32 pad;    /**< pad must be 0 */
+    __u64 offset;    /**<
+              * out, offset into the drm node to use for
+              * subsequent mmap call
+              */
+};
+
+/**
+ * @brief semaphore command
+ */
+struct qaic_sem {
+    __u16 val;    /**< in, Only lower 12 bits are valid */
+    __u8  index;    /**< in, Only lower 5 bits are valid */
+    __u8  presync;    /**< in, 1 if presync operation, 0 if postsync */
+    __u8  cmd;    /**< in, See enum sem_cmd */
+    __u8  flags;    /**< in, See sem_flags for valid bits.  All others must be 0 */
+    __u16 pad;    /**< pad must be 0 */
+};
+
+struct qaic_attach_slice_entry {
+    __u64 size;        /**< in, Size memory to allocate for this BO slice */
+    struct qaic_sem    sem0;    /**< in, Must be zero if not valid */
+    struct qaic_sem    sem1;    /**< in, Must be zero if not valid */
+    struct qaic_sem    sem2;    /**< in, Must be zero if not valid */
+    struct qaic_sem    sem3;    /**< in, Must be zero if not valid */
+    __u64 dev_addr;        /**< in, Address in device to/from which data is copied */
+    __u64 db_addr;        /**< in, Doorbell address */
+    __u32 db_data;        /**< in, Data to write to doorbell */
+    __u32 db_len;        /**<
+                  * in, Doorbell length - 32, 16, or 8 bits.
+                  * 0 means doorbell is inactive
+                  */
+    __u64 offset;        /**< in, Offset from start of buffer */
+};
+
+struct qaic_attach_slice_hdr {
+    __u32 count;    /**< in, Number of slices for this BO */
+    __u32 dbc_id;    /**< in, Associate this BO with this DMA Bridge channel */ +    __u32 handle;    /**< in, Handle of BO to which slicing information is to be attached */ +    __u32 dir;    /**< in, Direction of data: 1 = DMA_TO_DEVICE, 2 = DMA_FROM_DEVICE */
+    __u64 size;    /**<
+              * in, Total length of BO
+              * If BO is imported (DMABUF/PRIME) then this size
+              * should not exceed the size of DMABUF provided.
+              * If BO is allocated using DRM_IOCTL_QAIC_CREATE_BO
+              * then this size should be exactly same as the size
+              * provided during DRM_IOCTL_QAIC_CREATE_BO.
+              */
+};
+
+struct qaic_attach_slice {
+    struct qaic_attach_slice_hdr hdr;
+    __u64 data;    /**<
+              * in, Pointer to a buffer which is container of
+              * struct qaic_attach_slice_entry[]
+              */
+};
+
+struct qaic_execute_entry {
+    __u32 handle;    /**< in, buffer handle */
+    __u32 dir;    /**< in, 1 = to device, 2 = from device */
+};
+
+struct qaic_partial_execute_entry {
+    __u32 handle;    /**< in, buffer handle */
+    __u32 dir;    /**< in, 1 = to device, 2 = from device */
+    __u64 resize;    /**< in, 0 = no resize */
+};
+
+struct qaic_execute_hdr {
+    __u32 count;    /**< in, number of executes following this header */ +    __u32 dbc_id;    /**< in, Identifier of assigned DMA Bridge channel */
+};
+
+struct qaic_execute {
+    struct qaic_execute_hdr hdr;
+    __u64 data;    /**< in, qaic_execute_entry or qaic_partial_execute_entry container */
+};
+
+struct qaic_wait {
+    __u32 handle;    /**< in, handle to wait on until execute is complete */
+    __u32 timeout;    /**< in, timeout for wait(in ms) */
+    __u32 dbc_id;    /**< in, Identifier of assigned DMA Bridge channel */
+    __u32 pad;    /**< pad must be 0 */
+};
+
+struct qaic_perf_stats_hdr {
+    __u16 count;    /**< in, Total number BOs requested */
+    __u16 pad;    /**< pad must be 0 */
+    __u32 dbc_id;    /**< in, Identifier of assigned DMA Bridge channel */
+};
+
+struct qaic_perf_stats {
+    struct qaic_perf_stats_hdr hdr;
+    __u64 data;    /**< in, qaic_perf_stats_entry container */
+};
+
+struct qaic_perf_stats_entry {
+    __u32 handle;            /**< in, Handle of the memory request */
+    __u32 queue_level_before;    /**<
+                      * out, Number of elements in queue
+                      * before submission given memory request
+                      */
+    __u32 num_queue_element;    /**<
+                      * out, Number of elements to add in the
+                      * queue for given memory request
+                      */
+    __u32 submit_latency_us;    /**<
+                      * out, Time taken by kernel to submit
+                      * the request to device
+                      */
+    __u32 device_latency_us;    /**<
+                      * out, Time taken by device to execute the
+                      * request. 0 if request is not completed
+                      */
+    __u32 pad;            /**< pad must be 0 */
+};
+
+struct qaic_part_dev {
+    __u32 partition_id;    /**< in, reservation id */
+    __u16 remove;        /**< in, 1 - Remove device 0 - Create device */
+    __u16 pad;        /**< pad must be 0 */
+};
+
+#define DRM_QAIC_MANAGE                0x00
+#define DRM_QAIC_CREATE_BO            0x01
+#define DRM_QAIC_MMAP_BO            0x02

I know that MMAP_BO ioctl is common in drm drivers but in my opinion it is a very poor name. I suggest naming it BO_INFO so in future you could extend it with other bo params besides
mmap offset.

I see your point, but I don't see how to implement it.

Let us assume this IOCTL is renamed DRM_QAIC_BO_INFO, and struct qaic_mmap_bo is named qaic_bo_info.

qaic_bo_info is part of the uAPI.  If, "tomorrow" we need to add a field to it, we cannot.  That changes the structure size and breaks the uAPI.

I can't add reserved fields in anticipation of future use since GregKH had said not to do that - https://lore.kernel.org/all/20200514163734.GB3154055@xxxxxxxxx/

So, I'm thinking the only approach would be a linked list similar to EXECUTE_BO where qaic_bo_info holds a userspace pointer that is the head of the list, and new list nodes can be defined in future.  That seems like an overengineered solution to some hypothetical future which may not come to pass.  If this is the solution, then I think I'd prefer to leave this ioctl as is, and deprecate it if the extension usecase ever comes to pass.

Thoughts?  Am I missing something?


+#define DRM_QAIC_ATTACH_SLICE_BO        0x03
+#define DRM_QAIC_EXECUTE_BO            0x04
+#define DRM_QAIC_PARTIAL_EXECUTE_BO        0x05
+#define DRM_QAIC_WAIT_BO            0x06
+#define DRM_QAIC_PERF_STATS_BO            0x07
+#define DRM_QAIC_PART_DEV            0x08
+
+#define DRM_IOCTL_QAIC_MANAGE            DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_MANAGE, struct qaic_manage_msg) +#define DRM_IOCTL_QAIC_CREATE_BO        DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_CREATE_BO,    struct qaic_create_bo) +#define DRM_IOCTL_QAIC_MMAP_BO            DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_MMAP_BO, struct qaic_mmap_bo) +#define DRM_IOCTL_QAIC_ATTACH_SLICE_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_ATTACH_SLICE_BO, struct qaic_attach_slice) +#define DRM_IOCTL_QAIC_EXECUTE_BO        DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_EXECUTE_BO,    struct qaic_execute) +#define DRM_IOCTL_QAIC_PARTIAL_EXECUTE_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_PARTIAL_EXECUTE_BO,    struct qaic_execute) +#define DRM_IOCTL_QAIC_WAIT_BO            DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_WAIT_BO, struct qaic_wait) +#define DRM_IOCTL_QAIC_PERF_STATS_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_PERF_STATS_BO, struct qaic_perf_stats) +#define DRM_IOCTL_QAIC_PART_DEV DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_PART_DEV, struct qaic_part_dev)
+
+#if defined(__CPLUSPLUS)

Use lowercase here: __cplusplus

Will do.


+}
+#endif
+
+#endif /* QAIC_ACCEL_H_ */

Regards,
Jacek






[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux