@@ -8,3 +8,4 @@ obj-$(CONFIG_INTEL_DLB) := dlb.o
dlb-objs := dlb_main.o
dlb-objs += dlb_pf_ops.o dlb_resource.o dlb_ioctl.o
+dlb-objs += dlb_file.o
new file mode 100644
@@ -0,0 +1,149 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright(C) 2016-2020 Intel Corporation. All rights reserved. */
+
+#include <linux/anon_inodes.h>
+#include <linux/file.h>
+#include <linux/module.h>
+#include <linux/mount.h>
+#include <linux/pseudo_fs.h>
+
+#include "dlb_main.h"
+
+/*
+ * dlb tracks its memory mappings so it can revoke them when an FLR is
+ * requested and user-space cannot be allowed to access the device. To achieve
+ * that, the driver creates a single inode through which all driver-created
+ * files can share a struct address_space, and unmaps the inode's address space
+ * during the reset preparation phase. Since the anon inode layer shares its
+ * inode with multiple kernel components, we cannot use that here.
+ *
+ * Doing so requires a custom pseudo-filesystem to allocate the inode. The FS
+ * and the inode are allocated on demand when a file is created, and both are
+ * freed when the last such file is closed.
+ *
+ * This is inspired by other drivers (cxl, dax, mem) and the anon inode layer.
+ */
+static int dlb_fs_cnt;
+static struct vfsmount *dlb_vfs_mount;
+
+#define DLBFS_MAGIC 0x444C4232 /* ASCII for DLB */
+static int dlb_init_fs_context(struct fs_context *fc)
+{
+ return init_pseudo(fc, DLBFS_MAGIC) ? 0 : -ENOMEM;
+}
+
+static struct file_system_type dlb_fs_type = {
+ .name = "dlb",
+ .owner = THIS_MODULE,
+ .init_fs_context = dlb_init_fs_context,
+ .kill_sb = kill_anon_super,
+};
+
+/* Allocate an anonymous inode. Must hold the resource mutex while calling. */
+static struct inode *dlb_alloc_inode(struct dlb *dlb)
+{
+ struct inode *inode;
+ int ret;
+
+ /* Increment the pseudo-FS's refcnt and (if not already) mount it. */
+ ret = simple_pin_fs(&dlb_fs_type, &dlb_vfs_mount, &dlb_fs_cnt);
+ if (ret < 0) {
+ dev_err(dlb->dev,
+ "[%s()] Cannot mount pseudo filesystem: %d\n",
+ __func__, ret);
+ return ERR_PTR(ret);
+ }
+
+ dlb->inode_cnt++;
+
+ if (dlb->inode_cnt > 1) {
+ /*
+ * Return the previously allocated inode. In this case, there
+ * is guaranteed >= 1 reference and so ihold() is safe to call.
+ */
+ ihold(dlb->inode);
+ return dlb->inode;
+ }
+
+ inode = alloc_anon_inode(dlb_vfs_mount->mnt_sb);
+ if (IS_ERR(inode)) {
+ dev_err(dlb->dev,
+ "[%s()] Cannot allocate inode: %ld\n",
+ __func__, PTR_ERR(inode));
+ dlb->inode_cnt = 0;
+ simple_release_fs(&dlb_vfs_mount, &dlb_fs_cnt);
+ }
+
+ dlb->inode = inode;
+
+ return inode;
+}
+
+/*
+ * Decrement the inode reference count and release the FS. Intended for
+ * unwinding dlb_alloc_inode(). Must hold the resource mutex while calling.
+ */
+static void dlb_free_inode(struct inode *inode)
+{
+ iput(inode);
+ simple_release_fs(&dlb_vfs_mount, &dlb_fs_cnt);
+}
+
+/*
+ * Release the FS. Intended for use in a file_operations release callback,
+ * which decrements the inode reference count separately. Must hold the
+ * resource mutex while calling.
+ */
+void dlb_release_fs(struct dlb *dlb)
+{
+ mutex_lock(&dlb_driver_mutex);
+
+ simple_release_fs(&dlb_vfs_mount, &dlb_fs_cnt);
+
+ dlb->inode_cnt--;
+
+ /* When the fs refcnt reaches zero, the inode has been freed */
+ if (dlb->inode_cnt == 0)
+ dlb->inode = NULL;
+
+ mutex_unlock(&dlb_driver_mutex);
+}
+
+/*
+ * Allocate a file with the requested flags, file operations, and name that
+ * uses the device's shared inode. Must hold the resource mutex while calling.
+ *
+ * Caller must separately allocate an fd and install the file in that fd.
+ */
+struct file *dlb_getfile(struct dlb *dlb,
+ int flags,
+ const struct file_operations *fops,
+ const char *name)
+{
+ struct inode *inode;
+ struct file *f;
+
+ if (!try_module_get(THIS_MODULE))
+ return ERR_PTR(-ENOENT);
+
+ mutex_lock(&dlb_driver_mutex);
+
+ inode = dlb_alloc_inode(dlb);
+ if (IS_ERR(inode)) {
+ mutex_unlock(&dlb_driver_mutex);
+ module_put(THIS_MODULE);
+ return ERR_CAST(inode);
+ }
+
+ f = alloc_file_pseudo(inode, dlb_vfs_mount, name, flags, fops);
+ if (IS_ERR(f)) {
+ dlb_free_inode(inode);
+ mutex_unlock(&dlb_driver_mutex);
+ module_put(THIS_MODULE);
+ return f;
+ }
+
+ mutex_unlock(&dlb_driver_mutex);
+
+ return f;
+}
@@ -66,12 +66,12 @@
#define DLB_LDB_PP_STRIDE 0x1000
#define DLB_LDB_PP_BOUND (DLB_LDB_PP_BASE + \
DLB_LDB_PP_STRIDE * DLB_MAX_NUM_LDB_PORTS)
-#define DLB_LDB_PP_OFFS(id) (DLB_LDB_PP_BASE + (id) * DLB_PP_SIZE)
+#define DLB_LDB_PP_OFFSET(id) (DLB_LDB_PP_BASE + (id) * DLB_PP_SIZE)
#define DLB_DIR_PP_BASE 0x2000000
#define DLB_DIR_PP_STRIDE 0x1000
#define DLB_DIR_PP_BOUND (DLB_DIR_PP_BASE + \
DLB_DIR_PP_STRIDE * DLB_MAX_NUM_DIR_PORTS)
-#define DLB_DIR_PP_OFFS(id) (DLB_DIR_PP_BASE + (id) * DLB_PP_SIZE)
+#define DLB_DIR_PP_OFFSET(id) (DLB_DIR_PP_BASE + (id) * DLB_PP_SIZE)
struct dlb_resource_id {
u32 phys_id;
@@ -161,6 +161,141 @@ static int dlb_domain_ioctl_create_dir_port(struct dlb *dlb,
return ret;
}
+static int dlb_create_port_fd(struct dlb *dlb, const char *prefix, u32 id,
+ const struct file_operations *fops,
+ int *fd, struct file **f)
+{
+ char *name;
+ int ret;
+
+ ret = get_unused_fd_flags(O_RDWR);
+ if (ret < 0)
+ return ret;
+
+ *fd = ret;
+
+ name = kasprintf(GFP_KERNEL, "%s:%d", prefix, id);
+ if (!name) {
+ put_unused_fd(*fd);
+ return -ENOMEM;
+ }
+
+ *f = dlb_getfile(dlb, O_RDWR | O_CLOEXEC, fops, name);
+
+ kfree(name);
+
+ if (IS_ERR(*f)) {
+ put_unused_fd(*fd);
+ return PTR_ERR(*f);
+ }
+
+ return 0;
+}
+
+static int dlb_domain_get_port_fd(struct dlb *dlb, struct dlb_domain *domain,
+ unsigned long user_arg, const char *name,
+ const struct file_operations *fops,
+ bool is_ldb)
+{
+ struct dlb_get_port_fd_args __user *uarg;
+ struct dlb_cmd_response response = {0};
+ struct dlb_get_port_fd_args arg;
+ struct dlb_port *port;
+ struct file *file;
+ int ret, fd;
+
+ uarg = (void __user *)user_arg;
+ if (copy_from_user(&arg, uarg, sizeof(arg)))
+ return -EFAULT;
+
+ mutex_lock(&dlb->resource_mutex);
+
+ if ((is_ldb &&
+ dlb->ops->ldb_port_owned_by_domain(&dlb->hw, domain->id,
+ arg.port_id) != 1)) {
+ response.status = DLB_ST_INVALID_PORT_ID;
+ ret = -EINVAL;
+ goto end;
+ }
+
+ if (!is_ldb &&
+ dlb->ops->dir_port_owned_by_domain(&dlb->hw, domain->id,
+ arg.port_id) != 1) {
+ response.status = DLB_ST_INVALID_PORT_ID;
+ ret = -EINVAL;
+ goto end;
+ }
+
+ port = (is_ldb) ? &dlb->ldb_port[arg.port_id] :
+ &dlb->dir_port[arg.port_id];
+
+ if (!port->valid) {
+ response.status = DLB_ST_INVALID_PORT_ID;
+ ret = -EINVAL;
+ goto end;
+ }
+
+ ret = dlb_create_port_fd(dlb, name, arg.port_id, fops, &fd, &file);
+ if (ret < 0)
+ goto end;
+
+ file->private_data = port;
+
+ response.id = fd;
+
+end:
+ BUILD_BUG_ON(offsetof(typeof(arg), response) != 0);
+
+ if (copy_to_user((void __user *)&uarg->response, &response, sizeof(response)))
+ ret = -EFAULT;
+
+ /*
+ * Save fd_install() until after the last point of failure. The domain
+ * refcnt is decremented in the close callback.
+ */
+ if (ret == 0) {
+ kref_get(&domain->refcnt);
+
+ fd_install(fd, file);
+ }
+
+ mutex_unlock(&dlb->resource_mutex);
+
+ return ret;
+}
+
+static int dlb_domain_ioctl_get_ldb_port_pp_fd(struct dlb *dlb,
+ struct dlb_domain *domain,
+ unsigned long user_arg)
+{
+ return dlb_domain_get_port_fd(dlb, domain, user_arg,
+ "dlb_ldb_pp:", &dlb_pp_fops, true);
+}
+
+static int dlb_domain_ioctl_get_ldb_port_cq_fd(struct dlb *dlb,
+ struct dlb_domain *domain,
+ unsigned long user_arg)
+{
+ return dlb_domain_get_port_fd(dlb, domain, user_arg,
+ "dlb_ldb_cq:", &dlb_cq_fops, true);
+}
+
+static int dlb_domain_ioctl_get_dir_port_pp_fd(struct dlb *dlb,
+ struct dlb_domain *domain,
+ unsigned long user_arg)
+{
+ return dlb_domain_get_port_fd(dlb, domain, user_arg,
+ "dlb_dir_pp:", &dlb_pp_fops, false);
+}
+
+static int dlb_domain_ioctl_get_dir_port_cq_fd(struct dlb *dlb,
+ struct dlb_domain *domain,
+ unsigned long user_arg)
+{
+ return dlb_domain_get_port_fd(dlb, domain, user_arg,
+ "dlb_dir_cq:", &dlb_cq_fops, false);
+}
+
long dlb_domain_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
{
struct dlb_domain *dom = f->private_data;
@@ -179,6 +314,14 @@ long dlb_domain_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
return dlb_domain_ioctl_create_ldb_port(dlb, dom, arg);
case DLB_IOC_CREATE_DIR_PORT:
return dlb_domain_ioctl_create_dir_port(dlb, dom, arg);
+ case DLB_IOC_GET_LDB_PORT_PP_FD:
+ return dlb_domain_ioctl_get_ldb_port_pp_fd(dlb, dom, arg);
+ case DLB_IOC_GET_LDB_PORT_CQ_FD:
+ return dlb_domain_ioctl_get_ldb_port_cq_fd(dlb, dom, arg);
+ case DLB_IOC_GET_DIR_PORT_PP_FD:
+ return dlb_domain_ioctl_get_dir_port_pp_fd(dlb, dom, arg);
+ case DLB_IOC_GET_DIR_PORT_CQ_FD:
+ return dlb_domain_ioctl_get_dir_port_cq_fd(dlb, dom, arg);
default:
return -ENOTTY;
}
@@ -17,6 +17,9 @@
MODULE_LICENSE("GPL v2");
MODULE_DESCRIPTION("Intel(R) Dynamic Load Balancer (DLB) Driver");
+/* The driver mutex protects data structures that used by multiple devices. */
+DEFINE_MUTEX(dlb_driver_mutex);
+
static struct class *dlb_class;
static struct cdev dlb_cdev;
static dev_t dlb_devt;
@@ -233,6 +236,121 @@ const struct file_operations dlb_domain_fops = {
.compat_ioctl = compat_ptr_ioctl,
};
+static unsigned long dlb_get_pp_addr(struct dlb *dlb, struct dlb_port *port)
+{
+ unsigned long pgoff = dlb->hw.func_phys_addr;
+
+ if (port->is_ldb)
+ pgoff += DLB_LDB_PP_OFFSET(port->id);
+ else
+ pgoff += DLB_DIR_PP_OFFSET(port->id);
+
+ return pgoff;
+}
+
+static int dlb_pp_mmap(struct file *f, struct vm_area_struct *vma)
+{
+ struct dlb_port *port = f->private_data;
+ struct dlb_domain *domain = port->domain;
+ struct dlb *dlb = domain->dlb;
+ unsigned long pgoff;
+ pgprot_t pgprot;
+ int ret;
+
+ mutex_lock(&dlb->resource_mutex);
+
+ if ((vma->vm_end - vma->vm_start) != DLB_PP_SIZE) {
+ ret = -EINVAL;
+ goto end;
+ }
+
+ pgprot = pgprot_noncached(vma->vm_page_prot);
+
+ pgoff = dlb_get_pp_addr(dlb, port);
+ ret = io_remap_pfn_range(vma,
+ vma->vm_start,
+ pgoff >> PAGE_SHIFT,
+ vma->vm_end - vma->vm_start,
+ pgprot);
+
+end:
+ mutex_unlock(&dlb->resource_mutex);
+
+ return ret;
+}
+
+static int dlb_cq_mmap(struct file *f, struct vm_area_struct *vma)
+{
+ struct dlb_port *port = f->private_data;
+ struct dlb_domain *domain = port->domain;
+ struct dlb *dlb = domain->dlb;
+ struct page *page;
+ int ret;
+
+ mutex_lock(&dlb->resource_mutex);
+
+ if ((vma->vm_end - vma->vm_start) != DLB_CQ_SIZE) {
+ ret = -EINVAL;
+ goto end;
+ }
+
+ page = virt_to_page(port->cq_base);
+
+ ret = remap_pfn_range(vma,
+ vma->vm_start,
+ page_to_pfn(page),
+ vma->vm_end - vma->vm_start,
+ vma->vm_page_prot);
+end:
+ mutex_unlock(&dlb->resource_mutex);
+
+ return ret;
+}
+
+static void dlb_port_unmap(struct dlb *dlb, struct dlb_port *port)
+{
+ if (!port->cq_base) {
+ unmap_mapping_range(dlb->inode->i_mapping,
+ (unsigned long)port->cq_base,
+ DLB_CQ_SIZE, 1);
+ } else {
+ unmap_mapping_range(dlb->inode->i_mapping,
+ dlb_get_pp_addr(dlb, port),
+ DLB_PP_SIZE, 1);
+ }
+}
+
+static int dlb_port_close(struct inode *i, struct file *f)
+{
+ struct dlb_port *port = f->private_data;
+ struct dlb_domain *domain = port->domain;
+ struct dlb *dlb = domain->dlb;
+
+ mutex_lock(&dlb->resource_mutex);
+
+ kref_put(&domain->refcnt, dlb_free_domain);
+
+ dlb_port_unmap(dlb, port);
+ /* Decrement the refcnt of the pseudo-FS used to allocate the inode */
+ dlb_release_fs(dlb);
+
+ mutex_unlock(&dlb->resource_mutex);
+
+ return 0;
+}
+
+const struct file_operations dlb_pp_fops = {
+ .owner = THIS_MODULE,
+ .release = dlb_port_close,
+ .mmap = dlb_pp_mmap,
+};
+
+const struct file_operations dlb_cq_fops = {
+ .owner = THIS_MODULE,
+ .release = dlb_port_close,
+ .mmap = dlb_cq_mmap,
+};
+
/**********************************/
/****** PCI driver callbacks ******/
/**********************************/
@@ -11,6 +11,7 @@
#include <linux/mutex.h>
#include <linux/pci.h>
#include <linux/types.h>
+#include <linux/file.h>
#include <uapi/linux/dlb.h>
@@ -27,6 +28,8 @@
#define DLB_NUM_FUNCS_PER_DEVICE (1 + DLB_MAX_NUM_VDEVS)
#define DLB_MAX_NUM_DEVICES (DLB_MAX_NUM_PFS * DLB_NUM_FUNCS_PER_DEVICE)
+extern struct mutex dlb_driver_mutex;
+
enum dlb_device_type {
DLB_PF,
};
@@ -63,6 +66,12 @@ struct dlb_device_ops {
int (*get_num_resources)(struct dlb_hw *hw,
struct dlb_get_num_resources_args *args);
int (*reset_domain)(struct dlb_hw *hw, u32 domain_id);
+ int (*ldb_port_owned_by_domain)(struct dlb_hw *hw,
+ u32 domain_id,
+ u32 port_id);
+ int (*dir_port_owned_by_domain)(struct dlb_hw *hw,
+ u32 domain_id,
+ u32 port_id);
int (*get_ldb_queue_depth)(struct dlb_hw *hw,
u32 domain_id,
struct dlb_get_ldb_queue_depth_args *args,
@@ -78,6 +87,8 @@ struct dlb_device_ops {
extern struct dlb_device_ops dlb_pf_ops;
extern const struct file_operations dlb_domain_fops;
+extern const struct file_operations dlb_pp_fops;
+extern const struct file_operations dlb_cq_fops;
struct dlb_port {
void *cq_base;
@@ -103,6 +114,11 @@ struct dlb {
struct file *f;
struct dlb_port ldb_port[DLB_MAX_NUM_LDB_PORTS];
struct dlb_port dir_port[DLB_MAX_NUM_DIR_PORTS];
+ /*
+ * Anonymous inode used to share an address_space for all domain
+ * device file mappings.
+ */
+ struct inode *inode;
/*
* The resource mutex serializes access to driver data structures and
* hardware registers.
@@ -110,6 +126,7 @@ struct dlb {
struct mutex resource_mutex;
enum dlb_device_type type;
int id;
+ u32 inode_cnt;
dev_t dev_number;
u8 domain_reset_failed;
};
@@ -118,6 +135,13 @@ struct dlb {
long dlb_ioctl(struct file *f, unsigned int cmd, unsigned long arg);
long dlb_domain_ioctl(struct file *f, unsigned int cmd, unsigned long arg);
+/* Prototypes for dlb_file.c */
+void dlb_release_fs(struct dlb *dlb);
+struct file *dlb_getfile(struct dlb *dlb,
+ int flags,
+ const struct file_operations *fops,
+ const char *name);
+
int dlb_init_domain(struct dlb *dlb, u32 domain_id);
void dlb_free_domain(struct kref *kref);
@@ -201,6 +201,22 @@ dlb_pf_query_cq_poll_mode(struct dlb *dlb, struct dlb_cmd_response *user_resp)
return 0;
}
+/**************************************/
+/****** Resource query callbacks ******/
+/**************************************/
+
+static int
+dlb_pf_ldb_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id)
+{
+ return dlb_ldb_port_owned_by_domain(hw, domain_id, port_id, false, 0);
+}
+
+static int
+dlb_pf_dir_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id)
+{
+ return dlb_dir_port_owned_by_domain(hw, domain_id, port_id, false, 0);
+}
+
/********************************/
/****** DLB PF Device Ops ******/
/********************************/
@@ -218,6 +234,8 @@ struct dlb_device_ops dlb_pf_ops = {
.create_dir_port = dlb_pf_create_dir_port,
.get_num_resources = dlb_pf_get_num_resources,
.reset_domain = dlb_pf_reset_domain,
+ .ldb_port_owned_by_domain = dlb_pf_ldb_port_owned_by_domain,
+ .dir_port_owned_by_domain = dlb_pf_dir_port_owned_by_domain,
.get_ldb_queue_depth = dlb_pf_get_ldb_queue_depth,
.get_dir_queue_depth = dlb_pf_get_dir_queue_depth,
.init_hardware = dlb_pf_init_hardware,
@@ -236,6 +236,32 @@ static struct dlb_hw_domain *dlb_get_domain_from_id(struct dlb_hw *hw, u32 id,
return NULL;
}
+static struct dlb_ldb_port *
+dlb_get_domain_ldb_port(u32 id, bool vdev_req, struct dlb_hw_domain *domain)
+{
+ struct dlb_ldb_port *port;
+ int i;
+
+ if (id >= DLB_MAX_NUM_LDB_PORTS)
+ return NULL;
+
+ for (i = 0; i < DLB_NUM_COS_DOMAINS; i++) {
+ list_for_each_entry(port, &domain->used_ldb_ports[i], domain_list) {
+ if ((!vdev_req && port->id.phys_id == id) ||
+ (vdev_req && port->id.virt_id == id))
+ return port;
+ }
+
+ list_for_each_entry(port, &domain->avail_ldb_ports[i], domain_list) {
+ if ((!vdev_req && port->id.phys_id == id) ||
+ (vdev_req && port->id.virt_id == id))
+ return port;
+ }
+ }
+
+ return NULL;
+}
+
static struct dlb_dir_pq_pair *
dlb_get_domain_used_dir_pq(u32 id, bool vdev_req, struct dlb_hw_domain *domain)
{
@@ -253,6 +279,29 @@ dlb_get_domain_used_dir_pq(u32 id, bool vdev_req, struct dlb_hw_domain *domain)
return NULL;
}
+static struct dlb_dir_pq_pair *
+dlb_get_domain_dir_pq(u32 id, bool vdev_req, struct dlb_hw_domain *domain)
+{
+ struct dlb_dir_pq_pair *port;
+
+ if (id >= DLB_MAX_NUM_DIR_PORTS)
+ return NULL;
+
+ list_for_each_entry(port, &domain->used_dir_pq_pairs, domain_list) {
+ if ((!vdev_req && port->id.phys_id == id) ||
+ (vdev_req && port->id.virt_id == id))
+ return port;
+ }
+
+ list_for_each_entry(port, &domain->avail_dir_pq_pairs, domain_list) {
+ if ((!vdev_req && port->id.phys_id == id) ||
+ (vdev_req && port->id.virt_id == id))
+ return port;
+ }
+
+ return NULL;
+}
+
static struct dlb_ldb_queue *
dlb_get_domain_ldb_queue(u32 id, bool vdev_req, struct dlb_hw_domain *domain)
{
@@ -3394,6 +3443,88 @@ int dlb_reset_domain(struct dlb_hw *hw, u32 domain_id, bool vdev_req,
return dlb_domain_reset_software_state(hw, domain);
}
+/**
+ * dlb_ldb_port_owned_by_domain() - query whether a port is owned by a domain
+ * @hw: dlb_hw handle for a particular device.
+ * @domain_id: domain ID.
+ * @port_id: port ID.
+ * @vdev_req: indicates whether this request came from a vdev.
+ * @vdev_id: If vdev_req is true, this contains the vdev's ID.
+ *
+ * This function returns whether a load-balanced port is owned by a specified
+ * domain.
+ *
+ * A vdev can be either an SR-IOV virtual function or a Scalable IOV virtual
+ * device.
+ *
+ * Return:
+ * Returns 0 if false, 1 if true, <0 otherwise.
+ *
+ * EINVAL - Invalid domain or port ID, or the domain is not configured.
+ */
+int dlb_ldb_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id,
+ bool vdev_req, unsigned int vdev_id)
+{
+ struct dlb_hw_domain *domain;
+ struct dlb_ldb_port *port;
+
+ if (vdev_req && vdev_id >= DLB_MAX_NUM_VDEVS)
+ return -EINVAL;
+
+ domain = dlb_get_domain_from_id(hw, domain_id, vdev_req, vdev_id);
+
+ if (!domain || !domain->configured)
+ return -EINVAL;
+
+ port = dlb_get_domain_ldb_port(port_id, vdev_req, domain);
+
+ if (!port)
+ return -EINVAL;
+
+ return port->domain_id.phys_id == domain->id.phys_id;
+}
+
+/**
+ * dlb_dir_port_owned_by_domain() - query whether a port is owned by a domain
+ * @hw: dlb_hw handle for a particular device.
+ * @domain_id: domain ID.
+ * @port_id: port ID.
+ * @vdev_req: indicates whether this request came from a vdev.
+ * @vdev_id: If vdev_req is true, this contains the vdev's ID.
+ *
+ * This function returns whether a directed port is owned by a specified
+ * domain.
+ *
+ * A vdev can be either an SR-IOV virtual function or a Scalable IOV virtual
+ * device.
+ *
+ * Return:
+ * Returns 0 if false, 1 if true, <0 otherwise.
+ *
+ * EINVAL - Invalid domain or port ID, or the domain is not configured.
+ */
+int dlb_dir_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id,
+ bool vdev_req, unsigned int vdev_id)
+{
+ struct dlb_dir_pq_pair *port;
+ struct dlb_hw_domain *domain;
+
+ if (vdev_req && vdev_id >= DLB_MAX_NUM_VDEVS)
+ return -EINVAL;
+
+ domain = dlb_get_domain_from_id(hw, domain_id, vdev_req, vdev_id);
+
+ if (!domain || !domain->configured)
+ return -EINVAL;
+
+ port = dlb_get_domain_dir_pq(port_id, vdev_req, domain);
+
+ if (!port)
+ return -EINVAL;
+
+ return port->domain_id.phys_id == domain->id.phys_id;
+}
+
/**
* dlb_hw_get_num_resources() - query the PCI function's available resources
* @hw: dlb_hw handle for a particular device.
@@ -44,6 +44,12 @@ int dlb_hw_create_ldb_port(struct dlb_hw *hw, u32 domain_id,
int dlb_reset_domain(struct dlb_hw *hw, u32 domain_id, bool vdev_req,
unsigned int vdev_id);
+int dlb_ldb_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id,
+ bool vdev_req, unsigned int vdev_id);
+
+int dlb_dir_port_owned_by_domain(struct dlb_hw *hw, u32 domain_id, u32 port_id,
+ bool vdev_req, unsigned int vdev_id);
+
int dlb_hw_get_num_resources(struct dlb_hw *hw,
struct dlb_get_num_resources_args *arg,
bool vdev_req, unsigned int vdev_id);
@@ -374,6 +374,40 @@ struct dlb_create_dir_port_args {
__s32 queue_id;
};
+/*
+ * DLB_CMD_GET_LDB_PORT_PP_FD: Get file descriptor to mmap a load-balanced
+ * port's producer port (PP).
+ * DLB_CMD_GET_LDB_PORT_CQ_FD: Get file descriptor to mmap a load-balanced
+ * port's consumer queue (CQ).
+ *
+ * The load-balanced port must have been previously created with the ioctl
+ * DLB_CMD_CREATE_LDB_PORT. The fd is used to mmap the PP/CQ region.
+ *
+ * DLB_CMD_GET_DIR_PORT_PP_FD: Get file descriptor to mmap a directed port's
+ * producer port (PP).
+ * DLB_CMD_GET_DIR_PORT_CQ_FD: Get file descriptor to mmap a directed port's
+ * consumer queue (CQ).
+ *
+ * The directed port must have been previously created with the ioctl
+ * DLB_CMD_CREATE_DIR_PORT. The fd is used to mmap PP/CQ region.
+ *
+ * Output parameters:
+ * @response.status: Detailed error code. In certain cases, such as if the
+ * ioctl request arg is invalid, the driver won't set status.
+ * @response.id: fd.
+ *
+ * Input parameters:
+ * @port_id: port ID.
+ * @padding0: Reserved for future use.
+ */
+struct dlb_get_port_fd_args {
+ /* Output parameters */
+ struct dlb_cmd_response response;
+ /* Input parameters */
+ __u32 port_id;
+ __u32 padding0;
+};
+
enum dlb_domain_user_interface_commands {
DLB_DOMAIN_CMD_CREATE_LDB_QUEUE,
DLB_DOMAIN_CMD_CREATE_DIR_QUEUE,
@@ -381,12 +415,21 @@ enum dlb_domain_user_interface_commands {
DLB_DOMAIN_CMD_GET_DIR_QUEUE_DEPTH,
DLB_DOMAIN_CMD_CREATE_LDB_PORT,
DLB_DOMAIN_CMD_CREATE_DIR_PORT,
+ DLB_DOMAIN_CMD_GET_LDB_PORT_PP_FD,
+ DLB_DOMAIN_CMD_GET_LDB_PORT_CQ_FD,
+ DLB_DOMAIN_CMD_GET_DIR_PORT_PP_FD,
+ DLB_DOMAIN_CMD_GET_DIR_PORT_CQ_FD,
/* NUM_DLB_DOMAIN_CMD must be last */
NUM_DLB_DOMAIN_CMD,
};
+/*
+ * Mapping sizes for memory mapping the consumer queue (CQ) memory space, and
+ * producer port (PP) MMIO space.
+ */
#define DLB_CQ_SIZE 65536
+#define DLB_PP_SIZE 4096
/********************/
/* dlb ioctl codes */
@@ -434,5 +477,21 @@ enum dlb_domain_user_interface_commands {
_IOWR(DLB_IOC_MAGIC, \
DLB_DOMAIN_CMD_CREATE_DIR_PORT, \
struct dlb_create_dir_port_args)
+#define DLB_IOC_GET_LDB_PORT_PP_FD \
+ _IOWR(DLB_IOC_MAGIC, \
+ DLB_DOMAIN_CMD_GET_LDB_PORT_PP_FD, \
+ struct dlb_get_port_fd_args)
+#define DLB_IOC_GET_LDB_PORT_CQ_FD \
+ _IOWR(DLB_IOC_MAGIC, \
+ DLB_DOMAIN_CMD_GET_LDB_PORT_CQ_FD, \
+ struct dlb_get_port_fd_args)
+#define DLB_IOC_GET_DIR_PORT_PP_FD \
+ _IOWR(DLB_IOC_MAGIC, \
+ DLB_DOMAIN_CMD_GET_DIR_PORT_PP_FD, \
+ struct dlb_get_port_fd_args)
+#define DLB_IOC_GET_DIR_PORT_CQ_FD \
+ _IOWR(DLB_IOC_MAGIC, \
+ DLB_DOMAIN_CMD_GET_DIR_PORT_CQ_FD, \
+ struct dlb_get_port_fd_args)
#endif /* __DLB_H */
Once a port is created, the application can mmap the corresponding DMA memory and MMIO into user-space. This allows user-space applications to do (performance-sensitive) enqueue and dequeue independent of the kernel driver. The mmap callback is only available through special port files: a producer port (PP) file and a consumer queue (CQ) file. User-space gets an fd for these files by calling a new ioctl, DLB_DOMAIN_CMD_GET_{LDB, DIR}_PORT_{PP, CQ}_FD, and passing in a port ID. If the ioctl succeeds, the returned fd can be used to mmap that port's PP/CQ. Device reset requires first unmapping all user-space mappings, to prevent applications from interfering with the reset operation. To this end, the driver uses a single inode -- allocated when the first PP/CQ file is created, and freed when the last such file is closed -- and attaches all port files to this common inode, as done elsewhere in Linux (e.g. cxl, dax). Allocating this inode requires creating a pseudo-filesystem. The driver initializes this FS when the inode is allocated, and frees the FS after the inode is freed. The driver doesn't use anon_inode_getfd() for these port mmap files because the anon inode layer uses a single inode that is shared with other kernel components -- calling unmap_mapping_range() on that shared inode would likely break the kernel. Signed-off-by: Gage Eads <gage.eads@intel.com> Signed-off-by: Mike Ximing Chen <mike.ximing.chen@intel.com> Reviewed-by: Magnus Karlsson <magnus.karlsson@intel.com> Reviewed-by: Dan Williams <dan.j.williams@intel.com> --- drivers/misc/dlb/Makefile | 1 + drivers/misc/dlb/dlb_file.c | 149 ++++++++++++++++++++++++++++++++ drivers/misc/dlb/dlb_hw_types.h | 4 +- drivers/misc/dlb/dlb_ioctl.c | 143 ++++++++++++++++++++++++++++++ drivers/misc/dlb/dlb_main.c | 118 +++++++++++++++++++++++++ drivers/misc/dlb/dlb_main.h | 24 +++++ drivers/misc/dlb/dlb_pf_ops.c | 18 ++++ drivers/misc/dlb/dlb_resource.c | 131 ++++++++++++++++++++++++++++ drivers/misc/dlb/dlb_resource.h | 6 ++ include/uapi/linux/dlb.h | 59 +++++++++++++ 10 files changed, 651 insertions(+), 2 deletions(-) create mode 100644 drivers/misc/dlb/dlb_file.c