misc: mic: SCIF RMA list operations
authorSudeep Dutt <sudeep.dutt@intel.com>
Wed, 30 Sep 2015 01:15:44 +0000 (18:15 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 4 Oct 2015 11:54:54 +0000 (12:54 +0100)
This patch adds the implementation for operations performed on the
list of SCIF windows. Examples of such operations includes adding the
windows to the list of registered (or cached) windows, querying the
list of self or remote windows and unregistering windows. The query
operation is used by SCIF APIs which initiate DMAs, CPU copies or
fences to ensure that a window remains valid during a transfer.

Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Nikhil Rao <nikhil.rao@intel.com>
Signed-off-by: Sudeep Dutt <sudeep.dutt@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/misc/mic/scif/scif_rma_list.c [new file with mode: 0644]
drivers/misc/mic/scif/scif_rma_list.h [new file with mode: 0644]

diff --git a/drivers/misc/mic/scif/scif_rma_list.c b/drivers/misc/mic/scif/scif_rma_list.c
new file mode 100644 (file)
index 0000000..e1ef8da
--- /dev/null
@@ -0,0 +1,291 @@
+/*
+ * Intel MIC Platform Software Stack (MPSS)
+ *
+ * Copyright(c) 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * Intel SCIF driver.
+ *
+ */
+#include "scif_main.h"
+#include <linux/mmu_notifier.h>
+#include <linux/highmem.h>
+
+/*
+ * scif_insert_tcw:
+ *
+ * Insert a temp window to the temp registration list sorted by va_for_temp.
+ * RMA lock must be held.
+ */
+void scif_insert_tcw(struct scif_window *window, struct list_head *head)
+{
+       struct scif_window *curr = NULL;
+       struct scif_window *prev = list_entry(head, struct scif_window, list);
+       struct list_head *item;
+
+       INIT_LIST_HEAD(&window->list);
+       /* Compare with tail and if the entry is new tail add it to the end */
+       if (!list_empty(head)) {
+               curr = list_entry(head->prev, struct scif_window, list);
+               if (curr->va_for_temp < window->va_for_temp) {
+                       list_add_tail(&window->list, head);
+                       return;
+               }
+       }
+       list_for_each(item, head) {
+               curr = list_entry(item, struct scif_window, list);
+               if (curr->va_for_temp > window->va_for_temp)
+                       break;
+               prev = curr;
+       }
+       list_add(&window->list, &prev->list);
+}
+
+/*
+ * scif_insert_window:
+ *
+ * Insert a window to the self registration list sorted by offset.
+ * RMA lock must be held.
+ */
+void scif_insert_window(struct scif_window *window, struct list_head *head)
+{
+       struct scif_window *curr = NULL, *prev = NULL;
+       struct list_head *item;
+
+       INIT_LIST_HEAD(&window->list);
+       list_for_each(item, head) {
+               curr = list_entry(item, struct scif_window, list);
+               if (curr->offset > window->offset)
+                       break;
+               prev = curr;
+       }
+       if (!prev)
+               list_add(&window->list, head);
+       else
+               list_add(&window->list, &prev->list);
+       scif_set_window_ref(window, window->nr_pages);
+}
+
+/*
+ * scif_query_tcw:
+ *
+ * Query the temp cached registration list of ep for an overlapping window
+ * in case of permission mismatch, destroy the previous window. if permissions
+ * match and overlap is partial, destroy the window but return the new range
+ * RMA lock must be held.
+ */
+int scif_query_tcw(struct scif_endpt *ep, struct scif_rma_req *req)
+{
+       struct list_head *item, *temp, *head = req->head;
+       struct scif_window *window;
+       u64 start_va_window, start_va_req = req->va_for_temp;
+       u64 end_va_window, end_va_req = start_va_req + req->nr_bytes;
+
+       if (!req->nr_bytes)
+               return -EINVAL;
+       /*
+        * Avoid traversing the entire list to find out that there
+        * is no entry that matches
+        */
+       if (!list_empty(head)) {
+               window = list_last_entry(head, struct scif_window, list);
+               end_va_window = window->va_for_temp +
+                       (window->nr_pages << PAGE_SHIFT);
+               if (start_va_req > end_va_window)
+                       return -ENXIO;
+       }
+       list_for_each_safe(item, temp, head) {
+               window = list_entry(item, struct scif_window, list);
+               start_va_window = window->va_for_temp;
+               end_va_window = window->va_for_temp +
+                       (window->nr_pages << PAGE_SHIFT);
+               if (start_va_req < start_va_window &&
+                   end_va_req < start_va_window)
+                       break;
+               if (start_va_req >= end_va_window)
+                       continue;
+               if ((window->prot & req->prot) == req->prot) {
+                       if (start_va_req >= start_va_window &&
+                           end_va_req <= end_va_window) {
+                               *req->out_window = window;
+                               return 0;
+                       }
+                       /* expand window */
+                       if (start_va_req < start_va_window) {
+                               req->nr_bytes +=
+                                       start_va_window - start_va_req;
+                               req->va_for_temp = start_va_window;
+                       }
+                       if (end_va_req >= end_va_window)
+                               req->nr_bytes += end_va_window - end_va_req;
+               }
+               /* Destroy the old window to create a new one */
+               __scif_rma_destroy_tcw_helper(window);
+               break;
+       }
+       return -ENXIO;
+}
+
+/*
+ * scif_query_window:
+ *
+ * Query the registration list and check if a valid contiguous
+ * range of windows exist.
+ * RMA lock must be held.
+ */
+int scif_query_window(struct scif_rma_req *req)
+{
+       struct list_head *item;
+       struct scif_window *window;
+       s64 end_offset, offset = req->offset;
+       u64 tmp_min, nr_bytes_left = req->nr_bytes;
+
+       if (!req->nr_bytes)
+               return -EINVAL;
+
+       list_for_each(item, req->head) {
+               window = list_entry(item, struct scif_window, list);
+               end_offset = window->offset +
+                       (window->nr_pages << PAGE_SHIFT);
+               if (offset < window->offset)
+                       /* Offset not found! */
+                       return -ENXIO;
+               if (offset >= end_offset)
+                       continue;
+               /* Check read/write protections. */
+               if ((window->prot & req->prot) != req->prot)
+                       return -EPERM;
+               if (nr_bytes_left == req->nr_bytes)
+                       /* Store the first window */
+                       *req->out_window = window;
+               tmp_min = min((u64)end_offset - offset, nr_bytes_left);
+               nr_bytes_left -= tmp_min;
+               offset += tmp_min;
+               /*
+                * Range requested encompasses
+                * multiple windows contiguously.
+                */
+               if (!nr_bytes_left) {
+                       /* Done for partial window */
+                       if (req->type == SCIF_WINDOW_PARTIAL ||
+                           req->type == SCIF_WINDOW_SINGLE)
+                               return 0;
+                       /* Extra logic for full windows */
+                       if (offset == end_offset)
+                               /* Spanning multiple whole windows */
+                               return 0;
+                               /* Not spanning multiple whole windows */
+                       return -ENXIO;
+               }
+               if (req->type == SCIF_WINDOW_SINGLE)
+                       break;
+       }
+       dev_err(scif_info.mdev.this_device,
+               "%s %d ENXIO\n", __func__, __LINE__);
+       return -ENXIO;
+}
+
+/*
+ * scif_rma_list_unregister:
+ *
+ * Traverse the self registration list starting from window:
+ * 1) Call scif_unregister_window(..)
+ * RMA lock must be held.
+ */
+int scif_rma_list_unregister(struct scif_window *window,
+                            s64 offset, int nr_pages)
+{
+       struct scif_endpt *ep = (struct scif_endpt *)window->ep;
+       struct list_head *head = &ep->rma_info.reg_list;
+       s64 end_offset;
+       int err = 0;
+       int loop_nr_pages;
+       struct scif_window *_window;
+
+       list_for_each_entry_safe_from(window, _window, head, list) {
+               end_offset = window->offset + (window->nr_pages << PAGE_SHIFT);
+               loop_nr_pages = min((int)((end_offset - offset) >> PAGE_SHIFT),
+                                   nr_pages);
+               err = scif_unregister_window(window);
+               if (err)
+                       return err;
+               nr_pages -= loop_nr_pages;
+               offset += (loop_nr_pages << PAGE_SHIFT);
+               if (!nr_pages)
+                       break;
+       }
+       return 0;
+}
+
+/*
+ * scif_unmap_all_window:
+ *
+ * Traverse all the windows in the self registration list and:
+ * 1) Delete any DMA mappings created
+ */
+void scif_unmap_all_windows(scif_epd_t epd)
+{
+       struct list_head *item, *tmp;
+       struct scif_window *window;
+       struct scif_endpt *ep = (struct scif_endpt *)epd;
+       struct list_head *head = &ep->rma_info.reg_list;
+
+       mutex_lock(&ep->rma_info.rma_lock);
+       list_for_each_safe(item, tmp, head) {
+               window = list_entry(item, struct scif_window, list);
+               scif_unmap_window(ep->remote_dev, window);
+       }
+       mutex_unlock(&ep->rma_info.rma_lock);
+}
+
+/*
+ * scif_unregister_all_window:
+ *
+ * Traverse all the windows in the self registration list and:
+ * 1) Call scif_unregister_window(..)
+ * RMA lock must be held.
+ */
+int scif_unregister_all_windows(scif_epd_t epd)
+{
+       struct list_head *item, *tmp;
+       struct scif_window *window;
+       struct scif_endpt *ep = (struct scif_endpt *)epd;
+       struct list_head *head = &ep->rma_info.reg_list;
+       int err = 0;
+
+       mutex_lock(&ep->rma_info.rma_lock);
+retry:
+       item = NULL;
+       tmp = NULL;
+       list_for_each_safe(item, tmp, head) {
+               window = list_entry(item, struct scif_window, list);
+               ep->rma_info.async_list_del = 0;
+               err = scif_unregister_window(window);
+               if (err)
+                       dev_err(scif_info.mdev.this_device,
+                               "%s %d err %d\n",
+                               __func__, __LINE__, err);
+               /*
+                * Need to restart list traversal if there has been
+                * an asynchronous list entry deletion.
+                */
+               if (ACCESS_ONCE(ep->rma_info.async_list_del))
+                       goto retry;
+       }
+       mutex_unlock(&ep->rma_info.rma_lock);
+       if (!list_empty(&ep->rma_info.mmn_list)) {
+               spin_lock(&scif_info.rmalock);
+               list_add_tail(&ep->mmu_list, &scif_info.mmu_notif_cleanup);
+               spin_unlock(&scif_info.rmalock);
+               schedule_work(&scif_info.mmu_notif_work);
+       }
+       return err;
+}
diff --git a/drivers/misc/mic/scif/scif_rma_list.h b/drivers/misc/mic/scif/scif_rma_list.h
new file mode 100644 (file)
index 0000000..7d58d1d
--- /dev/null
@@ -0,0 +1,57 @@
+/*
+ * Intel MIC Platform Software Stack (MPSS)
+ *
+ * Copyright(c) 2015 Intel Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * Intel SCIF driver.
+ *
+ */
+#ifndef SCIF_RMA_LIST_H
+#define SCIF_RMA_LIST_H
+
+/*
+ * struct scif_rma_req - Self Registration list RMA Request query
+ *
+ * @out_window - Returns the window if found
+ * @offset: Starting offset
+ * @nr_bytes: number of bytes
+ * @prot: protection requested i.e. read or write or both
+ * @type: Specify single, partial or multiple windows
+ * @head: Head of list on which to search
+ * @va_for_temp: VA for searching temporary cached windows
+ */
+struct scif_rma_req {
+       struct scif_window **out_window;
+       union {
+               s64 offset;
+               unsigned long va_for_temp;
+       };
+       size_t nr_bytes;
+       int prot;
+       enum scif_window_type type;
+       struct list_head *head;
+};
+
+/* Insert */
+void scif_insert_window(struct scif_window *window, struct list_head *head);
+void scif_insert_tcw(struct scif_window *window,
+                    struct list_head *head);
+/* Query */
+int scif_query_window(struct scif_rma_req *request);
+int scif_query_tcw(struct scif_endpt *ep, struct scif_rma_req *request);
+/* Called from close to unregister all self windows */
+int scif_unregister_all_windows(scif_epd_t epd);
+void scif_unmap_all_windows(scif_epd_t epd);
+/* Traverse list and unregister */
+int scif_rma_list_unregister(struct scif_window *window, s64 offset,
+                            int nr_pages);
+#endif /* SCIF_RMA_LIST_H */