GRU Driver: driver initialization, file & vma ops
authorJack Steiner <steiner@sgi.com>
Wed, 30 Jul 2008 05:33:56 +0000 (22:33 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 30 Jul 2008 16:41:47 +0000 (09:41 -0700)
This file contains the functions for initializing the driver, handling
file & vma operations and for processing IOCTL requests from the user.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/misc/sgi-gru/grufile.c [new file with mode: 0644]

diff --git a/drivers/misc/sgi-gru/grufile.c b/drivers/misc/sgi-gru/grufile.c
new file mode 100644 (file)
index 0000000..09c9c65
--- /dev/null
@@ -0,0 +1,481 @@
+/*
+ * SN Platform GRU Driver
+ *
+ *              FILE OPERATIONS & DRIVER INITIALIZATION
+ *
+ * This file supports the user system call for file open, close, mmap, etc.
+ * This also incudes the driver initialization code.
+ *
+ *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/io.h>
+#include <linux/smp_lock.h>
+#include <linux/spinlock.h>
+#include <linux/device.h>
+#include <linux/miscdevice.h>
+#include <linux/interrupt.h>
+#include <linux/proc_fs.h>
+#include <linux/uaccess.h>
+#include "gru.h"
+#include "grulib.h"
+#include "grutables.h"
+
+#if defined CONFIG_X86_64
+#include <asm/genapic.h>
+#include <asm/irq.h>
+#define IS_UV()                is_uv_system()
+#elif defined CONFIG_IA64
+#include <asm/system.h>
+#include <asm/sn/simulator.h>
+/* temp support for running on hardware simulator */
+#define IS_UV()                IS_MEDUSA() || ia64_platform_is("uv")
+#else
+#define IS_UV()                0
+#endif
+
+#include <asm/uv/uv_hub.h>
+#include <asm/uv/uv_mmrs.h>
+
+struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
+unsigned long gru_start_paddr, gru_end_paddr __read_mostly;
+struct gru_stats_s gru_stats;
+
+/* Guaranteed user available resources on each node */
+static int max_user_cbrs, max_user_dsr_bytes;
+
+static struct file_operations gru_fops;
+static struct miscdevice gru_miscdev;
+
+
+/*
+ * gru_vma_close
+ *
+ * Called when unmapping a device mapping. Frees all gru resources
+ * and tables belonging to the vma.
+ */
+static void gru_vma_close(struct vm_area_struct *vma)
+{
+       struct gru_vma_data *vdata;
+       struct gru_thread_state *gts;
+       struct list_head *entry, *next;
+
+       if (!vma->vm_private_data)
+               return;
+
+       vdata = vma->vm_private_data;
+       vma->vm_private_data = NULL;
+       gru_dbg(grudev, "vma %p, file %p, vdata %p\n", vma, vma->vm_file,
+                               vdata);
+       list_for_each_safe(entry, next, &vdata->vd_head) {
+               gts =
+                   list_entry(entry, struct gru_thread_state, ts_next);
+               list_del(&gts->ts_next);
+               mutex_lock(&gts->ts_ctxlock);
+               if (gts->ts_gru)
+                       gru_unload_context(gts, 0);
+               mutex_unlock(&gts->ts_ctxlock);
+               gts_drop(gts);
+       }
+       kfree(vdata);
+       STAT(vdata_free);
+}
+
+/*
+ * gru_file_mmap
+ *
+ * Called when mmaping the device.  Initializes the vma with a fault handler
+ * and private data structure necessary to allocate, track, and free the
+ * underlying pages.
+ */
+static int gru_file_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       if ((vma->vm_flags & (VM_SHARED | VM_WRITE)) != (VM_SHARED | VM_WRITE))
+               return -EPERM;
+
+       vma->vm_flags |=
+           (VM_IO | VM_DONTCOPY | VM_LOCKED | VM_DONTEXPAND | VM_PFNMAP |
+                       VM_RESERVED);
+       vma->vm_page_prot = PAGE_SHARED;
+       vma->vm_ops = &gru_vm_ops;
+
+       vma->vm_private_data = gru_alloc_vma_data(vma, 0);
+       if (!vma->vm_private_data)
+               return -ENOMEM;
+
+       gru_dbg(grudev, "file %p, vaddr 0x%lx, vma %p, vdata %p\n",
+               file, vma->vm_start, vma, vma->vm_private_data);
+       return 0;
+}
+
+/*
+ * Create a new GRU context
+ */
+static int gru_create_new_context(unsigned long arg)
+{
+       struct gru_create_context_req req;
+       struct vm_area_struct *vma;
+       struct gru_vma_data *vdata;
+       int ret = -EINVAL;
+
+
+       if (copy_from_user(&req, (void __user *)arg, sizeof(req)))
+               return -EFAULT;
+
+       if (req.data_segment_bytes == 0 ||
+                               req.data_segment_bytes > max_user_dsr_bytes)
+               return -EINVAL;
+       if (!req.control_blocks || !req.maximum_thread_count ||
+                               req.control_blocks > max_user_cbrs)
+               return -EINVAL;
+
+       if (!(req.options & GRU_OPT_MISS_MASK))
+               req.options |= GRU_OPT_MISS_FMM_INTR;
+
+       down_write(&current->mm->mmap_sem);
+       vma = gru_find_vma(req.gseg);
+       if (vma) {
+               vdata = vma->vm_private_data;
+               vdata->vd_user_options = req.options;
+               vdata->vd_dsr_au_count =
+                   GRU_DS_BYTES_TO_AU(req.data_segment_bytes);
+               vdata->vd_cbr_au_count = GRU_CB_COUNT_TO_AU(req.control_blocks);
+               ret = 0;
+       }
+       up_write(&current->mm->mmap_sem);
+
+       return ret;
+}
+
+/*
+ * Get GRU configuration info (temp - for emulator testing)
+ */
+static long gru_get_config_info(unsigned long arg)
+{
+       struct gru_config_info info;
+       int nodesperblade;
+
+       if (num_online_nodes() > 1 &&
+                       (uv_node_to_blade_id(1) == uv_node_to_blade_id(0)))
+               nodesperblade = 2;
+       else
+               nodesperblade = 1;
+       info.cpus = num_online_cpus();
+       info.nodes = num_online_nodes();
+       info.blades = info.nodes / nodesperblade;
+       info.chiplets = GRU_CHIPLETS_PER_BLADE * info.blades;
+
+       if (copy_to_user((void __user *)arg, &info, sizeof(info)))
+               return -EFAULT;
+       return 0;
+}
+
+/*
+ * Get GRU chiplet status
+ */
+static long gru_get_chiplet_status(unsigned long arg)
+{
+       struct gru_state *gru;
+       struct gru_chiplet_info info;
+
+       if (copy_from_user(&info, (void __user *)arg, sizeof(info)))
+               return -EFAULT;
+
+       if (info.node == -1)
+               info.node = numa_node_id();
+       if (info.node >= num_possible_nodes() ||
+                       info.chiplet >= GRU_CHIPLETS_PER_HUB ||
+                       info.node < 0 || info.chiplet < 0)
+               return -EINVAL;
+
+       info.blade = uv_node_to_blade_id(info.node);
+       gru = get_gru(info.blade, info.chiplet);
+
+       info.total_dsr_bytes = GRU_NUM_DSR_BYTES;
+       info.total_cbr = GRU_NUM_CB;
+       info.total_user_dsr_bytes = GRU_NUM_DSR_BYTES -
+               gru->gs_reserved_dsr_bytes;
+       info.total_user_cbr = GRU_NUM_CB - gru->gs_reserved_cbrs;
+       info.free_user_dsr_bytes = hweight64(gru->gs_dsr_map) *
+                       GRU_DSR_AU_BYTES;
+       info.free_user_cbr = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
+
+       if (copy_to_user((void __user *)arg, &info, sizeof(info)))
+               return -EFAULT;
+       return 0;
+}
+
+/*
+ * gru_file_unlocked_ioctl
+ *
+ * Called to update file attributes via IOCTL calls.
+ */
+static long gru_file_unlocked_ioctl(struct file *file, unsigned int req,
+                                   unsigned long arg)
+{
+       int err = -EBADRQC;
+
+       gru_dbg(grudev, "file %p\n", file);
+
+       switch (req) {
+       case GRU_CREATE_CONTEXT:
+               err = gru_create_new_context(arg);
+               break;
+       case GRU_SET_TASK_SLICE:
+               err = gru_set_task_slice(arg);
+               break;
+       case GRU_USER_GET_EXCEPTION_DETAIL:
+               err = gru_get_exception_detail(arg);
+               break;
+       case GRU_USER_UNLOAD_CONTEXT:
+               err = gru_user_unload_context(arg);
+               break;
+       case GRU_GET_CHIPLET_STATUS:
+               err = gru_get_chiplet_status(arg);
+               break;
+       case GRU_USER_FLUSH_TLB:
+               err = gru_user_flush_tlb(arg);
+               break;
+       case GRU_USER_CALL_OS:
+               err = gru_handle_user_call_os(arg);
+               break;
+       case GRU_GET_CONFIG_INFO:
+               err = gru_get_config_info(arg);
+               break;
+       }
+       return err;
+}
+
+/*
+ * Called at init time to build tables for all GRUs that are present in the
+ * system.
+ */
+static void gru_init_chiplet(struct gru_state *gru, unsigned long paddr,
+                            void *vaddr, int nid, int bid, int grunum)
+{
+       spin_lock_init(&gru->gs_lock);
+       spin_lock_init(&gru->gs_asid_lock);
+       gru->gs_gru_base_paddr = paddr;
+       gru->gs_gru_base_vaddr = vaddr;
+       gru->gs_gid = bid * GRU_CHIPLETS_PER_BLADE + grunum;
+       gru->gs_blade = gru_base[bid];
+       gru->gs_blade_id = bid;
+       gru->gs_cbr_map = (GRU_CBR_AU == 64) ? ~0 : (1UL << GRU_CBR_AU) - 1;
+       gru->gs_dsr_map = (1UL << GRU_DSR_AU) - 1;
+       gru_tgh_flush_init(gru);
+       gru_dbg(grudev, "bid %d, nid %d, gru %x, vaddr %p (0x%lx)\n",
+               bid, nid, gru->gs_gid, gru->gs_gru_base_vaddr,
+               gru->gs_gru_base_paddr);
+       gru_kservices_init(gru);
+}
+
+static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
+{
+       int pnode, nid, bid, chip;
+       int cbrs, dsrbytes, n;
+       int order = get_order(sizeof(struct gru_blade_state));
+       struct page *page;
+       struct gru_state *gru;
+       unsigned long paddr;
+       void *vaddr;
+
+       max_user_cbrs = GRU_NUM_CB;
+       max_user_dsr_bytes = GRU_NUM_DSR_BYTES;
+       for_each_online_node(nid) {
+               bid = uv_node_to_blade_id(nid);
+               pnode = uv_node_to_pnode(nid);
+               if (gru_base[bid])
+                       continue;
+               page = alloc_pages_node(nid, GFP_KERNEL, order);
+               if (!page)
+                       goto fail;
+               gru_base[bid] = page_address(page);
+               memset(gru_base[bid], 0, sizeof(struct gru_blade_state));
+               gru_base[bid]->bs_lru_gru = &gru_base[bid]->bs_grus[0];
+               spin_lock_init(&gru_base[bid]->bs_lock);
+
+               dsrbytes = 0;
+               cbrs = 0;
+               for (gru = gru_base[bid]->bs_grus, chip = 0;
+                               chip < GRU_CHIPLETS_PER_BLADE;
+                               chip++, gru++) {
+                       paddr = gru_chiplet_paddr(gru_base_paddr, pnode, chip);
+                       vaddr = gru_chiplet_vaddr(gru_base_vaddr, pnode, chip);
+                       gru_init_chiplet(gru, paddr, vaddr, bid, nid, chip);
+                       n = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
+                       cbrs = max(cbrs, n);
+                       n = hweight64(gru->gs_dsr_map) * GRU_DSR_AU_BYTES;
+                       dsrbytes = max(dsrbytes, n);
+               }
+               max_user_cbrs = min(max_user_cbrs, cbrs);
+               max_user_dsr_bytes = min(max_user_dsr_bytes, dsrbytes);
+       }
+
+       return 0;
+
+fail:
+       for (nid--; nid >= 0; nid--)
+               free_pages((unsigned long)gru_base[nid], order);
+       return -ENOMEM;
+}
+
+#ifdef CONFIG_IA64
+
+static int get_base_irq(void)
+{
+       return IRQ_GRU;
+}
+
+#elif defined CONFIG_X86_64
+
+static void noop(unsigned int irq)
+{
+}
+
+static struct irq_chip gru_chip = {
+       .name           = "gru",
+       .mask           = noop,
+       .unmask         = noop,
+       .ack            = noop,
+};
+
+static int get_base_irq(void)
+{
+       set_irq_chip(IRQ_GRU, &gru_chip);
+       set_irq_chip(IRQ_GRU + 1, &gru_chip);
+       return IRQ_GRU;
+}
+#endif
+
+/*
+ * gru_init
+ *
+ * Called at boot or module load time to initialize the GRUs.
+ */
+static int __init gru_init(void)
+{
+       int ret, irq, chip;
+       char id[10];
+       void *gru_start_vaddr;
+
+       if (!IS_UV())
+               return 0;
+
+#if defined CONFIG_IA64
+       gru_start_paddr = 0xd000000000UL; /* ZZZZZZZZZZZZZZZZZZZ fixme */
+#else
+       gru_start_paddr = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR) &
+                               0x7fffffffffffUL;
+
+#endif
+       gru_start_vaddr = __va(gru_start_paddr);
+       gru_end_paddr = gru_start_paddr + MAX_NUMNODES * GRU_SIZE;
+       printk(KERN_INFO "GRU space: 0x%lx - 0x%lx\n",
+              gru_start_paddr, gru_end_paddr);
+       irq = get_base_irq();
+       for (chip = 0; chip < GRU_CHIPLETS_PER_BLADE; chip++) {
+               ret = request_irq(irq + chip, gru_intr, 0, id, NULL);
+               if (ret) {
+                       printk(KERN_ERR "%s: request_irq failed\n",
+                              GRU_DRIVER_ID_STR);
+                       goto exit1;
+               }
+       }
+
+       ret = misc_register(&gru_miscdev);
+       if (ret) {
+               printk(KERN_ERR "%s: misc_register failed\n",
+                      GRU_DRIVER_ID_STR);
+               goto exit1;
+       }
+
+       ret = gru_proc_init();
+       if (ret) {
+               printk(KERN_ERR "%s: proc init failed\n", GRU_DRIVER_ID_STR);
+               goto exit2;
+       }
+
+       ret = gru_init_tables(gru_start_paddr, gru_start_vaddr);
+       if (ret) {
+               printk(KERN_ERR "%s: init tables failed\n", GRU_DRIVER_ID_STR);
+               goto exit3;
+       }
+
+       printk(KERN_INFO "%s: v%s\n", GRU_DRIVER_ID_STR,
+              GRU_DRIVER_VERSION_STR);
+       return 0;
+
+exit3:
+       gru_proc_exit();
+exit2:
+       misc_deregister(&gru_miscdev);
+exit1:
+       for (--chip; chip >= 0; chip--)
+               free_irq(irq + chip, NULL);
+       return ret;
+
+}
+
+static void __exit gru_exit(void)
+{
+       int i, bid;
+       int order = get_order(sizeof(struct gru_state) *
+                             GRU_CHIPLETS_PER_BLADE);
+
+       for (i = 0; i < GRU_CHIPLETS_PER_BLADE; i++)
+               free_irq(IRQ_GRU + i, NULL);
+
+       for (bid = 0; bid < GRU_MAX_BLADES; bid++)
+               free_pages((unsigned long)gru_base[bid], order);
+
+       misc_deregister(&gru_miscdev);
+       gru_proc_exit();
+}
+
+static struct file_operations gru_fops = {
+       .owner          = THIS_MODULE,
+       .unlocked_ioctl = gru_file_unlocked_ioctl,
+       .mmap           = gru_file_mmap,
+};
+
+static struct miscdevice gru_miscdev = {
+       .minor          = MISC_DYNAMIC_MINOR,
+       .name           = "gru",
+       .fops           = &gru_fops,
+};
+
+struct vm_operations_struct gru_vm_ops = {
+       .close          = gru_vma_close,
+       .fault          = gru_fault,
+};
+
+module_init(gru_init);
+module_exit(gru_exit);
+
+module_param(options, ulong, 0644);
+MODULE_PARM_DESC(options, "Various debug options");
+
+MODULE_AUTHOR("Silicon Graphics, Inc.");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION(GRU_DRIVER_ID_STR GRU_DRIVER_VERSION_STR);
+MODULE_VERSION(GRU_DRIVER_VERSION_STR);
+