This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / arch / cris / arch-v32 / kernel / arbiter.c
diff --git a/arch/cris/arch-v32/kernel/arbiter.c b/arch/cris/arch-v32/kernel/arbiter.c
new file mode 100644 (file)
index 0000000..3870d2f
--- /dev/null
@@ -0,0 +1,297 @@
+/*
+ * Memory arbiter functions. Allocates bandwith through the
+ * arbiter and sets up arbiter breakpoints.
+ *
+ * The algorithm first assigns slots to the clients that has specified
+ * bandwith (e.g. ethernet) and then the remaining slots are divided
+ * on all the active clients.
+ *
+ * Copyright (c) 2004, 2005 Axis Communications AB.
+ */
+
+#include <linux/config.h>
+#include <asm/arch/hwregs/reg_map.h>
+#include <asm/arch/hwregs/reg_rdwr.h>
+#include <asm/arch/hwregs/marb_defs.h>
+#include <asm/arch/arbiter.h>
+#include <asm/arch/hwregs/intr_vect.h>
+#include <linux/interrupt.h>
+#include <linux/signal.h>
+#include <linux/errno.h>
+#include <linux/spinlock.h>
+#include <asm/io.h>
+
+struct crisv32_watch_entry
+{
+  unsigned long instance;
+  watch_callback* cb;
+  unsigned long start;
+  unsigned long end;
+  int used;
+};
+
+#define NUMBER_OF_BP 4
+#define NBR_OF_CLIENTS 14
+#define NBR_OF_SLOTS 64
+#define SDRAM_BANDWIDTH 100000000 /* Some kind of expected value */
+#define INTMEM_BANDWIDTH 400000000
+#define NBR_OF_REGIONS 2
+
+static struct crisv32_watch_entry watches[NUMBER_OF_BP] =
+{
+  {regi_marb_bp0},
+  {regi_marb_bp1},
+  {regi_marb_bp2},
+  {regi_marb_bp3}
+};
+
+static int requested_slots[NBR_OF_REGIONS][NBR_OF_CLIENTS];
+static int active_clients[NBR_OF_REGIONS][NBR_OF_CLIENTS];
+static int max_bandwidth[NBR_OF_REGIONS] = {SDRAM_BANDWIDTH, INTMEM_BANDWIDTH};
+
+DEFINE_SPINLOCK(arbiter_lock);
+
+static irqreturn_t
+crisv32_arbiter_irq(int irq, void* dev_id, struct pt_regs* regs);
+
+static void crisv32_arbiter_config(int region)
+{
+       int slot;
+       int client;
+       int interval = 0;
+       int val[NBR_OF_SLOTS];
+
+       for (slot = 0; slot < NBR_OF_SLOTS; slot++)
+           val[slot] = NBR_OF_CLIENTS + 1;
+
+       for (client = 0; client < NBR_OF_CLIENTS; client++)
+       {
+           int pos;
+           if (!requested_slots[region][client])
+              continue;
+           interval = NBR_OF_SLOTS / requested_slots[region][client];
+           pos = 0;
+           while (pos < NBR_OF_SLOTS)
+           {
+               if (val[pos] != NBR_OF_CLIENTS + 1)
+                  pos++;
+               else
+               {
+                       val[pos] = client;
+                       pos += interval;
+               }
+           }
+       }
+
+       client = 0;
+       for (slot = 0; slot < NBR_OF_SLOTS; slot++)
+       {
+               if (val[slot] == NBR_OF_CLIENTS + 1)
+               {
+                       int first = client;
+                       while(!active_clients[region][client]) {
+                               client = (client + 1) % NBR_OF_CLIENTS;
+                               if (client == first)
+                                  break;
+                       }
+                       val[slot] = client;
+                       client = (client + 1) % NBR_OF_CLIENTS;
+               }
+               if (region == EXT_REGION)
+                  REG_WR_INT_VECT(marb, regi_marb, rw_ext_slots, slot, val[slot]);
+               else if (region == INT_REGION)
+                  REG_WR_INT_VECT(marb, regi_marb, rw_int_slots, slot, val[slot]);
+       }
+}
+
+extern char _stext, _etext;
+
+static void crisv32_arbiter_init(void)
+{
+       static int initialized = 0;
+
+       if (initialized)
+               return;
+
+       initialized = 1;
+
+       /* CPU caches are active. */
+       active_clients[EXT_REGION][10] = active_clients[EXT_REGION][11] = 1;
+        crisv32_arbiter_config(EXT_REGION);
+        crisv32_arbiter_config(INT_REGION);
+
+       if (request_irq(MEMARB_INTR_VECT, crisv32_arbiter_irq, SA_INTERRUPT,
+                        "arbiter", NULL))
+               printk(KERN_ERR "Couldn't allocate arbiter IRQ\n");
+
+#ifndef CONFIG_ETRAX_KGDB
+        /* Global watch for writes to kernel text segment. */
+        crisv32_arbiter_watch(virt_to_phys(&_stext), &_etext - &_stext,
+                              arbiter_all_clients, arbiter_all_write, NULL);
+#endif
+}
+
+
+
+int crisv32_arbiter_allocate_bandwith(int client, int region,
+                                     unsigned long bandwidth)
+{
+       int i;
+       int total_assigned = 0;
+       int total_clients = 0;
+       int req;
+
+       crisv32_arbiter_init();
+
+       for (i = 0; i < NBR_OF_CLIENTS; i++)
+       {
+               total_assigned += requested_slots[region][i];
+               total_clients += active_clients[region][i];
+       }
+       req = NBR_OF_SLOTS / (max_bandwidth[region] / bandwidth);
+
+       if (total_assigned + total_clients + req + 1 > NBR_OF_SLOTS)
+          return -ENOMEM;
+
+       active_clients[region][client] = 1;
+       requested_slots[region][client] = req;
+       crisv32_arbiter_config(region);
+
+       return 0;
+}
+
+int crisv32_arbiter_watch(unsigned long start, unsigned long size,
+                          unsigned long clients, unsigned long accesses,
+                          watch_callback* cb)
+{
+       int i;
+
+       crisv32_arbiter_init();
+
+       if (start > 0x80000000) {
+               printk("Arbiter: %lX doesn't look like a physical address", start);
+               return -EFAULT;
+       }
+
+       spin_lock(&arbiter_lock);
+
+       for (i = 0; i < NUMBER_OF_BP; i++) {
+               if (!watches[i].used) {
+                       reg_marb_rw_intr_mask intr_mask = REG_RD(marb, regi_marb, rw_intr_mask);
+
+                       watches[i].used = 1;
+                       watches[i].start = start;
+                       watches[i].end = start + size;
+                       watches[i].cb = cb;
+
+                       REG_WR_INT(marb_bp, watches[i].instance, rw_first_addr, watches[i].start);
+                       REG_WR_INT(marb_bp, watches[i].instance, rw_last_addr, watches[i].end);
+                       REG_WR_INT(marb_bp, watches[i].instance, rw_op, accesses);
+                       REG_WR_INT(marb_bp, watches[i].instance, rw_clients, clients);
+
+                       if (i == 0)
+                               intr_mask.bp0 = regk_marb_yes;
+                       else if (i == 1)
+                               intr_mask.bp1 = regk_marb_yes;
+                       else if (i == 2)
+                               intr_mask.bp2 = regk_marb_yes;
+                       else if (i == 3)
+                               intr_mask.bp3 = regk_marb_yes;
+
+                       REG_WR(marb, regi_marb, rw_intr_mask, intr_mask);
+                       spin_unlock(&arbiter_lock);
+
+                       return i;
+               }
+       }
+       spin_unlock(&arbiter_lock);
+       return -ENOMEM;
+}
+
+int crisv32_arbiter_unwatch(int id)
+{
+       reg_marb_rw_intr_mask intr_mask = REG_RD(marb, regi_marb, rw_intr_mask);
+
+       crisv32_arbiter_init();
+
+       spin_lock(&arbiter_lock);
+
+       if ((id < 0) || (id >= NUMBER_OF_BP) || (!watches[id].used)) {
+               spin_unlock(&arbiter_lock);
+               return -EINVAL;
+       }
+
+       memset(&watches[id], 0, sizeof(struct crisv32_watch_entry));
+
+       if (id == 0)
+               intr_mask.bp0 = regk_marb_no;
+       else if (id == 1)
+               intr_mask.bp2 = regk_marb_no;
+       else if (id == 2)
+               intr_mask.bp2 = regk_marb_no;
+       else if (id == 3)
+               intr_mask.bp3 = regk_marb_no;
+
+       REG_WR(marb, regi_marb, rw_intr_mask, intr_mask);
+
+       spin_unlock(&arbiter_lock);
+       return 0;
+}
+
+extern void show_registers(struct pt_regs *regs);
+
+static irqreturn_t
+crisv32_arbiter_irq(int irq, void* dev_id, struct pt_regs* regs)
+{
+       reg_marb_r_masked_intr masked_intr = REG_RD(marb, regi_marb, r_masked_intr);
+       reg_marb_bp_r_brk_clients r_clients;
+       reg_marb_bp_r_brk_addr r_addr;
+       reg_marb_bp_r_brk_op r_op;
+       reg_marb_bp_r_brk_first_client r_first;
+       reg_marb_bp_r_brk_size r_size;
+       reg_marb_bp_rw_ack ack = {0};
+       reg_marb_rw_ack_intr ack_intr = {.bp0=1,.bp1=1,.bp2=1,.bp3=1};
+       struct crisv32_watch_entry* watch;
+
+       if (masked_intr.bp0) {
+               watch = &watches[0];
+               ack_intr.bp0 = regk_marb_yes;
+       } else if (masked_intr.bp1) {
+               watch = &watches[1];
+               ack_intr.bp1 = regk_marb_yes;
+       } else if (masked_intr.bp2) {
+               watch = &watches[2];
+               ack_intr.bp2 = regk_marb_yes;
+       } else if (masked_intr.bp3) {
+               watch = &watches[3];
+               ack_intr.bp3 = regk_marb_yes;
+       } else {
+               return IRQ_NONE;
+       }
+
+       /* Retrieve all useful information and print it. */
+       r_clients = REG_RD(marb_bp, watch->instance, r_brk_clients);
+       r_addr = REG_RD(marb_bp, watch->instance, r_brk_addr);
+       r_op = REG_RD(marb_bp, watch->instance, r_brk_op);
+       r_first = REG_RD(marb_bp, watch->instance, r_brk_first_client);
+       r_size = REG_RD(marb_bp, watch->instance, r_brk_size);
+
+       printk("Arbiter IRQ\n");
+       printk("Clients %X addr %X op %X first %X size %X\n",
+              REG_TYPE_CONV(int, reg_marb_bp_r_brk_clients, r_clients),
+              REG_TYPE_CONV(int, reg_marb_bp_r_brk_addr, r_addr),
+              REG_TYPE_CONV(int, reg_marb_bp_r_brk_op, r_op),
+              REG_TYPE_CONV(int, reg_marb_bp_r_brk_first_client, r_first),
+              REG_TYPE_CONV(int, reg_marb_bp_r_brk_size, r_size));
+
+       REG_WR(marb_bp, watch->instance, rw_ack, ack);
+       REG_WR(marb, regi_marb, rw_ack_intr, ack_intr);
+
+       printk("IRQ occured at %lX\n", regs->erp);
+
+       if (watch->cb)
+               watch->cb();
+
+
+       return IRQ_HANDLED;
+}