[MIPS] MT: Enable coexistence of AP/SP with VSMP and SMTC.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

6 files changed, 197 insertions, 169 deletions

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index 0893e084150e..3513e226837b 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -1377,17 +1377,6 @@ config MIPS_MT_SMTC
           This is a kernel model which is known a SMTC or lately has been
           marketesed into SMVP.
 
-config MIPS_VPE_LOADER
-        bool "VPE loader support."
-        depends on SYS_SUPPORTS_MULTITHREADING
-        select CPU_MIPSR2_IRQ_VI
-        select CPU_MIPSR2_IRQ_EI
-        select CPU_MIPSR2_SRS
-        select MIPS_MT
-        help
-          Includes a loader for loading an elf relocatable object
-          onto another VPE and running it.
-
 endchoice
 
 config MIPS_MT
@@ -1398,8 +1387,19 @@ config SYS_SUPPORTS_MULTITHREADING
 
 config MIPS_MT_FPAFF
         bool "Dynamic FPU affinity for FP-intensive threads"
-        depends on MIPS_MT
         default y
+        depends on MIPS_MT_SMP || MIPS_MT_SMTC
+
+config MIPS_VPE_LOADER
+        bool "VPE loader support."
+        depends on SYS_SUPPORTS_MULTITHREADING
+        select CPU_MIPSR2_IRQ_VI
+        select CPU_MIPSR2_IRQ_EI
+        select CPU_MIPSR2_SRS
+        select MIPS_MT
+        help
+          Includes a loader for loading an elf relocatable object
+          onto another VPE and running it.
 
 config MIPS_MT_SMTC_INSTANT_REPLAY
         bool "Low-latency Dispatch of Deferred SMTC IPIs"
diff --git a/arch/mips/kernel/kspd.c b/arch/mips/kernel/kspd.c
index c6580018c94b..cb9a14a1ca5b 100644
--- a/arch/mips/kernel/kspd.c
+++ b/arch/mips/kernel/kspd.c
@@ -89,7 +89,7 @@ static int sp_stopping = 0;
 #define MTSP_O_EXCL             0x0800
 #define MTSP_O_BINARY           0x8000
 
-#define SP_VPE 1
+extern int tclimit;
 
 struct apsp_table  {
         int sp;
@@ -225,8 +225,8 @@ void sp_work_handle_request(void)
         /* Run the syscall at the priviledge of the user who loaded the
            SP program */
 
-        if (vpe_getuid(SP_VPE))
-                sp_setfsuidgid( vpe_getuid(SP_VPE), vpe_getgid(SP_VPE));
+        if (vpe_getuid(tclimit))
+                sp_setfsuidgid(vpe_getuid(tclimit), vpe_getgid(tclimit));
 
         switch (sc.cmd) {
         /* needs the flags argument translating from SDE kit to
@@ -245,7 +245,7 @@ void sp_work_handle_request(void)
 
         case MTSP_SYSCALL_EXIT:
                 list_for_each_entry(n, &kspd_notifylist, list)
-                        n->kspd_sp_exit(SP_VPE);
+                        n->kspd_sp_exit(tclimit);
                 sp_stopping = 1;
 
                 printk(KERN_DEBUG "KSPD got exit syscall from SP exitcode %d\n",
@@ -255,7 +255,7 @@ void sp_work_handle_request(void)
         case MTSP_SYSCALL_OPEN:
                 generic.arg1 = translate_open_flags(generic.arg1);
 
-                vcwd = vpe_getcwd(SP_VPE);
+                vcwd = vpe_getcwd(tclimit);
 
                 /* change to the cwd of the process that loaded the SP program */
                 old_fs = get_fs();
@@ -283,7 +283,7 @@ void sp_work_handle_request(void)
                 break;
         } /* switch */
 
-        if (vpe_getuid(SP_VPE))
+        if (vpe_getuid(tclimit))
                 sp_setfsuidgid( 0, 0);
 
         old_fs = get_fs();
@@ -364,10 +364,9 @@ static void startwork(int vpe)
                 }
 
                 INIT_WORK(&work, sp_work);
-                queue_work(workqueue, &work);
-        } else
-                queue_work(workqueue, &work);
+        }
 
+        queue_work(workqueue, &work);
 }
 
 static void stopwork(int vpe)
@@ -389,7 +388,7 @@ static int kspd_module_init(void)
 
         notify.start = startwork;
         notify.stop = stopwork;
-        vpe_notify(SP_VPE, &notify);
+        vpe_notify(tclimit, &notify);
 
         return 0;
 }
diff --git a/arch/mips/kernel/mips-mt.c b/arch/mips/kernel/mips-mt.c
index 1a7d89231299..7169a4db37b8 100644
--- a/arch/mips/kernel/mips-mt.c
+++ b/arch/mips/kernel/mips-mt.c
@@ -21,6 +21,28 @@
 #include <asm/r4kcache.h>
 #include <asm/cacheflush.h>
 
+int vpelimit;
+
+static int __init maxvpes(char *str)
+{
+        get_option(&str, &vpelimit);
+
+        return 1;
+}
+
+__setup("maxvpes=", maxvpes);
+
+int tclimit;
+
+static int __init maxtcs(char *str)
+{
+        get_option(&str, &tclimit);
+
+        return 1;
+}
+
+__setup("maxtcs=", maxtcs);
+
 /*
  * Dump new MIPS MT state for the core. Does not leave TCs halted.
  * Takes an argument which taken to be a pre-call MVPControl value.
diff --git a/arch/mips/kernel/rtlx.c b/arch/mips/kernel/rtlx.c
index 8cf24d716d41..5c040060560e 100644
--- a/arch/mips/kernel/rtlx.c
+++ b/arch/mips/kernel/rtlx.c
@@ -40,12 +40,11 @@
 #include <asm/atomic.h>
 #include <asm/cpu.h>
 #include <asm/processor.h>
+#include <asm/mips_mt.h>
 #include <asm/system.h>
 #include <asm/vpe.h>
 #include <asm/rtlx.h>
 
-#define RTLX_TARG_VPE 1
-
 static struct rtlx_info *rtlx;
 static int major;
 static char module_name[] = "rtlx";
@@ -165,10 +164,10 @@ int rtlx_open(int index, int can_sleep)
         }
 
         if (rtlx == NULL) {
-                if( (p = vpe_get_shared(RTLX_TARG_VPE)) == NULL) {
+                if( (p = vpe_get_shared(tclimit)) == NULL) {
                         if (can_sleep) {
                                 __wait_event_interruptible(channel_wqs[index].lx_queue,
-                                                           (p = vpe_get_shared(RTLX_TARG_VPE)),
+                                                           (p = vpe_get_shared(tclimit)),
                                                            ret);
                                 if (ret)
                                         goto out_fail;
@@ -477,6 +476,19 @@ static int rtlx_module_init(void)
         struct device *dev;
         int i, err;
 
+        if (!cpu_has_mipsmt) {
+                printk("VPE loader: not a MIPS MT capable processor\n");
+                return -ENODEV;
+        }
+
+        if (tclimit == 0) {
+                printk(KERN_WARNING "No TCs reserved for AP/SP, not "
+                       "initializing RTLX.\nPass maxtcs=<n> argument as kernel "
+                       "argument\n");
+
+                return -ENODEV;
+        }
+
         major = register_chrdev(0, module_name, &rtlx_fops);
         if (major < 0) {
                 printk(register_chrdev_failed);
@@ -501,7 +513,7 @@ static int rtlx_module_init(void)
         /* set up notifiers */
         notify.start = starting;
         notify.stop = stopping;
-        vpe_notify(RTLX_TARG_VPE, &notify);
+        vpe_notify(tclimit, &notify);
 
         if (cpu_has_vint)
                 set_vi_handler(MIPS_CPU_RTLX_IRQ, rtlx_dispatch);
diff --git a/arch/mips/kernel/smtc.c b/arch/mips/kernel/smtc.c
index f2c7aed663e7..6b3c3ea8bc61 100644
--- a/arch/mips/kernel/smtc.c
+++ b/arch/mips/kernel/smtc.c
@@ -86,25 +86,11 @@ unsigned int smtc_status = 0;
 
 /* Boot command line configuration overrides */
 
-static int vpelimit = 0;
-static int tclimit = 0;
 static int ipibuffers = 0;
 static int nostlb = 0;
 static int asidmask = 0;
 unsigned long smtc_asid_mask = 0xff;
 
-static int __init maxvpes(char *str)
-{
-        get_option(&str, &vpelimit);
-        return 1;
-}
-
-static int __init maxtcs(char *str)
-{
-        get_option(&str, &tclimit);
-        return 1;
-}
-
 static int __init ipibufs(char *str)
 {
         get_option(&str, &ipibuffers);
@@ -137,8 +123,6 @@ static int __init asidmask_set(char *str)
         return 1;
 }
 
-__setup("maxvpes=", maxvpes);
-__setup("maxtcs=", maxtcs);
 __setup("ipibufs=", ipibufs);
 __setup("nostlb", stlb_disable);
 __setup("asidmask=", asidmask_set);
diff --git a/arch/mips/kernel/vpe.c b/arch/mips/kernel/vpe.c
index a2bee10f04cf..c726c47cd2c3 100644
--- a/arch/mips/kernel/vpe.c
+++ b/arch/mips/kernel/vpe.c
@@ -27,7 +27,6 @@
  * To load and run, simply cat a SP 'program file' to /dev/vpe1.
  * i.e cat spapp >/dev/vpe1.
  */
-
 #include <linux/kernel.h>
 #include <linux/device.h>
 #include <linux/module.h>
@@ -54,6 +53,7 @@
 #include <asm/system.h>
 #include <asm/vpe.h>
 #include <asm/kspd.h>
+#include <asm/mips_mt.h>
 
 typedef void *vpe_handle;
 
@@ -132,14 +132,9 @@ struct tc {
         enum tc_state state;
         int index;
 
-        /* parent VPE */
-        struct vpe *pvpe;
-
-        /* The list of TC's with this VPE */
-        struct list_head tc;
-
-        /* The global list of tc's */
-        struct list_head list;
+        struct vpe *pvpe;       /* parent VPE */
+        struct list_head tc;    /* The list of TC's with this VPE */
+        struct list_head list;  /* The global list of tc's */
 };
 
 struct {
@@ -217,18 +212,17 @@ struct vpe *alloc_vpe(int minor)
 /* allocate a tc. At startup only tc0 is running, all other can be halted. */
 struct tc *alloc_tc(int index)
 {
-        struct tc *t;
-
-        if ((t = kzalloc(sizeof(struct tc), GFP_KERNEL)) == NULL) {
-                return NULL;
-        }
+        struct tc *tc;
 
-        INIT_LIST_HEAD(&t->tc);
-        list_add_tail(&t->list, &vpecontrol.tc_list);
+        if ((tc = kzalloc(sizeof(struct tc), GFP_KERNEL)) == NULL)
+                goto out;
 
-        t->index = index;
+        INIT_LIST_HEAD(&tc->tc);
+        tc->index = index;
+        list_add_tail(&tc->list, &vpecontrol.tc_list);
 
-        return t;
+out:
+        return tc;
 }
 
 /* clean up and free everything */
@@ -663,66 +657,48 @@ static void dump_elfsymbols(Elf_Shdr * sechdrs, unsigned int symindex,
 }
 #endif
 
-static void dump_tc(struct tc *t)
-{
-        unsigned long val;
-
-        settc(t->index);
-        printk(KERN_DEBUG "VPE loader: TC index %d targtc %ld "
-               "TCStatus 0x%lx halt 0x%lx\n",
-               t->index, read_c0_vpecontrol() & VPECONTROL_TARGTC,
-               read_tc_c0_tcstatus(), read_tc_c0_tchalt());
-
-        printk(KERN_DEBUG " tcrestart 0x%lx\n", read_tc_c0_tcrestart());
-        printk(KERN_DEBUG " tcbind 0x%lx\n", read_tc_c0_tcbind());
-
-        val = read_c0_vpeconf0();
-        printk(KERN_DEBUG " VPEConf0 0x%lx MVP %ld\n", val,
-               (val & VPECONF0_MVP) >> VPECONF0_MVP_SHIFT);
-
-        printk(KERN_DEBUG " c0 status 0x%lx\n", read_vpe_c0_status());
-        printk(KERN_DEBUG " c0 cause 0x%lx\n", read_vpe_c0_cause());
-
-        printk(KERN_DEBUG " c0 badvaddr 0x%lx\n", read_vpe_c0_badvaddr());
-        printk(KERN_DEBUG " c0 epc 0x%lx\n", read_vpe_c0_epc());
-}
-
-static void dump_tclist(void)
-{
-        struct tc *t;
-
-        list_for_each_entry(t, &vpecontrol.tc_list, list) {
-                dump_tc(t);
-        }
-}
-
 /* We are prepared so configure and start the VPE... */
 static int vpe_run(struct vpe * v)
 {
+        unsigned long flags, val, dmt_flag;
         struct vpe_notifications *n;
-        unsigned long val, dmt_flag;
+        unsigned int vpeflags;
         struct tc *t;
 
         /* check we are the Master VPE */
+        local_irq_save(flags);
         val = read_c0_vpeconf0();
         if (!(val & VPECONF0_MVP)) {
                 printk(KERN_WARNING
                        "VPE loader: only Master VPE's are allowed to configure MT\n");
+                local_irq_restore(flags);
+
                 return -1;
         }
 
-        /* disable MT (using dvpe) */
-        dvpe();
+        dmt_flag = dmt();
+        vpeflags = dvpe();
 
         if (!list_empty(&v->tc)) {
                 if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
-                        printk(KERN_WARNING "VPE loader: TC %d is already in use.\n",
-                               t->index);
+                        evpe(vpeflags);
+                        emt(dmt_flag);
+                        local_irq_restore(flags);
+
+                        printk(KERN_WARNING
+                               "VPE loader: TC %d is already in use.\n",
+                               t->index);
                         return -ENOEXEC;
                 }
         } else {
-                printk(KERN_WARNING "VPE loader: No TC's associated with VPE %d\n",
+                evpe(vpeflags);
+                emt(dmt_flag);
+                local_irq_restore(flags);
+
+                printk(KERN_WARNING
+                       "VPE loader: No TC's associated with VPE %d\n",
                        v->minor);
+
                 return -ENOEXEC;
         }
 
@@ -733,21 +709,20 @@ static int vpe_run(struct vpe * v)
 
         /* should check it is halted, and not activated */
         if ((read_tc_c0_tcstatus() & TCSTATUS_A) || !(read_tc_c0_tchalt() & TCHALT_H)) {
-                printk(KERN_WARNING "VPE loader: TC %d is already doing something!\n",
+                evpe(vpeflags);
+                emt(dmt_flag);
+                local_irq_restore(flags);
+
+                printk(KERN_WARNING "VPE loader: TC %d is already active!\n",
                        t->index);
-                dump_tclist();
+
                 return -ENOEXEC;
         }
 
-        /*
-         * Disable multi-threaded execution whilst we activate, clear the
-         * halt bit and bound the tc to the other VPE...
-         */
-        dmt_flag = dmt();
-
         /* Write the address we want it to start running from in the TCPC register. */
         write_tc_c0_tcrestart((unsigned long)v->__start);
         write_tc_c0_tccontext((unsigned long)0);
+
         /*
          * Mark the TC as activated, not interrupt exempt and not dynamically
          * allocatable
@@ -763,15 +738,14 @@ static int vpe_run(struct vpe * v)
          * here...  Or set $a3 to zero and define DFLT_STACK_SIZE and
          * DFLT_HEAP_SIZE when you compile your program
          */
-        mttgpr(7, physical_memsize);
-
+        mttgpr(7, physical_memsize);
 
         /* set up VPE1 */
         /*
          * bind the TC to VPE 1 as late as possible so we only have the final
          * VPE registers to set up, and so an EJTAG probe can trigger on it
          */
-        write_tc_c0_tcbind((read_tc_c0_tcbind() & ~TCBIND_CURVPE) | v->minor);
+        write_tc_c0_tcbind((read_tc_c0_tcbind() & ~TCBIND_CURVPE) | 1);
 
         write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~(VPECONF0_VPA));
 
@@ -793,15 +767,16 @@ static int vpe_run(struct vpe * v)
         /* take system out of configuration state */
         clear_c0_mvpcontrol(MVPCONTROL_VPC);
 
-        /* now safe to re-enable multi-threading */
-        emt(dmt_flag);
-
-        /* set it running */
+#ifdef CONFIG_SMP
         evpe(EVPE_ENABLE);
+#else
+        evpe(vpeflags);
+#endif
+        emt(dmt_flag);
+        local_irq_restore(flags);
 
-        list_for_each_entry(n, &v->notify, list) {
-                n->start(v->minor);
-        }
+        list_for_each_entry(n, &v->notify, list)
+                n->start(minor);
 
         return 0;
 }
@@ -1023,23 +998,15 @@ static int vpe_elfload(struct vpe * v)
         return 0;
 }
 
-void __used dump_vpe(struct vpe * v)
-{
-        struct tc *t;
-
-        settc(v->minor);
-
-        printk(KERN_DEBUG "VPEControl 0x%lx\n", read_vpe_c0_vpecontrol());
-        printk(KERN_DEBUG "VPEConf0 0x%lx\n", read_vpe_c0_vpeconf0());
-
-        list_for_each_entry(t, &vpecontrol.tc_list, list)
-                dump_tc(t);
-}
-
 static void cleanup_tc(struct tc *tc)
 {
+        unsigned long flags;
+        unsigned int mtflags, vpflags;
         int tmp;
 
+        local_irq_save(flags);
+        mtflags = dmt();
+        vpflags = dvpe();
         /* Put MVPE's into 'configuration state' */
         set_c0_mvpcontrol(MVPCONTROL_VPC);
 
@@ -1054,9 +1021,12 @@ static void cleanup_tc(struct tc *tc)
         write_tc_c0_tchalt(TCHALT_H);
 
         /* bind it to anything other than VPE1 */
-        write_tc_c0_tcbind(read_tc_c0_tcbind() & ~TCBIND_CURVPE); // | TCBIND_CURVPE
+//      write_tc_c0_tcbind(read_tc_c0_tcbind() & ~TCBIND_CURVPE); // | TCBIND_CURVPE
 
         clear_c0_mvpcontrol(MVPCONTROL_VPC);
+        evpe(vpflags);
+        emt(mtflags);
+        local_irq_restore(flags);
 }
 
 static int getcwd(char *buff, int size)
@@ -1077,36 +1047,32 @@ static int getcwd(char *buff, int size)
 /* checks VPE is unused and gets ready to load program  */
 static int vpe_open(struct inode *inode, struct file *filp)
 {
-        int minor, ret;
         enum vpe_state state;
-        struct vpe *v;
         struct vpe_notifications *not;
+        struct vpe *v;
+        int ret;
 
-        /* assume only 1 device at the mo. */
-        if ((minor = iminor(inode)) != 1) {
+        if (minor != iminor(inode)) {
+                /* assume only 1 device at the moment. */
                 printk(KERN_WARNING "VPE loader: only vpe1 is supported\n");
                 return -ENODEV;
         }
 
-        if ((v = get_vpe(minor)) == NULL) {
+        if ((v = get_vpe(tclimit)) == NULL) {
                 printk(KERN_WARNING "VPE loader: unable to get vpe\n");
                 return -ENODEV;
         }
 
         state = xchg(&v->state, VPE_STATE_INUSE);
         if (state != VPE_STATE_UNUSED) {
-                dvpe();
-
                 printk(KERN_DEBUG "VPE loader: tc in use dumping regs\n");
 
-                dump_tc(get_tc(minor));
-
                 list_for_each_entry(not, &v->notify, list) {
-                        not->stop(minor);
+                        not->stop(tclimit);
                 }
 
                 release_progmem(v->load_addr);
-                cleanup_tc(get_tc(minor));
+                cleanup_tc(get_tc(tclimit));
         }
 
         /* this of-course trashes what was there before... */
@@ -1133,26 +1099,25 @@ static int vpe_open(struct inode *inode, struct file *filp)
 
         v->shared_ptr = NULL;
         v->__start = 0;
+
         return 0;
 }
 
 static int vpe_release(struct inode *inode, struct file *filp)
 {
-        int minor, ret = 0;
         struct vpe *v;
         Elf_Ehdr *hdr;
+        int ret = 0;
 
-        minor = iminor(inode);
-        if ((v = get_vpe(minor)) == NULL)
+        v = get_vpe(tclimit);
+        if (v == NULL)
                 return -ENODEV;
 
-        // simple case of fire and forget, so tell the VPE to run...
-
         hdr = (Elf_Ehdr *) v->pbuffer;
         if (memcmp(hdr->e_ident, ELFMAG, 4) == 0) {
-                if (vpe_elfload(v) >= 0)
+                if (vpe_elfload(v) >= 0) {
                         vpe_run(v);
-                else {
+                } else {
                         printk(KERN_WARNING "VPE loader: ELF load failed.\n");
                         ret = -ENOEXEC;
                 }
@@ -1179,12 +1144,14 @@ static int vpe_release(struct inode *inode, struct file *filp)
 static ssize_t vpe_write(struct file *file, const char __user * buffer,
                          size_t count, loff_t * ppos)
 {
-        int minor;
         size_t ret = count;
         struct vpe *v;
 
-        minor = iminor(file->f_path.dentry->d_inode);
-        if ((v = get_vpe(minor)) == NULL)
+        if (iminor(file->f_path.dentry->d_inode) != minor)
+                return -ENODEV;
+
+        v = get_vpe(tclimit);
+        if (v == NULL)
                 return -ENODEV;
 
         if (v->pbuffer == NULL) {
@@ -1370,17 +1337,34 @@ static struct device *vpe_dev;
 
 static int __init vpe_module_init(void)
 {
+        unsigned int mtflags, vpflags;
+        int hw_tcs, hw_vpes, tc, err = 0;
+        unsigned long flags, val;
         struct vpe *v = NULL;
         struct device *dev;
         struct tc *t;
-        unsigned long val;
-        int i, err;
 
         if (!cpu_has_mipsmt) {
                 printk("VPE loader: not a MIPS MT capable processor\n");
                 return -ENODEV;
         }
 
+        if (vpelimit == 0) {
+                printk(KERN_WARNING "No VPEs reserved for AP/SP, not "
+                       "initializing VPE loader.\nPass maxvpes=<n> argument as "
+                       "kernel argument\n");
+
+                return -ENODEV;
+        }
+
+        if (tclimit == 0) {
+                printk(KERN_WARNING "No TCs reserved for AP/SP, not "
+                       "initializing VPE loader.\nPass maxtcs=<n> argument as "
+                       "kernel argument\n");
+
+                return -ENODEV;
+        }
+
         major = register_chrdev(0, module_name, &vpe_fops);
         if (major < 0) {
                 printk("VPE loader: unable to register character device\n");
@@ -1388,40 +1372,61 @@ static int __init vpe_module_init(void)
         }
 
         dev = device_create(mt_class, NULL, MKDEV(major, minor),
-                            "tc%d", minor);
+                            "vpe%d", minor);
         if (IS_ERR(dev)) {
                 err = PTR_ERR(dev);
                 goto out_chrdev;
         }
         vpe_dev = dev;
 
-        dmt();
-        dvpe();
+        local_irq_save(flags);
+        mtflags = dmt();
+        vpflags = dvpe();
 
         /* Put MVPE's into 'configuration state' */
         set_c0_mvpcontrol(MVPCONTROL_VPC);
 
         /* dump_mtregs(); */
 
-
         val = read_c0_mvpconf0();
-        for (i = 0; i < ((val & MVPCONF0_PTC) + 1); i++) {
-                t = alloc_tc(i);
+        hw_tcs = (val & MVPCONF0_PTC) + 1;
+        hw_vpes = ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1;
+
+        for (tc = tclimit; tc < hw_tcs; tc++) {
+                /*
+                 * Must re-enable multithreading temporarily or in case we
+                 * reschedule send IPIs or similar we might hang.
+                 */
+                clear_c0_mvpcontrol(MVPCONTROL_VPC);
+                evpe(vpflags);
+                emt(mtflags);
+                local_irq_restore(flags);
+                t = alloc_tc(tc);
+                if (!t) {
+                        err = -ENOMEM;
+                        goto out;
+                }
+
+                local_irq_save(flags);
+                mtflags = dmt();
+                vpflags = dvpe();
+                set_c0_mvpcontrol(MVPCONTROL_VPC);
 
                 /* VPE's */
-                if (i < ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1) {
-                        settc(i);
+                if (tc < hw_tcs) {
+                        settc(tc);
 
-                        if ((v = alloc_vpe(i)) == NULL) {
+                        if ((v = alloc_vpe(tc)) == NULL) {
                                 printk(KERN_WARNING "VPE: unable to allocate VPE\n");
-                                return -ENODEV;
+
+                                goto out_reenable;
                         }
 
                         /* add the tc to the list of this vpe's tc's. */
                         list_add(&t->tc, &v->tc);
 
                         /* deactivate all but vpe0 */
-                        if (i != 0) {
+                        if (tc >= tclimit) {
                                 unsigned long tmp = read_vpe_c0_vpeconf0();
 
                                 tmp &= ~VPECONF0_VPA;
@@ -1434,7 +1439,7 @@ static int __init vpe_module_init(void)
                         /* disable multi-threading with TC's */
                         write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);
 
-                        if (i != 0) {
+                        if (tc >= vpelimit) {
                                 /*
                                  * Set config to be the same as vpe0,
                                  * particularly kseg0 coherency alg
@@ -1446,10 +1451,10 @@ static int __init vpe_module_init(void)
                 /* TC's */
                 t->pvpe = v;    /* set the parent vpe */
 
-                if (i != 0) {
+                if (tc >= tclimit) {
                         unsigned long tmp;
 
-                        settc(i);
+                        settc(tc);
 
                         /* Any TC that is bound to VPE0 gets left as is - in case
                            we are running SMTC on VPE0. A TC that is bound to any
@@ -1479,9 +1484,14 @@ static int __init vpe_module_init(void)
                 }
         }
 
+out_reenable:
         /* release config state */
         clear_c0_mvpcontrol(MVPCONTROL_VPC);
 
+        evpe(vpflags);
+        emt(mtflags);
+        local_irq_restore(flags);
+
 #ifdef CONFIG_MIPS_APSP_KSPD
         kspd_events.kspd_sp_exit = kspd_sp_exit;
 #endif
@@ -1490,6 +1500,7 @@ static int __init vpe_module_init(void)
 out_chrdev:
         unregister_chrdev(major, module_name);
 
+out:
         return err;
 }