vpe.c

linux-2.6.15.6

{
	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);

	printk(KERN_DEBUG "dump_elfsymbols: n %d\n", n);
	for (i = 1; i < n; i++) {
		printk(KERN_DEBUG " i %d name <%s> 0x%x\n", i,
		       strtab + sym[i].st_name, sym[i].st_value);
	}
}
#endif

static void dump_tc(struct tc *t)
{
	printk(KERN_WARNING "VPE: TC index %d TCStatus 0x%lx halt 0x%lx\n",
	       t->index, read_tc_c0_tcstatus(), read_tc_c0_tchalt());
	printk(KERN_WARNING "VPE: tcrestart 0x%lx\n", read_tc_c0_tcrestart());
}

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... */
int vpe_run(struct vpe * v)
{
	unsigned long val;
	struct tc *t;

	/* check we are the Master VPE */
	val = read_c0_vpeconf0();
	if (!(val & VPECONF0_MVP)) {
		printk(KERN_WARNING
		       "VPE: only Master VPE's are allowed to configure MT\n");
		return -1;
	}

	/* disable MT (using dvpe) */
	dvpe();

	/* Put MVPE's into 'configuration state' */
	set_c0_mvpcontrol(MVPCONTROL_VPC);

	if (!list_empty(&v->tc)) {
		if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
			printk(KERN_WARNING "VPE: TC %d is already in use.\n",
			       t->index);
			return -ENOEXEC;
		}
	} else {
		printk(KERN_WARNING "VPE: No TC's associated with VPE %d\n",
		       v->minor);
		return -ENOEXEC;
	}

	settc(t->index);

	val = read_vpe_c0_vpeconf0();

	/* 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: TC %d is already doing something!\n",
		       t->index);

		dump_tclist();
		return -ENOEXEC;
	}

	/* Write the address we want it to start running from in the TCPC register. */
	write_tc_c0_tcrestart((unsigned long)v->__start);

	/* write the sivc_info address to tccontext */
	write_tc_c0_tccontext((unsigned long)0);

	/* Set up the XTC bit in vpeconf0 to point at our tc */
	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() | (t->index << VPECONF0_XTC_SHIFT));

	/* mark the TC as activated, not interrupt exempt and not dynamically allocatable */
	val = read_tc_c0_tcstatus();
	val = (val & ~(TCSTATUS_DA | TCSTATUS_IXMT)) | TCSTATUS_A;
	write_tc_c0_tcstatus(val);

	write_tc_c0_tchalt(read_tc_c0_tchalt() & ~TCHALT_H);

	/* set up VPE1 */
	write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);	// no multiple TC's
	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() | VPECONF0_VPA);	// enable this VPE

	/*
	 * The sde-kit passes 'memsize' to __start in $a3, so set something
	 * here...
	 * Or set $a3 (register 7) to zero and define DFLT_STACK_SIZE and
	 * DFLT_HEAP_SIZE when you compile your program
	 */

	mttgpr(7, 0);

	/* set config to be the same as vpe0, particularly kseg0 coherency alg */
	write_vpe_c0_config(read_c0_config());

	/* clear out any left overs from a previous program */
	write_vpe_c0_cause(0);

	/* take system out of configuration state */
	clear_c0_mvpcontrol(MVPCONTROL_VPC);

	/* clear interrupts enabled IE, ERL, EXL, and KSU from c0 status */
	write_vpe_c0_status(read_vpe_c0_status() & ~(ST0_ERL | ST0_KSU | ST0_IE | ST0_EXL));

	/* set it running */
	evpe(EVPE_ENABLE);

	return 0;
}

static unsigned long find_vpe_symbols(struct vpe * v, Elf_Shdr * sechdrs,
				      unsigned int symindex, const char *strtab,
				      struct module *mod)
{
	Elf_Sym *sym = (void *)sechdrs[symindex].sh_addr;
	unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);

	for (i = 1; i < n; i++) {
		if (strcmp(strtab + sym[i].st_name, "__start") == 0) {
			v->__start = sym[i].st_value;
		}

		if (strcmp(strtab + sym[i].st_name, "vpe_shared") == 0) {
			v->shared_ptr = (void *)sym[i].st_value;
		}
	}

	return 0;
}

/*
 * Allocates a VPE with some program code space(the load address), copies
 * the contents of the program (p)buffer performing relocatations/etc,
 * free's it when finished.
*/
int vpe_elfload(struct vpe * v)
{
	Elf_Ehdr *hdr;
	Elf_Shdr *sechdrs;
	long err = 0;
	char *secstrings, *strtab = NULL;
	unsigned int len, i, symindex = 0, strindex = 0;

	struct module mod;	// so we can re-use the relocations code

	memset(&mod, 0, sizeof(struct module));
	strcpy(mod.name, "VPE dummy prog module");

	hdr = (Elf_Ehdr *) v->pbuffer;
	len = v->plen;

	/* Sanity checks against insmoding binaries or wrong arch,
	   weird elf version */
	if (memcmp(hdr->e_ident, ELFMAG, 4) != 0
	    || hdr->e_type != ET_REL || !elf_check_arch(hdr)
	    || hdr->e_shentsize != sizeof(*sechdrs)) {
		printk(KERN_WARNING
		       "VPE program, wrong arch or weird elf version\n");

		return -ENOEXEC;
	}

	if (len < hdr->e_shoff + hdr->e_shnum * sizeof(Elf_Shdr)) {
		printk(KERN_ERR "VPE program length %u truncated\n", len);
		return -ENOEXEC;
	}

	/* Convenience variables */
	sechdrs = (void *)hdr + hdr->e_shoff;
	secstrings = (void *)hdr + sechdrs[hdr->e_shstrndx].sh_offset;
	sechdrs[0].sh_addr = 0;

	/* And these should exist, but gcc whinges if we don't init them */
	symindex = strindex = 0;

	for (i = 1; i < hdr->e_shnum; i++) {

		if (sechdrs[i].sh_type != SHT_NOBITS
		    && len < sechdrs[i].sh_offset + sechdrs[i].sh_size) {
			printk(KERN_ERR "VPE program length %u truncated\n",
			       len);
			return -ENOEXEC;
		}

		/* Mark all sections sh_addr with their address in the
		   temporary image. */
		sechdrs[i].sh_addr = (size_t) hdr + sechdrs[i].sh_offset;

		/* Internal symbols and strings. */
		if (sechdrs[i].sh_type == SHT_SYMTAB) {
			symindex = i;
			strindex = sechdrs[i].sh_link;
			strtab = (char *)hdr + sechdrs[strindex].sh_offset;
		}
	}

	layout_sections(&mod, hdr, sechdrs, secstrings);

	v->load_addr = alloc_progmem(mod.core_size);
	memset(v->load_addr, 0, mod.core_size);

	printk("VPE elf_loader: loading to %p\n", v->load_addr);

	for (i = 0; i < hdr->e_shnum; i++) {
		void *dest;

		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
			continue;

		dest = v->load_addr + sechdrs[i].sh_entsize;

		if (sechdrs[i].sh_type != SHT_NOBITS)
			memcpy(dest, (void *)sechdrs[i].sh_addr,
			       sechdrs[i].sh_size);
		/* Update sh_addr to point to copy in image. */
		sechdrs[i].sh_addr = (unsigned long)dest;
	}

	/* Fix up syms, so that st_value is a pointer to location. */
	err =
		simplify_symbols(sechdrs, symindex, strtab, secstrings,
				 hdr->e_shnum, &mod);
	if (err < 0) {
		printk(KERN_WARNING "VPE: unable to simplify symbols\n");
		goto cleanup;
	}

	/* Now do relocations. */
	for (i = 1; i < hdr->e_shnum; i++) {
		const char *strtab = (char *)sechdrs[strindex].sh_addr;
		unsigned int info = sechdrs[i].sh_info;

		/* Not a valid relocation section? */
		if (info >= hdr->e_shnum)
			continue;

		/* Don't bother with non-allocated sections */
		if (!(sechdrs[info].sh_flags & SHF_ALLOC))
			continue;

		if (sechdrs[i].sh_type == SHT_REL)
			err =
				apply_relocations(sechdrs, strtab, symindex, i, &mod);
		else if (sechdrs[i].sh_type == SHT_RELA)
			err = apply_relocate_add(sechdrs, strtab, symindex, i,
						 &mod);
		if (err < 0) {
			printk(KERN_WARNING
			       "vpe_elfload: error in relocations err %ld\n",
			       err);
			goto cleanup;
		}
	}

	/* make sure it's physically written out */
	flush_icache_range((unsigned long)v->load_addr,
			   (unsigned long)v->load_addr + v->len);

	if ((find_vpe_symbols(v, sechdrs, symindex, strtab, &mod)) < 0) {

		printk(KERN_WARNING
		       "VPE: program doesn't contain __start or vpe_shared symbols\n");
		err = -ENOEXEC;
	}

	printk(" elf loaded\n");

cleanup:
	return err;
}

static void dump_vpe(struct vpe * v)
{
	struct tc *t;

	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);
	}
}

/* checks for VPE is unused and gets ready to load program	 */
static int vpe_open(struct inode *inode, struct file *filp)
{
	int minor;
	struct vpe *v;

	/* assume only 1 device at the mo. */
	if ((minor = MINOR(inode->i_rdev)) != 1) {
		printk(KERN_WARNING "VPE: only vpe1 is supported\n");
		return -ENODEV;
	}

	if ((v = get_vpe(minor)) == NULL) {
		printk(KERN_WARNING "VPE: unable to get vpe\n");
		return -ENODEV;
	}

	if (v->state != VPE_STATE_UNUSED) {
		unsigned long tmp;
		struct tc *t;

		printk(KERN_WARNING "VPE: device %d already in use\n", minor);

		dvpe();
		dump_vpe(v);

		printk(KERN_WARNING "VPE: re-initialising %d\n", minor);

		release_progmem(v->load_addr);

		t = get_tc(minor);
		settc(minor);
		tmp = read_tc_c0_tcstatus();

		/* mark not allocated and not dynamically allocatable */
		tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
		tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
		write_tc_c0_tcstatus(tmp);

		write_tc_c0_tchalt(TCHALT_H);

	}

	// allocate it so when we get write ops we know it's expected.
	v->state = VPE_STATE_INUSE;

	/* this of-course trashes what was there before... */
	v->pbuffer = vmalloc(P_SIZE);
	v->plen = P_SIZE;
	v->load_addr = NULL;
	v->len = 0;

	return 0;
}

static int vpe_release(struct inode *inode, struct file *filp)
{
	int minor, ret = 0;
	struct vpe *v;
	Elf_Ehdr *hdr;

	minor = MINOR(inode->i_rdev);
	if ((v = get_vpe(minor)) == 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)
			vpe_run(v);
		else {
			printk(KERN_WARNING "VPE: ELF load failed.\n");
			ret = -ENOEXEC;
		}
	} else {
		printk(KERN_WARNING "VPE: only elf files are supported\n");
		ret = -ENOEXEC;
	}

	// cleanup any temp buffers
	if (v->pbuffer)
		vfree(v->pbuffer);
	v->plen = 0;
	return ret;
}

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 = MINOR(file->f_dentry->d_inode->i_rdev);
	if ((v = get_vpe(minor)) == NULL)
		return -ENODEV;

	if (v->pbuffer == NULL) {
		printk(KERN_ERR "vpe_write: no pbuffer\n");
		return -ENOMEM;
	}

	if ((count + v->len) > v->plen) {
		printk(KERN_WARNING
		       "VPE Loader: elf size too big. Perhaps strip uneeded symbols\n");
		return -ENOMEM;
	}

	count -= copy_from_user(v->pbuffer + v->len, buffer, count);
	if (!count) {
		printk("vpe_write: copy_to_user failed\n");
		return -EFAULT;
	}

	v->len += count;
	return ret;
}

static struct file_operations vpe_fops = {
	.owner = THIS_MODULE,
	.open = vpe_open,
	.release = vpe_release,
	.write = vpe_write
};

/* module wrapper entry points */
/* give me a vpe */
vpe_handle vpe_alloc(void)
{
	int i;
	struct vpe *v;

	/* find a vpe */
	for (i = 1; i < MAX_VPES; i++) {
		if ((v = get_vpe(i)) != NULL) {
			v->state = VPE_STATE_INUSE;
			return v;
		}
	}
	return NULL;
}

EXPORT_SYMBOL(vpe_alloc);

/* start running from here */
int vpe_start(vpe_handle vpe, unsigned long start)
{
	struct vpe *v = vpe;

	v->__start = start;
	return vpe_run(v);
}

EXPORT_SYMBOL(vpe_start);

/* halt it for now */
int vpe_stop(vpe_handle vpe)
{
	struct vpe *v = vpe;
	struct tc *t;
	unsigned int evpe_flags;

	evpe_flags = dvpe();

	if ((t = list_entry(v->tc.next, struct tc, tc)) != NULL) {

		settc(t->index);
		write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);
	}

	evpe(evpe_flags);

	return 0;
}

EXPORT_SYMBOL(vpe_stop);

/* I've done with it thank you */
int vpe_free(vpe_handle vpe)
{
	struct vpe *v = vpe;
	struct tc *t;
	unsigned int evpe_flags;

	if ((t = list_entry(v->tc.next, struct tc, tc)) == NULL) {
		return -ENOEXEC;
	}

	evpe_flags = dvpe();

	/* Put MVPE's into 'configuration state' */
	set_c0_mvpcontrol(MVPCONTROL_VPC);

	settc(t->index);
	write_vpe_c0_vpeconf0(read_vpe_c0_vpeconf0() & ~VPECONF0_VPA);

	/* mark the TC unallocated and halt'ed */
	write_tc_c0_tcstatus(read_tc_c0_tcstatus() & ~TCSTATUS_A);
	write_tc_c0_tchalt(TCHALT_H);

	v->state = VPE_STATE_UNUSED;

	clear_c0_mvpcontrol(MVPCONTROL_VPC);
	evpe(evpe_flags);

	return 0;
}

EXPORT_SYMBOL(vpe_free);

void *vpe_get_shared(int index)
{
	struct vpe *v;

	if ((v = get_vpe(index)) == NULL) {
		printk(KERN_WARNING "vpe: invalid vpe index %d\n", index);
		return NULL;
	}

	return v->shared_ptr;
}

EXPORT_SYMBOL(vpe_get_shared);

static int __init vpe_module_init(void)
{
	struct vpe *v = NULL;
	struct tc *t;
	unsigned long val;
	int i;

	if (!cpu_has_mipsmt) {
		printk("VPE loader: not a MIPS MT capable processor\n");
		return -ENODEV;
	}

	if ((major = register_chrdev(0, module_name, &vpe_fops) < 0)) {
		printk("VPE loader: unable to register character device\n");
		return major;
	}

	dmt();
	dvpe();

	/* Put MVPE's into 'configuration state' */
	set_c0_mvpcontrol(MVPCONTROL_VPC);

	/* dump_mtregs(); */

	INIT_LIST_HEAD(&vpecontrol.vpe_list);
	INIT_LIST_HEAD(&vpecontrol.tc_list);

	val = read_c0_mvpconf0();
	for (i = 0; i < ((val & MVPCONF0_PTC) + 1); i++) {
		t = alloc_tc(i);

		/* VPE's */
		if (i < ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1) {
			settc(i);

			if ((v = alloc_vpe(i)) == NULL) {
				printk(KERN_WARNING "VPE: unable to allocate VPE\n");
				return -ENODEV;
			}

			list_add(&t->tc, &v->tc);	/* add the tc to the list of this vpe's tc's. */

			/* deactivate all but vpe0 */
			if (i != 0) {
				unsigned long tmp = read_vpe_c0_vpeconf0();

				tmp &= ~VPECONF0_VPA;

				/* master VPE */
				tmp |= VPECONF0_MVP;
				write_vpe_c0_vpeconf0(tmp);
			}

			/* disable multi-threading with TC's */
			write_vpe_c0_vpecontrol(read_vpe_c0_vpecontrol() & ~VPECONTROL_TE);

			if (i != 0) {
				write_vpe_c0_status((read_c0_status() &
						     ~(ST0_IM | ST0_IE | ST0_KSU))
						    | ST0_CU0);

				/* set config to be the same as vpe0, particularly kseg0 coherency alg */
				write_vpe_c0_config(read_c0_config());
			}

		}

		/* TC's */
		t->pvpe = v;	/* set the parent vpe */

		if (i != 0) {
			unsigned long tmp;

			/* tc 0 will of course be running.... */
			if (i == 0)
				t->state = TC_STATE_RUNNING;

			settc(i);

			/* bind a TC to each VPE, May as well put all excess TC's
			   on the last VPE */
			if (i >= (((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1))
				write_tc_c0_tcbind(read_tc_c0_tcbind() |
						   ((val & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT));
			else
				write_tc_c0_tcbind(read_tc_c0_tcbind() | i);

			tmp = read_tc_c0_tcstatus();

			/* mark not allocated and not dynamically allocatable */
			tmp &= ~(TCSTATUS_A | TCSTATUS_DA);
			tmp |= TCSTATUS_IXMT;	/* interrupt exempt */
			write_tc_c0_tcstatus(tmp);

			write_tc_c0_tchalt(TCHALT_H);
		}
	}

	/* release config state */
	clear_c0_mvpcontrol(MVPCONTROL_VPC);

	return 0;
}

static void __exit vpe_module_exit(void)
{
	struct vpe *v, *n;

	list_for_each_entry_safe(v, n, &vpecontrol.vpe_list, list) {
		if (v->state != VPE_STATE_UNUSED) {
			release_vpe(v);
		}
	}

	unregister_chrdev(major, module_name);
}

module_init(vpe_module_init);
module_exit(vpe_module_exit);
MODULE_DESCRIPTION("MIPS VPE Loader");
MODULE_AUTHOR("Elizabeth Clarke, MIPS Technologies, Inc");
MODULE_LICENSE("GPL");