sys/arch/evbppc/evbppc/bus_dma.c
author yamt <yamt@NetBSD.org>
Wed, 01 Mar 2006 12:38:10 +0000
branchtrunk
changeset 143952 9cf9eb9ab26a
parent 142000 a267994bcd74
permissions -rw-r--r--
merge yamt-uio_vmspace branch. - use vmspace rather than proc or lwp where appropriate. the latter is more natural to specify an address space. (and less likely to be abused for random purposes.) - fix a swdmover race.

/*	$NetBSD: bus_dma.c,v 1.12 2006/03/01 12:38:11 yamt Exp $	*/

/*-
 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
 * NASA Ames Research Center.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the NetBSD
 *	Foundation, Inc. and its contributors.
 * 4. Neither the name of The NetBSD Foundation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: bus_dma.c,v 1.12 2006/03/01 12:38:11 yamt Exp $");

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/device.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/mbuf.h>

#include <uvm/uvm_extern.h>

#define _IBM4XX_BUS_DMA_PRIVATE
#include <machine/bus.h>
#include <machine/intr.h>

int	_bus_dmamap_load_buffer(bus_dma_tag_t, bus_dmamap_t, void *,
	    bus_size_t, struct vmspace *, int, paddr_t *, int *, int);

/*
 * Common function for DMA map creation.  May be called by bus-specific
 * DMA map creation functions.
 */
int
_bus_dmamap_create(bus_dma_tag_t t, bus_size_t size, int nsegments,
		   bus_size_t maxsegsz, bus_size_t boundary, int flags,
		   bus_dmamap_t *dmamp)
{
	struct ibm4xx_bus_dmamap *map;
	void *mapstore;
	size_t mapsize;

	/*
	 * Allocate and initialize the DMA map.  The end of the map
	 * is a variable-sized array of segments, so we allocate enough
	 * room for them in one shot.
	 *
	 * Note we don't preserve the WAITOK or NOWAIT flags.  Preservation
	 * of ALLOCNOW notifies others that we've reserved these resources,
	 * and they are not to be freed.
	 *
	 * The bus_dmamap_t includes one bus_dma_segment_t, hence
	 * the (nsegments - 1).
	 */
	mapsize = sizeof(struct ibm4xx_bus_dmamap) +
	    (sizeof(bus_dma_segment_t) * (nsegments - 1));
	if ((mapstore = malloc(mapsize, M_DMAMAP,
	    (flags & BUS_DMA_NOWAIT) ? M_NOWAIT : M_WAITOK)) == NULL)
		return (ENOMEM);

	memset(mapstore, 0, mapsize);
	map = (struct ibm4xx_bus_dmamap *)mapstore;
	map->_dm_size = size;
	map->_dm_segcnt = nsegments;
	map->_dm_maxmaxsegsz = maxsegsz;
	map->_dm_boundary = boundary;
	map->_dm_bounce_thresh = t->_bounce_thresh;
	map->_dm_flags = flags & ~(BUS_DMA_WAITOK|BUS_DMA_NOWAIT);
	map->dm_maxsegsz = maxsegsz;
	map->dm_mapsize = 0;		/* no valid mappings */
	map->dm_nsegs = 0;

	*dmamp = map;
	return (0);
}

/*
 * Common function for DMA map destruction.  May be called by bus-specific
 * DMA map destruction functions.
 */
void
_bus_dmamap_destroy(bus_dma_tag_t t, bus_dmamap_t map)
{

	free(map, M_DMAMAP);
}

/*
 * Utility function to load a linear buffer.  lastaddrp holds state
 * between invocations (for multiple-buffer loads).  segp contains
 * the starting segment on entrance, and the ending segment on exit.
 * first indicates if this is the first invocation of this function.
 */
int
_bus_dmamap_load_buffer(bus_dma_tag_t t, bus_dmamap_t map, void *buf,
			bus_size_t buflen, struct vmspace *vm, int flags,
			paddr_t *lastaddrp, int *segp, int first)
{
	bus_size_t sgsize;
	bus_addr_t curaddr, lastaddr, baddr, bmask;
	vaddr_t vaddr = (vaddr_t)buf;
	int seg;

	lastaddr = *lastaddrp;
	bmask = ~(map->_dm_boundary - 1);

	for (seg = *segp; buflen > 0 ; ) {
		/*
		 * Get the physical address for this segment.
		 */
		if (!VMSPACE_IS_KERNEL(vm))
			(void) pmap_extract(vm_map_pmap(&vm->vm_map),
			    vaddr, (void *)&curaddr);
		else
			curaddr = vtophys(vaddr);

		/*
		 * If we're beyond the bounce threshold, notify
		 * the caller.
		 */
		if (map->_dm_bounce_thresh != 0 &&
		    curaddr >= map->_dm_bounce_thresh)
			return (EINVAL);

		/*
		 * Compute the segment size, and adjust counts.
		 */
		sgsize = PAGE_SIZE - ((u_long)vaddr & PGOFSET);
		if (buflen < sgsize)
			sgsize = buflen;

		/*
		 * Make sure we don't cross any boundaries.
		 */
		if (map->_dm_boundary > 0) {
			baddr = (curaddr + map->_dm_boundary) & bmask;
			if (sgsize > (baddr - curaddr))
				sgsize = (baddr - curaddr);
		}

		/*
		 * Insert chunk into a segment, coalescing with
		 * the previous segment if possible.
		 */
		if (first) {
			map->dm_segs[seg].ds_addr = curaddr;
			map->dm_segs[seg].ds_len = sgsize;
			first = 0;
		} else {
			if (curaddr == lastaddr &&
			    (map->dm_segs[seg].ds_len + sgsize) <=
			     map->dm_maxsegsz &&
			    (map->_dm_boundary == 0 ||
			     (map->dm_segs[seg].ds_addr & bmask) ==
			     (curaddr & bmask)))
				map->dm_segs[seg].ds_len += sgsize;
			else {
				if (++seg >= map->_dm_segcnt)
					break;
				map->dm_segs[seg].ds_addr = curaddr;
				map->dm_segs[seg].ds_len = sgsize;
			}
		}

		lastaddr = curaddr + sgsize;
		vaddr += sgsize;
		buflen -= sgsize;
	}

	*segp = seg;
	*lastaddrp = lastaddr;

	/*
	 * Did we fit?
	 */
	if (buflen != 0)
		return (EFBIG);		/* XXX better return value here? */

	return (0);
}

/*
 * Common function for loading a DMA map with a linear buffer.  May
 * be called by bus-specific DMA map load functions.
 */
int
_bus_dmamap_load(bus_dma_tag_t t, bus_dmamap_t map, void *buf,
		 bus_size_t buflen, struct proc *p, int flags)
{
	paddr_t lastaddr;
	int seg, error;
	struct vmspace *vm;

	/*
	 * Make sure that on error condition we return "no valid mappings".
	 */
	map->dm_mapsize = 0;
	map->dm_nsegs = 0;
	KASSERT(map->dm_maxsegsz <= map->_dm_maxmaxsegsz);

	if (buflen > map->_dm_size)
		return (EINVAL);

	if (p != NULL) {
		vm = p->p_vmspace;
	} else {
		vm = vmspace_kernel();
	}

	seg = 0;
	error = _bus_dmamap_load_buffer(t, map, buf, buflen, vm, flags,
		&lastaddr, &seg, 1);
	if (error == 0) {
		map->dm_mapsize = buflen;
		map->dm_nsegs = seg + 1;
	}
	return (error);
}

/*
 * Like _bus_dmamap_load(), but for mbufs.
 */
int
_bus_dmamap_load_mbuf(bus_dma_tag_t t, bus_dmamap_t map, struct mbuf *m0,
		      int flags)
{
	paddr_t lastaddr;
	int seg, error, first;
	struct mbuf *m;

	/*
	 * Make sure that on error condition we return "no valid mappings."
	 */
	map->dm_mapsize = 0;
	map->dm_nsegs = 0;
	KASSERT(map->dm_maxsegsz <= map->_dm_maxmaxsegsz);

#ifdef DIAGNOSTIC
	if ((m0->m_flags & M_PKTHDR) == 0)
		panic("_bus_dmamap_load_mbuf: no packet header");
#endif

	if (m0->m_pkthdr.len > map->_dm_size)
		return (EINVAL);

	first = 1;
	seg = 0;
	error = 0;
	for (m = m0; m != NULL && error == 0; m = m->m_next) {
		if (m->m_len == 0)
			continue;
		error = _bus_dmamap_load_buffer(t, map, m->m_data, m->m_len,
		    vmspace_kernel(), flags, &lastaddr, &seg, first);
		first = 0;
	}
	if (error == 0) {
		map->dm_mapsize = m0->m_pkthdr.len;
		map->dm_nsegs = seg + 1;
	}
	return (error);
}

/*
 * Like _bus_dmamap_load(), but for uios.
 */
int
_bus_dmamap_load_uio(bus_dma_tag_t t, bus_dmamap_t map, struct uio *uio,
		     int flags)
{
	paddr_t lastaddr;
	int seg, i, error, first;
	bus_size_t minlen, resid;
	struct iovec *iov;
	caddr_t addr;

	/*
	 * Make sure that on error condition we return "no valid mappings."
	 */
	map->dm_mapsize = 0;
	map->dm_nsegs = 0;
	KASSERT(map->dm_maxsegsz <= map->_dm_maxmaxsegsz);

	resid = uio->uio_resid;
	iov = uio->uio_iov;

	first = 1;
	seg = 0;
	error = 0;
	for (i = 0; i < uio->uio_iovcnt && resid != 0 && error == 0; i++) {
		/*
		 * Now at the first iovec to load.  Load each iovec
		 * until we have exhausted the residual count.
		 */
		minlen = resid < iov[i].iov_len ? resid : iov[i].iov_len;
		addr = (caddr_t)iov[i].iov_base;

		error = _bus_dmamap_load_buffer(t, map, addr, minlen,
		    uio->uio_vmspace, flags, &lastaddr, &seg, first);
		first = 0;

		resid -= minlen;
	}
	if (error == 0) {
		map->dm_mapsize = uio->uio_resid;
		map->dm_nsegs = seg + 1;
	}
	return (error);
}

/*
 * Like _bus_dmamap_load(), but for raw memory.
 */
int
_bus_dmamap_load_raw(bus_dma_tag_t t, bus_dmamap_t map, bus_dma_segment_t *segs,
		     int nsegs, bus_size_t size, int flags)
{

	panic("_bus_dmamap_load_raw: not implemented");
}

/*
 * Common function for unloading a DMA map.  May be called by
 * chipset-specific DMA map unload functions.
 */
void
_bus_dmamap_unload(bus_dma_tag_t t, bus_dmamap_t map)
{

	/*
	 * No resources to free; just mark the mappings as
	 * invalid.
	 */
	map->dm_maxsegsz = map->_dm_maxmaxsegsz;
	map->dm_mapsize = 0;
	map->dm_nsegs = 0;
}

/*
 * Common function for DMA map synchronization.  May be called
 * by chipset-specific DMA map synchronization functions.
 */
void
_bus_dmamap_sync(bus_dma_tag_t t, bus_dmamap_t map, bus_addr_t offset,
		 bus_size_t len, int ops)
{
	bus_addr_t addr, startline, endline, cachemask;
	bus_size_t cachestride;
	int i, minlen;

	/*
	 * Mixing PRE and POST operations is not allowed.
	 */
	if ((ops & (BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE)) != 0 &&
	    (ops & (BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE)) != 0)
		panic("_bus_dmamap_sync: mix PRE and POST");

#if 0
	/* We only care about PREREAD and PREWRITE */
	if ((ops & (BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE)) == 0)
		return;
#endif
	cachestride = curcpu()->ci_ci.dcache_line_size;
#ifdef DIAGNOSTIC
	if (cachestride<4)
		printf("_bus_dmamap_sync: WARNING line size %x\n", cachestride);
#endif
	if (cachestride < 4) cachestride = 4;
	cachemask = ~(cachestride - 1);

	for (i = 0; i < map->dm_nsegs && len != 0; i++) {
		/* Find the beginning segment. */
		if (offset >= map->dm_segs[i].ds_len) {
			offset -= map->dm_segs[i].ds_len;
			continue;
		}

		/*
		 * Now at the first segment to sync; nail
		 * each segment until we have exhausted the
		 * length.
		 */
		minlen = len < map->dm_segs[i].ds_len - offset ? len :
		map->dm_segs[i].ds_len - offset;

		addr = map->dm_segs[i].ds_addr;

#if 0    
		printf("bus_dmamap_sync: %s segment %d (%p + %x) \n",
		    (ops & BUS_DMASYNC_PREREAD) ? "invalidate": "flush",
		    i, addr + offset, minlen);
#endif
		/* XXX does not deal with address wrap */

		/* Cache line first byte belongs to */
		startline = (addr + offset) & cachemask; 
		/* Cache line last byte belongs to */
		endline = ( addr + offset + minlen - 1) & cachemask; 

		if (ops & BUS_DMASYNC_PREREAD) {
			/*
			 * Flush & Invalidate cache, so we'll re-read whatever's
			 * in memory now.
			 */
			for (addr = startline; addr <= endline;
			    addr += cachestride)
				__asm volatile("dcbf 0,%0"::"r"(addr));
		} else if (ops & BUS_DMASYNC_PREWRITE) {
			/*
			 * Flush cache so memory contains correct data.
			 */
			for (addr = startline; addr <= endline;
			    addr += cachestride)
				__asm volatile("dcbst 0,%0"::"r"(addr));
		} else if (ops & BUS_DMASYNC_POSTREAD){
			/*
			 * Invalidate cache w/o flush, so we'll read whatever's
			 * in memory.
			 */
			for (addr = startline; addr <= endline;
			    addr += cachestride)
				__asm volatile("dcbi 0,%0"::"r"(addr));
		}

		offset = 0;
		len -= minlen;
	}
	__asm volatile("eieio; sync; isync");
}

/*
 * Common function for DMA-safe memory allocation.  May be called
 * by bus-specific DMA memory allocation functions.
 */
int
_bus_dmamem_alloc(bus_dma_tag_t t, bus_size_t size, bus_size_t alignment,
		  bus_size_t boundary, bus_dma_segment_t *segs, int nsegs,
		  int *rsegs, int flags)
{
	paddr_t avail_start, avail_end;
	paddr_t curaddr, lastaddr, high;
	struct vm_page *m;    
	struct pglist mlist;
	int curseg, error;

	avail_start = vm_physmem[0].avail_start << PGSHIFT;
	avail_end = vm_physmem[vm_nphysseg - 1].avail_end << PGSHIFT;

	/* Always round the size. */
	size = round_page(size);

	high = avail_end - PAGE_SIZE;

	/*
	 * Allocate pages from the VM system.
	 */
	error = uvm_pglistalloc(size, avail_start, high, alignment, boundary,
	    &mlist, nsegs, (flags & BUS_DMA_NOWAIT) == 0);
	if (error)
		return (error);

	/*
	 * Compute the location, size, and number of segments actually
	 * returned by the VM code.
	 */
	m = mlist.tqh_first;
	curseg = 0;
	lastaddr = segs[curseg].ds_addr = VM_PAGE_TO_PHYS(m);
	segs[curseg].ds_len = PAGE_SIZE;
	m = m->pageq.tqe_next;

	for (; m != NULL; m = m->pageq.tqe_next) {
		curaddr = VM_PAGE_TO_PHYS(m);
#ifdef DIAGNOSTIC
		if (curaddr < avail_start || curaddr >= high) {
			printf("vm_page_alloc_memory returned non-sensical"
			    " address 0x%lx\n", curaddr);
			panic("_bus_dmamem_alloc");
		}
#endif
		if (curaddr == (lastaddr + PAGE_SIZE))
			segs[curseg].ds_len += PAGE_SIZE;
		else {
			curseg++;
			segs[curseg].ds_addr = curaddr;
			segs[curseg].ds_len = PAGE_SIZE;
		}
		lastaddr = curaddr;
	}

	*rsegs = curseg + 1;

	return (0);
}

/*
 * Common function for freeing DMA-safe memory.  May be called by
 * bus-specific DMA memory free functions.
 */
void
_bus_dmamem_free(bus_dma_tag_t t, bus_dma_segment_t *segs, int nsegs)
{
	struct vm_page *m;
	bus_addr_t addr;
	struct pglist mlist;
	int curseg;

	/*
	 * Build a list of pages to free back to the VM system.
	 */
	TAILQ_INIT(&mlist);
	for (curseg = 0; curseg < nsegs; curseg++) {
		for (addr = segs[curseg].ds_addr;
		    addr < (segs[curseg].ds_addr + segs[curseg].ds_len);
		    addr += PAGE_SIZE) {
			m = PHYS_TO_VM_PAGE(addr);
			TAILQ_INSERT_TAIL(&mlist, m, pageq);
		}
	}

	uvm_pglistfree(&mlist);
}

/*
 * Common function for mapping DMA-safe memory.  May be called by
 * bus-specific DMA memory map functions.
 */
int
_bus_dmamem_map(bus_dma_tag_t t, bus_dma_segment_t *segs, int nsegs,
		size_t size, caddr_t *kvap, int flags)
{
	vaddr_t va;
	bus_addr_t addr;
	int curseg;
	const uvm_flag_t kmflags =
	    (flags & BUS_DMA_NOWAIT) != 0 ? UVM_KMF_NOWAIT : 0;

	size = round_page(size);

	va = uvm_km_alloc(kernel_map, size, 0, UVM_KMF_VAONLY | kmflags);

	if (va == 0)
		return (ENOMEM);

	*kvap = (caddr_t)va;

	for (curseg = 0; curseg < nsegs; curseg++) {
		for (addr = segs[curseg].ds_addr;
		    addr < (segs[curseg].ds_addr + segs[curseg].ds_len);
		    addr += PAGE_SIZE, va += PAGE_SIZE, size -= PAGE_SIZE) {
			if (size == 0)
				panic("_bus_dmamem_map: size botch");
			pmap_enter(pmap_kernel(), va, addr,
			    VM_PROT_READ | VM_PROT_WRITE,
			    VM_PROT_READ | VM_PROT_WRITE | PMAP_WIRED |
			    PME_NOCACHE);
		}
	}

	return (0);
}

/*
 * Common function for unmapping DMA-safe memory.  May be called by
 * bus-specific DMA memory unmapping functions.
 */
void
_bus_dmamem_unmap(bus_dma_tag_t t, caddr_t kva, size_t size)
{

#ifdef DIAGNOSTIC
	if ((u_long)kva & PGOFSET)
		panic("_bus_dmamem_unmap");
#endif

	size = round_page(size);

	pmap_remove(pmap_kernel(), (vaddr_t)kva, (vaddr_t)kva + size);
	pmap_update(pmap_kernel());
	uvm_km_free(kernel_map, (vaddr_t)kva, size, UVM_KMF_VAONLY);
}

/*
 * Common functin for mmap(2)'ing DMA-safe memory.  May be called by
 * bus-specific DMA mmap(2)'ing functions.
 */
paddr_t
_bus_dmamem_mmap(bus_dma_tag_t t, bus_dma_segment_t *segs, int nsegs, off_t off,
		 int prot, int flags)
{
	int i;

	for (i = 0; i < nsegs; i++) {
#ifdef DIAGNOSTIC
		if (off & PGOFSET)
			panic("_bus_dmamem_mmap: offset unaligned");
		if (segs[i].ds_addr & PGOFSET)
			panic("_bus_dmamem_mmap: segment unaligned");
		if (segs[i].ds_len & PGOFSET)
			panic("_bus_dmamem_mmap: segment size not multiple"
			    " of page size");
#endif
		if (off >= segs[i].ds_len) {
			off -= segs[i].ds_len;
			continue;
		}

		return (segs[i].ds_addr + off);
	}

	/* Page not found. */
	return (-1);
}


/*
 * "generic" DMA struct, nothing special.
 */
struct ibm4xx_bus_dma_tag ibm4xx_default_bus_dma_tag = {
	0,			/* _bounce_thresh */
	_bus_dmamap_create, 
	_bus_dmamap_destroy,
	_bus_dmamap_load,
	_bus_dmamap_load_mbuf,
	_bus_dmamap_load_uio,
	_bus_dmamap_load_raw,
	_bus_dmamap_unload,
	_bus_dmamap_sync,
	_bus_dmamem_alloc,
	_bus_dmamem_free,
	_bus_dmamem_map,
	_bus_dmamem_unmap,
	_bus_dmamem_mmap,
};