Module Name: src
Committed By: prlw1
Date: Mon Jan 7 15:07:41 UTC 2013
Modified Files:
src/sys/dev: video.c
src/sys/dev/usb: ehci.c usb_mem.c usb_mem.h
Log Message:
Allow USB memory allocation by multiple segments in scatter/gather lists
rather than in a single contiguous block which causes problems with
large USB video frames.
Based on a patch by Jeremy Morse in the thread
http://mail-index.netbsd.org/current-users/2011/01/26/msg015532.html
Tested by developing http://code.opencv.org/issues/2360
OK jmcneill@
To generate a diff of this commit:
cvs rdiff -u -r1.28 -r1.29 src/sys/dev/video.c
cvs rdiff -u -r1.196 -r1.197 src/sys/dev/usb/ehci.c
cvs rdiff -u -r1.54 -r1.55 src/sys/dev/usb/usb_mem.c
cvs rdiff -u -r1.28 -r1.29 src/sys/dev/usb/usb_mem.h
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/dev/video.c
diff -u src/sys/dev/video.c:1.28 src/sys/dev/video.c:1.29
--- src/sys/dev/video.c:1.28 Thu Feb 2 17:21:18 2012
+++ src/sys/dev/video.c Mon Jan 7 15:07:40 2013
@@ -1,4 +1,4 @@
-/* $NetBSD: video.c,v 1.28 2012/02/02 17:21:18 drochner Exp $ */
+/* $NetBSD: video.c,v 1.29 2013/01/07 15:07:40 prlw1 Exp $ */
/*
* Copyright (c) 2008 Patrick Mahoney <[email protected]>
@@ -36,7 +36,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: video.c,v 1.28 2012/02/02 17:21:18 drochner Exp $");
+__KERNEL_RCSID(0, "$NetBSD: video.c,v 1.29 2013/01/07 15:07:40 prlw1 Exp $");
#include "video.h"
#if NVIDEO > 0
@@ -1604,7 +1604,8 @@ videoopen(dev_t dev, int flags, int ifmt
sc = device_private(device_lookup(&video_cd, VIDEOUNIT(dev)));
if (sc == NULL) {
- DPRINTF(("videoopen: failed to get softc\n"));
+ DPRINTF(("videoopen: failed to get softc for unit %d\n",
+ VIDEOUNIT(dev)));
return ENXIO;
}
Index: src/sys/dev/usb/ehci.c
diff -u src/sys/dev/usb/ehci.c:1.196 src/sys/dev/usb/ehci.c:1.197
--- src/sys/dev/usb/ehci.c:1.196 Sat Jan 5 23:34:16 2013
+++ src/sys/dev/usb/ehci.c Mon Jan 7 15:07:40 2013
@@ -1,4 +1,4 @@
-/* $NetBSD: ehci.c,v 1.196 2013/01/05 23:34:16 christos Exp $ */
+/* $NetBSD: ehci.c,v 1.197 2013/01/07 15:07:40 prlw1 Exp $ */
/*
* Copyright (c) 2004-2012 The NetBSD Foundation, Inc.
@@ -53,7 +53,7 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: ehci.c,v 1.196 2013/01/05 23:34:16 christos Exp $");
+__KERNEL_RCSID(0, "$NetBSD: ehci.c,v 1.197 2013/01/07 15:07:40 prlw1 Exp $");
#include "ohci.h"
#include "uhci.h"
@@ -1336,12 +1336,18 @@ ehci_allocm(struct usbd_bus *bus, usb_dm
struct ehci_softc *sc = bus->hci_private;
usbd_status err;
- err = usb_allocmem(&sc->sc_bus, size, 0, dma);
+ err = usb_allocmem_flags(&sc->sc_bus, size, 0, dma, USBMALLOC_MULTISEG);
+#ifdef EHCI_DEBUG
+ if (err)
+ printf("ehci_allocm: usb_allocmem_flags()= %s (%d)\n",
+ usbd_errstr(err), err);
+#endif
if (err == USBD_NOMEM)
err = usb_reserve_allocm(&sc->sc_dma_reserve, dma, size);
#ifdef EHCI_DEBUG
if (err)
- printf("ehci_allocm: usb_allocmem()=%d\n", err);
+ printf("ehci_allocm: usb_reserve_allocm()= %s (%d)\n",
+ usbd_errstr(err), err);
#endif
return (err);
}
@@ -2727,18 +2733,20 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
ehci_soft_qtd_t **sp, ehci_soft_qtd_t **ep)
{
ehci_soft_qtd_t *next, *cur;
- ehci_physaddr_t dataphys, dataphyspage, dataphyslastpage, nextphys;
+ ehci_physaddr_t nextphys;
u_int32_t qtdstatus;
int len, curlen, mps;
int i, tog;
+ int pages, pageoffs;
+ bus_size_t curoffs;
+ vaddr_t va, va_offs;
usb_dma_t *dma = &xfer->dmabuf;
u_int16_t flags = xfer->flags;
+ paddr_t a;
DPRINTFN(alen<4*4096,("ehci_alloc_sqtd_chain: start len=%d\n", alen));
len = alen;
- dataphys = DMAADDR(dma, 0);
- dataphyslastpage = EHCI_PAGE(dataphys + len - 1);
qtdstatus = EHCI_QTD_ACTIVE |
EHCI_QTD_SET_PID(rd ? EHCI_QTD_PID_IN : EHCI_QTD_PID_OUT) |
EHCI_QTD_SET_CERR(3)
@@ -2756,28 +2764,18 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
usb_syncmem(dma, 0, alen,
rd ? BUS_DMASYNC_PREREAD : BUS_DMASYNC_PREWRITE);
+ curoffs = 0;
for (;;) {
- dataphyspage = EHCI_PAGE(dataphys);
/* The EHCI hardware can handle at most 5 pages. */
- if (dataphyslastpage - dataphyspage <
- EHCI_QTD_NBUFFERS * EHCI_PAGE_SIZE) {
+ va_offs = (vaddr_t)KERNADDR(dma, curoffs);
+ va_offs = EHCI_PAGE_OFFSET(va_offs);
+ if (len-curoffs < EHCI_QTD_NBUFFERS*EHCI_PAGE_SIZE - va_offs) {
/* we can handle it in this QTD */
- curlen = len;
+ curlen = len - curoffs;
} else {
/* must use multiple TDs, fill as much as possible. */
- curlen = EHCI_QTD_NBUFFERS * EHCI_PAGE_SIZE -
- EHCI_PAGE_OFFSET(dataphys);
-#ifdef DIAGNOSTIC
- if (curlen > len) {
- printf("ehci_alloc_sqtd_chain: curlen=0x%x "
- "len=0x%x offs=0x%x\n", curlen, len,
- EHCI_PAGE_OFFSET(dataphys));
- printf("lastpage=0x%x page=0x%x phys=0x%x\n",
- dataphyslastpage, dataphyspage,
- dataphys);
- curlen = len;
- }
-#endif
+ curlen = EHCI_QTD_NBUFFERS * EHCI_PAGE_SIZE - va_offs;
+
/* the length must be a multiple of the max size */
curlen -= curlen % mps;
DPRINTFN(1,("ehci_alloc_sqtd_chain: multiple QTDs, "
@@ -2787,18 +2785,16 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
panic("ehci_alloc_sqtd_chain: curlen == 0");
#endif
}
- DPRINTFN(4,("ehci_alloc_sqtd_chain: dataphys=0x%08x "
- "dataphyslastpage=0x%08x len=%d curlen=%d\n",
- dataphys, dataphyslastpage,
- len, curlen));
- len -= curlen;
+ DPRINTFN(4,("ehci_alloc_sqtd_chain: len=%d curlen=%d "
+ "curoffs=%d\n", len, curlen, curoffs));
/*
* Allocate another transfer if there's more data left,
* or if force last short transfer flag is set and we're
* allocating a multiple of the max packet size.
*/
- if (len != 0 ||
+
+ if (curoffs + curlen != len ||
((curlen % mps) == 0 && !rd && curlen != 0 &&
(flags & USBD_FORCE_SHORT_XFER))) {
next = ehci_alloc_sqtd(sc);
@@ -2810,20 +2806,21 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
nextphys = EHCI_NULL;
}
- for (i = 0; i * EHCI_PAGE_SIZE <
- curlen + EHCI_PAGE_OFFSET(dataphys); i++) {
- ehci_physaddr_t a = dataphys + i * EHCI_PAGE_SIZE;
- if (i != 0) /* use offset only in first buffer */
- a = EHCI_PAGE(a);
- if (i >= EHCI_QTD_NBUFFERS) {
-#ifdef DIAGNOSTIC
- printf("ehci_alloc_sqtd_chain: i=%d\n", i);
-#endif
- goto nomem;
- }
- cur->qtd.qtd_buffer[i] = htole32(a);
- cur->qtd.qtd_buffer_hi[i] = 0;
+ /* Find number of pages we'll be using, insert dma addresses */
+ pages = EHCI_PAGE(curlen + EHCI_PAGE_SIZE -1) >> 12;
+ KASSERT(pages <= EHCI_QTD_NBUFFERS);
+ pageoffs = EHCI_PAGE(curoffs);
+ for (i = 0; i < pages; i++) {
+ a = DMAADDR(dma, pageoffs + i * EHCI_PAGE_SIZE);
+ cur->qtd.qtd_buffer[i] = htole32(a & 0xFFFFF000);
+ /* Cast up to avoid compiler warnings */
+ cur->qtd.qtd_buffer_hi[i] = htole32((uint64_t)a >> 32);
}
+
+ /* First buffer pointer requires a page offset to start at */
+ va = (vaddr_t)KERNADDR(dma, curoffs);
+ cur->qtd.qtd_buffer[0] |= htole32(EHCI_PAGE_OFFSET(va));
+
cur->nextqtd = next;
cur->qtd.qtd_next = cur->qtd.qtd_altnext = nextphys;
cur->qtd.qtd_status =
@@ -2832,7 +2829,8 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
cur->len = curlen;
DPRINTFN(10,("ehci_alloc_sqtd_chain: cbp=0x%08x end=0x%08x\n",
- dataphys, dataphys + curlen));
+ curoffs, curoffs + curlen));
+
/* adjust the toggle based on the number of packets in this
qtd */
if (((curlen + mps - 1) / mps) & 1) {
@@ -2845,7 +2843,7 @@ ehci_alloc_sqtd_chain(struct ehci_pipe *
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
DPRINTFN(10,("ehci_alloc_sqtd_chain: extend chain\n"));
if (len)
- dataphys += curlen;
+ curoffs += curlen;
cur = next;
}
cur->qtd.qtd_status |= htole32(EHCI_QTD_IOC);
Index: src/sys/dev/usb/usb_mem.c
diff -u src/sys/dev/usb/usb_mem.c:1.54 src/sys/dev/usb/usb_mem.c:1.55
--- src/sys/dev/usb/usb_mem.c:1.54 Sat Jan 5 23:34:20 2013
+++ src/sys/dev/usb/usb_mem.c Mon Jan 7 15:07:41 2013
@@ -1,4 +1,4 @@
-/* $NetBSD: usb_mem.c,v 1.54 2013/01/05 23:34:20 christos Exp $ */
+/* $NetBSD: usb_mem.c,v 1.55 2013/01/07 15:07:41 prlw1 Exp $ */
/*
* Copyright (c) 1998 The NetBSD Foundation, Inc.
@@ -38,12 +38,12 @@
*/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: usb_mem.c,v 1.54 2013/01/05 23:34:20 christos Exp $");
+__KERNEL_RCSID(0, "$NetBSD: usb_mem.c,v 1.55 2013/01/07 15:07:41 prlw1 Exp $");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
-#include <sys/malloc.h>
+#include <sys/kmem.h>
#include <sys/queue.h>
#include <sys/device.h> /* for usbdivar.h */
#include <sys/bus.h>
@@ -82,7 +82,7 @@ struct usb_frag_dma {
};
Static usbd_status usb_block_allocmem(bus_dma_tag_t, size_t, size_t,
- usb_dma_block_t **);
+ usb_dma_block_t **, bool);
Static void usb_block_freemem(usb_dma_block_t *);
LIST_HEAD(usb_dma_block_qh, usb_dma_block);
@@ -113,13 +113,20 @@ usb_mem_init(void)
Static usbd_status
usb_block_allocmem(bus_dma_tag_t tag, size_t size, size_t align,
- usb_dma_block_t **dmap)
+ usb_dma_block_t **dmap, bool multiseg)
{
usb_dma_block_t *b;
int error;
DPRINTFN(5, ("usb_block_allocmem: size=%zu align=%zu\n", size, align));
+ if (size == 0) {
+#ifdef DIAGNOSTIC
+ printf("usb_block_allocmem: called with size==0\n");
+#endif
+ return USBD_INVAL;
+ }
+
#ifdef DIAGNOSTIC
if (cpu_intr_p()) {
printf("usb_block_allocmem: in interrupt context, size=%lu\n",
@@ -131,6 +138,9 @@ usb_block_allocmem(bus_dma_tag_t tag, si
/* First check the free list. */
LIST_FOREACH(b, &usb_blk_freelist, next) {
+ /* Don't allocate multiple segments to unwilling callers */
+ if (b->nsegs != 1 && !multiseg)
+ continue;
if (b->tag == tag && b->size >= size && b->align >= align) {
LIST_REMOVE(b, next);
usb_blk_nfree--;
@@ -149,15 +159,27 @@ usb_block_allocmem(bus_dma_tag_t tag, si
#endif
DPRINTFN(6, ("usb_block_allocmem: no free\n"));
- b = malloc(sizeof *b, M_USB, M_NOWAIT | M_ZERO);
+ b = kmem_zalloc(sizeof *b, KM_SLEEP);
if (b == NULL)
return (USBD_NOMEM);
b->tag = tag;
b->size = size;
b->align = align;
+
+ b->nsegs = (size + (PAGE_SIZE-1)) / PAGE_SIZE;
+ if (!multiseg)
+ /* Caller wants one segment */
+ b->nsegs = 1;
+
+ b->segs = kmem_alloc(b->nsegs * sizeof(*b->segs), KM_SLEEP);
+ if (b->segs == NULL) {
+ kmem_free(b, sizeof *b);
+ return USBD_NOMEM;
+ }
+
error = bus_dmamem_alloc(tag, b->size, align, 0,
- b->segs, __arraycount(b->segs),
+ b->segs, b->nsegs,
&b->nsegs, BUS_DMA_NOWAIT);
if (error)
goto free0;
@@ -167,7 +189,7 @@ usb_block_allocmem(bus_dma_tag_t tag, si
if (error)
goto free1;
- error = bus_dmamap_create(tag, b->size, 1, b->size,
+ error = bus_dmamap_create(tag, b->size, b->nsegs, b->size,
0, BUS_DMA_NOWAIT, &b->map);
if (error)
goto unmap;
@@ -181,6 +203,7 @@ usb_block_allocmem(bus_dma_tag_t tag, si
#ifdef USB_FRAG_DMA_WORKAROUND
memset(b->kaddr, 0, b->size);
#endif
+
return (USBD_NORMAL_COMPLETION);
destroy:
@@ -190,7 +213,8 @@ usb_block_allocmem(bus_dma_tag_t tag, si
free1:
bus_dmamem_free(tag, b->segs, b->nsegs);
free0:
- free(b, M_USB);
+ kmem_free(b->segs, b->nsegs * sizeof(*b->segs));
+ kmem_free(b, sizeof *b);
return (USBD_NOMEM);
}
@@ -208,7 +232,8 @@ usb_block_real_freemem(usb_dma_block_t *
bus_dmamap_destroy(b->tag, b->map);
bus_dmamem_unmap(b->tag, b->kaddr, b->size);
bus_dmamem_free(b->tag, b->segs, b->nsegs);
- free(p, M_USB);
+ kmem_free(b->segs, b->nsegs * sizeof(*b->segs));
+ kmem_free(b, sizeof *b);
}
#endif
@@ -247,21 +272,31 @@ usb_block_freemem(usb_dma_block_t *b)
usbd_status
usb_allocmem(usbd_bus_handle bus, size_t size, size_t align, usb_dma_t *p)
{
+ return usb_allocmem_flags(bus, size, align, p, 0);
+}
+
+usbd_status
+usb_allocmem_flags(usbd_bus_handle bus, size_t size, size_t align, usb_dma_t *p,
+ int flags)
+{
bus_dma_tag_t tag = bus->dmatag;
usbd_status err;
struct usb_frag_dma *f;
usb_dma_block_t *b;
int i;
static ONCE_DECL(init_control);
+ bool frag;
RUN_ONCE(&init_control, usb_mem_init);
+ frag = (flags & USBMALLOC_MULTISEG);
+
/* If the request is large then just use a full block. */
if (size > USB_MEM_SMALL || align > USB_MEM_SMALL) {
DPRINTFN(1, ("usb_allocmem: large alloc %d\n", (int)size));
size = (size + USB_MEM_BLOCK - 1) & ~(USB_MEM_BLOCK - 1);
mutex_enter(&usb_blk_lock);
- err = usb_block_allocmem(tag, size, align, &p->block);
+ err = usb_block_allocmem(tag, size, align, &p->block, frag);
if (!err) {
#ifdef DEBUG
LIST_INSERT_HEAD(&usb_blk_fulllist, p->block, next);
@@ -284,7 +319,8 @@ usb_allocmem(usbd_bus_handle bus, size_t
}
if (f == NULL) {
DPRINTFN(1, ("usb_allocmem: adding fragments\n"));
- err = usb_block_allocmem(tag, USB_MEM_BLOCK, USB_MEM_SMALL,&b);
+ err = usb_block_allocmem(tag, USB_MEM_BLOCK, USB_MEM_SMALL, &b,
+ false);
if (err) {
mutex_exit(&usb_blk_lock);
return (err);
@@ -313,6 +349,7 @@ usb_allocmem(usbd_bus_handle bus, size_t
LIST_REMOVE(f, next);
mutex_exit(&usb_blk_lock);
DPRINTFN(5, ("usb_allocmem: use frag=%p size=%d\n", f, (int)size));
+
return (USBD_NORMAL_COMPLETION);
}
@@ -349,6 +386,38 @@ usb_freemem(usbd_bus_handle bus, usb_dma
DPRINTFN(5, ("usb_freemem: frag=%p\n", f));
}
+bus_addr_t
+usb_dmaaddr(usb_dma_t *dma, unsigned int offset)
+{
+ unsigned int i;
+ bus_size_t seg_offs;
+
+ offset += dma->offs;
+
+ KASSERT(offset < dma->block->size);
+
+ if (dma->block->nsegs == 1) {
+ KASSERT(dma->block->map->dm_segs[0].ds_len > offset);
+ return dma->block->map->dm_segs[0].ds_addr + offset;
+ }
+
+ /* Search for a bus_segment_t corresponding to this offset. With no
+ * record of the offset in the map to a particular dma_segment_t, we
+ * have to iterate from the start of the list each time. Could be
+ * improved */
+ seg_offs = 0;
+ for (i = 0; i < dma->block->nsegs; i++) {
+ if (seg_offs + dma->block->map->dm_segs[i].ds_len > offset)
+ break;
+
+ seg_offs += dma->block->map->dm_segs[i].ds_len;
+ }
+
+ KASSERT(i != dma->block->nsegs);
+ offset -= seg_offs;
+ return dma->block->map->dm_segs[i].ds_addr + offset;
+}
+
void
usb_syncmem(usb_dma_t *p, bus_addr_t offset, bus_size_t len, int ops)
{
@@ -367,7 +436,7 @@ usb_reserve_allocm(struct usb_dma_reserv
if (rs->vaddr == 0 || size > USB_MEM_RESERVE)
return USBD_NOMEM;
- dma->block = malloc(sizeof *dma->block, M_USB, M_ZERO | M_NOWAIT);
+ dma->block = kmem_zalloc(sizeof *dma->block, KM_SLEEP);
if (dma->block == NULL)
return USBD_NOMEM;
@@ -403,7 +472,8 @@ usb_reserve_freem(struct usb_dma_reserve
error = extent_free(rs->extent,
(u_long)(rs->paddr + dma->offs), dma->block->size, 0);
- free(dma->block, M_USB);
+ /* XXXPW correct that segs[0] is not used? */
+ kmem_free(dma->block, dma->block->size);
}
int
Index: src/sys/dev/usb/usb_mem.h
diff -u src/sys/dev/usb/usb_mem.h:1.28 src/sys/dev/usb/usb_mem.h:1.29
--- src/sys/dev/usb/usb_mem.h:1.28 Fri Feb 24 06:48:28 2012
+++ src/sys/dev/usb/usb_mem.h Mon Jan 7 15:07:41 2013
@@ -1,4 +1,4 @@
-/* $NetBSD: usb_mem.h,v 1.28 2012/02/24 06:48:28 mrg Exp $ */
+/* $NetBSD: usb_mem.h,v 1.29 2013/01/07 15:07:41 prlw1 Exp $ */
/* $FreeBSD: src/sys/dev/usb/usb_mem.h,v 1.9 1999/11/17 22:33:47 n_hibma Exp $ */
/*
@@ -35,7 +35,7 @@ typedef struct usb_dma_block {
bus_dma_tag_t tag;
bus_dmamap_t map;
void *kaddr;
- bus_dma_segment_t segs[1];
+ bus_dma_segment_t *segs;
int nsegs;
size_t size;
size_t align;
@@ -45,14 +45,20 @@ typedef struct usb_dma_block {
LIST_ENTRY(usb_dma_block) next;
} usb_dma_block_t;
-#define DMAADDR(dma, o) ((dma)->block->map->dm_segs[0].ds_addr + (dma)->offs + (o))
-#define KERNADDR(dma, o) \
- ((void *)((char *)(dma)->block->kaddr + (dma)->offs + (o)))
+#define USBMALLOC_MULTISEG 1
-usbd_status usb_allocmem(usbd_bus_handle,size_t,size_t, usb_dma_t *);
+usbd_status usb_allocmem(usbd_bus_handle, size_t, size_t, usb_dma_t *);
+usbd_status usb_allocmem_flags(usbd_bus_handle, size_t, size_t, usb_dma_t *,
+ int);
void usb_freemem(usbd_bus_handle, usb_dma_t *);
void usb_syncmem(usb_dma_t *, bus_addr_t, bus_size_t, int ops);
+bus_addr_t usb_dmaaddr(usb_dma_t *, unsigned int);
+
+#define DMAADDR(dma, o) usb_dmaaddr((dma), (o))
+#define KERNADDR(dma, o) \
+ ((void *)((char *)(dma)->block->kaddr + (dma)->offs + (o)))
+
struct extent;
struct usb_dma_reserve {
