* Re: [PATCH v2]Resend:videobuf_dma_sg: a new implementation for mmap
[not found] ` <201007301131.17638.laurent.pinchart@ideasonboard.com>
@ 2010-07-30 11:38 ` Figo.zhang
2010-07-30 11:39 ` Figo.zhang
1 sibling, 0 replies; 2+ messages in thread
From: Figo.zhang @ 2010-07-30 11:38 UTC (permalink / raw)
To: npiggin, Mauro Carvalho Chehab, akpm
Cc: linux-media, Laurent Pinchart, lkml, Ingo Molnar
On Fri, 2010-07-30 at 11:31 +0200, Laurent Pinchart wrote:
> Hi,
>
> On Friday 30 July 2010 02:08:02 Figo.zhang wrote:
> > a mmap issue for videobuf-dma-sg: it will alloc a new page for mmaping when
> > it encounter page fault at video_vm_ops->fault(). pls see
> > http://www.spinics.net/lists/linux-media/msg21243.html
> >
> > a new implementation for mmap, it translate to vmalloc to page at
> > video_vm_ops->fault().
> >
> > in v2, if mem->dma.vmalloc is NULL at video_vm_ops->fault(), it will alloc
> > memory by vmlloc_32().
>
> You're replacing allocation in videobuf_vm_fault by allocationg in
> videobuf_vm_fault. I don't see the point. videobuf_vm_fault needs to go away
> completely.
in now videobuf code, the mmap alloc a new buf, the capture dma buffer
using vmalloc() alloc buffer, how is
relationship with them? in usrspace , the mmap region will not the
actual capture dma data, how is work?
>
> This has been discussed previously: fixing videobuf is not really possible. A
> videobuf2 implementation is needed and is (slowly) being worked on.
hi Mauro, do you think this is a issue for videobuf-dma-sg?
> I wouldn't
> bother with this patch, just drop it.
>
> > Signed-off-by: Figo.zhang <figo1802@gmail.com>
> > ---
> > drivers/media/video/videobuf-dma-sg.c | 50
> > +++++++++++++++++++++++++++------ 1 files changed, 41 insertions(+), 9
> > deletions(-)
> >
> > diff --git a/drivers/media/video/videobuf-dma-sg.c
> > b/drivers/media/video/videobuf-dma-sg.c index 8359e6b..f7295da 100644
> > --- a/drivers/media/video/videobuf-dma-sg.c
> > +++ b/drivers/media/video/videobuf-dma-sg.c
> > @@ -201,10 +201,11 @@ int videobuf_dma_init_kernel(struct videobuf_dmabuf
> > *dma, int direction, dprintk(1, "init kernel [%d pages]\n", nr_pages);
> >
> > dma->direction = direction;
> > - dma->vmalloc = vmalloc_32(nr_pages << PAGE_SHIFT);
> > - if (NULL == dma->vmalloc) {
> > - dprintk(1, "vmalloc_32(%d pages) failed\n", nr_pages);
> > - return -ENOMEM;
> > + if (!dma->vmalloc)
> > + dma->vmalloc = vmalloc_32(nr_pages << PAGE_SHIFT);
> > + if (NULL == dma->vmalloc) {
> > + dprintk(1, "vmalloc_32(%d pages) failed\n", nr_pages);
> > + return -ENOMEM;
> > }
> >
> > dprintk(1, "vmalloc is at addr 0x%08lx, size=%d\n",
> > @@ -397,16 +398,47 @@ static void videobuf_vm_close(struct vm_area_struct
> > *vma) */
> > static int videobuf_vm_fault(struct vm_area_struct *vma, struct vm_fault
> > *vmf) {
> > - struct page *page;
> > + struct page *page = NULL;
> > + struct videobuf_mapping *map = vma->vm_private_data;
> > + struct videobuf_queue *q = map->q;
> > + struct videobuf_dma_sg_memory *mem = NULL;
> > +
> > + unsigned long offset;
> > + unsigned long page_nr;
> > + int first;
> >
> > dprintk(3, "fault: fault @ %08lx [vma %08lx-%08lx]\n",
> > (unsigned long)vmf->virtual_address,
> > vma->vm_start, vma->vm_end);
> >
> > - page = alloc_page(GFP_USER | __GFP_DMA32);
> > - if (!page)
> > - return VM_FAULT_OOM;
> > - clear_user_highpage(page, (unsigned long)vmf->virtual_address);
> > + mutex_lock(&q->vb_lock);
> > + offset = (unsigned long)vmf->virtual_address - vma->vm_start;
> > + page_nr = offset >> PAGE_SHIFT;
> > +
> > + for (first = 0; first < VIDEO_MAX_FRAME; first++) {
> > + if (NULL == q->bufs[first])
> > + continue;
> > +
> > + MAGIC_CHECK(mem->magic, MAGIC_SG_MEM);
> > +
> > + if (q->bufs[first]->map == map)
> > + break;
> > + }
> > +
> > + mem = q->bufs[first]->priv;
> > + if (!mem)
> > + return VM_FAULT_SIGBUS;
> > + if (!mem->dma.vmalloc) {
> > + mem->dma.vmalloc = vmalloc_32(PAGE_ALIGN(q->bufs[first]->size));
> > + if (NULL == mem->dma.vmalloc) {
> > + dprintk(1, "%s: vmalloc_32() failed\n", __func__);
> > + return VM_FAULT_OOM;
> > + }
> > + } else
> > + page = vmalloc_to_page(mem->dma.vmalloc+
> > + (offset & (~PAGE_MASK)));
> > + mutex_unlock(&q->vb_lock);
> > +
> > vmf->page = page;
> >
> > return 0;
>
^ permalink raw reply [flat|nested] 2+ messages in thread* Re: [PATCH v2]Resend:videobuf_dma_sg: a new implementation for mmap
[not found] ` <201007301131.17638.laurent.pinchart@ideasonboard.com>
2010-07-30 11:38 ` [PATCH v2]Resend:videobuf_dma_sg: a new implementation for mmap Figo.zhang
@ 2010-07-30 11:39 ` Figo.zhang
1 sibling, 0 replies; 2+ messages in thread
From: Figo.zhang @ 2010-07-30 11:39 UTC (permalink / raw)
To: npiggin, Mauro Carvalho Chehab, akpm, Greg Kroah-Hartman
Cc: linux-media, Laurent Pinchart, lkml, Ingo Molnar
On Fri, 2010-07-30 at 11:31 +0200, Laurent Pinchart wrote:
> Hi,
>
> On Friday 30 July 2010 02:08:02 Figo.zhang wrote:
> > a mmap issue for videobuf-dma-sg: it will alloc a new page for mmaping when
> > it encounter page fault at video_vm_ops->fault(). pls see
> > http://www.spinics.net/lists/linux-media/msg21243.html
> >
> > a new implementation for mmap, it translate to vmalloc to page at
> > video_vm_ops->fault().
> >
> > in v2, if mem->dma.vmalloc is NULL at video_vm_ops->fault(), it will alloc
> > memory by vmlloc_32().
>
> You're replacing allocation in videobuf_vm_fault by allocationg in
> videobuf_vm_fault. I don't see the point. videobuf_vm_fault needs to go away
> completely.
in now videobuf code, the mmap alloc a new buf, the capture dma buffer
using vmalloc() alloc buffer, how is
relationship with them? in usrspace , the mmap region will not the
actual capture dma data, how is work?
>
> This has been discussed previously: fixing videobuf is not really possible. A
> videobuf2 implementation is needed and is (slowly) being worked on.
hi Mauro, do you think this is a issue for videobuf-dma-sg?
> I wouldn't
> bother with this patch, just drop it.
>
> > Signed-off-by: Figo.zhang <figo1802@gmail.com>
> > ---
> > drivers/media/video/videobuf-dma-sg.c | 50
> > +++++++++++++++++++++++++++------ 1 files changed, 41 insertions(+), 9
> > deletions(-)
> >
> > diff --git a/drivers/media/video/videobuf-dma-sg.c
> > b/drivers/media/video/videobuf-dma-sg.c index 8359e6b..f7295da 100644
> > --- a/drivers/media/video/videobuf-dma-sg.c
> > +++ b/drivers/media/video/videobuf-dma-sg.c
> > @@ -201,10 +201,11 @@ int videobuf_dma_init_kernel(struct videobuf_dmabuf
> > *dma, int direction, dprintk(1, "init kernel [%d pages]\n", nr_pages);
> >
> > dma->direction = direction;
> > - dma->vmalloc = vmalloc_32(nr_pages << PAGE_SHIFT);
> > - if (NULL == dma->vmalloc) {
> > - dprintk(1, "vmalloc_32(%d pages) failed\n", nr_pages);
> > - return -ENOMEM;
> > + if (!dma->vmalloc)
> > + dma->vmalloc = vmalloc_32(nr_pages << PAGE_SHIFT);
> > + if (NULL == dma->vmalloc) {
> > + dprintk(1, "vmalloc_32(%d pages) failed\n", nr_pages);
> > + return -ENOMEM;
> > }
> >
> > dprintk(1, "vmalloc is at addr 0x%08lx, size=%d\n",
> > @@ -397,16 +398,47 @@ static void videobuf_vm_close(struct vm_area_struct
> > *vma) */
> > static int videobuf_vm_fault(struct vm_area_struct *vma, struct vm_fault
> > *vmf) {
> > - struct page *page;
> > + struct page *page = NULL;
> > + struct videobuf_mapping *map = vma->vm_private_data;
> > + struct videobuf_queue *q = map->q;
> > + struct videobuf_dma_sg_memory *mem = NULL;
> > +
> > + unsigned long offset;
> > + unsigned long page_nr;
> > + int first;
> >
> > dprintk(3, "fault: fault @ %08lx [vma %08lx-%08lx]\n",
> > (unsigned long)vmf->virtual_address,
> > vma->vm_start, vma->vm_end);
> >
> > - page = alloc_page(GFP_USER | __GFP_DMA32);
> > - if (!page)
> > - return VM_FAULT_OOM;
> > - clear_user_highpage(page, (unsigned long)vmf->virtual_address);
> > + mutex_lock(&q->vb_lock);
> > + offset = (unsigned long)vmf->virtual_address - vma->vm_start;
> > + page_nr = offset >> PAGE_SHIFT;
> > +
> > + for (first = 0; first < VIDEO_MAX_FRAME; first++) {
> > + if (NULL == q->bufs[first])
> > + continue;
> > +
> > + MAGIC_CHECK(mem->magic, MAGIC_SG_MEM);
> > +
> > + if (q->bufs[first]->map == map)
> > + break;
> > + }
> > +
> > + mem = q->bufs[first]->priv;
> > + if (!mem)
> > + return VM_FAULT_SIGBUS;
> > + if (!mem->dma.vmalloc) {
> > + mem->dma.vmalloc = vmalloc_32(PAGE_ALIGN(q->bufs[first]->size));
> > + if (NULL == mem->dma.vmalloc) {
> > + dprintk(1, "%s: vmalloc_32() failed\n", __func__);
> > + return VM_FAULT_OOM;
> > + }
> > + } else
> > + page = vmalloc_to_page(mem->dma.vmalloc+
> > + (offset & (~PAGE_MASK)));
> > + mutex_unlock(&q->vb_lock);
> > +
> > vmf->page = page;
> >
> > return 0;
>
^ permalink raw reply [flat|nested] 2+ messages in thread