how to mmap in videobuf-dma-sg.c

hi,all,
I am puzzle that how to mmap ( V4L2_MEMORY_MMAP) in videobuf-dma-sg.c?

In this file, it alloc the momery using vmalloc_32() , and put this momery into sglist table,and then use dma_map_sg() to create sg dma at __videobuf_iolock() function. but in __videobuf_mmap_mapper(), i canot understand how it do the mmap?
why it not use the remap_vmalloc_range() to do the mmap?

Thanks ,

Em Thu, 21 May 2009 12:46:04 +0800
"Figo.zhang" <[email protected]> escreveu:

The answer is simple: remap_vmalloc_range() is newer than videobuf code. This part of the code was written back to kernel 2.4, and nobody cared to update it to use those newer functions, and simplify its code.

If you want, feel free to propose some cleanups on it

Cheers,
Mauro

hi mauro,
Thank you!
But i canot found the similar function code of remap_vmalloc_range() in the videobuf-dma-contig.c file. So i want to know the how is work in __videobuf_mmap_mapper() function?

hi mauro:
Thank you. But i still have a puzzle question about mmap() in videobuf-dma-sg.c. I canot find how to remap the dma buffer (which alloc by vmalloc_32()) to the vma area in __videobuf_mmap_mapper()?

there is my test driver code about dma-sg,it work well. i use remap_pfn_range() to remap the dma buffer to vma area in mmap method.

so would you like to give me some detail about it?

Best Regards,

Figo.zhang

static int mydev_alloc_dma_sg(struct mydev_device *dev, struct mydev_buf *buf)

{
	int nr_pages;
	
	int i;
	struct page *pg;
	unsigned char * virt;

	nr_pages = mydev_buffer_pages(buf->size);

	buf->nr_pages = nr_pages;

	dprintk("%s:: buf->nr_pages =%d\n", __func__, buf->nr_pages);
	
	buf->sglist = kcalloc(buf->nr_pages, sizeof(struct scatterlist),
GFP_KERNEL);
	if (NULL == buf->sglist)
		return NULL;
	
	sg_init_table(buf->sglist, buf->nr_pages);

	buf->vmalloc = vmalloc_32(buf->nr_pages << PAGE_SHIFT);

	memset(buf->vmalloc,0,buf->nr_pages << PAGE_SHIFT);

	virt = buf->vmalloc;
	
	for(i = 0; i< buf->nr_pages; i++,virt += PAGE_SIZE){
		pg = vmalloc_to_page(virt);
		if (NULL == pg)
			goto nopage;
		BUG_ON(PageHighMem(pg));
		sg_set_page(&buf->sglist[i], pg, PAGE_SIZE, 0);
		
		}

	buf->sglen = dma_map_sg(&dev->pci->dev, buf->sglist,
					buf->nr_pages, DMA_FROM_DEVICE);

	return 0;

nopage:

	dprintk("sgl: oops - no page\n");
	kfree(buf->sglist);
	return 0;
}

in mmap();

static int do_mmap_sg(struct mydev_device *dev, struct vm_area_struct * vma)

{
	struct videobuf_mapping *map;
	unsigned long pos,page;
	unsigned long start = vma->vm_start;
	unsigned long size = vma->vm_end - vma->vm_start;
	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
	unsigned int first, last, end;

	int retval = -EINVAL;
	int i;

	/* look for first buffer to map */
	for (first = 0; first < VIDEO_MAX_FRAME; first++) {
	
		if (dev->ktbuf[first].boff == (vma->vm_pgoff << PAGE_SHIFT))
			break;
	}
	if (VIDEO_MAX_FRAME == first) {
		dprintk("mmap app bug: offset invalid [offset=0x%lx]\n",
			(vma->vm_pgoff << PAGE_SHIFT));
		goto done;
	}
	/* create mapping + update buffer list */
	retval = -ENOMEM;
	map = kmalloc(sizeof(struct videobuf_mapping),GFP_KERNEL);
	if (NULL == map)
		goto done;

		pos = (unsigned long)dev->ktbuf[first].vmalloc;

	while (size > 0) {
		page = vmalloc_to_pfn((void *)pos);
		if (remap_pfn_range(vma, start, page, PAGE_SIZE, PAGE_SHARED)) {
			return -EAGAIN;
		}
		start += PAGE_SIZE;
		pos += PAGE_SIZE;
		if (size > PAGE_SIZE)
			size -= PAGE_SIZE;
		else
			size = 0;
	}

	map->count    = 1;
	map->start    = vma->vm_start;
	map->end      = vma->vm_end;
	vma->vm_ops   = &mydev_vm_ops;
	vma->vm_flags |=/* VM_DONTEXPAND |*/ VM_RESERVED;
vma->vm_flags &= ~VM_IO; /* using shared anonymous pages */
	vma->vm_private_data = map;

	mydev_vm_open(vma);
	retval = 0;

done:

	return retval;
}

thanks, in __videobuf_mmap_mapper(), it define a videobuf_vm_ops->fault, it
will alloc a new page for mmaping when it encounter page fault (do_page_fault),
so how the mmap() can mmap the vmalloc memory which had allocted before using __videobuf_iolock()/vmalloc_
32() ?

Thanks,
Figo.zhang

Em 22-07-2010 23:31, figo zhang escreveu:

Sorry for not answering earlier.

The current videobuf implementation has some problems. Laurent and Pawel are working on a new implementation that will likely solve such issues. Not sure if they already submitted the patches, since I just return back from vacations, and I'm still trying to handle all the backlogs on my inboxes. I haven't look at LMML posts yet.

Cheers,
Mauro.

Hi Mauro,

On Sunday 25 July 2010 19:46:49 Mauro Carvalho Chehab wrote:

[snip]

No patches submitted yet. It will probably still be weeks before we can submit something.

Regards,

Laurent Pinchart