Re: [PATCH 03/18] MM: use ->swap_rw for reads from SWP_FS_OPS swap-space

[Christoph Hellwig <hch@infradead.org>]

> +int sio_pool_init(void)
> +{
> +	if (!sio_pool)
> +		sio_pool = mempool_create_kmalloc_pool(
> +			SWAP_CLUSTER_MAX, sizeof(struct swap_iocb));

I can't see anything serializing access here, so we'll need a lock or
cmpxchg dance.

> +	if (sio_pool)
> +		return 0;
> +	else
> +		return -ENOMEM;

Nit: This would flow much nicer as:

	if (!sio_pool)
		return -ENOMEM;
	return 0;

>  int swap_readpage(struct page *page, bool synchronous)
>  {
>  	struct bio *bio;
> @@ -378,13 +412,25 @@ int swap_readpage(struct page *page, bool synchronous)
>  	}
>  
>  	if (data_race(sis->flags & SWP_FS_OPS)) {
> -		//struct file *swap_file = sis->swap_file;
> -		//struct address_space *mapping = swap_file->f_mapping;

This should not be left by the previous patch.  In fact I suspect the
part of the previous patch that adds ->swap_rw should probably be folded
into this patch.

> +		struct file *swap_file = sis->swap_file;
> +		struct address_space *mapping = swap_file->f_mapping;
> +		struct iov_iter from;
> +		struct swap_iocb *sio;
> +		loff_t pos = page_file_offset(page);
> +
> +		sio = mempool_alloc(sio_pool, GFP_KERNEL);
> +		init_sync_kiocb(&sio->iocb, swap_file);
> +		sio->iocb.ki_pos = pos;
> +		sio->iocb.ki_complete = sio_read_complete;
> +		sio->bvec.bv_page = page;
> +		sio->bvec.bv_len = PAGE_SIZE;
> +		sio->bvec.bv_offset = 0;
> +
> +		iov_iter_bvec(&from, READ, &sio->bvec, 1, PAGE_SIZE);
> +		ret = mapping->a_ops->swap_rw(&sio->iocb, &from);
> +		if (ret != -EIOCBQUEUED)
> +			sio_read_complete(&sio->iocb, ret);
>  
>  		goto out;

I'd be tempted to split the SWP_FS_OPS into a helper to keep the
code tidy.