[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] Re: [RFC PATCH 33/33] Add Xen virtual block device driver.



On Tue, 2006-07-18 at 00:00 -0700, Chris Wright wrote:
> plain text document attachment (blkfront)
> The block device frontend driver allows the kernel to access block
> devices exported exported by a virtual machine containing a physical
> block device driver.

Hi,

as first general comment, I think that some of the memory allocation
GFP_ flags are possibly incorrect; I would expect several places to use
GFP_NOIO rather than GFP_KERNEL, to avoid recursion/deadlocks

> +static void blkif_recover(struct blkfront_info *info)
> +{
> +     int i;
> +     struct blkif_request *req;
> +     struct blk_shadow *copy;
> +     int j;
> +
> +     /* Stage 1: Make a safe copy of the shadow state. */
> +     copy = kmalloc(sizeof(info->shadow), GFP_KERNEL | __GFP_NOFAIL);

like here..

> +     memcpy(copy, info->shadow, sizeof(info->shadow));

and __GFP_NOFAIL is usually horrid; is this because error recovery was
an afterthought, or because it's physically impossible? In addition
__GFP_NOFAIL in a block device driver is... an interesting way to add
OOM deadlocks... have the VM guys looked into this yet?

> +#if 1
> +#define IPRINTK(fmt, args...) \
> +    printk(KERN_INFO "xen_blk: " fmt, ##args)
> +#else
> +#define IPRINTK(fmt, args...) ((void)0)
> +#endif

hmm isn't this a duplication of the pr_debug() and dev_dbg()
infrastructure? Please don't reinvent new ones..





_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.