[PATCH RFC] 1/2 central workspace for zlib
dwmw2 at infradead.org
Mon Jun 2 15:36:13 EDT 2003
On Mon, 2003-06-02 at 17:37, Jörn Engel wrote:
> On Mon, 2 June 2003 16:59:25 +0100, David Woodhouse wrote:
> > On Mon, 2003-06-02 at 16:53, Jörn Engel wrote:
> > > Maybe lazy allocation. vmalloc() it with the first write(), which
> > > should be never in production use. So the extra overhead doesn't
> > > really matter.
> > Seems reasonable.
> Patch is in CVS.
> Not 100% sure about the correct return code, if the lazy allocation
> fails. Can you check that?
Thanks. The return code doesn't matter. If you look at the caller you
see it's only used as uptodate = !!ret anyway. It can't go any further.
I was concerned briefly that the allocation could happen when we're
trying to evict a page on memory pressure -- but in fact that's
virtually impossible since you'd have to mount the file system and mmap
your file without actually writing anything to the block device.
And anyway, refer to earlier comments about anyone using the naïve
read/erase/modify/writeback mtdblock translation layer for writing in
production wanting to be shot for it :)
More information about the linux-mtd