From: Christoph Hellwig <hch@xxxxxxxxxxxxx> Sent: Tuesday, November 21, 2023 10:26 PM > > On Tue, Nov 21, 2023 at 01:20:11PM -0800, mhkelley58@xxxxxxxxx wrote: > > From: Michael Kelley <mhklinux@xxxxxxxxxxx> > > > > The mm subsystem currently provides no mechanism to map memory pages > > to a specified virtual address range. A virtual address range can be > > allocated using get_vm_area(), but the only function available for > > mapping memory pages to a caller-specified address in that range is > > ioremap_page_range(), which is inappropriate for system memory. > > > > Fix this by allowing vmap_pages_range() to be used by callers outside > > of vmalloc.c. > > I really do not want to expose vmap_pages_range. Please try to come up > with a good way to encapsulate your map at a certain address primitive > and implement it in vmalloc.c. To clarify, is your concern narrowly about vmap_pages_range() specifically? Or is your concern more generally about having two separate steps as applied to system memory: 1) allocate the virtual address space and 2) do the mapping? The two separate steps are already available for MMIO space. Doing the equivalent for system memory should be straightforward. Conversely, combining the two steps into a single new vmap() variant would be a little messy, but can probably be made to work. This combined approach will be less efficient since my use case does a single allocation of virtual address space and repeatedly maps/unmaps the same page in that space. I would need to take some measurements to see if the inefficiency actually matters. Michael
