Date: Thu, 20 May 2004 08:54:33 -0400 (EDT) From: Andrew Gallatin <gallatin@cs.duke.edu> To: Bruce M Simpson <bms@spc.org> Cc: hackers@freebsd.org Subject: Re: api for sharing memory from kernel to userspace? Message-ID: <16556.43657.379465.516798@grasshopper.cs.duke.edu> In-Reply-To: <20040520101721.GB30196@empiric.dek.spc.org> References: <20040519122907.GO81341@elvis.mu.org> <200405192229.i4JMT95J087312@wattres.Watt.COM> <20040520101721.GB30196@empiric.dek.spc.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Bruce M Simpson writes: > On Wed, May 19, 2004 at 03:29:09PM -0700, Steve Watt wrote: > > >On Wed, May 19, 2004 at 05:29:07AM -0700, Alfred Perlstein wrote: > > >+> I need to share about 100megs of memory between kernel and userspace. > > >+> > > >+> The memory can not be paged and should appear contig in the process's > > >+> address space. Any suggestions? > > >+> > > > > The way we accomplished this in some other operating system was to > > create a shared memory segment with an implementation-reserved > > name, and then have the application shm_open the name and mmap it in. > > > > Shouldn't be hard with a device driver. > > This sounds like the way to go. > > The way I achieved this with a PCMCIA memory card was to kmem_alloc_nofault() > a range and then pmap_map() it, then handle mmap() as per the regular > character device case. Why not simply use malloc(9) to allocate the memory and then mmap() it out via a normal character driver's mmap() routine? > Do bear in mind that 100 megs is quite large in relative terms, so it's > possible that kmem_alloc_nofault() would fail. Speaking of this, is there any method that a driver can use to allocate unmapped physical memory in the kernel? For a variety of reasons, I need to allocate this memory in the kernel and mmap() it out. It can not be paged out, and there can be a lot of it. I'm currently using malloc(9), but I run into kmem size limits, and pmap 4MB page bugs on x86. I was thinking about just using vm_page_alloc(..,VM_ALLOC_NOOBJ) and mmap'ing out the memory. Would that work? Or could it still be paged? Drew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?16556.43657.379465.516798>