From: Longpeng > Sent: 16 August 2021 01:26 > Hi David, > > 在 2021/8/15 4:07, David Laight 写道: > > ... > >>>>> Let me describe my use case more clearly (just ignore if you're not > >>>>> interested in it): > >>>>> > >>>>> 1. Prog A mmap() 4GB memory (anon or file-mapping), suppose the > >>>>> allocated VA range is [0x40000000,0x140000000) > >>>>> > >>>>> 2. Prog A specifies [0x48000000,0x50000000) and > >>>>> [0x80000000,0x100000000) will be shared by its child. > >>>>> > >>>>> 3. Prog A fork() Prog B and then Prog B exec() a new ELF binary. > >>>>> > >>>>> 4. Prog B notice the shared ranges (e.g. by input parameters or > >>>>> ...) > >>>>> and remap them to a continuous VA range. > > > > Remapping to contiguous VA is going to be difficult in the > > general case for (IIRC) VIVT caches. > > The required cache coherence may only be attainable by > > using uncached mappings. > > > > The Prog B uses mremap() syscall to remapping the shared ranges to other places, > this is a common case for mremap in userspace. > The cache coherence should already be processed in mremap core logic, otherwise > there's maybe something wrong in mremap(). Maybe it does, and probably mremap() makes it work. But with VIVT caches if a pages gets mapped in two processes at the same time at different offsets into the cache then then both mappings end up being uncached. This was always a problem with normal file mmap. I don't know if Linux manages to pick the VA after finding the page has another mapping - SVR4 didn't. David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
