hi, > YAMAMOTO Takashi <y...@mwd.biglobe.ne.jp> wrote: > >> > This avoids deadlocks in the following situations: >> > 1) when memory is low: ioflush waits the fileystem, the fielsystem waits >> > for memory >> can you explain how it is a problem? > > As I understand, one way to free memory is to flush vnode backed pages > to the backend storage. If ioflush calls VOP_FSYNC on a memory-starved > userland filesystem, it will get stuck until the filesystem gets memory > again, and while it is stuck, it does not help freeing memory.
as i told you a few times, ioflush is not a thread to free memory. pagedaemon is. please read ufs_bmaparray and grep "uvm.pagedaemon_lwp" in src/sys/kern/ to see what other filesystems do. i don't think this approach works for puffs because it's almost impossible to say in which cases an operation needs memory allocation to complete, though. YAMAMOTO Takashi > > -- > Emmanuel Dreyfus > http://hcpnet.free.fr/pubz > m...@netbsd.org
