On 2021-09-23 16:06:58 [-0700], Minchan Kim wrote: > On Thu, Sep 23, 2021 at 07:01:21PM +0200, Sebastian Andrzej Siewior wrote: > > For efficiency reasons, zsmalloc is using a slim `handle'. The value is > > the address of a memory allocation of 4 or 8 bytes depending on the size > > of the long data type. The lowest bit in that allocated memory is used > > as a bit spin lock. > > The usage of the bit spin lock is problematic because with the bit spin > > lock held zsmalloc acquires a rwlock_t and spinlock_t which are both > > sleeping locks on PREEMPT_RT and therefore must not be acquired with > > disabled preemption. > > I am not sure how long the preemption disabled section takes since it > just disable for a page copy mostly. after the bit-spin-lock, there are sleeping locks. These are problematic. > > > > There is a patch which extends the handle on PREEMPT_RT so that a full > > spinlock_t fits (even with lockdep enabled) and then eliminates the bit > > spin lock. I'm not sure how sensible zsmalloc on PREEMPT_RT is given > > that it is used to store compressed user memory. > > I don't see what's relation between PREEMPT_RT and compressed user > memory so you can reach such conclustion. Disable zsmalloc also makes > disable zram. I think in-compress memory swap rather than storage > swap/block sometimes would be useful for even RT. The only user of zsmalloc I found was zswap which compressed user pages. Maybe I didn't collect all the dots. > > Disable ZSMALLOC on PREEMPT_RT. If there is need for it, we can try to > > get it to work. > > Please send the patch which extends handle with spin_lock rather than > simply disabing. Okay. I will clean it up a little and post it then. Sebastian
