On Fri, Jul 06, 2018 at 07:48:12PM +0800, Guo Ren wrote: > On Thu, Jul 05, 2018 at 08:00:08PM +0200, Peter Zijlstra wrote: > > On Mon, Jul 02, 2018 at 01:30:14AM +0800, Guo Ren wrote: > > > +#ifdef CONFIG_CPU_HAS_LDSTEX > > > +ENTRY(csky_cmpxchg) > > > + USPTOKSP > > > + mfcr a3, epc > > > + INCTRAP a3 > > > + > > > + subi sp, 8 > > > + stw a3, (sp, 0) > > > + mfcr a3, epsr > > > + stw a3, (sp, 4) > > > + > > > + psrset ee > > > +1: > > > + ldex a3, (a2) > > > + cmpne a0, a3 > > > + bt16 2f > > > + mov a3, a1 > > > + stex a3, (a2) > > > + bez a3, 1b > > > +2: > > > + sync.is > > > + mvc a0 > > > + ldw a3, (sp, 0) > > > + mtcr a3, epc > > > + ldw a3, (sp, 4) > > > + mtcr a3, epsr > > > + addi sp, 8 > > > + KSPTOUSP > > > + rte > > > +END(csky_cmpxchg) > > > +#else > > > > Please explain... if the CPU has LDEX/STEX, then _why_ do you need this? > Our libc use csky_cmpxchg and we want it compatible. Of course, we'll > also implement the ldex/stex atomic operations in libs in future.
I would strongly suggest not providing this syscall on SMP systems from the get go. There is no point in starting with legacy problems without an actual legacy.
