On Tue, Dec 05, 2017 at 10:28:22PM +0000, James Hogan wrote:
> Since commit 68923cdc2eb3 ("MIPS: CM: Add cluster & block args to
> mips_cm_lock_other()"), mips_smp_send_ipi_mask() has used
> mips_cm_lock_other_cpu() with each CPU number, rather than
> mips_cm_lock_other() with the first VPE in each core. Prior to r6,
> multicore multithreaded systems such as dual-core dual-thread
> interAptivs with CPU Idle enabled (e.g. MIPS Creator Ci40) results in
> mips_cm_lock_other() repeatedly hitting WARN_ON(vp != 0).
>
> There doesn't appear to be anything fundamentally wrong about passing a
> non-zero VP/VPE number, even if it is a core's region that is locked
> into the other region before r6, so remove that particular WARN_ON().
>
> Fixes: 68923cdc2eb3 ("MIPS: CM: Add cluster & block args to mips_cm_lock_other()")
> Signed-off-by: James Hogan <jhogan@xxxxxxxxxx>
> Reviewed-by: Paul Burton <paul.burton@xxxxxxxx>
> Cc: Ralf Baechle <ralf@xxxxxxxxxxxxxx>
> Cc: linux-mips@xxxxxxxxxxxxxx
> Cc: <stable@xxxxxxxxxxxxxxx> # 4.14+
> ---
> arch/mips/kernel/mips-cm.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/arch/mips/kernel/mips-cm.c b/arch/mips/kernel/mips-cm.c
> index dd5567b1e305..8f5bd04f320a 100644
> --- a/arch/mips/kernel/mips-cm.c
> +++ b/arch/mips/kernel/mips-cm.c
> @@ -292,7 +292,6 @@ void mips_cm_lock_other(unsigned int cluster, unsigned int core,
> *this_cpu_ptr(&cm_core_lock_flags));
> } else {
> WARN_ON(cluster != 0);
> - WARN_ON(vp != 0);
I think the reason is that for a while the combination of SMP/CMP with
MT was at the bottom of priorities and nobody really cared about it so
a WARN_ON was thrown in. Which in this case might well itself be a bug!
Ralf