Re: [PATCH v3 09/33] iommu/mediatek: Remove for_each_m4u in tlb_sync_all

[AngeloGioacchino Del Regno]

Il 09/01/22 03:48, Yong Wu ha scritto:
> On Tue, 2022-01-04 at 16:55 +0100, AngeloGioacchino Del Regno wrote:
> > Il 23/09/21 13:58, Yong Wu ha scritto:
> > > The tlb_sync_all is called from these three functions:
> > > a) flush_iotlb_all: it will be called for each a iommu HW.
> > > b) tlb_flush_range_sync: it already has for_each_m4u.
> > > c) in irq: When IOMMU HW translation fault, Only need flush itself.
> > >
> > > Thus, No need for_each_m4u in this tlb_sync_all. Remove it.
> > >
> > > Signed-off-by: Yong Wu <yong.wu@xxxxxxxxxxxx>
> > > ---
> > >    drivers/iommu/mtk_iommu.c | 18 +++++++-----------
> > >    1 file changed, 7 insertions(+), 11 deletions(-)
> > >
> > > diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
> > > index 6f4f6624e3ac..0b4c30baa864 100644
> > > --- a/drivers/iommu/mtk_iommu.c
> > > +++ b/drivers/iommu/mtk_iommu.c
> > > @@ -206,19 +206,15 @@ static struct mtk_iommu_domain
> > > *to_mtk_domain(struct iommu_domain *dom)
> > >    
> > >    static void mtk_iommu_tlb_flush_all(struct mtk_iommu_data *data)
> > >    {
> > > -	struct list_head *head = data->hw_list;
> > > -
> > > -	for_each_m4u(data, head) {
> > > -		if (pm_runtime_get_if_in_use(data->dev) <= 0)
> > > -			continue;
> > > +	if (pm_runtime_get_if_in_use(data->dev) <= 0)
> > > +		return;
> > >    
> > > -		writel_relaxed(F_INVLD_EN1 | F_INVLD_EN0,
> > > -			       data->base + data->plat_data-
> > > > inv_sel_reg);
> > > -		writel_relaxed(F_ALL_INVLD, data->base +
> > > REG_MMU_INVALIDATE);
> > > -		wmb(); /* Make sure the tlb flush all done */
> > > +	writel_relaxed(F_INVLD_EN1 | F_INVLD_EN0,
> > > +		       data->base + data->plat_data->inv_sel_reg);
> > > +	writel_relaxed(F_ALL_INVLD, data->base + REG_MMU_INVALIDATE);
> > > +	wmb(); /* Make sure the tlb flush all done */
> >
> > There aren't a lot of writes here - not anymore, since you are no
> > longer doing
> > this for_each_m4u()...
> > ...so, please change writel_relaxed() to writel() calls, allowing you
> > to also
> > remove the write barrier at the end (since in the non relaxed
> > version, order is already ensured).
>
> In the "writel", the "__iowmb()" runs before "write_relaxed". Then how
> to guarantee the last register was wrote into the HW. Here the flush
> all don't have sync(waiting it complete)

That's right, I'm sorry for the invalid proposal.

Though, there's something else to mention here... if writing
(F_INVLD_EN1 | F_INVLD_EN0) to inv_sel_reg is *required* to happen before
writing F_ALL_INVLD to REG_MMU_INVALIDATE (which I think is exactly the
case here), then, in order to ensure write ordering, you should still use
writel() instead of the relaxed accessor; after which, since (as you mentioned)
there is no sync readback loop, you can keep that wmb() at the end.

> > >    
> > > -		pm_runtime_put(data->dev);
> > > -	}
> > > +	pm_runtime_put(data->dev);
> > >    }
> > >    
> > >    static void mtk_iommu_tlb_flush_range_sync(unsigned long iova,
> > > size_t size,