Re: [PATCH v10 09/23] iommu/io-pgtable-arm-v7s: Extend to support PA[33:32] for MediaTek

From: Will Deacon
Date: Thu Aug 22 2019 - 09:45:44 EST


On Thu, Aug 22, 2019 at 11:57:11AM +0100, Robin Murphy wrote:
> On 2019-08-22 11:17 am, Will Deacon wrote:
> > On Thu, Aug 22, 2019 at 11:08:58AM +0100, Robin Murphy wrote:
> > > On 2019-08-22 9:56 am, Yong Wu wrote:
> > > > On Wed, 2019-08-21 at 16:24 +0100, Will Deacon wrote:
> > > > > On Wed, Aug 21, 2019 at 09:53:12PM +0800, Yong Wu wrote:
> > > > > > MediaTek extend the arm v7s descriptor to support up to 34 bits PA where
> > > > > > the bit32 and bit33 are encoded in the bit9 and bit4 of the PTE
> > > > > > respectively. Meanwhile the iova still is 32bits.
> > > > > >
> > > > > > Regarding whether the pagetable address could be over 4GB, the mt8183
> > > > > > support it while the previous mt8173 don't, thus keep it as is.
> > > > > >
> > > > > > Signed-off-by: Yong Wu <yong.wu@xxxxxxxxxxxx>
> > > > > > ---
> > > > > > drivers/iommu/io-pgtable-arm-v7s.c | 32 +++++++++++++++++++++++++-------
> > > > > > include/linux/io-pgtable.h | 7 +++----
> > > > > > 2 files changed, 28 insertions(+), 11 deletions(-)
> > > > >
> > > > > [...]
> > > > >
> > > > > > @@ -731,7 +747,9 @@ static struct io_pgtable *arm_v7s_alloc_pgtable(struct io_pgtable_cfg *cfg,
> > > > > > {
> > > > > > struct arm_v7s_io_pgtable *data;
> > > > > > - if (cfg->ias > ARM_V7S_ADDR_BITS || cfg->oas > ARM_V7S_ADDR_BITS)
> > > > > > + if (cfg->ias > ARM_V7S_ADDR_BITS ||
> > > > > > + (cfg->oas > ARM_V7S_ADDR_BITS &&
> > > > > > + !(cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_EXT)))
> > > > >
> > > > > Please can you instead change arm_v7s_alloc_pgtable() so that it allows an
> > > > > ias of up to 34 when the IO_PGTABLE_QUIRK_ARM_MTK_EXT is set?
> > > >
> > > > Here I only simply skip the oas checking for our case. then which way do
> > > > your prefer? something like you commented before:?
> > > >
> > > >
> > > > if (cfg->ias > ARM_V7S_ADDR_BITS)
> > > > return NULL;
> > > >
> > > > if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_EXT) {
> > > > if (!IS_ENABLED(CONFIG_PHYS_ADDR_T_64BIT))
> > > > cfg->oas = min(cfg->oas, ARM_V7S_ADDR_BITS);
> > > > else if (cfg->oas > 34)
> > > > return NULL;
> > > > } else if (cfg->oas > ARM_V7S_ADDR_BITS) {
> > > > return NULL;
> > > > }
> > >
> > > All it should take is something like:
> > >
> > > if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_EXT)
> > > max_oas = 34;
> > > else
> > > max_oas = 32;
> > > if (cfg->oas > max_oas)
> > > return NULL;
> > >
> > > or even just:
> > >
> > > if (cfg->oas > 32 ||
> > > (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_EXT && cfg->oas > 34))
> > > return NULL;
> > >
> > > (and if we prefer the latter style, perhaps we could introduce some kind of
> > > "is_mtk_4gb()" helper to save on verbosity)
> >
> > I wondered the same thing, but another place we'd want the check is in
> > iopte_to_paddr() which probably needs the PHYS_ADDR_T check to avoid GCC
> > warnings, although I didn't try it.
>
> I'm pretty sure I confirmed that "paddr |= BIT_ULL(32)" doesn't warn when
> phys_addt_t is 32-bit - it's well-defined unsigned integer truncation after
> all, and if GCC starts warning about all the valid no-op code it optimises
> away then it's going to run up against IS_ENABLED() first and foremost ;)

You're quite right, although we live in a world where GCC shouts at us about
missing comments in switch statements so I think my worry was justified!

> > So if we did:
> >
> > static bool cfg_mtk_ext_enabled(struct io_pgtable_cfg *cfg)
> > {
> > return IS_ENABLED(CONFIG_PHYS_ADDR_T_64BIT) &&
> > cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_EXT;
> > }
> >
> > Then I suppose we could do this in _alloc():
> >
> > if (cfg->oas > cfg_mtk_ext_enabled(cfg) ? 34 : ARM_V7S_ADDR_BITS)
> > return NULL;

^^ Apparantly, I left the bracketting here as an exercise to the reader.

> >
> > and then this in iopte_to_paddr():
> >
> > [...]
> >
> > paddr = pte & mask;
> > if (!cfg_mtk_ext_enabled(cfg))
> > return paddr;
> >
> > if (pte & ARM_V7S_ATTR_MTK_PA_BIT32)
> > paddr |= ...
> >
> > [...]
> >
> > What do you reckon?
>
> Yeah, that's the general shape of things I was picturing - I'm not that
> fussed about the PHYS_ADDR_T_64BIT thing, especially if it's wrapped up in
> just one place, so if you do want to keep it as belt-and-braces I'll just
> consider it a slight code size optimisation for 32-bit builds.

Ok, great. Yong Wu -- are you ok respinning with the above + missing
brackets?

Will