| Message ID | 698173b487383735e470a28e5cca4f9db22703de.1572024120.git.robin.murphy@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | iommu/io-pgtable: Cleanup and prep for split tables | expand |
On Fri, Oct 25, 2019 at 07:08:35PM +0100, Robin Murphy wrote: > The nature of the LPAE format means that data->pg_shift is always > redundant with data->bits_per_level, since they represent the size of a > page and the number of PTEs per page respectively, and the size of a PTE > is constant. Thus it works out more efficient to only store the latter, > and derive the former via a trivial addition where necessary. > > Signed-off-by: Robin Murphy <robin.murphy@arm.com> > --- > drivers/iommu/io-pgtable-arm.c | 29 +++++++++++++---------------- > 1 file changed, 13 insertions(+), 16 deletions(-) > > diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c > index 4b1483eb0ccf..15b4927ce36b 100644 > --- a/drivers/iommu/io-pgtable-arm.c > +++ b/drivers/iommu/io-pgtable-arm.c > @@ -36,10 +36,11 @@ > * in a virtual address mapped by the pagetable in d. > */ > #define ARM_LPAE_LVL_SHIFT(l,d) \ > - (((ARM_LPAE_MAX_LEVELS - 1 - (l)) * (d)->bits_per_level) + \ > - (d)->pg_shift) > + (((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level) + \ > + ilog2(sizeof(arm_lpae_iopte))) > > -#define ARM_LPAE_GRANULE(d) (1UL << (d)->pg_shift) > +#define ARM_LPAE_GRANULE(d) \ > + (sizeof(arm_lpae_iopte) << (d)->bits_per_level) > #define ARM_LPAE_PGD_SIZE(d) \ > (sizeof(arm_lpae_iopte) << (d)->pgd_bits) > > @@ -55,9 +56,7 @@ > ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1)) > > /* Calculate the block/page mapping size at level l for pagetable in d. */ > -#define ARM_LPAE_BLOCK_SIZE(l,d) \ > - (1ULL << (ilog2(sizeof(arm_lpae_iopte)) + \ > - ((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level))) > +#define ARM_LPAE_BLOCK_SIZE(l,d) (1ULL << ARM_LPAE_LVL_SHIFT(l,d)) > > /* Page table bits */ > #define ARM_LPAE_PTE_TYPE_SHIFT 0 > @@ -175,8 +174,7 @@ struct arm_lpae_io_pgtable { > > int pgd_bits; > int start_level; > - unsigned long pg_shift; > - unsigned long bits_per_level; > + int bits_per_level; > > void *pgd; > }; > @@ -206,7 +204,7 @@ static phys_addr_t iopte_to_paddr(arm_lpae_iopte pte, > { > u64 paddr = pte & ARM_LPAE_PTE_ADDR_MASK; > > - if (data->pg_shift < 16) > + if (data->bits_per_level < 13) /* i.e. 64K granule */ nit, but: if (ARM_LPAE_GRANULE(data) < SZ_64K) might be clearer and avoid the need for a comment? (I can make the change locally if you agree) Will
On 04/11/2019 18:17, Will Deacon wrote: > On Fri, Oct 25, 2019 at 07:08:35PM +0100, Robin Murphy wrote: >> The nature of the LPAE format means that data->pg_shift is always >> redundant with data->bits_per_level, since they represent the size of a >> page and the number of PTEs per page respectively, and the size of a PTE >> is constant. Thus it works out more efficient to only store the latter, >> and derive the former via a trivial addition where necessary. >> >> Signed-off-by: Robin Murphy <robin.murphy@arm.com> >> --- >> drivers/iommu/io-pgtable-arm.c | 29 +++++++++++++---------------- >> 1 file changed, 13 insertions(+), 16 deletions(-) >> >> diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c >> index 4b1483eb0ccf..15b4927ce36b 100644 >> --- a/drivers/iommu/io-pgtable-arm.c >> +++ b/drivers/iommu/io-pgtable-arm.c >> @@ -36,10 +36,11 @@ >> * in a virtual address mapped by the pagetable in d. >> */ >> #define ARM_LPAE_LVL_SHIFT(l,d) \ >> - (((ARM_LPAE_MAX_LEVELS - 1 - (l)) * (d)->bits_per_level) + \ >> - (d)->pg_shift) >> + (((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level) + \ >> + ilog2(sizeof(arm_lpae_iopte))) >> >> -#define ARM_LPAE_GRANULE(d) (1UL << (d)->pg_shift) >> +#define ARM_LPAE_GRANULE(d) \ >> + (sizeof(arm_lpae_iopte) << (d)->bits_per_level) >> #define ARM_LPAE_PGD_SIZE(d) \ >> (sizeof(arm_lpae_iopte) << (d)->pgd_bits) >> >> @@ -55,9 +56,7 @@ >> ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1)) >> >> /* Calculate the block/page mapping size at level l for pagetable in d. */ >> -#define ARM_LPAE_BLOCK_SIZE(l,d) \ >> - (1ULL << (ilog2(sizeof(arm_lpae_iopte)) + \ >> - ((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level))) >> +#define ARM_LPAE_BLOCK_SIZE(l,d) (1ULL << ARM_LPAE_LVL_SHIFT(l,d)) >> >> /* Page table bits */ >> #define ARM_LPAE_PTE_TYPE_SHIFT 0 >> @@ -175,8 +174,7 @@ struct arm_lpae_io_pgtable { >> >> int pgd_bits; >> int start_level; >> - unsigned long pg_shift; >> - unsigned long bits_per_level; >> + int bits_per_level; >> >> void *pgd; >> }; >> @@ -206,7 +204,7 @@ static phys_addr_t iopte_to_paddr(arm_lpae_iopte pte, >> { >> u64 paddr = pte & ARM_LPAE_PTE_ADDR_MASK; >> >> - if (data->pg_shift < 16) >> + if (data->bits_per_level < 13) /* i.e. 64K granule */ > > nit, but: > > if (ARM_LPAE_GRANULE(data) < SZ_64K) > > might be clearer and avoid the need for a comment? Unfortunately GCC doesn't treat the two as directly equivalent (presumably due to boundary conditions) so will emit the additional faff to actually compute and compare the intermediate value every time, rather than just trivially testing the shift. I figured the minor I$/register pressure win was worth the small price of a comment. Robin. > (I can make the change locally if you agree) > > Will >
On Mon, Nov 04, 2019 at 06:36:51PM +0000, Robin Murphy wrote: > On 04/11/2019 18:17, Will Deacon wrote: > > On Fri, Oct 25, 2019 at 07:08:35PM +0100, Robin Murphy wrote: > > > The nature of the LPAE format means that data->pg_shift is always > > > redundant with data->bits_per_level, since they represent the size of a > > > page and the number of PTEs per page respectively, and the size of a PTE > > > is constant. Thus it works out more efficient to only store the latter, > > > and derive the former via a trivial addition where necessary. > > > > > > Signed-off-by: Robin Murphy <robin.murphy@arm.com> > > > --- > > > drivers/iommu/io-pgtable-arm.c | 29 +++++++++++++---------------- > > > 1 file changed, 13 insertions(+), 16 deletions(-) > > > > > > diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c > > > index 4b1483eb0ccf..15b4927ce36b 100644 > > > --- a/drivers/iommu/io-pgtable-arm.c > > > +++ b/drivers/iommu/io-pgtable-arm.c > > > @@ -36,10 +36,11 @@ > > > * in a virtual address mapped by the pagetable in d. > > > */ > > > #define ARM_LPAE_LVL_SHIFT(l,d) \ > > > - (((ARM_LPAE_MAX_LEVELS - 1 - (l)) * (d)->bits_per_level) + \ > > > - (d)->pg_shift) > > > + (((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level) + \ > > > + ilog2(sizeof(arm_lpae_iopte))) > > > -#define ARM_LPAE_GRANULE(d) (1UL << (d)->pg_shift) > > > +#define ARM_LPAE_GRANULE(d) \ > > > + (sizeof(arm_lpae_iopte) << (d)->bits_per_level) > > > #define ARM_LPAE_PGD_SIZE(d) \ > > > (sizeof(arm_lpae_iopte) << (d)->pgd_bits) > > > @@ -55,9 +56,7 @@ > > > ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1)) > > > /* Calculate the block/page mapping size at level l for pagetable in d. */ > > > -#define ARM_LPAE_BLOCK_SIZE(l,d) \ > > > - (1ULL << (ilog2(sizeof(arm_lpae_iopte)) + \ > > > - ((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level))) > > > +#define ARM_LPAE_BLOCK_SIZE(l,d) (1ULL << ARM_LPAE_LVL_SHIFT(l,d)) > > > /* Page table bits */ > > > #define ARM_LPAE_PTE_TYPE_SHIFT 0 > > > @@ -175,8 +174,7 @@ struct arm_lpae_io_pgtable { > > > int pgd_bits; > > > int start_level; > > > - unsigned long pg_shift; > > > - unsigned long bits_per_level; > > > + int bits_per_level; > > > void *pgd; > > > }; > > > @@ -206,7 +204,7 @@ static phys_addr_t iopte_to_paddr(arm_lpae_iopte pte, > > > { > > > u64 paddr = pte & ARM_LPAE_PTE_ADDR_MASK; > > > - if (data->pg_shift < 16) > > > + if (data->bits_per_level < 13) /* i.e. 64K granule */ > > > > nit, but: > > > > if (ARM_LPAE_GRANULE(data) < SZ_64K) > > > > might be clearer and avoid the need for a comment? > > Unfortunately GCC doesn't treat the two as directly equivalent (presumably > due to boundary conditions) so will emit the additional faff to actually > compute and compare the intermediate value every time, rather than just > trivially testing the shift. I figured the minor I$/register pressure win > was worth the small price of a comment. Bet ya can't measure the difference ;) I'd prefer the readable version in the absence of numbers. Will
diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c index 4b1483eb0ccf..15b4927ce36b 100644 --- a/drivers/iommu/io-pgtable-arm.c +++ b/drivers/iommu/io-pgtable-arm.c @@ -36,10 +36,11 @@ * in a virtual address mapped by the pagetable in d. */ #define ARM_LPAE_LVL_SHIFT(l,d) \ - (((ARM_LPAE_MAX_LEVELS - 1 - (l)) * (d)->bits_per_level) + \ - (d)->pg_shift) + (((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level) + \ + ilog2(sizeof(arm_lpae_iopte))) -#define ARM_LPAE_GRANULE(d) (1UL << (d)->pg_shift) +#define ARM_LPAE_GRANULE(d) \ + (sizeof(arm_lpae_iopte) << (d)->bits_per_level) #define ARM_LPAE_PGD_SIZE(d) \ (sizeof(arm_lpae_iopte) << (d)->pgd_bits) @@ -55,9 +56,7 @@ ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1)) /* Calculate the block/page mapping size at level l for pagetable in d. */ -#define ARM_LPAE_BLOCK_SIZE(l,d) \ - (1ULL << (ilog2(sizeof(arm_lpae_iopte)) + \ - ((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level))) +#define ARM_LPAE_BLOCK_SIZE(l,d) (1ULL << ARM_LPAE_LVL_SHIFT(l,d)) /* Page table bits */ #define ARM_LPAE_PTE_TYPE_SHIFT 0 @@ -175,8 +174,7 @@ struct arm_lpae_io_pgtable { int pgd_bits; int start_level; - unsigned long pg_shift; - unsigned long bits_per_level; + int bits_per_level; void *pgd; }; @@ -206,7 +204,7 @@ static phys_addr_t iopte_to_paddr(arm_lpae_iopte pte, { u64 paddr = pte & ARM_LPAE_PTE_ADDR_MASK; - if (data->pg_shift < 16) + if (data->bits_per_level < 13) /* i.e. 64K granule */ return paddr; /* Rotate the packed high-order bits back to the top */ @@ -742,9 +740,8 @@ static void arm_lpae_restrict_pgsizes(struct io_pgtable_cfg *cfg) static struct arm_lpae_io_pgtable * arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg) { - unsigned long va_bits; struct arm_lpae_io_pgtable *data; - int levels; + int levels, va_bits, pg_shift; arm_lpae_restrict_pgsizes(cfg); @@ -766,10 +763,10 @@ arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg) if (!data) return NULL; - data->pg_shift = __ffs(cfg->pgsize_bitmap); - data->bits_per_level = data->pg_shift - ilog2(sizeof(arm_lpae_iopte)); + pg_shift = __ffs(cfg->pgsize_bitmap); + data->bits_per_level = pg_shift - ilog2(sizeof(arm_lpae_iopte)); - va_bits = cfg->ias - data->pg_shift; + va_bits = cfg->ias - pg_shift; levels = DIV_ROUND_UP(va_bits, data->bits_per_level); data->start_level = ARM_LPAE_MAX_LEVELS - levels; @@ -1138,9 +1135,9 @@ static void __init arm_lpae_dump_ops(struct io_pgtable_ops *ops) pr_err("cfg: pgsize_bitmap 0x%lx, ias %u-bit\n", cfg->pgsize_bitmap, cfg->ias); - pr_err("data: %d levels, 0x%zx pgd_size, %lu pg_shift, %lu bits_per_level, pgd @ %p\n", + pr_err("data: %d levels, 0x%zx pgd_size, %u pg_shift, %u bits_per_level, pgd @ %p\n", ARM_LPAE_MAX_LEVELS - data->start_level, ARM_LPAE_PGD_SIZE(data), - data->pg_shift, data->bits_per_level, data->pgd); + ilog2(ARM_LPAE_GRANULE(data)), data->bits_per_level, data->pgd); } #define __FAIL(ops, i) ({ \
The nature of the LPAE format means that data->pg_shift is always redundant with data->bits_per_level, since they represent the size of a page and the number of PTEs per page respectively, and the size of a PTE is constant. Thus it works out more efficient to only store the latter, and derive the former via a trivial addition where necessary. Signed-off-by: Robin Murphy <robin.murphy@arm.com> --- drivers/iommu/io-pgtable-arm.c | 29 +++++++++++++---------------- 1 file changed, 13 insertions(+), 16 deletions(-)