[PATCH] iommu/sun50i: Allow page sizes between 4K and 1M

[PATCH] iommu/sun50i: Allow page sizes between 4K and 1M

[Jernej Skrabec]

While peripheral supports only 4K page sizes, we can easily emulate
support for bigger page sizes, up to 1M. This is done by making multiple
entries in map function or clearing multiple entries in unmap.

Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
---
 drivers/iommu/sun50i-iommu.c | 49 +++++++++++++++++++++++-------------
 1 file changed, 31 insertions(+), 18 deletions(-)

diff --git a/drivers/iommu/sun50i-iommu.c b/drivers/iommu/sun50i-iommu.c
index 74c5cb93e900..93077575d60f 100644
--- a/drivers/iommu/sun50i-iommu.c
+++ b/drivers/iommu/sun50i-iommu.c
@@ -95,6 +95,10 @@
 
 #define SPAGE_SIZE			4096
 
+#define SUN50I_IOMMU_PGSIZES		(SZ_4K | SZ_8K | SZ_16K | SZ_32K | \
+					 SZ_64K | SZ_128K | SZ_256K | \
+					 SZ_512K | SZ_1M)
+
 struct sun50i_iommu {
 	struct iommu_device iommu;
 
@@ -593,10 +597,12 @@ static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
 {
 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
 	struct sun50i_iommu *iommu = sun50i_domain->iommu;
-	u32 pte_index;
+	u32 pte_index, pages, i;
 	u32 *page_table, *pte_addr;
 	int ret = 0;
 
+	pages = size / SPAGE_SIZE;
+
 	page_table = sun50i_dte_get_page_table(sun50i_domain, iova, gfp);
 	if (IS_ERR(page_table)) {
 		ret = PTR_ERR(page_table);
@@ -604,18 +610,22 @@ static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
 	}
 
 	pte_index = sun50i_iova_get_pte_index(iova);
-	pte_addr = &page_table[pte_index];
-	if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
-		phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
-		dev_err(iommu->dev,
-			"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
-			&iova, &page_phys, &paddr, prot);
-		ret = -EBUSY;
-		goto out;
+	for (i = 0; i < pages; i++) {
+		pte_addr = &page_table[pte_index + i];
+		if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
+			phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
+
+			dev_err(iommu->dev,
+				"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
+				&iova, &page_phys, &paddr, prot);
+			ret = -EBUSY;
+			goto out;
+		}
+		*pte_addr = sun50i_mk_pte(paddr, prot);
+		paddr += SPAGE_SIZE;
 	}
 
-	*pte_addr = sun50i_mk_pte(paddr, prot);
-	sun50i_table_flush(sun50i_domain, pte_addr, 1);
+	sun50i_table_flush(sun50i_domain, &page_table[pte_index], pages);
 
 out:
 	return ret;
@@ -626,8 +636,10 @@ static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova
 {
 	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
 	phys_addr_t pt_phys;
+	u32 dte, pages, i;
 	u32 *pte_addr;
-	u32 dte;
+
+	pages = size / SPAGE_SIZE;
 
 	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
 	if (!sun50i_dte_is_pt_valid(dte))
@@ -636,13 +648,14 @@ static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova
 	pt_phys = sun50i_dte_get_pt_address(dte);
 	pte_addr = (u32 *)phys_to_virt(pt_phys) + sun50i_iova_get_pte_index(iova);
 
-	if (!sun50i_pte_is_page_valid(*pte_addr))
-		return 0;
+	for (i = 0; i < pages; i++)
+		if (!sun50i_pte_is_page_valid(pte_addr[i]))
+			return 0;
 
-	memset(pte_addr, 0, sizeof(*pte_addr));
-	sun50i_table_flush(sun50i_domain, pte_addr, 1);
+	memset(pte_addr, 0, sizeof(*pte_addr) * pages);
+	sun50i_table_flush(sun50i_domain, pte_addr, pages);
 
-	return SZ_4K;
+	return size;
 }
 
 static phys_addr_t sun50i_iommu_iova_to_phys(struct iommu_domain *domain,
@@ -827,7 +840,7 @@ static int sun50i_iommu_of_xlate(struct device *dev,
 }
 
 static const struct iommu_ops sun50i_iommu_ops = {
-	.pgsize_bitmap	= SZ_4K,
+	.pgsize_bitmap	= SUN50I_IOMMU_PGSIZES,
 	.device_group	= sun50i_iommu_device_group,
 	.domain_alloc	= sun50i_iommu_domain_alloc,
 	.of_xlate	= sun50i_iommu_of_xlate,
-- 
2.42.0


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH] iommu/sun50i: Allow page sizes between 4K and 1M

[Robin Murphy]

On 2023-09-11 20:02, Jernej Skrabec wrote:
> While peripheral supports only 4K page sizes, we can easily emulate
> support for bigger page sizes, up to 1M. This is done by making multiple
> entries in map function or clearing multiple entries in unmap.

No, it's done by converting over to the {map,unmap}_pages ops. There's 
no need to resurrect this horrible bodge.

Thanks,
Robin.

> Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
> ---
>   drivers/iommu/sun50i-iommu.c | 49 +++++++++++++++++++++++-------------
>   1 file changed, 31 insertions(+), 18 deletions(-)
> 
> diff --git a/drivers/iommu/sun50i-iommu.c b/drivers/iommu/sun50i-iommu.c
> index 74c5cb93e900..93077575d60f 100644
> --- a/drivers/iommu/sun50i-iommu.c
> +++ b/drivers/iommu/sun50i-iommu.c
> @@ -95,6 +95,10 @@
>   
>   #define SPAGE_SIZE			4096
>   
> +#define SUN50I_IOMMU_PGSIZES		(SZ_4K | SZ_8K | SZ_16K | SZ_32K | \
> +					 SZ_64K | SZ_128K | SZ_256K | \
> +					 SZ_512K | SZ_1M)
> +
>   struct sun50i_iommu {
>   	struct iommu_device iommu;
>   
> @@ -593,10 +597,12 @@ static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
>   {
>   	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
>   	struct sun50i_iommu *iommu = sun50i_domain->iommu;
> -	u32 pte_index;
> +	u32 pte_index, pages, i;
>   	u32 *page_table, *pte_addr;
>   	int ret = 0;
>   
> +	pages = size / SPAGE_SIZE;
> +
>   	page_table = sun50i_dte_get_page_table(sun50i_domain, iova, gfp);
>   	if (IS_ERR(page_table)) {
>   		ret = PTR_ERR(page_table);
> @@ -604,18 +610,22 @@ static int sun50i_iommu_map(struct iommu_domain *domain, unsigned long iova,
>   	}
>   
>   	pte_index = sun50i_iova_get_pte_index(iova);
> -	pte_addr = &page_table[pte_index];
> -	if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
> -		phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
> -		dev_err(iommu->dev,
> -			"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
> -			&iova, &page_phys, &paddr, prot);
> -		ret = -EBUSY;
> -		goto out;
> +	for (i = 0; i < pages; i++) {
> +		pte_addr = &page_table[pte_index + i];
> +		if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
> +			phys_addr_t page_phys = sun50i_pte_get_page_address(*pte_addr);
> +
> +			dev_err(iommu->dev,
> +				"iova %pad already mapped to %pa cannot remap to %pa prot: %#x\n",
> +				&iova, &page_phys, &paddr, prot);
> +			ret = -EBUSY;
> +			goto out;
> +		}
> +		*pte_addr = sun50i_mk_pte(paddr, prot);
> +		paddr += SPAGE_SIZE;
>   	}
>   
> -	*pte_addr = sun50i_mk_pte(paddr, prot);
> -	sun50i_table_flush(sun50i_domain, pte_addr, 1);
> +	sun50i_table_flush(sun50i_domain, &page_table[pte_index], pages);
>   
>   out:
>   	return ret;
> @@ -626,8 +636,10 @@ static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova
>   {
>   	struct sun50i_iommu_domain *sun50i_domain = to_sun50i_domain(domain);
>   	phys_addr_t pt_phys;
> +	u32 dte, pages, i;
>   	u32 *pte_addr;
> -	u32 dte;
> +
> +	pages = size / SPAGE_SIZE;
>   
>   	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
>   	if (!sun50i_dte_is_pt_valid(dte))
> @@ -636,13 +648,14 @@ static size_t sun50i_iommu_unmap(struct iommu_domain *domain, unsigned long iova
>   	pt_phys = sun50i_dte_get_pt_address(dte);
>   	pte_addr = (u32 *)phys_to_virt(pt_phys) + sun50i_iova_get_pte_index(iova);
>   
> -	if (!sun50i_pte_is_page_valid(*pte_addr))
> -		return 0;
> +	for (i = 0; i < pages; i++)
> +		if (!sun50i_pte_is_page_valid(pte_addr[i]))
> +			return 0;
>   
> -	memset(pte_addr, 0, sizeof(*pte_addr));
> -	sun50i_table_flush(sun50i_domain, pte_addr, 1);
> +	memset(pte_addr, 0, sizeof(*pte_addr) * pages);
> +	sun50i_table_flush(sun50i_domain, pte_addr, pages);
>   
> -	return SZ_4K;
> +	return size;
>   }
>   
>   static phys_addr_t sun50i_iommu_iova_to_phys(struct iommu_domain *domain,
> @@ -827,7 +840,7 @@ static int sun50i_iommu_of_xlate(struct device *dev,
>   }
>   
>   static const struct iommu_ops sun50i_iommu_ops = {
> -	.pgsize_bitmap	= SZ_4K,
> +	.pgsize_bitmap	= SUN50I_IOMMU_PGSIZES,
>   	.device_group	= sun50i_iommu_device_group,
>   	.domain_alloc	= sun50i_iommu_domain_alloc,
>   	.of_xlate	= sun50i_iommu_of_xlate,

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH] iommu/sun50i: Allow page sizes between 4K and 1M

[Jernej Škrabec]

On Monday, September 11, 2023 9:06:12 PM CEST Robin Murphy wrote:
> On 2023-09-11 20:02, Jernej Skrabec wrote:
> > While peripheral supports only 4K page sizes, we can easily emulate
> > support for bigger page sizes, up to 1M. This is done by making multiple
> > entries in map function or clearing multiple entries in unmap.
> 
> No, it's done by converting over to the {map,unmap}_pages ops. There's
> no need to resurrect this horrible bodge.

Missed those, thanks! In any case, logic to implement those isn't that 
different than here.

Best regards,
Jernej

> 
> Thanks,
> Robin.
> 
> > Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
> > ---
> > 
> >   drivers/iommu/sun50i-iommu.c | 49 +++++++++++++++++++++++-------------
> >   1 file changed, 31 insertions(+), 18 deletions(-)
> > 
> > diff --git a/drivers/iommu/sun50i-iommu.c b/drivers/iommu/sun50i-iommu.c
> > index 74c5cb93e900..93077575d60f 100644
> > --- a/drivers/iommu/sun50i-iommu.c
> > +++ b/drivers/iommu/sun50i-iommu.c
> > @@ -95,6 +95,10 @@
> > 
> >   #define SPAGE_SIZE			4096
> > 
> > +#define SUN50I_IOMMU_PGSIZES		(SZ_4K | SZ_8K | SZ_16K | SZ_32K | 
\
> > +					 SZ_64K | SZ_128K | 
SZ_256K | \
> > +					 SZ_512K | SZ_1M)
> > +
> > 
> >   struct sun50i_iommu {
> >   
> >   	struct iommu_device iommu;
> > 
> > @@ -593,10 +597,12 @@ static int sun50i_iommu_map(struct iommu_domain
> > *domain, unsigned long iova,> 
> >   {
> >   
> >   	struct sun50i_iommu_domain *sun50i_domain = 
to_sun50i_domain(domain);
> >   	struct sun50i_iommu *iommu = sun50i_domain->iommu;
> > 
> > -	u32 pte_index;
> > +	u32 pte_index, pages, i;
> > 
> >   	u32 *page_table, *pte_addr;
> >   	int ret = 0;
> > 
> > +	pages = size / SPAGE_SIZE;
> > +
> > 
> >   	page_table = sun50i_dte_get_page_table(sun50i_domain, iova, gfp);
> >   	if (IS_ERR(page_table)) {
> >   	
> >   		ret = PTR_ERR(page_table);
> > 
> > @@ -604,18 +610,22 @@ static int sun50i_iommu_map(struct iommu_domain
> > *domain, unsigned long iova,> 
> >   	}
> >   	
> >   	pte_index = sun50i_iova_get_pte_index(iova);
> > 
> > -	pte_addr = &page_table[pte_index];
> > -	if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
> > -		phys_addr_t page_phys = 
sun50i_pte_get_page_address(*pte_addr);
> > -		dev_err(iommu->dev,
> > -			"iova %pad already mapped to %pa cannot 
remap to %pa prot: %#x\n",
> > -			&iova, &page_phys, &paddr, prot);
> > -		ret = -EBUSY;
> > -		goto out;
> > +	for (i = 0; i < pages; i++) {
> > +		pte_addr = &page_table[pte_index + i];
> > +		if (unlikely(sun50i_pte_is_page_valid(*pte_addr))) {
> > +			phys_addr_t page_phys = 
sun50i_pte_get_page_address(*pte_addr);
> > +
> > +			dev_err(iommu->dev,
> > +				"iova %pad already mapped to %pa 
cannot remap to %pa prot: %#x\n",
> > +				&iova, &page_phys, &paddr, prot);
> > +			ret = -EBUSY;
> > +			goto out;
> > +		}
> > +		*pte_addr = sun50i_mk_pte(paddr, prot);
> > +		paddr += SPAGE_SIZE;
> > 
> >   	}
> > 
> > -	*pte_addr = sun50i_mk_pte(paddr, prot);
> > -	sun50i_table_flush(sun50i_domain, pte_addr, 1);
> > +	sun50i_table_flush(sun50i_domain, &page_table[pte_index], pages);
> > 
> >   out:
> >   	return ret;
> > 
> > @@ -626,8 +636,10 @@ static size_t sun50i_iommu_unmap(struct iommu_domain
> > *domain, unsigned long iova> 
> >   {
> >   
> >   	struct sun50i_iommu_domain *sun50i_domain = 
to_sun50i_domain(domain);
> >   	phys_addr_t pt_phys;
> > 
> > +	u32 dte, pages, i;
> > 
> >   	u32 *pte_addr;
> > 
> > -	u32 dte;
> > +
> > +	pages = size / SPAGE_SIZE;
> > 
> >   	dte = sun50i_domain->dt[sun50i_iova_get_dte_index(iova)];
> >   	if (!sun50i_dte_is_pt_valid(dte))
> > 
> > @@ -636,13 +648,14 @@ static size_t sun50i_iommu_unmap(struct iommu_domain
> > *domain, unsigned long iova> 
> >   	pt_phys = sun50i_dte_get_pt_address(dte);
> >   	pte_addr = (u32 *)phys_to_virt(pt_phys) +
> >   	sun50i_iova_get_pte_index(iova);
> > 
> > -	if (!sun50i_pte_is_page_valid(*pte_addr))
> > -		return 0;
> > +	for (i = 0; i < pages; i++)
> > +		if (!sun50i_pte_is_page_valid(pte_addr[i]))
> > +			return 0;
> > 
> > -	memset(pte_addr, 0, sizeof(*pte_addr));
> > -	sun50i_table_flush(sun50i_domain, pte_addr, 1);
> > +	memset(pte_addr, 0, sizeof(*pte_addr) * pages);
> > +	sun50i_table_flush(sun50i_domain, pte_addr, pages);
> > 
> > -	return SZ_4K;
> > +	return size;
> > 
> >   }
> >   
> >   static phys_addr_t sun50i_iommu_iova_to_phys(struct iommu_domain
> >   *domain,
> > 
> > @@ -827,7 +840,7 @@ static int sun50i_iommu_of_xlate(struct device *dev,
> > 
> >   }
> >   
> >   static const struct iommu_ops sun50i_iommu_ops = {
> > 
> > -	.pgsize_bitmap	= SZ_4K,
> > +	.pgsize_bitmap	= SUN50I_IOMMU_PGSIZES,
> > 
> >   	.device_group	= sun50i_iommu_device_group,
> >   	.domain_alloc	= sun50i_iommu_domain_alloc,
> >   	.of_xlate	= sun50i_iommu_of_xlate,





_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH] iommu/sun50i: Allow page sizes between 4K and 1M

[Robin Murphy]

On 2023-09-11 20:29, Jernej Škrabec wrote:
> On Monday, September 11, 2023 9:06:12 PM CEST Robin Murphy wrote:
>> On 2023-09-11 20:02, Jernej Skrabec wrote:
>>> While peripheral supports only 4K page sizes, we can easily emulate
>>> support for bigger page sizes, up to 1M. This is done by making multiple
>>> entries in map function or clearing multiple entries in unmap.
>>
>> No, it's done by converting over to the {map,unmap}_pages ops. There's
>> no need to resurrect this horrible bodge.
> 
> Missed those, thanks! In any case, logic to implement those isn't that
> different than here.

Indeed, if anything it should be even simpler since you don't need to 
worry about specific sizes at all - any tricky boundary you don't want 
to iterate through, just return short at that point and let the core 
call you back again with the remainder.

(In fact this reminds me one of the low-priority cleanup jobs 
outstanding was to trivially convert all the remaining drivers and 
retire the old interfaces altogether)

Cheers,
Robin.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel