diff mbox

[U-Boot,v4,18/23] ARM: Implement non-cached memory support

Message ID 1418189127-27407-19-git-send-email-sjg@chromium.org
State Awaiting Upstream
Delegated to: Albert ARIBAUD
Headers show

Commit Message

Simon Glass Dec. 10, 2014, 5:25 a.m. UTC 
From: Thierry Reding <treding@nvidia.com>

Implement an API that can be used by drivers to allocate memory from a
pool that is mapped uncached. This is useful if drivers would otherwise
need to do extensive cache maintenance (or explicitly maintaining the
cache isn't safe).

The API is protected using the new CONFIG_SYS_NONCACHED_MEMORY setting.
Boards can set this to the size to be used for the non-cached area. The
area will typically be right below the malloc() area, but architectures
should take care of aligning the beginning and end of the area to honor
any mapping restrictions. Architectures must also ensure that mappings
established for this area do not overlap with the malloc() area (which
should remain cached for improved performance).

While the API is currently only implemented for ARM v7, it should be
generic enough to allow other architectures to implement it as well.

Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
---

Changes in v4: None
Changes in v3:
- Avoid build error with noncached_init() when the dcache is disabled

 README                        | 19 +++++++++++++++++++
 arch/arm/include/asm/system.h |  5 +++++
 arch/arm/lib/cache.c          | 44 +++++++++++++++++++++++++++++++++++++++++++
 common/board_r.c              | 11 +++++++++++
 4 files changed, 79 insertions(+)

Comments

Simon Glass Dec. 10, 2014, 3:03 p.m. UTC | #1 
Hi Albert,

On 9 December 2014 at 22:25, Simon Glass <sjg@chromium.org> wrote:
> From: Thierry Reding <treding@nvidia.com>
>
> Implement an API that can be used by drivers to allocate memory from a
> pool that is mapped uncached. This is useful if drivers would otherwise
> need to do extensive cache maintenance (or explicitly maintaining the
> cache isn't safe).
>
> The API is protected using the new CONFIG_SYS_NONCACHED_MEMORY setting.
> Boards can set this to the size to be used for the non-cached area. The
> area will typically be right below the malloc() area, but architectures
> should take care of aligning the beginning and end of the area to honor
> any mapping restrictions. Architectures must also ensure that mappings
> established for this area do not overlap with the malloc() area (which
> should remain cached for improved performance).
>
> While the API is currently only implemented for ARM v7, it should be
> generic enough to allow other architectures to implement it as well.
>
> Signed-off-by: Thierry Reding <treding@nvidia.com>
> Signed-off-by: Simon Glass <sjg@chromium.org>
> ---
>
> Changes in v4: None
> Changes in v3:
> - Avoid build error with noncached_init() when the dcache is disabled
>
>  README                        | 19 +++++++++++++++++++
>  arch/arm/include/asm/system.h |  5 +++++
>  arch/arm/lib/cache.c          | 44 +++++++++++++++++++++++++++++++++++++++++++
>  common/board_r.c              | 11 +++++++++++
>  4 files changed, 79 insertions(+)

I think you have applied the other cache patches. Are you OK with this
one? I would like to get this series applied very soon (after DM I2C)
as it's been around since August.

>
> diff --git a/README b/README
> index 4ca04d0..5af345b 100644
> --- a/README
> +++ b/README
> @@ -4007,6 +4007,25 @@ Configuration Settings:
>                 boards which do not use the full malloc in SPL (which is
>                 enabled with CONFIG_SYS_SPL_MALLOC_START).
>
> +- CONFIG_SYS_NONCACHED_MEMORY:
> +               Size of non-cached memory area. This area of memory will be
> +               typically located right below the malloc() area and mapped
> +               uncached in the MMU. This is useful for drivers that would
> +               otherwise require a lot of explicit cache maintenance. For
> +               some drivers it's also impossible to properly maintain the
> +               cache. For example if the regions that need to be flushed
> +               are not a multiple of the cache-line size, *and* padding
> +               cannot be allocated between the regions to align them (i.e.
> +               if the HW requires a contiguous array of regions, and the
> +               size of each region is not cache-aligned), then a flush of
> +               one region may result in overwriting data that hardware has
> +               written to another region in the same cache-line. This can
> +               happen for example in network drivers where descriptors for
> +               buffers are typically smaller than the CPU cache-line (e.g.
> +               16 bytes vs. 32 or 64 bytes).
> +
> +               Non-cached memory is only supported on 32-bit ARM at present.
> +
>  - CONFIG_SYS_BOOTM_LEN:
>                 Normally compressed uImages are limited to an
>                 uncompressed size of 8 MBytes. If this is not enough,
> diff --git a/arch/arm/include/asm/system.h b/arch/arm/include/asm/system.h
> index 61e2914..89f2294 100644
> --- a/arch/arm/include/asm/system.h
> +++ b/arch/arm/include/asm/system.h
> @@ -212,6 +212,11 @@ void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
>   */
>  void mmu_page_table_flush(unsigned long start, unsigned long stop);
>
> +#ifdef CONFIG_SYS_NONCACHED_MEMORY
> +void noncached_init(void);
> +phys_addr_t noncached_alloc(size_t size, size_t align);
> +#endif /* CONFIG_SYS_NONCACHED_MEMORY */
> +
>  #endif /* __ASSEMBLY__ */
>
>  #define arch_align_stack(x) (x)
> diff --git a/arch/arm/lib/cache.c b/arch/arm/lib/cache.c
> index f1c0792..9cedeac 100644
> --- a/arch/arm/lib/cache.c
> +++ b/arch/arm/lib/cache.c
> @@ -8,6 +8,7 @@
>  /* for now: just dummy functions to satisfy the linker */
>
>  #include <common.h>
> +#include <malloc.h>
>
>  __weak void flush_cache(unsigned long start, unsigned long size)
>  {
> @@ -49,3 +50,46 @@ __weak void enable_caches(void)
>  {
>         puts("WARNING: Caches not enabled\n");
>  }
> +
> +#ifdef CONFIG_SYS_NONCACHED_MEMORY
> +/*
> + * Reserve one MMU section worth of address space below the malloc() area that
> + * will be mapped uncached.
> + */
> +static unsigned long noncached_start;
> +static unsigned long noncached_end;
> +static unsigned long noncached_next;
> +
> +void noncached_init(void)
> +{
> +       phys_addr_t start, end;
> +       size_t size;
> +
> +       end = ALIGN(mem_malloc_start, MMU_SECTION_SIZE) - MMU_SECTION_SIZE;
> +       size = ALIGN(CONFIG_SYS_NONCACHED_MEMORY, MMU_SECTION_SIZE);
> +       start = end - size;
> +
> +       debug("mapping memory %pa-%pa non-cached\n", &start, &end);
> +
> +       noncached_start = start;
> +       noncached_end = end;
> +       noncached_next = start;
> +
> +#ifndef CONFIG_SYS_DCACHE_OFF
> +       mmu_set_region_dcache_behaviour(noncached_start, size, DCACHE_OFF);
> +#endif
> +}
> +
> +phys_addr_t noncached_alloc(size_t size, size_t align)
> +{
> +       phys_addr_t next = ALIGN(noncached_next, align);
> +
> +       if (next >= noncached_end || (noncached_end - next) < size)
> +               return 0;
> +
> +       debug("allocated %zu bytes of uncached memory @%pa\n", size, &next);
> +       noncached_next = next + size;
> +
> +       return next;
> +}
> +#endif /* CONFIG_SYS_NONCACHED_MEMORY */
> diff --git a/common/board_r.c b/common/board_r.c
> index 4eb7a02..a301cc2 100644
> --- a/common/board_r.c
> +++ b/common/board_r.c
> @@ -265,6 +265,14 @@ static int initr_malloc(void)
>         return 0;
>  }
>
> +#ifdef CONFIG_SYS_NONCACHED_MEMORY
> +static int initr_noncached(void)
> +{
> +       noncached_init();
> +       return 0;
> +}
> +#endif
> +
>  #ifdef CONFIG_DM
>  static int initr_dm(void)
>  {
> @@ -687,6 +695,9 @@ init_fnc_t init_sequence_r[] = {
>  #endif
>         initr_barrier,
>         initr_malloc,
> +#ifdef CONFIG_SYS_NONCACHED_MEMORY
> +       initr_noncached,
> +#endif
>         bootstage_relocate,
>  #ifdef CONFIG_DM
>         initr_dm,
> --
> 2.2.0.rc0.207.ga3a616c
>

Regards,
Simon
Albert ARIBAUD Dec. 18, 2014, 8:17 p.m. UTC | #2 
Hello Simon,

On Wed, 10 Dec 2014 08:03:29 -0700, Simon Glass <sjg@chromium.org>
wrote:
> Hi Albert,
> 
> On 9 December 2014 at 22:25, Simon Glass <sjg@chromium.org> wrote:
> > From: Thierry Reding <treding@nvidia.com>
> >
> > Implement an API that can be used by drivers to allocate memory from a
> > pool that is mapped uncached. This is useful if drivers would otherwise
> > need to do extensive cache maintenance (or explicitly maintaining the
> > cache isn't safe).
> >
> > The API is protected using the new CONFIG_SYS_NONCACHED_MEMORY setting.
> > Boards can set this to the size to be used for the non-cached area. The
> > area will typically be right below the malloc() area, but architectures
> > should take care of aligning the beginning and end of the area to honor
> > any mapping restrictions. Architectures must also ensure that mappings
> > established for this area do not overlap with the malloc() area (which
> > should remain cached for improved performance).
> >
> > While the API is currently only implemented for ARM v7, it should be
> > generic enough to allow other architectures to implement it as well.
> >
> > Signed-off-by: Thierry Reding <treding@nvidia.com>
> > Signed-off-by: Simon Glass <sjg@chromium.org>
> > ---
> >
> > Changes in v4: None
> > Changes in v3:
> > - Avoid build error with noncached_init() when the dcache is disabled
> >
> >  README                        | 19 +++++++++++++++++++
> >  arch/arm/include/asm/system.h |  5 +++++
> >  arch/arm/lib/cache.c          | 44 +++++++++++++++++++++++++++++++++++++++++++
> >  common/board_r.c              | 11 +++++++++++
> >  4 files changed, 79 insertions(+)
> 
> I think you have applied the other cache patches. Are you OK with this
> one? I would like to get this series applied very soon (after DM I2C)
> as it's been around since August.

This is delegated to Tom Warren in Patchwork. I'm delegating it to
myself and applying it; if Tom has applied it too, this should not
cause a major merge issue.

Amicalement,
Simon Glass Dec. 18, 2014, 8:39 p.m. UTC | #3 
Hi Albert,

On 18 December 2014 at 13:17, Albert ARIBAUD <albert.u.boot@aribaud.net> wrote:
> Hello Simon,
>
> On Wed, 10 Dec 2014 08:03:29 -0700, Simon Glass <sjg@chromium.org>
> wrote:
>> Hi Albert,
>>
>> On 9 December 2014 at 22:25, Simon Glass <sjg@chromium.org> wrote:
>> > From: Thierry Reding <treding@nvidia.com>
>> >
>> > Implement an API that can be used by drivers to allocate memory from a
>> > pool that is mapped uncached. This is useful if drivers would otherwise
>> > need to do extensive cache maintenance (or explicitly maintaining the
>> > cache isn't safe).
>> >
>> > The API is protected using the new CONFIG_SYS_NONCACHED_MEMORY setting.
>> > Boards can set this to the size to be used for the non-cached area. The
>> > area will typically be right below the malloc() area, but architectures
>> > should take care of aligning the beginning and end of the area to honor
>> > any mapping restrictions. Architectures must also ensure that mappings
>> > established for this area do not overlap with the malloc() area (which
>> > should remain cached for improved performance).
>> >
>> > While the API is currently only implemented for ARM v7, it should be
>> > generic enough to allow other architectures to implement it as well.
>> >
>> > Signed-off-by: Thierry Reding <treding@nvidia.com>
>> > Signed-off-by: Simon Glass <sjg@chromium.org>
>> > ---
>> >
>> > Changes in v4: None
>> > Changes in v3:
>> > - Avoid build error with noncached_init() when the dcache is disabled
>> >
>> >  README                        | 19 +++++++++++++++++++
>> >  arch/arm/include/asm/system.h |  5 +++++
>> >  arch/arm/lib/cache.c          | 44 +++++++++++++++++++++++++++++++++++++++++++
>> >  common/board_r.c              | 11 +++++++++++
>> >  4 files changed, 79 insertions(+)
>>
>> I think you have applied the other cache patches. Are you OK with this
>> one? I would like to get this series applied very soon (after DM I2C)
>> as it's been around since August.
>
> This is delegated to Tom Warren in Patchwork. I'm delegating it to
> myself and applying it; if Tom has applied it too, this should not
> cause a major merge issue.

Agreed, sounds good, thanks.

Regards,
Simon
Albert ARIBAUD Dec. 19, 2014, 7:02 a.m. UTC | #4 
Hello Tom,

On Thu, 18 Dec 2014 21:33:09 +0000, Tom Warren <TWarren@nvidia.com>
wrote:
> Albert,
> 
> Please pull u-boot-tegra.git/master into ARM master. ./MAKEALL -s tegra is clean.

I assume this is actually for Tom Rini, as non-core ARM trees now merge
directly into mainline.1

Amicalement,
Simon Glass Dec. 29, 2014, 7:39 p.m. UTC | #5 
Hi Albert,

On 19 December 2014 at 00:02, Albert ARIBAUD <albert.u.boot@aribaud.net> wrote:
> Hello Tom,
>
> On Thu, 18 Dec 2014 21:33:09 +0000, Tom Warren <TWarren@nvidia.com>
> wrote:
>> Albert,
>>
>> Please pull u-boot-tegra.git/master into ARM master. ./MAKEALL -s tegra is clean.
>
> I assume this is actually for Tom Rini, as non-core ARM trees now merge
> directly into mainline.1

Yes I think Tom Rini will apply this when he gets back.

Regards,
Simon
diff mbox

Patch

diff --git a/README b/README
index 4ca04d0..5af345b 100644
--- a/README
+++ b/README
@@ -4007,6 +4007,25 @@  Configuration Settings:
 		boards which do not use the full malloc in SPL (which is
 		enabled with CONFIG_SYS_SPL_MALLOC_START).
 
+- CONFIG_SYS_NONCACHED_MEMORY:
+		Size of non-cached memory area. This area of memory will be
+		typically located right below the malloc() area and mapped
+		uncached in the MMU. This is useful for drivers that would
+		otherwise require a lot of explicit cache maintenance. For
+		some drivers it's also impossible to properly maintain the
+		cache. For example if the regions that need to be flushed
+		are not a multiple of the cache-line size, *and* padding
+		cannot be allocated between the regions to align them (i.e.
+		if the HW requires a contiguous array of regions, and the
+		size of each region is not cache-aligned), then a flush of
+		one region may result in overwriting data that hardware has
+		written to another region in the same cache-line. This can
+		happen for example in network drivers where descriptors for
+		buffers are typically smaller than the CPU cache-line (e.g.
+		16 bytes vs. 32 or 64 bytes).
+
+		Non-cached memory is only supported on 32-bit ARM at present.
+
 - CONFIG_SYS_BOOTM_LEN:
 		Normally compressed uImages are limited to an
 		uncompressed size of 8 MBytes. If this is not enough,
diff --git a/arch/arm/include/asm/system.h b/arch/arm/include/asm/system.h
index 61e2914..89f2294 100644
--- a/arch/arm/include/asm/system.h
+++ b/arch/arm/include/asm/system.h
@@ -212,6 +212,11 @@  void mmu_set_region_dcache_behaviour(phys_addr_t start, size_t size,
  */
 void mmu_page_table_flush(unsigned long start, unsigned long stop);
 
+#ifdef CONFIG_SYS_NONCACHED_MEMORY
+void noncached_init(void);
+phys_addr_t noncached_alloc(size_t size, size_t align);
+#endif /* CONFIG_SYS_NONCACHED_MEMORY */
+
 #endif /* __ASSEMBLY__ */
 
 #define arch_align_stack(x) (x)
diff --git a/arch/arm/lib/cache.c b/arch/arm/lib/cache.c
index f1c0792..9cedeac 100644
--- a/arch/arm/lib/cache.c
+++ b/arch/arm/lib/cache.c
@@ -8,6 +8,7 @@ 
 /* for now: just dummy functions to satisfy the linker */
 
 #include <common.h>
+#include <malloc.h>
 
 __weak void flush_cache(unsigned long start, unsigned long size)
 {
@@ -49,3 +50,46 @@  __weak void enable_caches(void)
 {
 	puts("WARNING: Caches not enabled\n");
 }
+
+#ifdef CONFIG_SYS_NONCACHED_MEMORY
+/*
+ * Reserve one MMU section worth of address space below the malloc() area that
+ * will be mapped uncached.
+ */
+static unsigned long noncached_start;
+static unsigned long noncached_end;
+static unsigned long noncached_next;
+
+void noncached_init(void)
+{
+	phys_addr_t start, end;
+	size_t size;
+
+	end = ALIGN(mem_malloc_start, MMU_SECTION_SIZE) - MMU_SECTION_SIZE;
+	size = ALIGN(CONFIG_SYS_NONCACHED_MEMORY, MMU_SECTION_SIZE);
+	start = end - size;
+
+	debug("mapping memory %pa-%pa non-cached\n", &start, &end);
+
+	noncached_start = start;
+	noncached_end = end;
+	noncached_next = start;
+
+#ifndef CONFIG_SYS_DCACHE_OFF
+	mmu_set_region_dcache_behaviour(noncached_start, size, DCACHE_OFF);
+#endif
+}
+
+phys_addr_t noncached_alloc(size_t size, size_t align)
+{
+	phys_addr_t next = ALIGN(noncached_next, align);
+
+	if (next >= noncached_end || (noncached_end - next) < size)
+		return 0;
+
+	debug("allocated %zu bytes of uncached memory @%pa\n", size, &next);
+	noncached_next = next + size;
+
+	return next;
+}
+#endif /* CONFIG_SYS_NONCACHED_MEMORY */
diff --git a/common/board_r.c b/common/board_r.c
index 4eb7a02..a301cc2 100644
--- a/common/board_r.c
+++ b/common/board_r.c
@@ -265,6 +265,14 @@  static int initr_malloc(void)
 	return 0;
 }
 
+#ifdef CONFIG_SYS_NONCACHED_MEMORY
+static int initr_noncached(void)
+{
+	noncached_init();
+	return 0;
+}
+#endif
+
 #ifdef CONFIG_DM
 static int initr_dm(void)
 {
@@ -687,6 +695,9 @@  init_fnc_t init_sequence_r[] = {
 #endif
 	initr_barrier,
 	initr_malloc,
+#ifdef CONFIG_SYS_NONCACHED_MEMORY
+	initr_noncached,
+#endif
 	bootstage_relocate,
 #ifdef CONFIG_DM
 	initr_dm,