[PATCH 0/2] fix mempool when --no-huge option is set

[PATCH 0/2] fix mempool when --no-huge option is set

[Damien Millescamps]

This set of patch permits to use the --no-huge option to create valid mempools.
However, drivers necessitating a physically contiguous memory larger than a standard
page won't work in this mode.
Only vmxnet3-usermap will work in this mode among the available open source PMD.

This can be useful for old kernels (< 2.6.33) and VM with limited amount of mempory.

Damien Millescamps (2):
  mem: get hugepages config
  mem: fix mempool for --no-huge

 lib/librte_eal/common/include/rte_eal.h  |   13 +++++++-
 lib/librte_eal/linuxapp/eal/eal.c        |    4 ++
 lib/librte_eal/linuxapp/eal/eal_memory.c |    2 +-
 lib/librte_mempool/rte_mempool.c         |   54 +++++++++++++++++++++++++++++-
 lib/librte_mempool/rte_mempool.h         |   20 +++++++----
 5 files changed, 83 insertions(+), 10 deletions(-)

-- 
1.7.2.5

[PATCH 1/2] mem: get hugepages config

[Damien Millescamps]

Allow external libraries and applications to know if hugepages
are enabled.

Signed-off-by: Thomas Monjalon <thomas.monjalon-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
---
 lib/librte_eal/common/include/rte_eal.h |   13 ++++++++++++-
 lib/librte_eal/linuxapp/eal/eal.c       |    4 ++++
 2 files changed, 16 insertions(+), 1 deletions(-)

diff --git a/lib/librte_eal/common/include/rte_eal.h b/lib/librte_eal/common/include/rte_eal.h
index 240530d..58a6f90 100644
--- a/lib/librte_eal/common/include/rte_eal.h
+++ b/lib/librte_eal/common/include/rte_eal.h
@@ -214,7 +214,18 @@ int rte_eal_init(int argc, char **argv);
 	}								\
 } while(0)
 
- 
+/**
+ * Whether EAL is using huge pages (disabled by --no-huge option).
+ * The no-huge mode cannot be used with UIO poll-mode drivers like igb/ixgbe.
+ * It is useful for NIC drivers (e.g. librte_pmd_mlx4, librte_pmd_vmxnet3) or
+ * crypto drivers (e.g. librte_crypto_nitrox) provided by third-parties such
+ * as 6WIND.
+ *
+ * @return
+ *   Nonzero if hugepages are enabled.
+ */
+int rte_eal_has_hugepages(void);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/lib/librte_eal/linuxapp/eal/eal.c b/lib/librte_eal/linuxapp/eal/eal.c
index ed0e9b1..52de2b6 100644
--- a/lib/librte_eal/linuxapp/eal/eal.c
+++ b/lib/librte_eal/linuxapp/eal/eal.c
@@ -945,3 +945,7 @@ rte_eal_process_type(void)
 	return (rte_config.process_type);
 }
 
+int rte_eal_has_hugepages(void)
+{
+	return ! internal_config.no_hugetlbfs;
+}
-- 
1.7.2.5

Re: [PATCH 1/2] mem: get hugepages config

[Adrien Mazarguil]

On Fri, Jul 26, 2013 at 04:39:12PM +0200, Damien Millescamps wrote:
> Allow external libraries and applications to know if hugepages
> are enabled.
> 
> Signed-off-by: Thomas Monjalon <thomas.monjalon-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> ---
>  lib/librte_eal/common/include/rte_eal.h |   13 ++++++++++++-
>  lib/librte_eal/linuxapp/eal/eal.c       |    4 ++++
>  2 files changed, 16 insertions(+), 1 deletions(-)

Acked-by: Adrien Mazarguil <adrien.mazarguil-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>

-- 
Adrien Mazarguil
6WIND

Re: [PATCH 1/2] mem: get hugepages config

[Thomas Monjalon]

26/07/2013 16:59, Adrien Mazarguil :
> On Fri, Jul 26, 2013 at 04:39:12PM +0200, Damien Millescamps wrote:
> > Allow external libraries and applications to know if hugepages
> > are enabled.
> > 
> > Signed-off-by: Thomas Monjalon <thomas.monjalon-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> > Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> > ---
> > 
> >  lib/librte_eal/common/include/rte_eal.h |   13 ++++++++++++-
> >  lib/librte_eal/linuxapp/eal/eal.c       |    4 ++++
> >  2 files changed, 16 insertions(+), 1 deletions(-)
> 
> Acked-by: Adrien Mazarguil <adrien.mazarguil-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>

applied

-- 
Thomas

[PATCH 2/2] mem: fix mempool for --no-huge

[Damien Millescamps]

In --no-huge mode, mempool provides objects with their associated
header/trailer fitting in a standard page (usually 4KB).
This means all non-UIO driver should work correctly in this mode,
since UIO drivers allocate ring sizes that cannot fit in a page.

Extend rte_mempool_virt2phy to obtain the correct physical address when
elements of the pool are not on the same physically contiguous memory region.
This is a first step for enhancement PR #29696.

Reason for this patch is to be able to run on a kernel < 2.6.37 without
the need to patch it, since all kernel below are either bugged or don't
have huge page support at all (< 2.6.28).

Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
---
 lib/librte_eal/linuxapp/eal/eal_memory.c |    2 +-
 lib/librte_mempool/rte_mempool.c         |   54 +++++++++++++++++++++++++++++-
 lib/librte_mempool/rte_mempool.h         |   20 +++++++----
 3 files changed, 67 insertions(+), 9 deletions(-)

diff --git a/lib/librte_eal/linuxapp/eal/eal_memory.c b/lib/librte_eal/linuxapp/eal/eal_memory.c
index ce0c2d8..00b5952 100644
--- a/lib/librte_eal/linuxapp/eal/eal_memory.c
+++ b/lib/librte_eal/linuxapp/eal/eal_memory.c
@@ -780,7 +780,7 @@ rte_eal_hugepage_init(void)
 	/* get pointer to global configuration */
 	mcfg = rte_eal_get_configuration()->mem_config;
 
-	/* for debug purposes, hugetlbfs can be disabled */
+	/* hugetlbfs can be disabled */
 	if (internal_config.no_hugetlbfs) {
 		addr = malloc(internal_config.memory);
 		mcfg->memseg[0].phys_addr = (phys_addr_t)(uintptr_t)addr;
diff --git a/lib/librte_mempool/rte_mempool.c b/lib/librte_mempool/rte_mempool.c
index aa8e76a..87183df 100644
--- a/lib/librte_mempool/rte_mempool.c
+++ b/lib/librte_mempool/rte_mempool.c
@@ -36,6 +36,7 @@
 #include <string.h>
 #include <stdint.h>
 #include <stdarg.h>
+#include <unistd.h>
 #include <inttypes.h>
 #include <errno.h>
 #include <sys/queue.h>
@@ -139,6 +140,8 @@ rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
 	uint32_t header_size, trailer_size;
 	unsigned i;
 	void *obj;
+	void *startaddr;
+	int page_size = getpagesize();
 
 	/* compilation-time checks */
 	RTE_BUILD_BUG_ON((sizeof(struct rte_mempool) &
@@ -227,6 +230,20 @@ rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
 						trailer_size);
 		trailer_size = new_size - header_size - elt_size;
 	}
+	if (! rte_eal_has_hugepages()) {
+		/*
+		 * compute trailer size so that pool elements fit exactly in
+		 * a standard page
+		 */
+		int new_size = page_size - header_size - elt_size;
+		if (new_size < 0 || (unsigned int)new_size < trailer_size) {
+			printf("When hugepages are disabled, pool objects "
+			       "can't exceed PAGE_SIZE: %d + %d + %d > %d\n",
+			       header_size, elt_size, trailer_size, page_size);
+			return NULL;
+		}
+		trailer_size = new_size;
+	}
 
 	/* this is the size of an object, including header and trailer */
 	total_elt_size = header_size + elt_size + trailer_size;
@@ -235,8 +252,31 @@ rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
 	 * cache-aligned */
 	private_data_size = (private_data_size +
 			     CACHE_LINE_MASK) & (~CACHE_LINE_MASK);
+
+	if (! rte_eal_has_hugepages()) {
+		/*
+		 * expand private data size to a whole page, so that the
+		 * first pool element will start on a new standard page
+		 */
+		int head = sizeof(struct rte_mempool);
+		int new_size = (private_data_size + head) % page_size;
+		if (new_size) {
+			private_data_size += page_size - new_size;
+		}
+	}
+
 	mempool_size = total_elt_size * n +
 		sizeof(struct rte_mempool) + private_data_size;
+
+	if (! rte_eal_has_hugepages()) {
+		/*
+		 * we want the memory pool to start on a page boundary,
+		 * because pool elements crossing page boundaries would
+		 * result in discontiguous physical addresses
+		 */
+		mempool_size += page_size;
+	}
+
 	rte_snprintf(mz_name, sizeof(mz_name), "MP_%s", name);
 
 	mz = rte_memzone_reserve(mz_name, mempool_size, socket_id, mz_flags);
@@ -248,8 +288,20 @@ rte_mempool_create(const char *name, unsigned n, unsigned elt_size,
 	if (mz == NULL)
 		goto exit;
 
+	if (rte_eal_has_hugepages()) {
+		startaddr = (void*)mz->addr;
+	} else {
+		/* align memory pool start address on a page boundary */
+		unsigned long addr = (unsigned long)mz->addr;
+		if (addr & (page_size - 1)) {
+			addr += page_size;
+			addr &= ~(page_size - 1);
+		}
+		startaddr = (void*)addr;
+	}
+
 	/* init the mempool structure */
-	mp = mz->addr;
+	mp = startaddr;
 	memset(mp, 0, sizeof(*mp));
 	rte_snprintf(mp->name, sizeof(mp->name), "%s", name);
 	mp->phys_addr = mz->phys_addr;
diff --git a/lib/librte_mempool/rte_mempool.h b/lib/librte_mempool/rte_mempool.h
index 67d7f09..33ce35c 100644
--- a/lib/librte_mempool/rte_mempool.h
+++ b/lib/librte_mempool/rte_mempool.h
@@ -950,16 +950,22 @@ rte_mempool_empty(const struct rte_mempool *mp)
  * @return
  *   The physical address of the elt element.
  */
-static inline phys_addr_t rte_mempool_virt2phy(const struct rte_mempool *mp,
-	const void *elt)
+static inline phys_addr_t
+rte_mempool_virt2phy(const struct rte_mempool *mp, const void *elt)
 {
-	uintptr_t off;
-
-	off = (const char *)elt - (const char *)mp;
-	return mp->phys_addr + off;
+	if (rte_eal_has_hugepages()) {
+		uintptr_t offset = (const char *)elt - (const char *)mp;
+		return mp->phys_addr + offset;
+	} else {
+		/*
+		 * If huge pages are disabled, we cannot assume the
+		 * memory region to be physically contiguous.
+		 * Lookup for each element.
+		 */
+		return rte_mem_virt2phy(elt);
+	}
 }
 
-
 /**
  * Check the consistency of mempool objects.
  *
-- 
1.7.2.5

Re: [PATCH 2/2] mem: fix mempool for --no-huge

[Adrien Mazarguil]

On Fri, Jul 26, 2013 at 04:39:13PM +0200, Damien Millescamps wrote:
> In --no-huge mode, mempool provides objects with their associated
> header/trailer fitting in a standard page (usually 4KB).
> This means all non-UIO driver should work correctly in this mode,
> since UIO drivers allocate ring sizes that cannot fit in a page.
> 
> Extend rte_mempool_virt2phy to obtain the correct physical address when
> elements of the pool are not on the same physically contiguous memory region.
> This is a first step for enhancement PR #29696.
> 
> Reason for this patch is to be able to run on a kernel < 2.6.37 without
> the need to patch it, since all kernel below are either bugged or don't
> have huge page support at all (< 2.6.28).
> 
> Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> ---
>  lib/librte_eal/linuxapp/eal/eal_memory.c |    2 +-
>  lib/librte_mempool/rte_mempool.c         |   54 +++++++++++++++++++++++++++++-
>  lib/librte_mempool/rte_mempool.h         |   20 +++++++----
>  3 files changed, 67 insertions(+), 9 deletions(-)

Acked-by: Adrien Mazarguil <adrien.mazarguil-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>

-- 
Adrien Mazarguil
6WIND

Re: [PATCH 2/2] mem: fix mempool for --no-huge

[Thomas Monjalon]

26/07/2013 16:59, Adrien Mazarguil :
> On Fri, Jul 26, 2013 at 04:39:13PM +0200, Damien Millescamps wrote:
> > In --no-huge mode, mempool provides objects with their associated
> > header/trailer fitting in a standard page (usually 4KB).
> > This means all non-UIO driver should work correctly in this mode,
> > since UIO drivers allocate ring sizes that cannot fit in a page.
> > 
> > Extend rte_mempool_virt2phy to obtain the correct physical address when
> > elements of the pool are not on the same physically contiguous memory
> > region. This is a first step for enhancement PR #29696.
> > 
> > Reason for this patch is to be able to run on a kernel < 2.6.37 without
> > the need to patch it, since all kernel below are either bugged or don't
> > have huge page support at all (< 2.6.28).
> > 
> > Signed-off-by: Damien Millescamps <damien.millescamps-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>
> > ---
> > 
> >  lib/librte_eal/linuxapp/eal/eal_memory.c |    2 +-
> >  lib/librte_mempool/rte_mempool.c         |   54
> >  +++++++++++++++++++++++++++++- lib/librte_mempool/rte_mempool.h        
> >  |   20 +++++++----
> >  3 files changed, 67 insertions(+), 9 deletions(-)
> 
> Acked-by: Adrien Mazarguil <adrien.mazarguil-pdR9zngts4EAvxtiuMwx3w@public.gmane.org>

applied

-- 
Thomas