Merge branch 'netdev_page_frags'

Alexander Duyck says:

====================
Refactor netdev page frags and move them into mm/

This patch series addresses several things.

First I found an issue in the performance of the pfmemalloc check from
build_skb.  To work around it I have provided a cached copy of pfmemalloc
to be used in __netdev_alloc_skb and __napi_alloc_skb.

Second I moved the page fragment allocation logic into the mm tree and
added functionality for freeing page fragments.  I had to fix igb before I
could do this as it was using a reference to NETDEV_FRAG_PAGE_MAX_SIZE
incorrectly.

Finally I went through and replaced all of the duplicate code that was
calling put_page and replaced it with calls to skb_free_frag.

With these changes in place a simple receive and drop test increased from a
packet rate of 8.9Mpps to 9.8Mpps.  The gains breakdown as follows:

8.9Mpps	Before			9.8Mpps	After
------------------------	------------------------
7.8%	put_compound_page	9.1%	__free_page_frag
3.9%	skb_free_head
1.1%	put_page

4.9%	build_skb		3.8%	__napi_alloc_skb
2.5%	__alloc_rx_skb
1.9%	__napi_alloc_skb
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
David S. Miller 2015-05-12 10:39:27 -04:00
commit 8df2914598
12 changed files with 224 additions and 172 deletions

View File

@ -662,7 +662,7 @@ static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data)
{
if (fp->rx_frag_size)
put_page(virt_to_head_page(data));
skb_free_frag(data);
else
kfree(data);
}

View File

@ -6618,7 +6618,7 @@ static void tg3_tx(struct tg3_napi *tnapi)
static void tg3_frag_free(bool is_frag, void *data)
{
if (is_frag)
put_page(virt_to_head_page(data));
skb_free_frag(data);
else
kfree(data);
}

View File

@ -798,7 +798,7 @@ static void hip04_free_ring(struct net_device *ndev, struct device *d)
for (i = 0; i < RX_DESC_NUM; i++)
if (priv->rx_buf[i])
put_page(virt_to_head_page(priv->rx_buf[i]));
skb_free_frag(priv->rx_buf[i]);
for (i = 0; i < TX_DESC_NUM; i++)
if (priv->tx_skb[i])

View File

@ -2079,11 +2079,6 @@ static void *e1000_alloc_frag(const struct e1000_adapter *a)
return data;
}
static void e1000_free_frag(const void *data)
{
put_page(virt_to_head_page(data));
}
/**
* e1000_clean_rx_ring - Free Rx Buffers per Queue
* @adapter: board private structure
@ -2107,7 +2102,7 @@ static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
adapter->rx_buffer_len,
DMA_FROM_DEVICE);
if (buffer_info->rxbuf.data) {
e1000_free_frag(buffer_info->rxbuf.data);
skb_free_frag(buffer_info->rxbuf.data);
buffer_info->rxbuf.data = NULL;
}
} else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
@ -4594,28 +4589,28 @@ static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
data = e1000_alloc_frag(adapter);
/* Failed allocation, critical failure */
if (!data) {
e1000_free_frag(olddata);
skb_free_frag(olddata);
adapter->alloc_rx_buff_failed++;
break;
}
if (!e1000_check_64k_bound(adapter, data, bufsz)) {
/* give up */
e1000_free_frag(data);
e1000_free_frag(olddata);
skb_free_frag(data);
skb_free_frag(olddata);
adapter->alloc_rx_buff_failed++;
break;
}
/* Use new allocation */
e1000_free_frag(olddata);
skb_free_frag(olddata);
}
buffer_info->dma = dma_map_single(&pdev->dev,
data,
adapter->rx_buffer_len,
DMA_FROM_DEVICE);
if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
e1000_free_frag(data);
skb_free_frag(data);
buffer_info->dma = 0;
adapter->alloc_rx_buff_failed++;
break;
@ -4637,7 +4632,7 @@ static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
adapter->rx_buffer_len,
DMA_FROM_DEVICE);
e1000_free_frag(data);
skb_free_frag(data);
buffer_info->rxbuf.data = NULL;
buffer_info->dma = 0;

View File

@ -4974,6 +4974,7 @@ netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
struct igb_tx_buffer *first;
int tso;
u32 tx_flags = 0;
unsigned short f;
u16 count = TXD_USE_COUNT(skb_headlen(skb));
__be16 protocol = vlan_get_protocol(skb);
u8 hdr_len = 0;
@ -4984,14 +4985,8 @@ netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
* + 1 desc for context descriptor,
* otherwise try next time
*/
if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
unsigned short f;
for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
} else {
count += skb_shinfo(skb)->nr_frags;
}
for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
if (igb_maybe_stop_tx(tx_ring, count + 3)) {
/* this is a hard error */

View File

@ -1359,7 +1359,7 @@ static void *mvneta_frag_alloc(const struct mvneta_port *pp)
static void mvneta_frag_free(const struct mvneta_port *pp, void *data)
{
if (likely(pp->frag_size <= PAGE_SIZE))
put_page(virt_to_head_page(data));
skb_free_frag(data);
else
kfree(data);
}

View File

@ -537,7 +537,7 @@ int netcp_unregister_rxhook(struct netcp_intf *netcp_priv, int order,
static void netcp_frag_free(bool is_frag, void *ptr)
{
if (is_frag)
put_page(virt_to_head_page(ptr));
skb_free_frag(ptr);
else
kfree(ptr);
}

View File

@ -366,6 +366,11 @@ extern void free_pages(unsigned long addr, unsigned int order);
extern void free_hot_cold_page(struct page *page, bool cold);
extern void free_hot_cold_page_list(struct list_head *list, bool cold);
struct page_frag_cache;
extern void *__alloc_page_frag(struct page_frag_cache *nc,
unsigned int fragsz, gfp_t gfp_mask);
extern void __free_page_frag(void *addr);
extern void __free_kmem_pages(struct page *page, unsigned int order);
extern void free_kmem_pages(unsigned long addr, unsigned int order);

View File

@ -226,6 +226,24 @@ struct page_frag {
#endif
};
#define PAGE_FRAG_CACHE_MAX_SIZE __ALIGN_MASK(32768, ~PAGE_MASK)
#define PAGE_FRAG_CACHE_MAX_ORDER get_order(PAGE_FRAG_CACHE_MAX_SIZE)
struct page_frag_cache {
void * va;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
__u16 offset;
__u16 size;
#else
__u32 offset;
#endif
/* we maintain a pagecount bias, so that we dont dirty cache line
* containing page->_count every time we allocate a fragment.
*/
unsigned int pagecnt_bias;
bool pfmemalloc;
};
typedef unsigned long __nocast vm_flags_t;
/*

View File

@ -2128,10 +2128,6 @@ static inline void __skb_queue_purge(struct sk_buff_head *list)
kfree_skb(skb);
}
#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
void *netdev_alloc_frag(unsigned int fragsz);
struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
@ -2186,6 +2182,11 @@ static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
}
static inline void skb_free_frag(void *addr)
{
__free_page_frag(addr);
}
void *napi_alloc_frag(unsigned int fragsz);
struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
unsigned int length, gfp_t gfp_mask);

View File

@ -2966,6 +2966,104 @@ void free_pages(unsigned long addr, unsigned int order)
EXPORT_SYMBOL(free_pages);
/*
* Page Fragment:
* An arbitrary-length arbitrary-offset area of memory which resides
* within a 0 or higher order page. Multiple fragments within that page
* are individually refcounted, in the page's reference counter.
*
* The page_frag functions below provide a simple allocation framework for
* page fragments. This is used by the network stack and network device
* drivers to provide a backing region of memory for use as either an
* sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
*/
static struct page *__page_frag_refill(struct page_frag_cache *nc,
gfp_t gfp_mask)
{
struct page *page = NULL;
gfp_t gfp = gfp_mask;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
__GFP_NOMEMALLOC;
page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
PAGE_FRAG_CACHE_MAX_ORDER);
nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
#endif
if (unlikely(!page))
page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
nc->va = page ? page_address(page) : NULL;
return page;
}
void *__alloc_page_frag(struct page_frag_cache *nc,
unsigned int fragsz, gfp_t gfp_mask)
{
unsigned int size = PAGE_SIZE;
struct page *page;
int offset;
if (unlikely(!nc->va)) {
refill:
page = __page_frag_refill(nc, gfp_mask);
if (!page)
return NULL;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
/* if size can vary use size else just use PAGE_SIZE */
size = nc->size;
#endif
/* Even if we own the page, we do not use atomic_set().
* This would break get_page_unless_zero() users.
*/
atomic_add(size - 1, &page->_count);
/* reset page count bias and offset to start of new frag */
nc->pfmemalloc = page->pfmemalloc;
nc->pagecnt_bias = size;
nc->offset = size;
}
offset = nc->offset - fragsz;
if (unlikely(offset < 0)) {
page = virt_to_page(nc->va);
if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
goto refill;
#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
/* if size can vary use size else just use PAGE_SIZE */
size = nc->size;
#endif
/* OK, page count is 0, we can safely set it */
atomic_set(&page->_count, size);
/* reset page count bias and offset to start of new frag */
nc->pagecnt_bias = size;
offset = size - fragsz;
}
nc->pagecnt_bias--;
nc->offset = offset;
return nc->va + offset;
}
EXPORT_SYMBOL(__alloc_page_frag);
/*
* Frees a page fragment allocated out of either a compound or order 0 page.
*/
void __free_page_frag(void *addr)
{
struct page *page = virt_to_head_page(addr);
if (unlikely(put_page_testzero(page)))
__free_pages_ok(page, compound_order(page));
}
EXPORT_SYMBOL(__free_page_frag);
/*
* alloc_kmem_pages charges newly allocated pages to the kmem resource counter
* of the current memory cgroup.

View File

@ -347,94 +347,18 @@ struct sk_buff *build_skb(void *data, unsigned int frag_size)
}
EXPORT_SYMBOL(build_skb);
struct netdev_alloc_cache {
struct page_frag frag;
/* we maintain a pagecount bias, so that we dont dirty cache line
* containing page->_count every time we allocate a fragment.
*/
unsigned int pagecnt_bias;
};
static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
static DEFINE_PER_CPU(struct netdev_alloc_cache, napi_alloc_cache);
static struct page *__page_frag_refill(struct netdev_alloc_cache *nc,
gfp_t gfp_mask)
{
const unsigned int order = NETDEV_FRAG_PAGE_MAX_ORDER;
struct page *page = NULL;
gfp_t gfp = gfp_mask;
if (order) {
gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
__GFP_NOMEMALLOC;
page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
nc->frag.size = PAGE_SIZE << (page ? order : 0);
}
if (unlikely(!page))
page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
nc->frag.page = page;
return page;
}
static void *__alloc_page_frag(struct netdev_alloc_cache __percpu *cache,
unsigned int fragsz, gfp_t gfp_mask)
{
struct netdev_alloc_cache *nc = this_cpu_ptr(cache);
struct page *page = nc->frag.page;
unsigned int size;
int offset;
if (unlikely(!page)) {
refill:
page = __page_frag_refill(nc, gfp_mask);
if (!page)
return NULL;
/* if size can vary use frag.size else just use PAGE_SIZE */
size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
/* Even if we own the page, we do not use atomic_set().
* This would break get_page_unless_zero() users.
*/
atomic_add(size - 1, &page->_count);
/* reset page count bias and offset to start of new frag */
nc->pagecnt_bias = size;
nc->frag.offset = size;
}
offset = nc->frag.offset - fragsz;
if (unlikely(offset < 0)) {
if (!atomic_sub_and_test(nc->pagecnt_bias, &page->_count))
goto refill;
/* if size can vary use frag.size else just use PAGE_SIZE */
size = NETDEV_FRAG_PAGE_MAX_ORDER ? nc->frag.size : PAGE_SIZE;
/* OK, page count is 0, we can safely set it */
atomic_set(&page->_count, size);
/* reset page count bias and offset to start of new frag */
nc->pagecnt_bias = size;
offset = size - fragsz;
}
nc->pagecnt_bias--;
nc->frag.offset = offset;
return page_address(page) + offset;
}
static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
static DEFINE_PER_CPU(struct page_frag_cache, napi_alloc_cache);
static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
{
struct page_frag_cache *nc;
unsigned long flags;
void *data;
local_irq_save(flags);
data = __alloc_page_frag(&netdev_alloc_cache, fragsz, gfp_mask);
nc = this_cpu_ptr(&netdev_alloc_cache);
data = __alloc_page_frag(nc, fragsz, gfp_mask);
local_irq_restore(flags);
return data;
}
@ -454,7 +378,9 @@ EXPORT_SYMBOL(netdev_alloc_frag);
static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
{
return __alloc_page_frag(&napi_alloc_cache, fragsz, gfp_mask);
struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache);
return __alloc_page_frag(nc, fragsz, gfp_mask);
}
void *napi_alloc_frag(unsigned int fragsz)
@ -463,51 +389,6 @@ void *napi_alloc_frag(unsigned int fragsz)
}
EXPORT_SYMBOL(napi_alloc_frag);
/**
* __alloc_rx_skb - allocate an skbuff for rx
* @length: length to allocate
* @gfp_mask: get_free_pages mask, passed to alloc_skb
* @flags: If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
* allocations in case we have to fallback to __alloc_skb()
* If SKB_ALLOC_NAPI is set, page fragment will be allocated
* from napi_cache instead of netdev_cache.
*
* Allocate a new &sk_buff and assign it a usage count of one. The
* buffer has unspecified headroom built in. Users should allocate
* the headroom they think they need without accounting for the
* built in space. The built in space is used for optimisations.
*
* %NULL is returned if there is no free memory.
*/
static struct sk_buff *__alloc_rx_skb(unsigned int length, gfp_t gfp_mask,
int flags)
{
struct sk_buff *skb = NULL;
unsigned int fragsz = SKB_DATA_ALIGN(length) +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
void *data;
if (sk_memalloc_socks())
gfp_mask |= __GFP_MEMALLOC;
data = (flags & SKB_ALLOC_NAPI) ?
__napi_alloc_frag(fragsz, gfp_mask) :
__netdev_alloc_frag(fragsz, gfp_mask);
if (likely(data)) {
skb = build_skb(data, fragsz);
if (unlikely(!skb))
put_page(virt_to_head_page(data));
}
} else {
skb = __alloc_skb(length, gfp_mask,
SKB_ALLOC_RX, NUMA_NO_NODE);
}
return skb;
}
/**
* __netdev_alloc_skb - allocate an skbuff for rx on a specific device
* @dev: network device to receive on
@ -521,19 +402,52 @@ static struct sk_buff *__alloc_rx_skb(unsigned int length, gfp_t gfp_mask,
*
* %NULL is returned if there is no free memory.
*/
struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
unsigned int length, gfp_t gfp_mask)
struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
gfp_t gfp_mask)
{
struct page_frag_cache *nc;
unsigned long flags;
struct sk_buff *skb;
bool pfmemalloc;
void *data;
length += NET_SKB_PAD;
skb = __alloc_rx_skb(length, gfp_mask, 0);
len += NET_SKB_PAD;
if (likely(skb)) {
skb_reserve(skb, NET_SKB_PAD);
skb->dev = dev;
if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
(gfp_mask & (__GFP_WAIT | GFP_DMA)))
return __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
len = SKB_DATA_ALIGN(len);
if (sk_memalloc_socks())
gfp_mask |= __GFP_MEMALLOC;
local_irq_save(flags);
nc = this_cpu_ptr(&netdev_alloc_cache);
data = __alloc_page_frag(nc, len, gfp_mask);
pfmemalloc = nc->pfmemalloc;
local_irq_restore(flags);
if (unlikely(!data))
return NULL;
skb = __build_skb(data, len);
if (unlikely(!skb)) {
skb_free_frag(data);
return NULL;
}
/* use OR instead of assignment to avoid clearing of bits in mask */
if (pfmemalloc)
skb->pfmemalloc = 1;
skb->head_frag = 1;
skb_reserve(skb, NET_SKB_PAD);
skb->dev = dev;
return skb;
}
EXPORT_SYMBOL(__netdev_alloc_skb);
@ -551,19 +465,43 @@ EXPORT_SYMBOL(__netdev_alloc_skb);
*
* %NULL is returned if there is no free memory.
*/
struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
unsigned int length, gfp_t gfp_mask)
struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
gfp_t gfp_mask)
{
struct page_frag_cache *nc = this_cpu_ptr(&napi_alloc_cache);
struct sk_buff *skb;
void *data;
length += NET_SKB_PAD + NET_IP_ALIGN;
skb = __alloc_rx_skb(length, gfp_mask, SKB_ALLOC_NAPI);
len += NET_SKB_PAD + NET_IP_ALIGN;
if (likely(skb)) {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
skb->dev = napi->dev;
if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
(gfp_mask & (__GFP_WAIT | GFP_DMA)))
return __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
len = SKB_DATA_ALIGN(len);
if (sk_memalloc_socks())
gfp_mask |= __GFP_MEMALLOC;
data = __alloc_page_frag(nc, len, gfp_mask);
if (unlikely(!data))
return NULL;
skb = __build_skb(data, len);
if (unlikely(!skb)) {
skb_free_frag(data);
return NULL;
}
/* use OR instead of assignment to avoid clearing of bits in mask */
if (nc->pfmemalloc)
skb->pfmemalloc = 1;
skb->head_frag = 1;
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
skb->dev = napi->dev;
return skb;
}
EXPORT_SYMBOL(__napi_alloc_skb);
@ -611,10 +549,12 @@ static void skb_clone_fraglist(struct sk_buff *skb)
static void skb_free_head(struct sk_buff *skb)
{
unsigned char *head = skb->head;
if (skb->head_frag)
put_page(virt_to_head_page(skb->head));
skb_free_frag(head);
else
kfree(skb->head);
kfree(head);
}
static void skb_release_data(struct sk_buff *skb)