| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20231011182154.20699-1-gerhard@engleder-embedded.com>
Date: Wed, 11 Oct 2023 20:21:54 +0200
From: Gerhard Engleder <gerhard@...leder-embedded.com>
To: netdev@...r.kernel.org
Cc: davem@...emloft.net,
kuba@...nel.org,
edumazet@...gle.com,
pabeni@...hat.com,
Gerhard Engleder <gerhard@...leder-embedded.com>
Subject: [PATCH net-next] tsnep: Inline small fragments within TX descriptor
The tsnep network controller is able to extend the descriptor directly
with data to be transmitted. In this case no TX data DMA address is
necessary. Instead of the TX data DMA address the TX data buffer is
placed at the end of the descriptor.
The descriptor is read with a 64 bytes DMA read by the tsnep network
controller. If the sum of descriptor data and TX data is less than or
equal to 64 bytes, then no additional DMA read is necessary to read the
TX data. Therefore, it makes sense to inline small fragments up to this
limit within the descriptor ring.
Inlined fragments need to be copied to the descriptor ring. On the other
hand DMA mapping is not necessary. At most 40 bytes are copied, so
copying should be faster than DMA mapping.
For A53 1.2 GHz copying takes <100ns and DMA mapping takes >200ns. So
inlining small fragments should result in lower CPU load. Performance
improvement is small. Thus, comparision of CPU load with and without
inlining of small fragments did not show any significant difference.
With this optimization less DMA reads will be done, which decreases the
load of the interconnect.
Signed-off-by: Gerhard Engleder <gerhard@...leder-embedded.com>
---
drivers/net/ethernet/engleder/tsnep_hw.h | 2 +
drivers/net/ethernet/engleder/tsnep_main.c | 103 +++++++++++++++------
2 files changed, 79 insertions(+), 26 deletions(-)
diff --git a/drivers/net/ethernet/engleder/tsnep_hw.h b/drivers/net/ethernet/engleder/tsnep_hw.h
index 55e1caf193a6..64c97eb66f67 100644
--- a/drivers/net/ethernet/engleder/tsnep_hw.h
+++ b/drivers/net/ethernet/engleder/tsnep_hw.h
@@ -181,6 +181,8 @@ struct tsnep_gcl_operation {
#define TSNEP_DESC_SIZE 256
#define TSNEP_DESC_SIZE_DATA_AFTER 2048
#define TSNEP_DESC_OFFSET 128
+#define TSNEP_DESC_SIZE_DATA_AFTER_INLINE (64 - sizeof(struct tsnep_tx_desc) + \
+ sizeof_field(struct tsnep_tx_desc, tx))
#define TSNEP_DESC_OWNER_COUNTER_MASK 0xC0000000
#define TSNEP_DESC_OWNER_COUNTER_SHIFT 30
#define TSNEP_DESC_LENGTH_MASK 0x00003FFF
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index 77a3fcb821c8..f16b6b3e21a6 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -51,12 +51,22 @@
#define TSNEP_COALESCE_USECS_MAX ((ECM_INT_DELAY_MASK >> ECM_INT_DELAY_SHIFT) * \
ECM_INT_DELAY_BASE_US + ECM_INT_DELAY_BASE_US - 1)
-#define TSNEP_TX_TYPE_SKB BIT(0)
-#define TSNEP_TX_TYPE_SKB_FRAG BIT(1)
-#define TSNEP_TX_TYPE_XDP_TX BIT(2)
-#define TSNEP_TX_TYPE_XDP_NDO BIT(3)
-#define TSNEP_TX_TYPE_XDP (TSNEP_TX_TYPE_XDP_TX | TSNEP_TX_TYPE_XDP_NDO)
-#define TSNEP_TX_TYPE_XSK BIT(4)
+/* mapping type */
+#define TSNEP_TX_TYPE_MAP BIT(0)
+#define TSNEP_TX_TYPE_MAP_PAGE BIT(1)
+#define TSNEP_TX_TYPE_INLINE BIT(2)
+/* buffer type */
+#define TSNEP_TX_TYPE_SKB BIT(8)
+#define TSNEP_TX_TYPE_SKB_MAP (TSNEP_TX_TYPE_SKB | TSNEP_TX_TYPE_MAP)
+#define TSNEP_TX_TYPE_SKB_INLINE (TSNEP_TX_TYPE_SKB | TSNEP_TX_TYPE_INLINE)
+#define TSNEP_TX_TYPE_SKB_FRAG BIT(9)
+#define TSNEP_TX_TYPE_SKB_FRAG_MAP_PAGE (TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_MAP_PAGE)
+#define TSNEP_TX_TYPE_SKB_FRAG_INLINE (TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_INLINE)
+#define TSNEP_TX_TYPE_XDP_TX BIT(10)
+#define TSNEP_TX_TYPE_XDP_NDO BIT(11)
+#define TSNEP_TX_TYPE_XDP_NDO_MAP_PAGE (TSNEP_TX_TYPE_XDP_NDO | TSNEP_TX_TYPE_MAP_PAGE)
+#define TSNEP_TX_TYPE_XDP (TSNEP_TX_TYPE_XDP_TX | TSNEP_TX_TYPE_XDP_NDO)
+#define TSNEP_TX_TYPE_XSK BIT(12)
#define TSNEP_XDP_TX BIT(0)
#define TSNEP_XDP_REDIRECT BIT(1)
@@ -416,6 +426,8 @@ static void tsnep_tx_activate(struct tsnep_tx *tx, int index, int length,
entry->properties |= TSNEP_TX_DESC_OWNER_USER_FLAG;
entry->desc->more_properties =
__cpu_to_le32(entry->len & TSNEP_DESC_LENGTH_MASK);
+ if (entry->type & TSNEP_TX_TYPE_INLINE)
+ entry->properties |= TSNEP_TX_DESC_DATA_AFTER_DESC_FLAG;
/* descriptor properties shall be written last, because valid data is
* signaled there
@@ -433,39 +445,79 @@ static int tsnep_tx_desc_available(struct tsnep_tx *tx)
return tx->read - tx->write - 1;
}
+static int tsnep_tx_map_frag(skb_frag_t *frag, struct tsnep_tx_entry *entry,
+ struct device *dmadev, dma_addr_t *dma)
+{
+ unsigned int len;
+ int mapped;
+
+ len = skb_frag_size(frag);
+ if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
+ *dma = skb_frag_dma_map(dmadev, frag, 0, len, DMA_TO_DEVICE);
+ if (dma_mapping_error(dmadev, *dma))
+ return -ENOMEM;
+ entry->type = TSNEP_TX_TYPE_SKB_FRAG_MAP_PAGE;
+ mapped = 1;
+ } else {
+ void *fragdata = skb_frag_address_safe(frag);
+
+ if (likely(fragdata)) {
+ memcpy(&entry->desc->tx, fragdata, len);
+ } else {
+ struct page *page = skb_frag_page(frag);
+
+ fragdata = kmap_local_page(page);
+ memcpy(&entry->desc->tx, fragdata + skb_frag_off(frag),
+ len);
+ kunmap_local(fragdata);
+ }
+ entry->type = TSNEP_TX_TYPE_SKB_FRAG_INLINE;
+ mapped = 0;
+ }
+
+ return mapped;
+}
+
static int tsnep_tx_map(struct sk_buff *skb, struct tsnep_tx *tx, int count)
{
struct device *dmadev = tx->adapter->dmadev;
struct tsnep_tx_entry *entry;
unsigned int len;
- dma_addr_t dma;
int map_len = 0;
- int i;
+ dma_addr_t dma;
+ int i, mapped;
for (i = 0; i < count; i++) {
entry = &tx->entry[(tx->write + i) & TSNEP_RING_MASK];
if (!i) {
len = skb_headlen(skb);
- dma = dma_map_single(dmadev, skb->data, len,
- DMA_TO_DEVICE);
-
- entry->type = TSNEP_TX_TYPE_SKB;
+ if (likely(len > TSNEP_DESC_SIZE_DATA_AFTER_INLINE)) {
+ dma = dma_map_single(dmadev, skb->data, len,
+ DMA_TO_DEVICE);
+ if (dma_mapping_error(dmadev, dma))
+ return -ENOMEM;
+ entry->type = TSNEP_TX_TYPE_SKB_MAP;
+ mapped = 1;
+ } else {
+ memcpy(&entry->desc->tx, skb->data, len);
+ entry->type = TSNEP_TX_TYPE_SKB_INLINE;
+ mapped = 0;
+ }
} else {
- len = skb_frag_size(&skb_shinfo(skb)->frags[i - 1]);
- dma = skb_frag_dma_map(dmadev,
- &skb_shinfo(skb)->frags[i - 1],
- 0, len, DMA_TO_DEVICE);
+ skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
- entry->type = TSNEP_TX_TYPE_SKB_FRAG;
+ len = skb_frag_size(frag);
+ mapped = tsnep_tx_map_frag(frag, entry, dmadev, &dma);
+ if (mapped < 0)
+ return mapped;
}
- if (dma_mapping_error(dmadev, dma))
- return -ENOMEM;
entry->len = len;
- dma_unmap_addr_set(entry, dma, dma);
-
- entry->desc->tx = __cpu_to_le64(dma);
+ if (likely(mapped)) {
+ dma_unmap_addr_set(entry, dma, dma);
+ entry->desc->tx = __cpu_to_le64(dma);
+ }
map_len += len;
}
@@ -484,13 +536,12 @@ static int tsnep_tx_unmap(struct tsnep_tx *tx, int index, int count)
entry = &tx->entry[(index + i) & TSNEP_RING_MASK];
if (entry->len) {
- if (entry->type & TSNEP_TX_TYPE_SKB)
+ if (entry->type & TSNEP_TX_TYPE_MAP)
dma_unmap_single(dmadev,
dma_unmap_addr(entry, dma),
dma_unmap_len(entry, len),
DMA_TO_DEVICE);
- else if (entry->type &
- (TSNEP_TX_TYPE_SKB_FRAG | TSNEP_TX_TYPE_XDP_NDO))
+ else if (entry->type & TSNEP_TX_TYPE_MAP_PAGE)
dma_unmap_page(dmadev,
dma_unmap_addr(entry, dma),
dma_unmap_len(entry, len),
@@ -586,7 +637,7 @@ static int tsnep_xdp_tx_map(struct xdp_frame *xdpf, struct tsnep_tx *tx,
if (dma_mapping_error(dmadev, dma))
return -ENOMEM;
- entry->type = TSNEP_TX_TYPE_XDP_NDO;
+ entry->type = TSNEP_TX_TYPE_XDP_NDO_MAP_PAGE;
} else {
page = unlikely(frag) ? skb_frag_page(frag) :
virt_to_page(xdpf->data);
--
2.30.2
Powered by blists - more mailing lists