netcp: try to reduce type confusion in descriptors
Subsystems (2)
PATCH 1/2 netcp: try to reduce type confusion in descriptors
The netcp driver produces tons of warnings when CONFIG_LPAE is enabled on ARM:
drivers/net/ethernet/ti/netcp_core.c: In function 'netcp_tx_map_skb':drivers/net/ethernet/ti/netcp_core.c:1084:13: warning: passing argument 1 of 'set_words' from incompatible pointer type [-Wincompatible-pointer-types]
This is the result of trying to pass a pointer to a dma_addr_t to a function that expects a u32 pointer to copy that into a DMA descriptor.
Looking at that code in more detail to fix the warnings, I see multiple related problems:
* The conversion functions are not endian-safe, as the DMA descriptors are almost certainly fixed-endian, but the CPU is not.
* On 64-bit machines, passing a pointer through a u32 variable is a bug, accessing an indirect pointer as a u32 pointer even more so.
* The handling of epib and psdata mixes native-endian and device-endian data.
In this patch, I try to sort out the types for most accesses here, adding le32_to_cpu/cpu_to_le32 where appropriate, and passing pointers through two 32-bit words in the descriptor padding, to make it plausible that the driver does the right thing if compiled for big-endian or 64-bit systems.
--- drivers/net/ethernet/ti/netcp_core.c | 123 ++++++++++++++++++++--------------- include/linux/soc/ti/knav_dma.h | 22 +++---- 2 files changed, 82 insertions(+), 63 deletions(-)
drivers/net/ethernet/ti/netcp_core.c +71 -52
@@ -109,69 +109,80 @@ module_param(netcp_debug_level, int, 0);MODULE_PARM_DESC (netcp_debug_level,"Netcp debug level (NETIF_MSG bits) (0=none,...,16=all)" );/* Helper functions - Get/Set */ -static void get_pkt_info (u32 *buff,u32 *buff_len,u32 *ndesc, +static void get_pkt_info (dma_addr_t *buff,u32 *buff_len,dma_addr_t *ndesc,struct knav_dma_desc *desc) { - *buff_len = desc->buff_len; - *buff = desc->buff; - *ndesc = desc->next_desc; + *buff_len =le32_to_cpu (desc->buff_len); + *buff =le32_to_cpu (desc->buff); + *ndesc =le32_to_cpu (desc->next_desc); } -static void get_pad_info (u32 *pad0,u32 *pad1,struct knav_dma_desc *desc) +static void get_pad_info (u32 *pad0,u32 *pad1,u32 *pad2,struct knav_dma_desc *desc) { - *pad0 = desc->pad[0 ]; - *pad1 = desc->pad[1 ]; + *pad0 =le32_to_cpu (desc->pad[0 ]); + *pad1 =le32_to_cpu (desc->pad[1 ]); + *pad2 =le32_to_cpu (desc->pad[2 ]); } -static void get_org_pkt_info (u32 *buff,u32 *buff_len, +static void get_pad_ptr (void **padptr,struct knav_dma_desc *desc) +{ +u64 pad64; + + pad64 =le32_to_cpu (desc->pad[0 ]) + + ((u64 )le32_to_cpu (desc->pad[1 ]) <<32 ); + *padptr = (void *)(uintptr_t )pad64; +} + +static void get_org_pkt_info (dma_addr_t *buff,u32 *buff_len,struct knav_dma_desc *desc) { - *buff = desc->orig_buff; - *buff_len = desc->orig_len; + *buff =le32_to_cpu (desc->orig_buff); + *buff_len =le32_to_cpu (desc->orig_len); } -static void get_words (u32 *words,int num_words,u32 *desc) +static void get_words (dma_addr_t *words,int num_words,__le32 *desc) {int i;for (i =0 ; i < num_words; i++) - words[i] = desc[i]; + words[i] =le32_to_cpu (desc[i]); } -static void set_pkt_info (u32 buff,u32 buff_len,u32 ndesc, +static void set_pkt_info (dma_addr_t buff,u32 buff_len,u32 ndesc,struct knav_dma_desc *desc) { - desc->buff_len = buff_len; - desc->buff = buff; - desc->next_desc = ndesc; + desc->buff_len =cpu_to_le32 (buff_len); + desc->buff =cpu_to_le32 (buff); + desc->next_desc =cpu_to_le32 (ndesc); }static void set_desc_info (u32 desc_info,u32 pkt_info,struct knav_dma_desc *desc) { - desc->desc_info = desc_info; - desc->packet_info = pkt_info; + desc->desc_info =cpu_to_le32 (desc_info); + desc->packet_info =cpu_to_le32 (pkt_info); } -static void set_pad_info (u32 pad0,u32 pad1,struct knav_dma_desc *desc) +static void set_pad_info (u32 pad0,u32 pad1,u32 pad2,struct knav_dma_desc *desc) { - desc->pad[0 ] = pad0; - desc->pad[1 ] = pad1; + desc->pad[0 ] =cpu_to_le32 (pad0); + desc->pad[1 ] =cpu_to_le32 (pad1); + desc->pad[2 ] =cpu_to_le32 (pad1); } -static void set_org_pkt_info (u32 buff,u32 buff_len, +static void set_org_pkt_info (dma_addr_t buff,u32 buff_len,struct knav_dma_desc *desc) { - desc->orig_buff = buff; - desc->orig_len = buff_len; + desc->orig_buff =cpu_to_le32 (buff); + desc->orig_len =cpu_to_le32 (buff_len); } -static void set_words (u32 *words,int num_words,u32 *desc) +static void set_words (u32 *words,int num_words,__le32 *desc) {int i;for (i =0 ; i < num_words; i++) - desc[i] = words[i]; + desc[i] =cpu_to_le32 (words[i]); }/* Read the e-fuse value as 32 bit values to be endian independent */
@@ -570,7 +581,7 @@ static void netcp_free_rx_desc_chain(struct netcp_intf *netcp,dma_addr_t dma_desc, dma_buf;unsigned int buf_len, dma_sz =sizeof (*ndesc);void *buf_ptr; -u32 tmp; +u32 pad[2 ];get_words (&dma_desc,1 , &desc->next_desc);
@@ -580,14 +591,15 @@ static void netcp_free_rx_desc_chain(struct netcp_intf *netcp,dev_err (netcp->ndev_dev,"failed to unmap Rx desc\n" );break ; } -get_pkt_info (&dma_buf, &tmp, &dma_desc, ndesc); -get_pad_info ((u32 *)&buf_ptr, &tmp, ndesc); +get_pad_ptr (&buf_ptr, ndesc);dma_unmap_page (netcp->dev, dma_buf,PAGE_SIZE ,DMA_FROM_DEVICE );__free_page (buf_ptr);knav_pool_desc_put (netcp->rx_pool, desc); } -get_pad_info ((u32 *)&buf_ptr, &buf_len, desc); +get_pad_info (&pad[0 ], &pad[1 ], &buf_len, desc); + buf_ptr = (void *)(uintptr_t )(pad[0 ] + ((u64 )pad[1 ] <<32 )); +if (buf_ptr)netcp_frag_free (buf_len <=PAGE_SIZE , buf_ptr);knav_pool_desc_put (netcp->rx_pool, desc);
@@ -626,7 +638,6 @@ static int netcp_process_one_rx_packet(struct netcp_intf *netcp)struct netcp_packet p_info;struct sk_buff *skb;void *org_buf_ptr; -u32 tmp; dma_desc =knav_queue_pop (netcp->rx_queue, &dma_sz);if (!dma_desc)
@@ -639,7 +650,7 @@ static int netcp_process_one_rx_packet(struct netcp_intf *netcp) }get_pkt_info (&dma_buff, &buf_len, &dma_desc, desc); -get_pad_info ((u32 *)&org_buf_ptr, &org_buf_len, desc); +get_pad_ptr (&org_buf_ptr, desc);if (unlikely (!org_buf_ptr)) {dev_err (netcp->ndev_dev,"NULL bufptr in desc\n" );
@@ -664,6 +675,7 @@ static int netcp_process_one_rx_packet(struct netcp_intf *netcp)/* Fill in the page fragment list */ while (dma_desc) {struct page *page; +void *ptr; ndesc =knav_pool_desc_unmap (netcp->rx_pool, dma_desc, dma_sz);if (unlikely (!ndesc)) {
@@ -672,14 +684,15 @@ static int netcp_process_one_rx_packet(struct netcp_intf *netcp) }get_pkt_info (&dma_buff, &buf_len, &dma_desc, ndesc); -get_pad_info ((u32 *)&page, &tmp, ndesc); +get_pad_ptr (ptr, ndesc); + page = ptr;if (likely (dma_buff && buf_len && page)) {dma_unmap_page (netcp->dev, dma_buff,PAGE_SIZE ,DMA_FROM_DEVICE ); }else { -dev_err (netcp->ndev_dev,"Bad Rx desc dma_buff(%p), len(%d), page(%p)\n" , - (void *)dma_buff, buf_len, page); +dev_err (netcp->ndev_dev,"Bad Rx desc dma_buff(%pad), len(%d), page(%p)\n" , + &dma_buff, buf_len, page); goto free_desc; }
@@ -750,7 +763,6 @@ static void netcp_free_rx_buf(struct netcp_intf *netcp, int fdq)unsigned int buf_len, dma_sz;dma_addr_t dma;void *buf_ptr; -u32 tmp;/* Allocate descriptor */ while ((dma =knav_queue_pop (netcp->rx_fdq[fdq], &dma_sz))) {
@@ -761,7 +773,7 @@ static void netcp_free_rx_buf(struct netcp_intf *netcp, int fdq) }get_org_pkt_info (&dma, &buf_len, desc); -get_pad_info ((u32 *)&buf_ptr, &tmp, desc); +get_pad_ptr (buf_ptr, desc);if (unlikely (!dma)) {dev_err (netcp->ndev_dev,"NULL orig_buff in desc\n" );
@@ -813,7 +825,7 @@ static int netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq)struct page *page;dma_addr_t dma;void *bufptr; -u32 pad[2 ]; +u32 pad[3 ];/* Allocate descriptor */ hwdesc =knav_pool_desc_get (netcp->rx_pool);
@@ -830,7 +842,7 @@ static int netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq)SKB_DATA_ALIGN (sizeof (struct skb_shared_info )); bufptr =netdev_alloc_frag (primary_buf_len); - pad[1 ] = primary_buf_len; + pad[2 ] = primary_buf_len;if (unlikely (!bufptr)) { dev_warn_ratelimited(netcp->ndev_dev,
@@ -842,7 +854,8 @@ static int netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq)if (unlikely (dma_mapping_error (netcp->dev, dma))) goto fail; - pad[0 ] = (u32 )bufptr; + pad[0 ] =lower_32_bits ((uintptr_t )bufptr); + pad[1 ] =upper_32_bits ((uintptr_t )bufptr); }else {/* Allocate a secondary receive queue entry */
@@ -853,8 +866,9 @@ static int netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq) } buf_len =PAGE_SIZE ; dma =dma_map_page (netcp->dev, page,0 , buf_len,DMA_TO_DEVICE ); - pad[0 ] = (u32 )page; - pad[1 ] =0 ; + pad[0 ] =lower_32_bits (dma); + pad[1 ] =upper_32_bits (dma); + pad[2 ] =0 ; } desc_info =KNAV_DMA_DESC_PS_INFO_IN_DESC ;
@@ -864,7 +878,7 @@ static int netcp_allocate_rx_buf(struct netcp_intf *netcp, int fdq) pkt_info |= (netcp->rx_queue_id &KNAV_DMA_DESC_RETQ_MASK ) <<KNAV_DMA_DESC_RETQ_SHIFT ;set_org_pkt_info (dma, buf_len, hwdesc); -set_pad_info (pad[0 ], pad[1 ], hwdesc); +set_pad_info (pad[0 ], pad[1 ], pad[2 ], hwdesc);set_desc_info (desc_info, pkt_info, hwdesc);/* Push to FDQs */
@@ -935,8 +949,8 @@ static void netcp_free_tx_desc_chain(struct netcp_intf *netcp,dma_unmap_single (netcp->dev, dma_buf, buf_len,DMA_TO_DEVICE ); else - dev_warn(netcp->ndev_dev, "bad Tx desc buf(%p), len(%d)\n", - (void *)dma_buf, buf_len); + dev_warn(netcp->ndev_dev, "bad Tx desc buf(%pad), len(%d)\n", + &dma_buf, buf_len);knav_pool_desc_put (netcp->tx_pool, ndesc); ndesc =NULL ;
@@ -953,11 +967,11 @@ static int netcp_process_tx_compl_packets(struct netcp_intf *netcp, unsigned int budget) {struct knav_dma_desc *desc; +void *ptr;struct sk_buff *skb;unsigned int dma_sz;dma_addr_t dma;int pkts =0 ; -u32 tmp;while (budget--) { dma =knav_queue_pop (netcp->tx_compl_q, &dma_sz);
@@ -970,7 +984,8 @@ static int netcp_process_tx_compl_packets(struct netcp_intf *netcp,continue ; } -get_pad_info ((u32 *)&skb, &tmp, desc); +get_pad_ptr (&ptr, desc); + skb = ptr;netcp_free_tx_desc_chain (netcp, desc, dma_sz);if (!skb) {dev_err (netcp->ndev_dev,"No skb in Tx desc\n" );
@@ -1059,6 +1074,7 @@ netcp_tx_map_skb(struct sk_buff *skb, struct netcp_intf *netcp)u32 page_offset = frag->page_offset;u32 buf_len =skb_frag_size (frag);dma_addr_t desc_dma; +u32 desc_dma_32;u32 pkt_info; dma_addr = dma_map_page(dev, page, page_offset, buf_len,
@@ -1075,13 +1091,13 @@ netcp_tx_map_skb(struct sk_buff *skb, struct netcp_intf *netcp) goto free_descs; } - desc_dma =knav_pool_desc_virt_to_dma (netcp->tx_pool, - (void *)ndesc); + desc_dma =knav_pool_desc_virt_to_dma (netcp->tx_pool, ndesc); pkt_info = (netcp->tx_compl_qid &KNAV_DMA_DESC_RETQ_MASK ) <<KNAV_DMA_DESC_RETQ_SHIFT ;set_pkt_info (dma_addr, buf_len,0 , ndesc); -set_words (&desc_dma,1 , &pdesc->next_desc); + desc_dma_32 = (u32 )desc_dma; +set_words (&desc_dma_32,1 , &pdesc->next_desc); pkt_len += buf_len;if (pdesc != desc) knav_pool_desc_map(netcp->tx_pool, pdesc,
@@ -1173,11 +1189,14 @@ static int netcp_tx_submit_skb(struct netcp_intf *netcp, }set_words (&tmp,1 , &desc->packet_info); -set_words ((u32 *)&skb,1 , &desc->pad[0 ]); + tmp =lower_32_bits ((uintptr_t )&skb); +set_words (&tmp,1 , &desc->pad[0 ]); + tmp =upper_32_bits ((uintptr_t )&skb); +set_words (&tmp,1 , &desc->pad[1 ]);if (tx_pipe->flags &SWITCH_TO_PORT_IN_TAGINFO ) { tmp = tx_pipe->switch_to_port; -set_words ((u32 *)&tmp,1 , &desc->tag_info); +set_words (&tmp,1 , &desc->tag_info); }/* submit packet descriptor */
include/linux/soc/ti/knav_dma.h +11 -11
@@ -144,17 +144,17 @@ struct knav_dma_cfg {* @psdata: Protocol specific */ struct knav_dma_desc { -u32 desc_info; -u32 tag_info; -u32 packet_info; -u32 buff_len; -u32 buff; -u32 next_desc; -u32 orig_len; -u32 orig_buff; -u32 epib[KNAV_DMA_NUM_EPIB_WORDS ]; -u32 psdata[KNAV_DMA_NUM_PS_WORDS ]; -u32 pad[4 ]; +__le32 desc_info; +__le32 tag_info; +__le32 packet_info; +__le32 buff_len; +__le32 buff; +__le32 next_desc; +__le32 orig_len; +__le32 orig_buff; +__le32 epib[KNAV_DMA_NUM_EPIB_WORDS ]; +__le32 psdata[KNAV_DMA_NUM_PS_WORDS ]; +__le32 pad[4 ]; }____cacheline_aligned ;#if IS_ENABLED (CONFIG_KEYSTONE_NAVIGATOR_DMA )
Signature · 1 line
2.1.0.rc2
PATCH 2/2 netcp: add more __le32 annotations
The handling of epib and psdata remains a bit unclear in the driver, as we access the same fields both as CPU-endian and through DMA from the device.
Sparse warns about this:ti/netcp_core.c:1147:21: warning: incorrect type in assignment (different base types)ti/netcp_core.c:1147:21: expected unsigned int [usertype] *[assigned] epibti/netcp_core.c:1147:21: got restricted __le32 *<noident>
This uses __le32 types in a few places and uses __force where the code looks fishy. The previous patch should really have produced the correct behavior, but this second patch is needed to shut up the warnings about it. Ideally it would be slightly rewritten to not need those casts, but I don't dare do that without access to the hardware for proper testing.
--- drivers/net/ethernet/ti/netcp.h | 2 +- drivers/net/ethernet/ti/netcp_core.c | 9 +++++---- 2 files changed, 6 insertions(+), 5 deletions(-)
drivers/net/ethernet/ti/netcp.h +1 -1
@@ -113,7 +113,7 @@ struct netcp_intf {#define NETCP_PSDATA_LEN KNAV_DMA_NUM_PS_WORDSstruct netcp_packet {struct sk_buff *skb; -u32 *epib; +__le32 *epib;u32 *psdata;unsigned int psdata_len;struct netcp_intf *netcp;
drivers/net/ethernet/ti/netcp_core.c +5 -4
@@ -1145,8 +1145,8 @@ static int netcp_tx_submit_skb(struct netcp_intf *netcp, p_info.ts_context =NULL ; p_info.txtstamp_complete =NULL ; p_info.epib = desc->epib; - p_info.psdata = desc->psdata; -memset (p_info.epib,0 ,KNAV_DMA_NUM_EPIB_WORDS *sizeof (u32)); + p_info.psdata = (u32 __force *)desc->psdata; +memset (p_info.epib,0 ,KNAV_DMA_NUM_EPIB_WORDS *sizeof (__le32));/* Find out where to inject the packet for transmission */ list_for_each_entry (tx_hook, &netcp->txhook_list_head, list) {
@@ -1170,11 +1170,12 @@ static int netcp_tx_submit_skb(struct netcp_intf *netcp,/* update descriptor */ if (p_info.psdata_len) { -u32 *psdata = p_info.psdata; +/* psdata points to both native-endian and device-endian data */ +__le32 *psdata = (void __force *)p_info.psdata;memmove (p_info.psdata, p_info.psdata + p_info.psdata_len, p_info.psdata_len); -set_words (psdata, p_info.psdata_len, psdata); +set_words (p_info.psdata, p_info.psdata_len, psdata); tmp |= (p_info.psdata_len &KNAV_DMA_DESC_PSLEN_MASK ) <<KNAV_DMA_DESC_PSLEN_SHIFT ; }
Signature · 1 line
2.1.0.rc2
Re: [PATCH 2/2] netcp: add more __le32 annotations
From: Arnd Bergmann <[email protected]>
Date: Tue, 08 Dec 2015 16:32:59 +0100
Quoted 9 lines
The handling of epib and psdata remains a bit unclear in the driver, as we access the same fields both as CPU-endian and through DMA from the device.
Sparse warns about this:
ti/netcp_core.c:1147:21: warning: incorrect type in assignment (different base types)ti/netcp_core.c:1147:21: expected unsigned int [usertype] *[assigned] epibti/netcp_core.c:1147:21: got restricted__le32*<noident>
Quoted 8 lines
This uses
__le32types in a few places and uses__forcewhere the code looks fishy. The previous patch should really have produced the correct behavior, but this second patch is needed to shut up the warnings about it. Ideally it would be slightly rewritten to not need those casts, but I don't dare do that without access to the hardware for proper testing.Signed-off-by Arnd Bergmann <[email protected]>
Applied to net-next.
From: Arnd Bergmann <[email protected]>
Date: Tue, 08 Dec 2015 16:32:27 +0100
Quoted 21 lines
The netcp driver produces tons of warnings when
CONFIG_LPAEis enabled on ARM:
drivers/net/ethernet/ti/netcp_core.c: In function 'netcp_tx_map_skb':drivers/net/ethernet/ti/netcp_core.c:1084:13: warning: passing argument 1 of 'set_words' from incompatible pointer type [-Wincompatible-pointer-types]This is the result of trying to pass a pointer to a
dma_addr_tto a function that expects a u32 pointer to copy that into a DMA descriptor.Looking at that code in more detail to fix the warnings, I see multiple related problems:
* The conversion functions are not endian-safe, as the DMA descriptors are almost certainly fixed-endian, but the CPU is not.
* On 64-bit machines, passing a pointer through a u32 variable is a bug, accessing an indirect pointer as a u32 pointer even more so.
* The handling of epib and psdata mixes native-endian and device-endian data.
Quoted 7 lines
In this patch, I try to sort out the types for most accesses here, adding le32_to_cpu/cpu_to_le32 where appropriate, and passing pointers through two 32-bit words in the descriptor padding, to make it plausible that the driver does the right thing if compiled for big-endian or 64-bit systems.
Signed-off-by Arnd Bergmann <[email protected]>
Applied to net-next.