trace: support expression for blob length
PATCH 1/2 trace: support expression for blob length
Support any expression as a blob length by using an intermediate variable in the trace point emitter itself.
This also avoids any side effect on the passed variable.
--- lib/eal/include/rte_trace_point.h | 11 +++++---- lib/eal/include/rte_trace_point_register.h | 3 ++- lib/ethdev/ethdev_trace.h | 27 ++++++---------------- 3 files changed, 15 insertions(+), 26 deletions(-)
lib/eal/include/rte_trace_point.h +6 -5
@@ -392,13 +392,14 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ if (unlikely(in == NULL)) \ return; \ - if (len > RTE_TRACE_BLOB_LEN_MAX) \ - len = RTE_TRACE_BLOB_LEN_MAX; \ - __rte_trace_point_emit(len, uint8_t); \ - memcpy(mem, in, len); \ - memset(RTE_PTR_ADD(mem, len), 0, RTE_TRACE_BLOB_LEN_MAX - len); \ + if (size > RTE_TRACE_BLOB_LEN_MAX) \ + size = RTE_TRACE_BLOB_LEN_MAX; \ + __rte_trace_point_emit(size, uint8_t); \ + memcpy(mem, in, size); \ + memset(RTE_PTR_ADD(mem, size), 0, RTE_TRACE_BLOB_LEN_MAX - size); \ mem = RTE_PTR_ADD(mem, RTE_TRACE_BLOB_LEN_MAX); \ } while (0)
lib/eal/include/rte_trace_point_register.h +2 -1
@@ -51,8 +51,9 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ RTE_SET_USED(in); \ - __rte_trace_point_emit(len, uint8_t); \ + __rte_trace_point_emit(size, uint8_t); \ __rte_trace_point_emit_field(RTE_TRACE_BLOB_LEN_MAX, \ RTE_STR(in)"[" RTE_STR(RTE_TRACE_BLOB_LEN_MAX)"]", \ RTE_STR(uint8_t)); \
lib/ethdev/ethdev_trace.h +7 -20
@@ -860,10 +860,8 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_add,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr,uint32_t pool,int ret), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN );rte_trace_point_emit_u32 (pool);rte_trace_point_emit_int (ret); )
@@ -872,20 +870,16 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_remove,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )RTE_TRACE_POINT ( rte_ethdev_trace_default_mac_addr_set, RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; - rte_trace_point_emit_u16(port_id); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); ) RTE_TRACE_POINT(
@@ -1102,11 +1096,9 @@ RTE_TRACE_POINT( RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr, int ret), - uint8_t len = nb_mc_addr * RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id);rte_trace_point_emit_u32 (nb_mc_addr); -rte_trace_point_emit_blob (mc_addr_set, len); +rte_trace_point_emit_blob (mc_addr_set, nb_mc_addr *RTE_ETHER_ADDR_LEN );rte_trace_point_emit_int (ret); )
@@ -1214,13 +1206,10 @@ RTE_TRACE_POINT( rte_ethdev_trace_get_dcb_info,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_eth_dcb_info *dcb_info,int ret), - uint8_t num_user_priorities = RTE_ETH_DCB_NUM_USER_PRIORITIES; -uint8_t num_tcs =RTE_ETH_DCB_NUM_TCS ; -rte_trace_point_emit_u16 (port_id );rte_trace_point_emit_u8 (dcb_info->nb_tcs); -rte_trace_point_emit_blob (dcb_info->prio_tc, num_user_priorities); -rte_trace_point_emit_blob (dcb_info->tc_bws, num_tcs); +rte_trace_point_emit_blob (dcb_info->prio_tc,RTE_ETH_DCB_NUM_USER_PRIORITIES ); +rte_trace_point_emit_blob (dcb_info->tc_bws,RTE_ETH_DCB_NUM_TCS );rte_trace_point_emit_int (ret); )
@@ -2126,10 +2115,8 @@ RTE_TRACE_POINT_FP( rte_eth_trace_macaddr_get,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *mac_addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (mac_addr->addr_bytes, len); +rte_trace_point_emit_blob (mac_addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )/* Called in loop in examples/ip_pipeline */
Signature · 1 line
2.47.1
PATCH 2/2 dmadev: avoid copies in tracepoints
No need to copy values in intermediate variables.
Just use the right trace point emitters.
--- lib/dmadev/rte_dmadev_trace.h | 20 ++++++-------------- lib/dmadev/rte_dmadev_trace_fp.h | 12 ++++-------- 2 files changed, 10 insertions(+), 22 deletions(-)
lib/dmadev/rte_dmadev_trace.h +6 -14
@@ -46,11 +46,10 @@ RTE_TRACE_POINT(const struct rte_dma_conf __dev_conf = {0 }; dev_conf = &__dev_conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int enable_silent = (int )dev_conf->enable_silent;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority); -rte_trace_point_emit_int (enable_silent); +rte_trace_point_emit_u8 (dev_conf->enable_silent);rte_trace_point_emit_int (ret); )
@@ -83,21 +82,14 @@ RTE_TRACE_POINT(const struct rte_dma_vchan_conf __conf = {0 }; conf = &__conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int src_port_type = conf->src_port.port_type; -int dst_port_type = conf->dst_port.port_type; -int direction = conf->direction; -uint64_t src_pcie_cfg; -uint64_t dst_pcie_cfg;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (direction); +rte_trace_point_emit_int (conf->direction);rte_trace_point_emit_u16 (conf->nb_desc); -rte_trace_point_emit_int (src_port_type); -memcpy (&src_pcie_cfg, &conf->src_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_u64 (src_pcie_cfg); -memcpy (&dst_pcie_cfg, &conf->dst_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_int (dst_port_type); -rte_trace_point_emit_u64 (dst_pcie_cfg); +rte_trace_point_emit_int (conf->src_port.port_type); +rte_trace_point_emit_blob (&conf->src_port.pcie,sizeof (uint64_t )); +rte_trace_point_emit_int (conf->dst_port.port_type); +rte_trace_point_emit_blob (&conf->dst_port.pcie,sizeof (uint64_t ));rte_trace_point_emit_ptr (conf->auto_free.m2d.pool);rte_trace_point_emit_int (ret); )
lib/dmadev/rte_dmadev_trace_fp.h +4 -8
@@ -37,10 +37,9 @@ RTE_TRACE_POINT_FP(enum rte_dma_vchan_status __status =0 ; status = &__status;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int vchan_status = *status;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (vchan_status); +rte_trace_point_emit_int (*status);rte_trace_point_emit_int (ret); )
@@ -107,13 +106,11 @@ RTE_TRACE_POINT_FP( last_idx = &__last_idx; has_error = &__has_error;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int has_error_val = *has_error; -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); -rte_trace_point_emit_int (has_error_val); +rte_trace_point_emit_u16 (*last_idx); +rte_trace_point_emit_u8 (*has_error);rte_trace_point_emit_u16 (ret); )
@@ -126,11 +123,10 @@ RTE_TRACE_POINT_FP(uint16_t __last_idx =0 ; last_idx = &__last_idx;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); +rte_trace_point_emit_u16 (*last_idx);rte_trace_point_emit_ptr (status);rte_trace_point_emit_u16 (ret); )
Signature · 1 line
2.47.1
Quoted 19 lines
-----Original Message-----
From: David Marchand <[email protected]>
Sent: Friday, January 24, 2025 9:44 PM
To: [email protected]
Cc: Jerin Jacob <[email protected]>; Sunil Kumar Kori <[email protected]>; Tyler Retzlaff <[email protected]>; Thomas Monjalon <[email protected]>; Ferruh Yigit <[email protected]>; Andrew Rybchenko <[email protected]>
Subject: [EXTERNAL] [PATCH 1/2] trace: support expression for blob lengthSupport any expression as a blob length by using an intermediate variable in the trace point emitter itself. This also avoids any side effect on the passed variable. Signed-off-by: David Marchand <david. marchand@ redhat. com> --- lib/eal/include/rte_trace_point. h
Support any expression as a blob length by using an intermediate variable in the trace point emitter itself.
Quoted 3 lines
This also avoids any side effect on the passed variable.
Signed-off-by David Marchand <[email protected]>
PATCH v2 1/3 trace: support expression for blob length
In reply to David MarchandSupport any expression as a blob length by using an intermediate variable in the trace point emitter itself.
This also avoids any side effect on the passed variable.
---
Changes since v1:
- removed code relying on arguments in "registering" rte_trace_point_emit_blob implementation (see patch 3),
- moved build check in non registering __rte_trace_point_emit,
--- lib/eal/include/rte_trace_point.h | 12 ++++++---- lib/eal/include/rte_trace_point_register.h | 2 +- lib/ethdev/ethdev_trace.h | 27 ++++++---------------- 3 files changed, 15 insertions(+), 26 deletions(-)
lib/eal/include/rte_trace_point.h +7 -5
@@ -378,6 +378,7 @@ do { \#define __rte_trace_point_emit (in, type) \ do { \ + RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(in))); \ memcpy(mem, &(in), sizeof(in)); \ mem = RTE_PTR_ADD(mem, sizeof(in)); \ } while (0)
@@ -392,13 +393,14 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ if (unlikely(in == NULL)) \ return; \ - if (len > RTE_TRACE_BLOB_LEN_MAX) \ - len = RTE_TRACE_BLOB_LEN_MAX; \ - __rte_trace_point_emit(len, uint8_t); \ - memcpy(mem, in, len); \ - memset(RTE_PTR_ADD(mem, len), 0, RTE_TRACE_BLOB_LEN_MAX - len); \ + if (size > RTE_TRACE_BLOB_LEN_MAX) \ + size = RTE_TRACE_BLOB_LEN_MAX; \ + __rte_trace_point_emit(size, uint8_t); \ + memcpy(mem, in, size); \ + memset(RTE_PTR_ADD(mem, size), 0, RTE_TRACE_BLOB_LEN_MAX - size); \ mem = RTE_PTR_ADD(mem, RTE_TRACE_BLOB_LEN_MAX); \ } while (0)
lib/eal/include/rte_trace_point_register.h +1 -1
@@ -37,7 +37,7 @@ RTE_INIT(trace##_init) \#define __rte_trace_point_emit (in, type) \ do { \ - RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(in))); \ + RTE_SET_USED(in); \ __rte_trace_point_emit_field(sizeof(type), RTE_STR(in), \ RTE_STR(type)); \ } while (0)
lib/ethdev/ethdev_trace.h +7 -20
@@ -860,10 +860,8 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_add,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr,uint32_t pool,int ret), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN );rte_trace_point_emit_u32 (pool);rte_trace_point_emit_int (ret); )
@@ -872,20 +870,16 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_remove,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )RTE_TRACE_POINT ( rte_ethdev_trace_default_mac_addr_set, RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; - rte_trace_point_emit_u16(port_id); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); ) RTE_TRACE_POINT(
@@ -1102,11 +1096,9 @@ RTE_TRACE_POINT( RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr, int ret), - uint8_t len = nb_mc_addr * RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id);rte_trace_point_emit_u32 (nb_mc_addr); -rte_trace_point_emit_blob (mc_addr_set, len); +rte_trace_point_emit_blob (mc_addr_set, nb_mc_addr *RTE_ETHER_ADDR_LEN );rte_trace_point_emit_int (ret); )
@@ -1214,13 +1206,10 @@ RTE_TRACE_POINT( rte_ethdev_trace_get_dcb_info,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_eth_dcb_info *dcb_info,int ret), - uint8_t num_user_priorities = RTE_ETH_DCB_NUM_USER_PRIORITIES; -uint8_t num_tcs =RTE_ETH_DCB_NUM_TCS ; -rte_trace_point_emit_u16 (port_id );rte_trace_point_emit_u8 (dcb_info->nb_tcs); -rte_trace_point_emit_blob (dcb_info->prio_tc, num_user_priorities); -rte_trace_point_emit_blob (dcb_info->tc_bws, num_tcs); +rte_trace_point_emit_blob (dcb_info->prio_tc,RTE_ETH_DCB_NUM_USER_PRIORITIES ); +rte_trace_point_emit_blob (dcb_info->tc_bws,RTE_ETH_DCB_NUM_TCS );rte_trace_point_emit_int (ret); )
@@ -2126,10 +2115,8 @@ RTE_TRACE_POINT_FP( rte_eth_trace_macaddr_get,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *mac_addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (mac_addr->addr_bytes, len); +rte_trace_point_emit_blob (mac_addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )/* Called in loop in examples/ip_pipeline */
Signature · 1 line
2.48.1
PATCH v2 2/3 dmadev: avoid copies in tracepoints
No need to copy values in intermediate variables.
Just use the right trace point emitters.
--- lib/dmadev/rte_dmadev_trace.h | 20 ++++++-------------- lib/dmadev/rte_dmadev_trace_fp.h | 12 ++++-------- 2 files changed, 10 insertions(+), 22 deletions(-)
lib/dmadev/rte_dmadev_trace.h +6 -14
@@ -46,11 +46,10 @@ RTE_TRACE_POINT(const struct rte_dma_conf __dev_conf = {0 }; dev_conf = &__dev_conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int enable_silent = (int )dev_conf->enable_silent;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority); -rte_trace_point_emit_int (enable_silent); +rte_trace_point_emit_u8 (dev_conf->enable_silent);rte_trace_point_emit_int (ret); )
@@ -83,21 +82,14 @@ RTE_TRACE_POINT(const struct rte_dma_vchan_conf __conf = {0 }; conf = &__conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int src_port_type = conf->src_port.port_type; -int dst_port_type = conf->dst_port.port_type; -int direction = conf->direction; -uint64_t src_pcie_cfg; -uint64_t dst_pcie_cfg;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (direction); +rte_trace_point_emit_int (conf->direction);rte_trace_point_emit_u16 (conf->nb_desc); -rte_trace_point_emit_int (src_port_type); -memcpy (&src_pcie_cfg, &conf->src_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_u64 (src_pcie_cfg); -memcpy (&dst_pcie_cfg, &conf->dst_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_int (dst_port_type); -rte_trace_point_emit_u64 (dst_pcie_cfg); +rte_trace_point_emit_int (conf->src_port.port_type); +rte_trace_point_emit_blob (&conf->src_port.pcie,sizeof (uint64_t )); +rte_trace_point_emit_int (conf->dst_port.port_type); +rte_trace_point_emit_blob (&conf->dst_port.pcie,sizeof (uint64_t ));rte_trace_point_emit_ptr (conf->auto_free.m2d.pool);rte_trace_point_emit_int (ret); )
lib/dmadev/rte_dmadev_trace_fp.h +4 -8
@@ -37,10 +37,9 @@ RTE_TRACE_POINT_FP(enum rte_dma_vchan_status __status =0 ; status = &__status;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int vchan_status = *status;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (vchan_status); +rte_trace_point_emit_int (*status);rte_trace_point_emit_int (ret); )
@@ -107,13 +106,11 @@ RTE_TRACE_POINT_FP( last_idx = &__last_idx; has_error = &__has_error;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int has_error_val = *has_error; -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); -rte_trace_point_emit_int (has_error_val); +rte_trace_point_emit_u16 (*last_idx); +rte_trace_point_emit_u8 (*has_error);rte_trace_point_emit_u16 (ret); )
@@ -126,11 +123,10 @@ RTE_TRACE_POINT_FP(uint16_t __last_idx =0 ; last_idx = &__last_idx;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); +rte_trace_point_emit_u16 (*last_idx);rte_trace_point_emit_ptr (status);rte_trace_point_emit_u16 (ret); )
Signature · 1 line
2.48.1
PATCH v2 3/3 trace: fix undefined behavior in register
In reply to David MarchandRegistering a tracepoint handler was resulting so far in undefined behavior at runtime.
The RTE_TRACE_POINT_REGISTER() macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.
Instead, declare an additional inline _register() handler for each tracepoint and make sure that the emitting macros in rte_trace_point_register.h only work on arguments name and type.
The original tracepoint handler prototype is adjusted by adding a __rte_unused for each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.
With this change in place, the workaround in dmadev tracepoints can be removed.
--- lib/dmadev/rte_dmadev_trace.h | 12 ------- lib/dmadev/rte_dmadev_trace_fp.h | 14 -------- lib/eal/include/rte_trace_point.h | 4 +++ lib/eal/include/rte_trace_point_register.h | 42 +++++++++++++++++++--- 4 files changed, 42 insertions(+), 30 deletions(-)
lib/dmadev/rte_dmadev_trace.h +0 -12
@@ -22,10 +22,6 @@ extern "C" {RTE_TRACE_POINT ( rte_dma_trace_info_get, RTE_TRACE_POINT_ARGS(int16_t dev_id, struct rte_dma_info *dev_info), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - struct rte_dma_info __dev_info = {0}; - dev_info = &__dev_info; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16(dev_id);rte_trace_point_emit_string (dev_info->dev_name);rte_trace_point_emit_u64 (dev_info->dev_capa);
@@ -42,10 +38,6 @@ RTE_TRACE_POINT( rte_dma_trace_configure,RTE_TRACE_POINT_ARGS (int16_t dev_id,const struct rte_dma_conf *dev_conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_conf __dev_conf = {0 }; - dev_conf = &__dev_conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority);
@@ -78,10 +70,6 @@ RTE_TRACE_POINT( rte_dma_trace_vchan_setup,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,const struct rte_dma_vchan_conf *conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_vchan_conf __conf = {0 }; - conf = &__conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (conf->direction);
lib/dmadev/rte_dmadev_trace_fp.h +0 -14
@@ -33,10 +33,6 @@ RTE_TRACE_POINT_FP( rte_dma_trace_vchan_status,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,enum rte_dma_vchan_status *status,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -enum rte_dma_vchan_status __status =0 ; - status = &__status; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (*status);
@@ -100,12 +96,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, bool *has_error, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; -bool __has_error = false; - last_idx = &__last_idx; - has_error = &__has_error; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
@@ -119,10 +109,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, enum rte_dma_status_code *status, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; - last_idx = &__last_idx; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
lib/eal/include/rte_trace_point.h +4 -0
@@ -36,6 +36,8 @@ extern "C" {/** The tracepoint object. */ typedef RTE_ATOMIC (uint64_t )rte_trace_point_t ; +#ifndef _RTE_TRACE_POINT_REGISTER_H_ +/** * Macro to define the tracepoint arguments in RTE_TRACE_POINT macro.
@@ -53,6 +55,8 @@ _tp _args \ __VA_ARGS__ \ } +#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ +/** * Create a tracepoint. *
lib/eal/include/rte_trace_point_register.h +38 -4
72 lines not on this page · raw message
Signature · 1 line
2.48.1
Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Quoted 17 lines · On Thu, 30 Jan 2025 15:58:49 +0100 David Marchand <[email protected]>…
Registering a tracepoint handler was resulting so far in undefined behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.Instead, declare an additional inline
_register()handler for each tracepoint and make sure that the emitting macros inrte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.
Quoted 4 lines
With this change in place, the workaround in dmadev tracepoints can be removed.
Signed-off-by David Marchand <[email protected]>
When I build with -Db_santize=undefined the following warning shows up. It seems related.
In function ‘rte_ethdev_trace_get_dcb_info’,
inlined from ‘rte_eth_dev_get_dcb_info’ at ../lib/ethdev/rte_ethdev.c:6720:2:../lib/eal/include/rte_trace_point.h:381:9: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
381 | memcpy(mem, &(in), sizeof(in)); \
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../lib/eal/include/rte_trace_point.h:53:9: note: in definition of macro ‘__RTE_TRACE_POINT’
53 | __VA_ARGS__ \
| ^~~~~~~~~~~
../lib/ethdev/ethdev_trace.h:1213:1: note: in expansion of macro ‘RTE_TRACE_POINT’
1213 | RTE_TRACE_POINT(
| ^~~~~~~~~~~~~~~
../lib/eal/include/rte_trace_point.h:399:9: note: in expansion of macro ‘__rte_trace_point_emit’
399 | __rte_trace_point_emit(len, uint8_t); \
| ^~~~~~~~~~~~~~~~~~~~~~
../lib/ethdev/ethdev_trace.h:1223:9: note: in expansion of macro ‘rte_trace_point_emit_blob’
1223 | rte_trace_point_emit_blob(dcb_info->tc_bws, num_tcs);
| ^~~~~~~~~~~~~~~~~~~~~~~~~In function ‘rte_eth_dev_get_dcb_info’:
cc1: note: destination object is likely at address zero
In function ‘rte_ethdev_trace_get_dcb_info’,
inlined from ‘rte_eth_dev_get_dcb_info’ at ../lib/ethdev/rte_ethdev.c:6720:2:../lib/eal/include/rte_trace_point.h:401:9: warning: ‘memset’ offset [0, 55] is out of the bounds [0, 0] [-Warray-bounds=]
401 | memset(RTE_PTR_ADD(mem, len), 0, RTE_TRACE_BLOB_LEN_MAX - len); \
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../lib/eal/include/rte_trace_point.h:53:9: note: in definition of macro ‘__RTE_TRACE_POINT’
53 | __VA_ARGS__ \
| ^~~~~~~~~~~
../lib/ethdev/ethdev_trace.h:1213:1: note: in expansion of macro ‘RTE_TRACE_POINT’
1213 | RTE_TRACE_POINT(
| ^~~~~~~~~~~~~~~../lib/ethdev/ethdev_trace.h:1222:9: note: in expansion of macro ‘rte_trace_point_emit_blob’
1222 | rte_trace_point_emit_blob(dcb_info->prio_tc, num_user_priorities);
Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Quoted 43 lines · Stephen Hemminger
Quoted 17 lines · On Thu, 30 Jan 2025 15:58:49 +0100 David Marchand <[email protected]>…
Registering a tracepoint handler was resulting so far in undefined behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.Instead, declare an additional inline
_register()handler for each tracepoint and make sure that the emitting macros inrte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.Quoted 4 lines
With this change in place, the workaround in dmadev tracepoints can be removed.
Signed-off-by David Marchand <[email protected]>When I build with -Db_santize=undefined the following warning shows up. It seems related.
In function ‘rte_ethdev_trace_get_dcb_info’,
inlined from ‘rte_eth_dev_get_dcb_info’ at../lib/ethdev/rte_ethdev.c:6720:2:../lib/eal/include/rte_trace_point.h:381:9: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]381 | memcpy(mem, &(in), sizeof(in)); \ | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ../lib/eal/include/rte_trace_point.h:53:9: note: in definition of macro ‘__RTE_TRACE_POINT’ 53 | __VA_ARGS__ \ | ^~~~~~~~~~~ ../lib/ethdev/ethdev_trace.h:1213:1: note: in expansion of macro ‘RTE_TRACE_POINT’ 1213 | RTE_TRACE_POINT( | ^~~~~~~~~~~~~~~ ../lib/eal/include/rte_trace_point.h:399:9: note: in expansion of macro ‘__rte_trace_point_emit’ 399 | __rte_trace_point_emit(len, uint8_t); \ | ^~~~~~~~~~~~~~~~~~~~~~
Quoted 2 lines
../lib/ethdev/ethdev_trace.h:1223:9: note: in expansion of macro ‘rte_trace_point_emit_blob’
1223 |rte_trace_point_emit_blob(dcb_info->tc_bws, num_tcs);
This is something I saw with optimisations (like O2 or O3 iirc) too.
Compiling and running with optimisations made GHA vm go out of memory, so I have been testing only with O0 when it comes to ubsan.
In any case, this series is fixing just one undefined behavior. Fixing all build and runtime errors seen with ubsan requires more work.
Signature · 1 line
David Marchand
Re: [PATCH v2 3/3] trace: fix undefined behavior in register
In reply to David MarchandHello Jerin, Sunil,
Quoted 18 lines · David Marchand
Registering a tracepoint handler was resulting so far in undefined behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.Instead, declare an additional inline
_register()handler for each tracepoint and make sure that the emitting macros inrte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.
Quoted 4 lines
With this change in place, the workaround in dmadev tracepoints can be removed.
Signed-off-by David Marchand <[email protected]>
Can I have your opinion and review on this patch?
Thanks.
Signature · 1 line
David Marchand
RE: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Hi David,
I will look into this at earliest and provide feedback.
Thanks
From: David Marchand <[email protected]>
Sent: Friday, February 7, 2025 2:20 PM
To: Jerin Jacob <[email protected]>; Sunil Kumar Kori <[email protected]>
Cc: [email protected]; Chengwen Feng <[email protected]>; Kevin Laatz <[email protected]>; Bruce Richardson <[email protected]>; Tyler Retzlaff <[email protected]>; Andre Muezerie <[email protected]>; Thomas Monjalon <[email protected]>; Stephen Hemminger <[email protected]>
Subject: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Hello Jerin, Sunil, On Thu, Jan 30, 2025 at 3: 59 PM David Marchand <david. marchand@ redhat. com> wrote: > > Registering a tracepoint handler was resulting so far in undefined > behavior at runtime. > > The RTE_TRACE_POINT_REGISTER()
Hello Jerin, Sunil,
On Thu, Jan 30, 2025 at 3:59 PM David Marchand
Quoted 36 lines · <[email protected]<mailto:[email protected]>> wrote:
Registering a tracepoint handler was resulting so far in undefined
behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler(which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in
reading the current stack with no relation to this function prototype.
Instead, declare an additional inline
_register()handler for eachtracepoint and make sure that the emitting macros in
rte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothingwith them).
This last part introduces an implementation limit of 15 arguments.
Quoted 7 lines
With this change in place, the workaround in dmadev tracepoints can be
removed.
Can I have your opinion and review on this patch?
Thanks.
--
David Marchand
RE: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Hi David,
I tried validating series to see the trace results but babeltrace throws error while reading metadata file and traces can’t be generated. Can you please have a look into this ?
Thanks.
From: Sunil Kumar Kori
Sent: Friday, February 7, 2025 5:09 PM
To: David Marchand <[email protected]>; Jerin Jacob <[email protected]>
Cc: [email protected]; Chengwen Feng <[email protected]>; Kevin Laatz <[email protected]>; Bruce Richardson <[email protected]>; Tyler Retzlaff <[email protected]>; Andre Muezerie <[email protected]>; Thomas Monjalon <[email protected]>; Stephen Hemminger <[email protected]>
Subject: RE: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Hi David,
I will look into this at earliest and provide feedback.
Thanks
From: David Marchand <[email protected]<mailto:[email protected]>>
Sent: Friday, February 7, 2025 2:20 PM
To: Jerin Jacob <[email protected]<mailto:[email protected]>>; Sunil Kumar Kori <[email protected]<mailto:[email protected]>>
Cc: [email protected]<mailto:[email protected]>; Chengwen Feng <[email protected]<mailto:[email protected]>>; Kevin Laatz <[email protected]<mailto:[email protected]>>; Bruce Richardson <[email protected]<mailto:[email protected]>>; Tyler Retzlaff <[email protected]<mailto:[email protected]>>; Andre Muezerie <[email protected]<mailto:[email protected]>>; Thomas Monjalon <[email protected]<mailto:[email protected]>>; Stephen Hemminger <[email protected]<mailto:[email protected]>>
Subject: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Hello Jerin, Sunil, On Thu, Jan 30, 2025 at 3: 59 PM David Marchand <david. marchand@ redhat. com> wrote: > > Registering a tracepoint handler was resulting so far in undefined > behavior at runtime. > > The RTE_TRACE_POINT_REGISTER()
Hello Jerin, Sunil,
On Thu, Jan 30, 2025 at 3:59 PM David Marchand
Quoted 36 lines · <[email protected]<mailto:[email protected]>> wrote:
Registering a tracepoint handler was resulting so far in undefined
behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler(which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in
reading the current stack with no relation to this function prototype.
Instead, declare an additional inline
_register()handler for eachtracepoint and make sure that the emitting macros in
rte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothingwith them).
This last part introduces an implementation limit of 15 arguments.
Quoted 7 lines
With this change in place, the workaround in dmadev tracepoints can be
removed.
Can I have your opinion and review on this patch?
Thanks.
--
David Marchand
Re: [EXTERNAL] Re: [PATCH v2 3/3] trace: fix undefined behavior in register
Quoted 6 lines · Sunil Kumar Kori
Hi David,
I tried validating series to see the trace results but babeltrace throws error while reading metadata file and traces can’t be generated.
Can you please have a look into this ?
Ok, I see what is wrong, in the dma trace update.
Too bad we don't have something in the CI to catch such issue...
Anyway, do you have comments on this last patch of the series?
Signature · 1 line
David Marchand
RE: [EXTERNAL] [PATCH v2 3/3] trace: fix undefined behavior in register
In reply to David MarchandQuoted 19 lines
-----Original Message-----
From: David Marchand <[email protected]>
Sent: Thursday, January 30, 2025 8:29 PM
To: [email protected]
Cc: Chengwen Feng <[email protected]>; Kevin Laatz <[email protected]>; Bruce Richardson <[email protected]>; Jerin Jacob <[email protected]>; Sunil Kumar Kori <[email protected]>; Tyler Retzlaff <[email protected]>
Subject: [EXTERNAL] [PATCH v2 3/3] trace: fix undefined behavior in registerRegistering a tracepoint handler was resulting so far in undefined behavior at runtime. The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void). At runtime, calling this handler while
Registering a tracepoint handler was resulting so far in undefined behavior at runtime.
Quoted 4 lines
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.
To reduce the complexity of N number of variable argument, I thought reading. SP should be fine(Not writing and corrupting it). Looks like some optimization is dmadev Trace not liking this. I am good with your N arguments scheme.
Quoted 50 lines
+#define
RTE_TRACE_POINT_ARGS_COUNT_(_0,_1,_2,_3,_4,_5,_6,_7,_8,_9,_10,_11,_12, _13,_14,_15,N, ...)\lib/eal/include/rte_trace_point_register.h +4
-0+ N +#define RTE_TRACE_POINT_ARGS_COUNT (...) \ + +RTE_TRACE_POINT_ARGS_COUNT_(__VA_ARGS__,15,14,13,12,11,10,9,8,7,6,5,4,3
+,2,1,0)(quoted diff) +31
-0+ +#define RTE_TRACE_POINT_ARGS_1(a) __rte_unused a #define +RTE_TRACE_POINT_ARGS_2(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_1(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_3(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_2(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_4(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_3(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_5(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_4(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_6(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_5(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_7(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_6(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_8(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_7(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_9(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_8(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_10(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_9(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_11(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_10(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_12(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_11(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_13(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_12(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_14(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_13(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_15(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_14(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_FUNC(a) RTE_TRACE_POINT_ARGS_ ## a#define
(quoted diff) +3
-0+RTE_TRACE_POINT_ARGS_EXPAND(...) __VA_ARGS__ #define +RTE_TRACE_POINT_ARGS_(N, ...) \ +RTE_TRACE_POINT_ARGS_EXPAND(RTE_TRACE_POINT_ARGS_FUNC(N))
(__VA_ARGS__)(quoted diff) +2
-0+#define RTE_TRACE_POINT_ARGS(...) \ + (RTE_TRACE_POINT_ARGS_(RTE_TRACE_POINT_ARGS_COUNT(0,
__VA_ARGS__),
Review
Quoted 7 lines
(quoted diff) +7
-0+__VA_ARGS__)) + +#define __RTE_TRACE_POINT(_mode, _tp, _args, ...) \ extern +rte_trace_point_t __##_tp; \ static __rte_always_inline void _tp _args +{ } \ static __rte_always_inline void \ _tp ## _register (void) \ { \ + __rte_trace_point_emit_header_##_mode(&__##_tp); \ + __VA_ARGS__ \
I have not tested this code, Will this calling __rte_trace_point_emit* version of register? If not, CTF spec file not generated properly.
https://github.com/DPDK/dpdk/blob/main/lib/eal/include/rte_trace_point_register.h#L35
Re: [EXTERNAL] [PATCH v2 3/3] trace: fix undefined behavior in register
Quoted 10 lines · Jerin Jacob
lib/eal/include/rte_trace_point_register.h extern "C" { · line 55
Quoted 6 lines
lib/eal/include/rte_trace_point_register.h +6
-0+ +#define __RTE_TRACE_POINT (_mode, _tp, _args, ...) \ extern +rte_trace_point_t __##_tp; \ static __rte_always_inline void _tp _args +{ } \ static __rte_always_inline void \ _tp ## _register (void) \ { \ + __rte_trace_point_emit_header_##_mode(&__##_tp); \ + __VA_ARGS__ \I have not tested this code, Will this calling __rte_trace_point_emit* version of register? If not, CTF spec file not generated properly.
https://github.com/DPDK/dpdk/blob/main/lib/eal/include/rte_trace_point_register.h#L35
Yes, there are issues with this series that I did not catch until recently. This will be fixed in a new revision (that will also add babeltrace calls in CI..).
Signature · 1 line
David Marchand
Quoted 19 lines
-----Original Message-----
From: David Marchand <[email protected]>
Sent: Thursday, January 30, 2025 8:29 PM
To: [email protected]
Cc: Jerin Jacob <[email protected]>; Sunil Kumar Kori <[email protected]>; Tyler Retzlaff <[email protected]>; Thomas Monjalon <[email protected]>; Ferruh Yigit <[email protected]>; Andrew Rybchenko <[email protected]>
Subject: [EXTERNAL] [PATCH v2 1/3] trace: support expression for blob lengthSupport any expression as a blob length by using an intermediate variable in the trace point emitter itself. This also avoids any side effect on the passed variable. Signed-off-by: David Marchand <david. marchand@ redhat. com> --- Changes
Support any expression as a blob length by using an intermediate variable in the trace point emitter itself.
Quoted 3 lines
This also avoids any side effect on the passed variable.
Signed-off-by David Marchand <[email protected]>
v3 cover Trace point framework enhancement for dmadev
In reply to David Marchanddmadev trace points have been working around trace point framework limitations.
One of this workaround blocks enabling dmadev on MSVC.
Traces files were not checked in CI, so this series also adds a new check.
Signature · 27 lines
David Marchand Changes since v2: - added check of traces validity in GHA, - split patch 2 and fixed CTF metadata, David Marchand (6): ci: check traces validity trace: support dereferencing arguments trace: support expression for blob length trace: support dumping binary inside a struct dmadev: avoid copies in tracepoints trace: fix undefined behavior in register .ci/linux-build.sh | 10 +++++ .github/workflows/build.yml | 2 +- lib/dmadev/rte_dmadev_trace.h | 32 +++------------ lib/dmadev/rte_dmadev_trace_fp.h | 26 ++---------- lib/eal/common/eal_common_trace_ctf.c | 39 ++++++++++++------ lib/eal/include/rte_trace_point.h | 16 +++++--- lib/eal/include/rte_trace_point_register.h | 46 +++++++++++++++++++--- lib/ethdev/ethdev_trace.h | 27 ++++--------- 8 files changed, 107 insertions(+), 91 deletions(-) -- 2.48.1
PATCH v3 1/6 ci: check traces validity
We currently do not validate if generated traces can be parsed by a tracing tool.
Add some check.
---
Changes since v2:
- added this check, as v2 was breaking CTF metadata content,
--- .ci/linux-build.sh | 10 ++++++++++ .github/workflows/build.yml | 2 +- 2 files changed, 11 insertions(+), 1 deletion(-)
.ci/linux-build.sh +10 -0
@@ -44,6 +44,14 @@ catch_coredump() { return 1 } +check_traces() { + which babeltrace >/dev/null || return 0 + for file in $(sudo find $HOME -name metadata); do + ! sudo babeltrace $(dirname $file) >/dev/null 2>&1 || continue + sudo babeltrace $(dirname $file) + done +} + cross_file= if [ "$AARCH64" = "true" ]; then
@@ -133,6 +141,7 @@ if [ -z "$cross_file" ]; then configure_coredump devtools/test-null.sh || failed="true" catch_coredump + check_traces [ "$failed" != "true" ] fi
@@ -173,6 +182,7 @@ if [ "$RUN_TESTS" = "true" ]; then configure_coredump sudo meson test -C build --suite fast-tests -t 3 || failed="true" catch_coredump + check_traces [ "$failed" != "true" ] fi
.github/workflows/build.yml +1 -1
@@ -151,7 +151,7 @@ jobs: run: sudo apt install -y crossbuild-essential-riscv64 - name: Install test tools packages if: env.AARCH64 != 'true' || env.PPC64LE != 'true' || env.RISCV64 != 'true' || env.RUN_TESTS == 'true' - run: sudo apt install -y gdb jq + run: sudo apt install -y babeltrace gdb jq - name: Install doc generation packages if: env.BUILD_DOCS == 'true' run: sudo apt install -y doxygen graphviz man-db python3-sphinx
Signature · 1 line
2.48.1
PATCH v3 2/6 trace: support dereferencing arguments
In reply to David MarchandRather than use an intermediate variable, allow use of * to dereference a trace point argument.
Update dmadev traces accordingly (and adjust the emitter type).
---
Changes since v2:
- split this change out of patch 2, as it required updating CTF metadata fixup,
--- lib/dmadev/rte_dmadev_trace_fp.h | 12 +++------ lib/eal/common/eal_common_trace_ctf.c | 35 ++++++++++++++++++--------- 2 files changed, 27 insertions(+), 20 deletions(-)
lib/dmadev/rte_dmadev_trace_fp.h +4 -8
@@ -37,10 +37,9 @@ RTE_TRACE_POINT_FP(enum rte_dma_vchan_status __status =0 ; status = &__status;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int vchan_status = *status;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (vchan_status); +rte_trace_point_emit_int (*status);rte_trace_point_emit_int (ret); )
@@ -107,13 +106,11 @@ RTE_TRACE_POINT_FP( last_idx = &__last_idx; has_error = &__has_error;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int has_error_val = *has_error; -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); -rte_trace_point_emit_int (has_error_val); +rte_trace_point_emit_u16 (*last_idx); +rte_trace_point_emit_u8 (*has_error);rte_trace_point_emit_u16 (ret); )
@@ -126,11 +123,10 @@ RTE_TRACE_POINT_FP(uint16_t __last_idx =0 ; last_idx = &__last_idx;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); +rte_trace_point_emit_u16 (*last_idx);rte_trace_point_emit_ptr (status);rte_trace_point_emit_u16 (ret); )
lib/eal/common/eal_common_trace_ctf.c +23 -12
@@ -373,6 +373,11 @@ rte_trace_metadata_dump(FILE *f)char *trace_metadata_fixup_field (const char *field) { +static const char *const tokens[] = { +"." , +"->" , +"*" , + };const char *ctf_reserved_words[] = {"align" ,"event" ,
@@ -390,23 +395,29 @@ char *trace_metadata_fixup_field(const char *field)return out; } +for (i =0 ; i <RTE_DIM (tokens); i++) { +if (strstr (field, tokens[i]) ==NULL ) +continue ; + goto fixup; + } +/* nothing to replace, return early */ -if (strstr (field,"." ) ==NULL &&strstr (field,"->" ) ==NULL ) -return NULL ; +return NULL ; +fixup: out =strdup (field);if (out ==NULL )return NULL ; - p = out; -while ((p =strstr (p,"." )) !=NULL ) { - p[0 ] ='_' ; - p++; - } - p = out; -while ((p =strstr (p,"->" )) !=NULL ) { - p[0 ] ='_' ; - p++; -memmove (p, p +1 ,strlen (p)); +for (i =0 ; i <RTE_DIM (tokens); i++) { + p = out; +while ((p =strstr (p, tokens[i])) !=NULL ) { + p[0 ] ='_' ; + p++; +if (strlen (tokens[i]) !=1 ) { +memmove (p, p + (strlen (tokens[i]) -1 ), +strlen (p) - (strlen (tokens[i]) -2 )); + } + } }return out; }
Signature · 1 line
2.48.1
RE: [EXTERNAL] [PATCH v3 2/6] trace: support dereferencing arguments
lib/eal/common/eal_common_trace_ctf.c rte_trace_metadata_dump(FILE *f) · line 379
Quoted 13 lines
lib/eal/common/eal_common_trace_ctf.c +5
-0@@ -373,6 +373,11 @@ rte_trace_metadata_dump(FILE *f)char *trace_metadata_fixup_field (const char *field) { +static const char *const tokens[] = { +"." , +"->" , +"*" , + };
As this patch is intended to support dereferencing operation but as a side effect it applies to multiplication operator too. Hence for A * B, field is generated like A___B (one extra underscore). So IMO, multiplication string shouldn't be used.
Quoted 5 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0const char *ctf_reserved_words[] = {"align" ,"event" ,2.48.1
Re: [EXTERNAL] [PATCH v3 2/6] trace: support dereferencing arguments
Quoted 8 lines · Sunil Kumar Kori
Quoted 8 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0@@ -373,6 +373,11 @@ rte_trace_metadata_dump(FILE *f)
Quoted 7 lines
lib/eal/common/eal_common_trace_ctf.c rte_trace_metadata_dump(FILE *f) · line 380
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c +5
-0char *trace_metadata_fixup_field (const char *field) { +static const char *const tokens[] = { +"." , +"->" , +"*" , + };As this patch is intended to support dereferencing operation but as a side effect it applies to multiplication operator too.
Support for multiplication helps with the next patch too, when passing an array size as a blob length.
Quoted 1 line
Hence for A * B, field is generated like
A___B(one extra underscore). So IMO, multiplication string shouldn't be used.
I am not following the consequence.. why should it be prevented?
Signature · 1 line
David Marchand
RE: [EXTERNAL] [PATCH v3 2/6] trace: support dereferencing arguments
Quoted 19 lines
Quoted 8 lines · Sunil Kumar Kori
Quoted 8 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0@@ -373,6 +373,11 @@ rte_trace_metadata_dump(FILE *f)Quoted 7 lines
lib/eal/common/eal_common_trace_ctf.c rte_trace_metadata_dump(FILE *f) · line 380
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c +5
-0char *trace_metadata_fixup_field (const char *field) { +static const char *const tokens[] = { +"." , +"->" , +"*" , + };As this patch is intended to support dereferencing operation but as a side
effect it applies to multiplication operator too.
Quoted 7 lines
Support for multiplication helps with the next patch too, when passing an array size as a blob length.
Quoted 1 line
Hence for A * B, field is generated like
A___B(one extra underscore). SoIMO, multiplication string shouldn't be used.
I am not following the consequence.. why should it be prevented?
As such there is no consequence on functionality. It's just about readability. For example, num_entries * size will result in num_entries___size which actually not representing that this is used to size. Instead tot_size = num_entries * size and then using tot_size clearly specifies that it is the total size of array.
Quoted 4 lines
--
David Marchand
PATCH v3 3/6 trace: support expression for blob length
In reply to David MarchandSupport any expression as a blob length by using an intermediate variable in the trace point emitter itself.
This also avoids any side effect on the passed variable.
On the "register" side, prefix the length variable in the trace metadata with the name of the emitted argument.
With this, we can update ethdev traces and avoid intermediate variables.
---
Changes since v2:
- fixed length field in CTF metadata,
Changes since v1:
- removed code relying on arguments in "registering" rte_trace_point_emit_blob implementation (see patch 3),
- moved build check in non registering __rte_trace_point_emit,
--- lib/eal/common/eal_common_trace_ctf.c | 1 + lib/eal/include/rte_trace_point.h | 12 ++++++---- lib/eal/include/rte_trace_point_register.h | 7 ++++-- lib/ethdev/ethdev_trace.h | 27 ++++++---------------- 4 files changed, 20 insertions(+), 27 deletions(-)
lib/eal/common/eal_common_trace_ctf.c +1 -0
@@ -377,6 +377,7 @@ char *trace_metadata_fixup_field(const char *field)"." ,"->" ,"*" , + " ", };const char *ctf_reserved_words[] = { "align",
lib/eal/include/rte_trace_point.h +7 -5
@@ -378,6 +378,7 @@ do { \#define __rte_trace_point_emit (in, type) \ do { \ + RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(in))); \ memcpy(mem, &(in), sizeof(in)); \ mem = RTE_PTR_ADD(mem, sizeof(in)); \ } while (0)
@@ -392,13 +393,14 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ if (unlikely(in == NULL)) \ return; \ - if (len > RTE_TRACE_BLOB_LEN_MAX) \ - len = RTE_TRACE_BLOB_LEN_MAX; \ - __rte_trace_point_emit(len, uint8_t); \ - memcpy(mem, in, len); \ - memset(RTE_PTR_ADD(mem, len), 0, RTE_TRACE_BLOB_LEN_MAX - len); \ + if (size > RTE_TRACE_BLOB_LEN_MAX) \ + size = RTE_TRACE_BLOB_LEN_MAX; \ + __rte_trace_point_emit(size, uint8_t); \ + memcpy(mem, in, size); \ + memset(RTE_PTR_ADD(mem, size), 0, RTE_TRACE_BLOB_LEN_MAX - size); \ mem = RTE_PTR_ADD(mem, RTE_TRACE_BLOB_LEN_MAX); \ } while (0)
lib/eal/include/rte_trace_point_register.h +5 -2
@@ -37,7 +37,7 @@ RTE_INIT(trace##_init) \#define __rte_trace_point_emit (in, type) \ do { \ - RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(in))); \ + RTE_SET_USED(in); \ __rte_trace_point_emit_field(sizeof(type), RTE_STR(in), \ RTE_STR(type)); \ } while (0)
@@ -52,7 +52,10 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ RTE_SET_USED(in); \ - __rte_trace_point_emit(len, uint8_t); \ + RTE_SET_USED(len); \ + __rte_trace_point_emit_field(sizeof(uint8_t), \ + RTE_STR(in) "_" RTE_STR(len), \ + RTE_STR(uint8_t)); \ __rte_trace_point_emit_field(RTE_TRACE_BLOB_LEN_MAX, \ RTE_STR(in)"[" RTE_STR(RTE_TRACE_BLOB_LEN_MAX)"]", \ RTE_STR(uint8_t)); \
lib/ethdev/ethdev_trace.h +7 -20
@@ -860,10 +860,8 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_add,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr,uint32_t pool,int ret), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN );rte_trace_point_emit_u32 (pool);rte_trace_point_emit_int (ret); )
@@ -872,20 +870,16 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_remove,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )RTE_TRACE_POINT ( rte_ethdev_trace_default_mac_addr_set, RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; - rte_trace_point_emit_u16(port_id); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); ) RTE_TRACE_POINT(
@@ -1102,11 +1096,9 @@ RTE_TRACE_POINT( RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr, int ret), - uint8_t len = nb_mc_addr * RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id);rte_trace_point_emit_u32 (nb_mc_addr); -rte_trace_point_emit_blob (mc_addr_set, len); +rte_trace_point_emit_blob (mc_addr_set, nb_mc_addr *RTE_ETHER_ADDR_LEN );rte_trace_point_emit_int (ret); )
@@ -1214,13 +1206,10 @@ RTE_TRACE_POINT( rte_ethdev_trace_get_dcb_info,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_eth_dcb_info *dcb_info,int ret), - uint8_t num_user_priorities = RTE_ETH_DCB_NUM_USER_PRIORITIES; -uint8_t num_tcs =RTE_ETH_DCB_NUM_TCS ; -rte_trace_point_emit_u16 (port_id );rte_trace_point_emit_u8 (dcb_info->nb_tcs); -rte_trace_point_emit_blob (dcb_info->prio_tc, num_user_priorities); -rte_trace_point_emit_blob (dcb_info->tc_bws, num_tcs); +rte_trace_point_emit_blob (dcb_info->prio_tc,RTE_ETH_DCB_NUM_USER_PRIORITIES ); +rte_trace_point_emit_blob (dcb_info->tc_bws,RTE_ETH_DCB_NUM_TCS );rte_trace_point_emit_int (ret); )
@@ -2126,10 +2115,8 @@ RTE_TRACE_POINT_FP( rte_eth_trace_macaddr_get,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *mac_addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (mac_addr->addr_bytes, len); +rte_trace_point_emit_blob (mac_addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )/* Called in loop in examples/ip_pipeline */
Signature · 1 line
2.48.1
PATCH v3 4/6 trace: support dumping binary inside a struct
In reply to David MarchandMake it possible to dump any point of a structure by accepting & and () in the CTF metadata.
Update dmadev traces accordingly.
---
Changes since v2:
- split this change out of patch 2, as it required updating CTF metadata fixup,
--- lib/dmadev/rte_dmadev_trace.h | 8 ++------ lib/eal/common/eal_common_trace_ctf.c | 3 +++ 2 files changed, 5 insertions(+), 6 deletions(-)
lib/dmadev/rte_dmadev_trace.h +2 -6
@@ -86,18 +86,14 @@ RTE_TRACE_POINT(int src_port_type = conf->src_port.port_type;int dst_port_type = conf->dst_port.port_type;int direction = conf->direction; -uint64_t src_pcie_cfg; -uint64_t dst_pcie_cfg;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (direction);rte_trace_point_emit_u16 (conf->nb_desc);rte_trace_point_emit_int (src_port_type); -memcpy (&src_pcie_cfg, &conf->src_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_u64 (src_pcie_cfg); -memcpy (&dst_pcie_cfg, &conf->dst_port.pcie,sizeof (uint64_t )); +rte_trace_point_emit_blob (&conf->src_port.pcie,sizeof (uint64_t ));rte_trace_point_emit_int (dst_port_type); -rte_trace_point_emit_u64 (dst_pcie_cfg); +rte_trace_point_emit_blob (&conf->dst_port.pcie,sizeof (uint64_t ));rte_trace_point_emit_ptr (conf->auto_free.m2d.pool);rte_trace_point_emit_int (ret); )
lib/eal/common/eal_common_trace_ctf.c +3 -0
@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char *field)"->" ,"*" ," " , +"&" , +"(" , + ")", };const char *ctf_reserved_words[] = { "align",
Signature · 1 line
2.48.1
RE: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char *field) · line 383
Quoted 12 lines
lib/eal/common/eal_common_trace_ctf.c +3
-0@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char *field)"->" ,"*" ," " , +"&" , +"(" , + ")",
Adding brackets makes token names a bit complex. Same name is used in metadata file to dump the traces to the user. With this complex name, user might not understand the purpose of that information.
For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this User might not get that which information is provided.
This is the reason; we followed the existing naming convention which is user friendly.
Quoted 5 lines
lib/eal/common/eal_common_trace_ctf.c +0
-1};const char *ctf_reserved_words[] = {"align" , --2.48.1
Re: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct
Quoted 17 lines · Sunil Kumar Kori
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char *field) · line 383
Quoted 13 lines
lib/eal/common/eal_common_trace_ctf.c +3
-0@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char *field)"->" ,"*" ," " , +"&" , +"(" , + ")",Adding brackets makes token names a bit complex. Same name is used in metadata file to dump the traces to the user. With this complex name, user might not understand the purpose of that information.
Quoted 2 lines
For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this User might not get that which information is provided.
In practice, as there is no other documentation for a trace point arguments, a user needs to read the trace point definitions. So it seems trivial to me to link a variable name in the trace point emitter, and the metadata in the trace files.
Quoted 2 lines
This is the reason; we followed the existing naming convention which is user friendly.
User friendly? I don't see how this is different with '.' and '->'. What is missing is documentation, as nothing describes this sanitisation in the first place (even before this series).
Signature · 1 line
David Marchand
RE: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct
Quoted 30 lines
Quoted 7 lines · Sunil Kumar Kori
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char
Quoted 7 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char*field)
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char *field) · line 383
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c +3
-0"->" ,"*" ," " , +"&" , +"(" , + ")",Quoted 6 lines
Adding brackets makes token names a bit complex. Same name is used in metadata file to dump the traces to the user. With this complex name, user might not understand the purpose of that information.
For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this User might not get that
which information is provided.
In practice, as there is no other documentation for a trace point arguments, a user needs to read the trace point definitions.
So it seems trivial to me to link a variable name in the trace point emitter, and the metadata in the trace files.Quoted 1 line
Quoted 4 lines
Quoted 1 line
This is the reason; we followed the existing naming convention which is user
friendly.
User friendly? I don't see how this is different with '.' and '->'.
In general, structure fields are given a proper name to represent the purpose. When we use it directly in trace point using '.' or '->' then it remains a meaningful name. Adding more tokens in name, is making them complex and deviating from there meaning.
I am not saying that the mentioned support should not be there. I am just trying to convey that If it is possible to make meaningful names, then that will be more helpful.
Quoted 6 lines
What is missing is documentation, as nothing describes this sanitisation in the first place (even before this series).
--
David Marchand
Re: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct
Quoted 34 lines · Sunil Kumar Kori
Quoted 34 lines
Quoted 7 lines · Sunil Kumar Kori
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char
Quoted 7 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char*field)
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char *field) · line 383
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c +3
-0"->" ,"*" ," " , +"&" , +"(" , + ")",Quoted 6 lines
Adding brackets makes token names a bit complex. Same name is used in metadata file to dump the traces to the user. With this complex name, user might not understand the purpose of that information.
For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this User might not get that
which information is provided.
In practice, as there is no other documentation for a trace point arguments, a user needs to read the trace point definitions.
So it seems trivial to me to link a variable name in the trace point emitter, and the metadata in the trace files.Quoted 2 lines
This is the reason; we followed the existing naming convention which is user
friendly.
Quoted 7 lines
Quoted 1 line
User friendly? I don't see how this is different with '.' and '->'.
In general, structure fields are given a proper name to represent the purpose. When we use it directly in trace point using '.' or '->' then it remains a meaningful name. Adding more tokens in name, is making them complex and deviating from there meaning.
I am not saying that the mentioned support should not be there. I am just trying to convey that If it is possible to make meaningful names, then that will be more helpful.
Hard to preserve such information given the limitations of the C parser (which seems to apply to the CTF format).
I still think that interpretation of the metadata in the traces require looking at the source code, which means that the "readability" objection is weak.
Jerin, opinion please.
Signature · 1 line
David Marchand
RE: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a struct
Quoted 65 lines
-----Original Message-----
From: David Marchand <[email protected]>
Sent: Tuesday, February 18, 2025 7:58 PM
To: Sunil Kumar Kori <[email protected]>; Jerin Jacob <[email protected]>
Cc: [email protected]; [email protected]; Kevin Laatz <[email protected]>; Bruce Richardson <[email protected]>; Tyler Retzlaff <[email protected]>
Subject: Re: [EXTERNAL] [PATCH v3 4/6] trace: support dumping binary inside a structOn Wed, Feb 12, 2025 at 6: 14 AM Sunil Kumar Kori <skori@ marvell. com>
wrote: > > > On Tue, Feb 11, 2025 at 9: 53 AM Sunil Kumar Kori <skori@ marvell. com> > > wrote: > > > > > > > diff --git a/lib/eal/common/eal_common_trace_ctf. cQuoted 32 lines · Sunil Kumar Kori
Quoted 32 lines
Quoted 7 lines · Sunil Kumar Kori
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char
Quoted 7 lines
lib/eal/common/eal_common_trace_ctf.c +0
-0@@ -378,6 +378,9 @@ char *trace_metadata_fixup_field(const char*field)
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c char *trace_metadata_fixup_field(const char *field) · line 383
Quoted 6 lines
lib/eal/common/eal_common_trace_ctf.c +3
-0"->" ,"*" ," " , +"&" , +"(" , + ")",Quoted 7 lines
Adding brackets makes token names a bit complex. Same name is used in metadata file to dump the traces to the user. With this complex name, user might not understand the purpose of that information.
For example, _conf_src_port_pcie_sizeof_uint64_t_ is created in metafile and same will be dumped. But with this User might not get that
which information is provided.
In practice, as there is no other documentation for a trace point arguments, a user needs to read the trace point definitions. So it seems trivial to me to link a variable name in the trace point emitter, and the metadata in the trace files.
Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
This is the reason; we followed the existing naming convention which is user
friendly.
User friendly? I don't see how this is different with '.' and '->'.
In general, structure fields are given a proper name to represent the purpose. When we use it directly in trace point using '.' or '->' then it remains a
meaningful name.
Quoted 1 line
Adding more tokens in name, is making them complex and deviating from
there meaning.
Quoted 3 lines
I am not saying that the mentioned support should not be there. I am just trying to convey that If it is possible to make meaningful names, then that
will be more helpful.
Quoted 7 lines
Hard to preserve such information given the limitations of the C parser (which seems to apply to the CTF format).
I still think that interpretation of the metadata in the traces require looking at the source code, which means that the "readability"
objection is weak.Jerin, opinion please.
One side, it reduces the number of lines of code, leveraging all CTF syntax and other side it traces string becomes complex. IMO, We can go with new patch scheme and document the new syntax so that someone parsing babetrace or tracecompass can understand the meaning of _conf_src_port_pcie_sizeof_uint64_t_ (as example)
Quoted 4 lines
--
David Marchand
PATCH v3 5/6 dmadev: avoid copies in tracepoints
In reply to David MarchandNo need to copy values in intermediate variables.
Just use the right trace point emitters.
---
Changes since v2:
- split this change into multiple changes,
only kept trivial parts in this patch,
--- lib/dmadev/rte_dmadev_trace.h | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-)
lib/dmadev/rte_dmadev_trace.h +4 -8
@@ -46,11 +46,10 @@ RTE_TRACE_POINT(const struct rte_dma_conf __dev_conf = {0 }; dev_conf = &__dev_conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int enable_silent = (int )dev_conf->enable_silent;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority); -rte_trace_point_emit_int (enable_silent); +rte_trace_point_emit_u8 (dev_conf->enable_silent);rte_trace_point_emit_int (ret); )
@@ -83,16 +82,13 @@ RTE_TRACE_POINT(const struct rte_dma_vchan_conf __conf = {0 }; conf = &__conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int src_port_type = conf->src_port.port_type; -int dst_port_type = conf->dst_port.port_type; -int direction = conf->direction;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (direction); +rte_trace_point_emit_int (conf->direction);rte_trace_point_emit_u16 (conf->nb_desc); -rte_trace_point_emit_int (src_port_type); +rte_trace_point_emit_int (conf->src_port.port_type);rte_trace_point_emit_blob (&conf->src_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_int (dst_port_type); +rte_trace_point_emit_int (conf->dst_port.port_type);rte_trace_point_emit_blob (&conf->dst_port.pcie,sizeof (uint64_t ));rte_trace_point_emit_ptr (conf->auto_free.m2d.pool);rte_trace_point_emit_int (ret);
Signature · 1 line
2.48.1
PATCH v3 6/6 trace: fix undefined behavior in register
In reply to David MarchandRegistering a tracepoint handler was resulting so far in undefined behavior at runtime.
The RTE_TRACE_POINT_REGISTER() macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.
Instead, declare an additional inline _register() handler for each tracepoint and make sure that the emitting macros in rte_trace_point_register.h only work on arguments name and type.
The original tracepoint handler prototype is adjusted by adding a __rte_unused for each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.
With this change in place, the workaround in dmadev tracepoints can be removed.
--- lib/dmadev/rte_dmadev_trace.h | 12 ------ lib/dmadev/rte_dmadev_trace_fp.h | 14 ------- lib/eal/include/rte_trace_point.h | 4 ++ lib/eal/include/rte_trace_point_register.h | 43 +++++++++++++++++++--- 4 files changed, 42 insertions(+), 31 deletions(-)
lib/dmadev/rte_dmadev_trace.h +0 -12
@@ -22,10 +22,6 @@ extern "C" {RTE_TRACE_POINT ( rte_dma_trace_info_get, RTE_TRACE_POINT_ARGS(int16_t dev_id, struct rte_dma_info *dev_info), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - struct rte_dma_info __dev_info = {0}; - dev_info = &__dev_info; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16(dev_id);rte_trace_point_emit_string (dev_info->dev_name);rte_trace_point_emit_u64 (dev_info->dev_capa);
@@ -42,10 +38,6 @@ RTE_TRACE_POINT( rte_dma_trace_configure,RTE_TRACE_POINT_ARGS (int16_t dev_id,const struct rte_dma_conf *dev_conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_conf __dev_conf = {0 }; - dev_conf = &__dev_conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority);
@@ -78,10 +70,6 @@ RTE_TRACE_POINT( rte_dma_trace_vchan_setup,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,const struct rte_dma_vchan_conf *conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_vchan_conf __conf = {0 }; - conf = &__conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (conf->direction);
lib/dmadev/rte_dmadev_trace_fp.h +0 -14
@@ -33,10 +33,6 @@ RTE_TRACE_POINT_FP( rte_dma_trace_vchan_status,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,enum rte_dma_vchan_status *status,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -enum rte_dma_vchan_status __status =0 ; - status = &__status; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (*status);
@@ -100,12 +96,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, bool *has_error, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; -bool __has_error = false; - last_idx = &__last_idx; - has_error = &__has_error; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
@@ -119,10 +109,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, enum rte_dma_status_code *status, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; - last_idx = &__last_idx; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
lib/eal/include/rte_trace_point.h +4 -0
@@ -36,6 +36,8 @@ extern "C" {/** The tracepoint object. */ typedef RTE_ATOMIC (uint64_t )rte_trace_point_t ; +#ifndef _RTE_TRACE_POINT_REGISTER_H_ +/** * Macro to define the tracepoint arguments in RTE_TRACE_POINT macro.
@@ -53,6 +55,8 @@ _tp _args \ __VA_ARGS__ \ } +#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ +/** * Create a tracepoint. *
lib/eal/include/rte_trace_point_register.h +38 -5
73 lines not on this page · raw message
Signature · 1 line
2.48.1
RE: [EXTERNAL] [PATCH v3 0/6] Trace point framework enhancement for dmadev
In reply to David MarchandHi David,
I validated the series and its functional now. Traces are generated properly.
Quoted 34 lines
dmadev trace points have been working around trace point framework limitations.
One of this workaround blocks enabling dmadev on MSVC.Traces files were not checked in CI, so this series also adds a new check.
--
David MarchandChanges since v2:
- added check of traces validity in GHA,
- split patch 2 and fixed CTF metadata,David Marchand (6):
ci: check traces validity
trace: support dereferencing arguments
trace: support expression for blob length
trace: support dumping binary inside a struct
dmadev: avoid copies in tracepoints
trace: fix undefined behavior in register.ci/linux-build.sh | 10 +++++ .github/workflows/build.yml | 2 +- lib/dmadev/rte_dmadev_trace.h | 32 +++------------ lib/dmadev/rte_dmadev_trace_fp.h | 26 ++---------- lib/eal/common/eal_common_trace_ctf.c | 39 ++++++++++++------ lib/eal/include/rte_trace_point.h | 16 +++++--- lib/eal/include/rte_trace_point_register.h | 46 +++++++++++++++++++--- lib/ethdev/ethdev_trace.h | 27 ++++--------- 8 files changed, 107 insertions(+), 91 deletions(-)--
2.48.1
v4 cover Trace point framework enhancement for dmadev
In reply to David Marchanddmadev trace points have been working around trace point framework limitations.
One of this workaround blocks enabling dmadev on MSVC.
Traces files were not checked in CI, so this series also adds a new check.
Signature · 32 lines
David Marchand Changes since v3: - dropped changes to CTF metadata, - added new helpers for dumping content of pointed at memory, - changed dmadev to avoid more changes in trace framework, Changes since v2: - added check of traces validity in GHA, - split patch 2 and fixed CTF metadata, David Marchand (5): ci: check traces validity dmadev: avoid copies in tracepoints trace: support dereferencing pointers trace: support expression for blob length trace: fix undefined behavior in register .ci/linux-build.sh | 10 ++++ .github/workflows/build.yml | 2 +- devtools/libabigail.abignore | 5 ++ lib/dmadev/rte_dmadev.h | 29 +++++----- lib/dmadev/rte_dmadev_trace.h | 32 ++--------- lib/dmadev/rte_dmadev_trace_fp.h | 26 ++------- lib/eal/include/rte_trace_point.h | 66 ++++++++++++++-------- lib/eal/include/rte_trace_point_register.h | 51 ++++++++++++++--- lib/ethdev/ethdev_trace.h | 27 +++------ 9 files changed, 135 insertions(+), 113 deletions(-) -- 2.48.1
PATCH v4 1/5 ci: check traces validity
We currently do not validate if generated traces can be parsed by a tracing tool.
Add some check.
---
Changes since v2:
- added this check, as v2 was breaking CTF metadata content,
--- .ci/linux-build.sh | 10 ++++++++++ .github/workflows/build.yml | 2 +- 2 files changed, 11 insertions(+), 1 deletion(-)
.ci/linux-build.sh +10 -0
@@ -44,6 +44,14 @@ catch_coredump() { return 1 } +check_traces() { + which babeltrace >/dev/null || return 0 + for file in $(sudo find $HOME -name metadata); do + ! sudo babeltrace $(dirname $file) >/dev/null 2>&1 || continue + sudo babeltrace $(dirname $file) + done +} + cross_file= if [ "$AARCH64" = "true" ]; then
@@ -133,6 +141,7 @@ if [ -z "$cross_file" ]; then configure_coredump devtools/test-null.sh || failed="true" catch_coredump + check_traces [ "$failed" != "true" ] fi
@@ -173,6 +182,7 @@ if [ "$RUN_TESTS" = "true" ]; then configure_coredump sudo meson test -C build --suite fast-tests -t 3 || failed="true" catch_coredump + check_traces [ "$failed" != "true" ] fi
.github/workflows/build.yml +1 -1
@@ -151,7 +151,7 @@ jobs: run: sudo apt install -y crossbuild-essential-riscv64 - name: Install test tools packages if: env.AARCH64 != 'true' || env.PPC64LE != 'true' || env.RISCV64 != 'true' || env.RUN_TESTS == 'true' - run: sudo apt install -y gdb jq + run: sudo apt install -y babeltrace gdb jq - name: Install doc generation packages if: env.BUILD_DOCS == 'true' run: sudo apt install -y doxygen graphviz man-db python3-sphinx
Signature · 1 line
2.48.1
PATCH v4 2/5 dmadev: avoid copies in tracepoints
In reply to David MarchandNo need to copy values in intermediate variables.
Use the right trace point emitters.
Update the pcie struct to avoid aliasing warning.
---
Changes since v3:
- added anonymous union around pcie struct (which triggered an abidiff warning that needs waiving) and kept original call to rte_trace_point_emit_u64,
Changes since v2:
- split this change into multiple changes,
only kept trivial parts in this patch,
--- devtools/libabigail.abignore | 5 +++++ lib/dmadev/rte_dmadev.h | 29 ++++++++++++++++------------- lib/dmadev/rte_dmadev_trace.h | 20 ++++++-------------- 3 files changed, 27 insertions(+), 27 deletions(-)
devtools/libabigail.abignore +5 -0
@@ -36,3 +36,8 @@ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; Temporary exceptions till next major ABI version ; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; +[suppress_type] + name = rte_dma_port_param + type_kind = struct + has_size_change = no + has_data_member = {pcie}
lib/dmadev/rte_dmadev.h +16 -13
@@ -523,19 +523,22 @@ struct rte_dma_port_param {* and capabilities. */ __extension__ -struct { -uint64_t coreid :4 ;/**< PCIe core id used. */ -uint64_t pfid :8 ;/**< PF id used. */ -uint64_t vfen :1 ;/**< VF enable bit. */ -uint64_t vfid :16 ;/**< VF id used. */ -/** The pasid filed in TLP packet. */ -uint64_t pasid :20 ; -/** The attributes filed in TLP packet. */ -uint64_t attr :3 ; -/** The processing hint filed in TLP packet. */ -uint64_t ph :2 ; -/** The steering tag filed in TLP packet. */ -uint64_t st :16 ; +union { +struct { +uint64_t coreid :4 ;/**< PCIe core id used. */ +uint64_t pfid :8 ;/**< PF id used. */ +uint64_t vfen :1 ;/**< VF enable bit. */ +uint64_t vfid :16 ;/**< VF id used. */ +/** The pasid filed in TLP packet. */ +uint64_t pasid :20 ; +/** The attributes filed in TLP packet. */ +uint64_t attr :3 ; +/** The processing hint filed in TLP packet. */ +uint64_t ph :2 ; +/** The steering tag filed in TLP packet. */ +uint64_t st :16 ; + }; +uint64_t val; } pcie; };uint64_t reserved[2 ];/**< Reserved for future fields. */
lib/dmadev/rte_dmadev_trace.h +6 -14
@@ -46,11 +46,10 @@ RTE_TRACE_POINT(const struct rte_dma_conf __dev_conf = {0 }; dev_conf = &__dev_conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int enable_silent = (int )dev_conf->enable_silent;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority); -rte_trace_point_emit_int (enable_silent); +rte_trace_point_emit_u8 (dev_conf->enable_silent);rte_trace_point_emit_int (ret); )
@@ -83,21 +82,14 @@ RTE_TRACE_POINT(const struct rte_dma_vchan_conf __conf = {0 }; conf = &__conf;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int src_port_type = conf->src_port.port_type; -int dst_port_type = conf->dst_port.port_type; -int direction = conf->direction; -uint64_t src_pcie_cfg; -uint64_t dst_pcie_cfg;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (direction); +rte_trace_point_emit_int (conf->direction);rte_trace_point_emit_u16 (conf->nb_desc); -rte_trace_point_emit_int (src_port_type); -memcpy (&src_pcie_cfg, &conf->src_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_u64 (src_pcie_cfg); -memcpy (&dst_pcie_cfg, &conf->dst_port.pcie,sizeof (uint64_t )); -rte_trace_point_emit_int (dst_port_type); -rte_trace_point_emit_u64 (dst_pcie_cfg); +rte_trace_point_emit_int (conf->src_port.port_type); +rte_trace_point_emit_u64 (conf->src_port.pcie.val); +rte_trace_point_emit_int (conf->dst_port.port_type); +rte_trace_point_emit_u64 (conf->dst_port.pcie.val);rte_trace_point_emit_ptr (conf->auto_free.m2d.pool);rte_trace_point_emit_int (ret); )
Signature · 1 line
2.48.1
PATCH v4 3/5 trace: support dereferencing pointers
In reply to David MarchandRather than use an intermediate variable, add new helpers that handle dereferencing pointers.
Update dmadev traces accordingly (and adjust the emitter type).
---
Changes since v3:
- started from scratch with different approach, added new helpers to avoid touching ctf metadata,
Changes since v2:
- split this change out of patch 2, as it required updating CTF metadata fixup,
--- lib/dmadev/rte_dmadev_trace_fp.h | 12 ++--- lib/eal/include/rte_trace_point.h | 52 ++++++++++++++-------- lib/eal/include/rte_trace_point_register.h | 9 ++-- 3 files changed, 41 insertions(+), 32 deletions(-)
lib/dmadev/rte_dmadev_trace_fp.h +4 -8
@@ -37,10 +37,9 @@ RTE_TRACE_POINT_FP(enum rte_dma_vchan_status __status =0 ; status = &__status;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int vchan_status = *status;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan); -rte_trace_point_emit_int (vchan_status); +rte_trace_point_emit_int_ptr (status);rte_trace_point_emit_int (ret); )
@@ -107,13 +106,11 @@ RTE_TRACE_POINT_FP( last_idx = &__last_idx; has_error = &__has_error;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int has_error_val = *has_error; -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); -rte_trace_point_emit_int (has_error_val); +rte_trace_point_emit_u16_ptr (last_idx); +rte_trace_point_emit_u8_ptr (has_error);rte_trace_point_emit_u16 (ret); )
@@ -126,11 +123,10 @@ RTE_TRACE_POINT_FP(uint16_t __last_idx =0 ; last_idx = &__last_idx;#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ -int last_idx_val = *last_idx;rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls); -rte_trace_point_emit_int (last_idx_val); +rte_trace_point_emit_u16_ptr (last_idx);rte_trace_point_emit_ptr (status);rte_trace_point_emit_u16 (ret); )
lib/eal/include/rte_trace_point.h +33 -19
77 lines not on this page · raw message
lib/eal/include/rte_trace_point_register.h +4 -5
@@ -35,11 +35,10 @@ RTE_INIT(trace##_init) \#define __rte_trace_point_emit_header_fp (t) \ __rte_trace_point_emit_header_generic(t) -#define __rte_trace_point_emit (in, type) \ +#define __rte_trace_point_emit (name, in, type) \ do { \ - RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(in))); \ - __rte_trace_point_emit_field(sizeof(type), RTE_STR(in), \ - RTE_STR(type)); \ + RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(*in))); \ + __rte_trace_point_emit_field(sizeof(type), name, RTE_STR(type)); \ } while (0)#define rte_trace_point_emit_string (in) \
@@ -52,7 +51,7 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ RTE_SET_USED(in); \ - __rte_trace_point_emit(len, uint8_t); \ + __rte_trace_point_emit(RTE_STR(len), &len, uint8_t); \ __rte_trace_point_emit_field(RTE_TRACE_BLOB_LEN_MAX, \ RTE_STR(in)"[" RTE_STR(RTE_TRACE_BLOB_LEN_MAX)"]", \ RTE_STR(uint8_t)); \
Signature · 1 line
2.48.1
RE: [EXTERNAL] [PATCH v4 3/5] trace: support dereferencing pointers
Quoted 14 lines
Rather than use an intermediate variable, add new helpers that handle dereferencing pointers.
Update dmadev traces accordingly (and adjust the emitter type).Signed-off-by David Marchand <[email protected]>---
Changes since v3:
- started from scratch with different approach, added new helpers to avoid touching ctf metadata,Changes since v2:
- split this change out of patch 2, as it required updating CTF metadata fixup,
Quoted 7 lines
--- lib/dmadev/rte_dmadev_trace_fp.h | 12 ++--- lib/eal/include/rte_trace_point.h | 52 ++++++++++++++-------- lib/eal/include/rte_trace_point_register.h | 9 ++-- 3 files changed, 41 insertions(+), 32 deletions(-)2.48.1
PATCH v4 4/5 trace: support expression for blob length
In reply to David MarchandSupport any expression as a blob length by using an intermediate variable in the trace point emitter itself.
This also avoids any side effect on the passed variable.
On the "register" side, prefix the length variable in the trace metadata with the name of the emitted argument.
With this, we can update ethdev traces and avoid intermediate variables.
---
Changes since v2:
- fixed length field in CTF metadata,
Changes since v1:
- removed code relying on arguments in "registering" rte_trace_point_emit_blob implementation (see patch 3),
- moved build check in non registering __rte_trace_point_emit,
--- lib/eal/include/rte_trace_point.h | 12 ++++++---- lib/eal/include/rte_trace_point_register.h | 6 ++++- lib/ethdev/ethdev_trace.h | 27 ++++++---------------- 3 files changed, 19 insertions(+), 26 deletions(-)
lib/eal/include/rte_trace_point.h +7 -5
@@ -378,6 +378,7 @@ do { \#define __rte_trace_point_emit (name, in, type) \ do { \ + RTE_BUILD_BUG_ON(sizeof(type) != sizeof(typeof(*in))); \ memcpy(mem, in, sizeof(*in)); \ mem = RTE_PTR_ADD(mem, sizeof(*in)); \ } while (0)
@@ -392,13 +393,14 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ if (unlikely(in == NULL)) \ return; \ - if (len > RTE_TRACE_BLOB_LEN_MAX) \ - len = RTE_TRACE_BLOB_LEN_MAX; \ - __rte_trace_point_emit(RTE_STR(len), &len, uint8_t); \ - memcpy(mem, in, len); \ - memset(RTE_PTR_ADD(mem, len), 0, RTE_TRACE_BLOB_LEN_MAX - len); \ + if (size > RTE_TRACE_BLOB_LEN_MAX) \ + size = RTE_TRACE_BLOB_LEN_MAX; \ + __rte_trace_point_emit("size", &size, uint8_t); \ + memcpy(mem, in, size); \ + memset(RTE_PTR_ADD(mem, size), 0, RTE_TRACE_BLOB_LEN_MAX - size); \ mem = RTE_PTR_ADD(mem, RTE_TRACE_BLOB_LEN_MAX); \ } while (0)
lib/eal/include/rte_trace_point_register.h +5 -1
@@ -50,8 +50,12 @@ do { \#define rte_trace_point_emit_blob (in, len) \ do { \ + uint8_t size = len; \ RTE_SET_USED(in); \ - __rte_trace_point_emit(RTE_STR(len), &len, uint8_t); \ + RTE_SET_USED(size); \ + __rte_trace_point_emit_field(sizeof(uint8_t), \ + RTE_STR(in) "_size", \ + RTE_STR(uint8_t)); \ __rte_trace_point_emit_field(RTE_TRACE_BLOB_LEN_MAX, \ RTE_STR(in)"[" RTE_STR(RTE_TRACE_BLOB_LEN_MAX)"]", \ RTE_STR(uint8_t)); \
lib/ethdev/ethdev_trace.h +7 -20
@@ -860,10 +860,8 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_add,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr,uint32_t pool,int ret), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN );rte_trace_point_emit_u32 (pool);rte_trace_point_emit_int (ret); )
@@ -872,20 +870,16 @@ RTE_TRACE_POINT( rte_ethdev_trace_mac_addr_remove,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )RTE_TRACE_POINT ( rte_ethdev_trace_default_mac_addr_set, RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *addr), - uint8_t len = RTE_ETHER_ADDR_LEN; - rte_trace_point_emit_u16(port_id); -rte_trace_point_emit_blob (addr->addr_bytes, len); +rte_trace_point_emit_blob (addr->addr_bytes,RTE_ETHER_ADDR_LEN ); ) RTE_TRACE_POINT(
@@ -1102,11 +1096,9 @@ RTE_TRACE_POINT( RTE_TRACE_POINT_ARGS(uint16_t port_id, const struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr, int ret), - uint8_t len = nb_mc_addr * RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id);rte_trace_point_emit_u32 (nb_mc_addr); -rte_trace_point_emit_blob (mc_addr_set, len); +rte_trace_point_emit_blob (mc_addr_set, nb_mc_addr *RTE_ETHER_ADDR_LEN );rte_trace_point_emit_int (ret); )
@@ -1214,13 +1206,10 @@ RTE_TRACE_POINT( rte_ethdev_trace_get_dcb_info,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_eth_dcb_info *dcb_info,int ret), - uint8_t num_user_priorities = RTE_ETH_DCB_NUM_USER_PRIORITIES; -uint8_t num_tcs =RTE_ETH_DCB_NUM_TCS ; -rte_trace_point_emit_u16 (port_id );rte_trace_point_emit_u8 (dcb_info->nb_tcs); -rte_trace_point_emit_blob (dcb_info->prio_tc, num_user_priorities); -rte_trace_point_emit_blob (dcb_info->tc_bws, num_tcs); +rte_trace_point_emit_blob (dcb_info->prio_tc,RTE_ETH_DCB_NUM_USER_PRIORITIES ); +rte_trace_point_emit_blob (dcb_info->tc_bws,RTE_ETH_DCB_NUM_TCS );rte_trace_point_emit_int (ret); )
@@ -2126,10 +2115,8 @@ RTE_TRACE_POINT_FP( rte_eth_trace_macaddr_get,RTE_TRACE_POINT_ARGS (uint16_t port_id,const struct rte_ether_addr *mac_addr), - uint8_t len = RTE_ETHER_ADDR_LEN; -rte_trace_point_emit_u16 (port_id ); -rte_trace_point_emit_blob (mac_addr->addr_bytes, len); +rte_trace_point_emit_blob (mac_addr->addr_bytes,RTE_ETHER_ADDR_LEN ); )/* Called in loop in examples/ip_pipeline */
Signature · 1 line
2.48.1
Quoted 19 lines
Support any expression as a blob length by using an intermediate variable in the trace point emitter itself.
This also avoids any side effect on the passed variable.On the "register" side, prefix the length variable in the trace metadata with the name of the emitted argument.
With this, we can update ethdev traces and avoid intermediate variables.
Signed-off-by David Marchand <[email protected]>---
Changes since v2:
- fixed length field in CTF metadata,Changes since v1:
- removed code relying on arguments in "registering"rte_trace_point_emit_blobimplementation (see patch 3),
- moved build check in non registering__rte_trace_point_emit,
Quoted 7 lines
--- lib/eal/include/rte_trace_point.h | 12 ++++++---- lib/eal/include/rte_trace_point_register.h | 6 ++++- lib/ethdev/ethdev_trace.h | 27 ++++++---------------- 3 files changed, 19 insertions(+), 26 deletions(-)2.48.1
PATCH v4 5/5 trace: fix undefined behavior in register
In reply to David MarchandRegistering a tracepoint handler was resulting so far in undefined behavior at runtime.
The RTE_TRACE_POINT_REGISTER() macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.
Instead, declare an additional inline _register() handler for each tracepoint and make sure that the emitting macros in rte_trace_point_register.h only work on arguments name and type.
The original tracepoint handler prototype is adjusted by adding a __rte_unused for each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.
With this change in place, the workaround in dmadev tracepoints can be removed.
--- lib/dmadev/rte_dmadev_trace.h | 12 ------ lib/dmadev/rte_dmadev_trace_fp.h | 14 ------- lib/eal/include/rte_trace_point.h | 4 ++ lib/eal/include/rte_trace_point_register.h | 44 +++++++++++++++++++--- 4 files changed, 42 insertions(+), 32 deletions(-)
lib/dmadev/rte_dmadev_trace.h +0 -12
@@ -22,10 +22,6 @@ extern "C" {RTE_TRACE_POINT ( rte_dma_trace_info_get, RTE_TRACE_POINT_ARGS(int16_t dev_id, struct rte_dma_info *dev_info), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - struct rte_dma_info __dev_info = {0}; - dev_info = &__dev_info; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16(dev_id);rte_trace_point_emit_string (dev_info->dev_name);rte_trace_point_emit_u64 (dev_info->dev_capa);
@@ -42,10 +38,6 @@ RTE_TRACE_POINT( rte_dma_trace_configure,RTE_TRACE_POINT_ARGS (int16_t dev_id,const struct rte_dma_conf *dev_conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_conf __dev_conf = {0 }; - dev_conf = &__dev_conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (dev_conf->nb_vchans);rte_trace_point_emit_u16 (dev_conf->priority);
@@ -78,10 +70,6 @@ RTE_TRACE_POINT( rte_dma_trace_vchan_setup,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,const struct rte_dma_vchan_conf *conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_vchan_conf __conf = {0 }; - conf = &__conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (conf->direction);
lib/dmadev/rte_dmadev_trace_fp.h +0 -14
@@ -33,10 +33,6 @@ RTE_TRACE_POINT_FP( rte_dma_trace_vchan_status,RTE_TRACE_POINT_ARGS (int16_t dev_id,uint16_t vchan,enum rte_dma_vchan_status *status,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -enum rte_dma_vchan_status __status =0 ; - status = &__status; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int_ptr (status);
@@ -100,12 +96,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, bool *has_error, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; -bool __has_error = false; - last_idx = &__last_idx; - has_error = &__has_error; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
@@ -119,10 +109,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, enum rte_dma_status_code *status, uint16_t ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; - last_idx = &__last_idx; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);
lib/eal/include/rte_trace_point.h +4 -0
@@ -36,6 +36,8 @@ extern "C" {/** The tracepoint object. */ typedef RTE_ATOMIC (uint64_t )rte_trace_point_t ; +#ifndef _RTE_TRACE_POINT_REGISTER_H_ +/** * Macro to define the tracepoint arguments in RTE_TRACE_POINT macro.
@@ -53,6 +55,8 @@ _tp _args \ __VA_ARGS__ \ } +#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ +/** * Create a tracepoint. *
lib/eal/include/rte_trace_point_register.h +38 -6
73 lines not on this page · raw message
Signature · 1 line
2.48.1
RE: [EXTERNAL] [PATCH v4 5/5] trace: fix undefined behavior in register
Quoted 230 lines
Registering a tracepoint handler was resulting so far in undefined behavior at runtime.
The
RTE_TRACE_POINT_REGISTER()macro was casting the tracepoint handler (which expects arguments) to a void (*)(void).
At runtime, calling this handler while registering resulted in reading the current stack with no relation to this function prototype.Instead, declare an additional inline
_register()handler for each tracepoint and make sure that the emitting macros inrte_trace_point_register.honly work on arguments name and type.The original tracepoint handler prototype is adjusted by adding a
__rte_unusedfor each argument (since emitting macros do nothing with them).
This last part introduces an implementation limit of 15 arguments.With this change in place, the workaround in dmadev tracepoints can be removed.
Signed-off-by David Marchand <[email protected]>--- lib/dmadev/rte_dmadev_trace.h | 12 ------ lib/dmadev/rte_dmadev_trace_fp.h | 14 ------- lib/eal/include/rte_trace_point.h | 4 ++ lib/eal/include/rte_trace_point_register.h | 44 +++++++++++++++++++--- 4 files changed, 42 insertions(+), 32 deletions(-)(quoted diff) +0
-0
b/lib/dmadev/rte_dmadev_trace.hindex1beb938168..1de92655f2100644lib/dmadev/rte_dmadev_trace.h +0
-7@@ -22,10 +22,6 @@ extern "C" {RTE_TRACE_POINT ( rte_dma_trace_info_get, RTE_TRACE_POINT_ARGS(int16_t dev_id, struct rte_dma_info *dev_info), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ - struct rte_dma_info __dev_info = {0}; - dev_info = &__dev_info; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16(dev_id);rte_trace_point_emit_string (dev_info->dev_name);rte_trace_point_emit_u64 (dev_info->dev_capa);@@ -42,10 +38,6 @@ RTE_TRACE_POINT( rte_dma_trace_configure,RTE_TRACE_POINT_ARGS (int16_t dev_id,const struct rte_dma_conf *dev_conf,int ret), -#ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_conf __dev_conf = {0 }; - dev_conf = &__dev_conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id );rte_trace_point_emit_u16 (dev_conf->nb_vchans);lib/dmadev/rte_dmadev_trace.h +0
-3rte_trace_point_emit_u16 (dev_conf->priority);@@ -78,10 +70,6 @@ RTE_TRACE_POINT( rte_dma_trace_vchan_setup, RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const struct rte_dma_vchan_conf *conf, int ret), - #ifdef _RTE_TRACE_POINT_REGISTER_H_ -const struct rte_dma_vchan_conf __conf = {0 }; - conf = &__conf; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int (conf->direction);lib/dmadev/rte_dmadev_trace_fp.h +0
-3@@ -33,10 +33,6 @@ RTE_TRACE_POINT_FP( rte_dma_trace_vchan_status, RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, enum rte_dma_vchan_status *status, int ret), - #ifdef _RTE_TRACE_POINT_REGISTER_H_ -enum rte_dma_vchan_status __status =0 ; - status = &__status; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_int_ptr (status);@@ -100,12 +96,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, bool *has_error, uint16_t ret), -#ifdef_RTE_TRACE_POINT_REGISTER_H_
lib/dmadev/rte_dmadev_trace_fp.h +0
-8-uint16_t __last_idx =0 ; -bool __has_error = false; - last_idx = &__last_idx; - has_error = &__has_error; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);@@ -119,10 +109,6 @@ RTE_TRACE_POINT_FP( RTE_TRACE_POINT_ARGS(int16_t dev_id, uint16_t vchan, const uint16_t nb_cpls, uint16_t *last_idx, enum rte_dma_status_code *status, uint16_t ret), - #ifdef _RTE_TRACE_POINT_REGISTER_H_ - uint16_t __last_idx = 0; - last_idx = &__last_idx; -#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ rte_trace_point_emit_i16 (dev_id);rte_trace_point_emit_u16 (vchan);rte_trace_point_emit_u16 (nb_cpls);lib/eal/include/rte_trace_point.h +4
-0@@ -36,6 +36,8 @@ extern "C" {/** The tracepoint object. */ typedef RTE_ATOMIC (uint64_t )rte_trace_point_t ; +#ifndef _RTE_TRACE_POINT_REGISTER_H_ +/** * Macro to define the tracepoint arguments in RTE_TRACE_POINT macro. @@ -53,6 +55,8 @@ _tp _args \ __VA_ARGS__ \ } +#endif /* _RTE_TRACE_POINT_REGISTER_H_ */ +/** * Create a tracepoint. * lib/eal/include/rte_trace_point_register.h +1
-0@@ -18,6 +18,43 @@ extern "C" { RTE_DECLARE_PER_LCORE(volatile int, trace_point_sz); +#define
RTE_TRACE_POINT_ARGS_COUNT_(_0,_1,_2,_3,_4,_5,_6,_7,_8,_9,_10,_11,_1 2,_13,_14,_15,N, ...)\lib/eal/include/rte_trace_point_register.h +4
-0+ N +#define RTE_TRACE_POINT_ARGS_COUNT (...) \ + +RTE_TRACE_POINT_ARGS_COUNT_(__VA_ARGS__,15,14,13,12,11,10,9,8,7,6,5,4,3
+,2,1,0)lib/eal/include/rte_trace_point_register.h +31
-0+ +#define RTE_TRACE_POINT_ARGS_1(a) __rte_unused a #define +RTE_TRACE_POINT_ARGS_2(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_1(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_3(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_2(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_4(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_3(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_5(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_4(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_6(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_5(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_7(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_6(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_8(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_7(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_9(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_8(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_10(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_9(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_11(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_10(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_12(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_11(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_13(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_12(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_14(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_13(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_15(a, ...) __rte_unused a, +RTE_TRACE_POINT_ARGS_14(__VA_ARGS__) +#define RTE_TRACE_POINT_ARGS_FUNC(a) RTE_TRACE_POINT_ARGS_ ## a#define
lib/eal/include/rte_trace_point_register.h +3
-0+RTE_TRACE_POINT_ARGS_EXPAND(...) __VA_ARGS__ #define +RTE_TRACE_POINT_ARGS_ (N , ...) \ +RTE_TRACE_POINT_ARGS_EXPAND(RTE_TRACE_POINT_ARGS_FUNC(N))(__VA_ARGS__)
lib/eal/include/rte_trace_point_register.h +2
-0+#define RTE_TRACE_POINT_ARGS (...) \ + (RTE_TRACE_POINT_ARGS_(RTE_TRACE_POINT_ARGS_COUNT(0,
__VA_ARGS__),lib/eal/include/rte_trace_point_register.h +9
-0+__VA_ARGS__)) + +#define __RTE_TRACE_POINT (_mode, _tp, _args, ...) \ extern +rte_trace_point_t __##_tp; \ static __rte_always_inline void _tp _args +{ } \ static __rte_always_inline void \ _tp ## _register (void) \ { \ + __rte_trace_point_emit_header_##_mode(&__##_tp); \ + __VA_ARGS__ \ +} +#define RTE_TRACE_POINT_REGISTER (trace, name) \ rte_trace_point_t
__rte_section("__rte_trace_point")__##trace; \ static const char __##trace##_name[] =RTE_STR(name); \ @@ -26,7 +63,7 @@RTE_INIT(trace##_init)\lib/eal/include/rte_trace_point_register.h +1
-1if (!rte_trace_feature_is_enabled ()) \return ; \ __rte_trace_point_register(&__##trace, __##trace##_name, \ - (void (*)(void)) trace); \ + trace ## _register); \ }#define __rte_trace_point_emit_header_generic (t) \ @@ -37,22 +74,17 @@
RTE_INIT(trace##_init)\#define
__rte_trace_point_emit(name, in, type)\ do { \lib/eal/include/rte_trace_point_register.h +0
-1-RTE_BUILD_BUG_ON (sizeof (type) !=sizeof (typeof (*in))); \ __rte_trace_point_emit_field(sizeof(type), name, RTE_STR(type)); \ }while (0)
#define
rte_trace_point_emit_string(in)\ do { \lib/eal/include/rte_trace_point_register.h +0
-1-RTE_SET_USED (in); \__rte_trace_point_emit_field(__RTE_TRACE_EMIT_STRING_LEN_MAX
, \lib/eal/include/rte_trace_point_register.h +0
-4RTE_STR(in)"[32]", "string_bounded_t"); \ } while (0)#define rte_trace_point_emit_blob (in, len) \ do { \ - uint8_t size = len; \ - RTE_SET_USED(in); \ - RTE_SET_USED(size); \ __rte_trace_point_emit_field(sizeof(uint8_t), \ RTE_STR(in) "_size", \ RTE_STR(uint8_t)); \ --
Quoted 1 line
2.48.1
Re: [PATCH v4 0/5] Trace point framework enhancement for dmadev
In reply to David MarchandQuoted 28 lines · David Marchand
dmadev trace points have been working around trace point framework limitations.
One of this workaround blocks enabling dmadev on MSVC.Traces files were not checked in CI, so this series also adds a new check.
--
David MarchandChanges since v3:
- dropped changes to CTF metadata,
- added new helpers for dumping content of pointed at memory,
- changed dmadev to avoid more changes in trace framework,Changes since v2:
- added check of traces validity in GHA,
- split patch 2 and fixed CTF metadata,David Marchand (5):
ci: check traces validity
dmadev: avoid copies in tracepoints
trace: support dereferencing pointers
trace: support expression for blob length
trace: fix undefined behavior in register
Quoted 10 lines
.ci/linux-build.sh | 10 ++++ .github/workflows/build.yml | 2 +- devtools/libabigail.abignore | 5 ++ lib/dmadev/rte_dmadev.h | 29 +++++----- lib/dmadev/rte_dmadev_trace.h | 32 ++--------- lib/dmadev/rte_dmadev_trace_fp.h | 26 ++------- lib/eal/include/rte_trace_point.h | 66 ++++++++++++++-------- lib/eal/include/rte_trace_point_register.h | 51 ++++++++++++++--- lib/ethdev/ethdev_trace.h | 27 +++------ 9 files changed, 135 insertions(+), 113 deletions(-)
Series applied, thanks.
Signature · 1 line
David Marchand