[dpdk-dev] [PATCH v3 2/2] eal: fix side effects in ptr align macros
Wang, Haiyue
haiyue.wang at intel.com
Wed May 12 03:13:41 CEST 2021
> -----Original Message-----
> From: Pavan Nikhilesh Bhagavatula <pbhagavatula at marvell.com>
> Sent: Tuesday, May 11, 2021 16:45
> To: Wang, Haiyue <haiyue.wang at intel.com>; david.marchand at redhat.com; Matan Azrad <matan at nvidia.com>;
> Shahaf Shuler <shahafs at nvidia.com>; Viacheslav Ovsiienko <viacheslavo at nvidia.com>; Jiawen Wu
> <jiawenwu at trustnetic.com>; Jian Wang <jianwang at trustnetic.com>; Chautru, Nicolas
> <nicolas.chautru at intel.com>; Thomas Monjalon <thomas at monjalon.net>; Yigit, Ferruh
> <ferruh.yigit at intel.com>; Andrew Rybchenko <andrew.rybchenko at oktetlabs.ru>
> Cc: dev at dpdk.org; stable at dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v3 2/2] eal: fix side effects in ptr align macros
>
> >> -----Original Message-----
> >> From: pbhagavatula at marvell.com <pbhagavatula at marvell.com>
> >> Sent: Tuesday, May 11, 2021 03:40
> >> To: david.marchand at redhat.com; Wang, Haiyue
> ><haiyue.wang at intel.com>; Matan Azrad <matan at nvidia.com>;
> >> Shahaf Shuler <shahafs at nvidia.com>; Viacheslav Ovsiienko
> ><viacheslavo at nvidia.com>; Jiawen Wu
> >> <jiawenwu at trustnetic.com>; Jian Wang <jianwang at trustnetic.com>;
> >Chautru, Nicolas
> >> <nicolas.chautru at intel.com>; Thomas Monjalon
> ><thomas at monjalon.net>; Yigit, Ferruh
> >> <ferruh.yigit at intel.com>; Andrew Rybchenko
> ><andrew.rybchenko at oktetlabs.ru>
> >> Cc: dev at dpdk.org; Pavan Nikhilesh <pbhagavatula at marvell.com>;
> >stable at dpdk.org
> >> Subject: [dpdk-dev] [PATCH v3 2/2] eal: fix side effects in ptr align
> >macros
> >>
> >> From: Pavan Nikhilesh <pbhagavatula at marvell.com>
> >>
> >> Avoid expanding parameters inside RTE_*_ALIGN macros.
> >> Update common_autotest to detect macro side effects.
> >> Workaround static arrays relying on RTE_ALIGN macros.
> >>
> >> Fixes: af75078fece3 ("first public release")
> >> Cc: stable at dpdk.org
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavatula at marvell.com>
> >> Signed-off-by: David Marchand <david.marchand at redhat.com>
> >> ---
> >> app/test/test_common.c | 6 ++++++
> >> drivers/net/e1000/e1000_ethdev.h | 7 ++++---
> >> drivers/net/ixgbe/ixgbe_ethdev.h | 6 ++++--
> >> drivers/net/mlx5/mlx5_rxtx_vec.h | 7 ++++---
> >> drivers/net/txgbe/txgbe_ethdev.h | 6 ++++--
> >> examples/bbdev_app/main.c | 2 +-
> >> lib/eal/include/rte_common.h | 17 +++++++++++++----
> >> lib/ethdev/rte_eth_ctrl.h | 5 +++--
> >> 8 files changed, 39 insertions(+), 17 deletions(-)
> >>
> >> diff --git a/app/test/test_common.c b/app/test/test_common.c
> >> index 0dbb87e741..9efe3b10f9 100644
> >> --- a/app/test/test_common.c
> >> +++ b/app/test/test_common.c
> >> @@ -69,6 +69,12 @@ test_macros(int __rte_unused unused_parm)
> >> TEST_SIDE_EFFECT_2(RTE_PTR_ADD, void *, size_t);
> >> TEST_SIDE_EFFECT_2(RTE_PTR_DIFF, void *, void *);
> >> TEST_SIDE_EFFECT_2(RTE_PTR_SUB, void *, size_t);
> >> + TEST_SIDE_EFFECT_2(RTE_PTR_ALIGN, void *, size_t);
> >> + TEST_SIDE_EFFECT_2(RTE_PTR_ALIGN_CEIL, void *, size_t);
> >> + TEST_SIDE_EFFECT_2(RTE_PTR_ALIGN_FLOOR, void *, size_t);
> >> + TEST_SIDE_EFFECT_2(RTE_ALIGN, unsigned int, unsigned int);
> >> + TEST_SIDE_EFFECT_2(RTE_ALIGN_CEIL, unsigned int, unsigned
> >int);
> >> + TEST_SIDE_EFFECT_2(RTE_ALIGN_FLOOR, unsigned int,
> >unsigned int);
> >> TEST_SIDE_EFFECT_2(RTE_ALIGN_MUL_CEIL, unsigned int,
> >unsigned int);
> >> TEST_SIDE_EFFECT_2(RTE_ALIGN_MUL_FLOOR, unsigned int,
> >unsigned int);
> >> TEST_SIDE_EFFECT_2(RTE_ALIGN_MUL_NEAR, unsigned int,
> >unsigned int);
> >> diff --git a/drivers/net/e1000/e1000_ethdev.h
> >b/drivers/net/e1000/e1000_ethdev.h
> >> index 3b4d9c3ee6..155d825d89 100644
> >> --- a/drivers/net/e1000/e1000_ethdev.h
> >> +++ b/drivers/net/e1000/e1000_ethdev.h
> >> @@ -332,9 +332,10 @@ struct igb_eth_syn_filter_ele {
> >> };
> >>
> >> #define IGB_FLEX_FILTER_MAXLEN 128 /**< bytes to use in flex
> >filter. */
> >> -#define IGB_FLEX_FILTER_MASK_SIZE \
> >> - (RTE_ALIGN(IGB_FLEX_FILTER_MAXLEN, CHAR_BIT) /
> >CHAR_BIT)
> >> - /**< mask bytes in flex filter. */
> >> +#define IGB_FLEX_FILTER_MASK_SIZE \
> >> + (RTE_ALIGN_FLOOR(IGB_FLEX_FILTER_MAXLEN + (CHAR_BIT -
> >1), CHAR_BIT) / \
> >> + CHAR_BIT)
> >> +/**< mask bytes in flex filter. */
> >>
> >
> >Since:
> >RTE_ALIGN --> RTE_ALIGN_CEIL(val, align) --> RTE_ALIGN_FLOOR(...)
> >
> >Why only change 'RTE_ALIGN_CEIL', but then call 'RTE_ALIGN_FLOOR'
> >directly ?
> >Sorry, can't get the point. : - (
> >
>
> RTE_ALIGN_CEIL is prone to macro side effects since it uses align twice in the macro.
> To fix it RTE_ALIGN_CEIL first expands the val and align values to temporary values
> and then calls ALIGN_FLOOR.
> Due to this we can no longer use ALIGN/ALIGN_CEIL in static array declarations
>
> Example:
> In file included from ../lib/eal/x86/include/rte_atomic.h:13,
> from ../examples/bbdev_app/main.c:21:
> ../lib/eal/include/rte_common.h:312:15: error: braced-group within expression allowed only inside a
> function
> __extension__({ \
> ^
> ../examples/bbdev_app/main.c:48:18: note: in expansion of macro 'RTE_ALIGN_CEIL'
> #define NCB (3 * RTE_ALIGN_CEIL(K + 4, 32))
> ^~~~~~~~~~~~~~
> ../examples/bbdev_app/main.c:145:23: note: in expansion of macro 'NCB'
> uint8_t llr_temp_buf[NCB];
>
> To workaround this I had to call RTE_ALIGN_FLOOR for known constants directly.
>
Clear now, thanks.
For Intel PMDs,
Acked-by: Haiyue Wang <haiyue.wang at intel.com>
BTW, I'm wondering that we can explicitly define the macro with type style
like RTE_ALIGN_CEIL_TYPE, so that people can notice the side effects directly,
and choose the right ALIGN macro.
And can still use ALIGN/ALIGN_CEIL in static array declarations, which is
very simple.
Just my thinking. ;-)
> >Why not change 'RTE_ALIGN' if it has issue ?
> >
> >>
> >> diff --git a/lib/eal/include/rte_common.h
> >b/lib/eal/include/rte_common.h
> >> index a142596587..6acd067b5c 100644
> >> --- a/lib/eal/include/rte_common.h
> >> +++ b/lib/eal/include/rte_common.h
> >> @@ -294,8 +294,13 @@ static void
> >__attribute__((destructor(RTE_PRIO(prio)), used)) func(void)
> >> * point to an address no lower than the first parameter. Second
> >parameter
> >> * must be a power-of-two value.
> >> */
> >> -#define RTE_PTR_ALIGN_CEIL(ptr, align) \
> >> - RTE_PTR_ALIGN_FLOOR((typeof(ptr))RTE_PTR_ADD(ptr, (align)
> >- 1), align)
> >> +#define RTE_PTR_ALIGN_CEIL(ptr, align) \
> >> + __extension__({ \
> >> + typeof(ptr) _pc = (ptr); \
> >> + typeof(align) _ac = (align); \
> >> +
> > RTE_PTR_ALIGN_FLOOR((typeof(ptr))RTE_PTR_ADD(_pc, _ac -
> >1), \
> >> + _ac); \
> >> + })
> >>
> >> /**
> >> * Macro to align a value to a given power-of-two. The resultant value
> >> @@ -303,8 +308,12 @@ static void
> >__attribute__((destructor(RTE_PRIO(prio)), used)) func(void)
> >> * than the first parameter. Second parameter must be a power-of-
> >two
> >> * value.
> >> */
> >> -#define RTE_ALIGN_CEIL(val, align) \
> >> - RTE_ALIGN_FLOOR(((val) + ((typeof(val)) (align) - 1)), align)
> >> +#define RTE_ALIGN_CEIL(val, align) \
> >> + __extension__({ \
> >> + typeof(val) _vc = (val); \
> >> + typeof(val) _ac = (typeof(val))(align); \
> >> + RTE_ALIGN_FLOOR((_vc + _ac - 1), _ac); \
> >> + })
> >>
> >> /**
> >> * Macro to align a pointer to a given power-of-two. The resultant
> >> diff --git a/lib/ethdev/rte_eth_ctrl.h b/lib/ethdev/rte_eth_ctrl.h
> >> index 42652f9cce..863e56170b 100644
> >> --- a/lib/ethdev/rte_eth_ctrl.h
> >> +++ b/lib/ethdev/rte_eth_ctrl.h
> >> @@ -431,8 +431,9 @@ enum rte_fdir_mode {
> >> };
> >>
> >> #define UINT64_BIT (CHAR_BIT * sizeof(uint64_t))
> >> -#define RTE_FLOW_MASK_ARRAY_SIZE \
> >> - (RTE_ALIGN(RTE_ETH_FLOW_MAX, UINT64_BIT)/UINT64_BIT)
> >> +#define RTE_FLOW_MASK_ARRAY_SIZE \
> >> + (RTE_ALIGN_FLOOR(RTE_ETH_FLOW_MAX + (UINT64_BIT - 1),
> >UINT64_BIT) / \
> >> + UINT64_BIT)
> >>
> >> /**
> >> * A structure used to get the information of flow director filter.
> >> --
> >> 2.17.1
More information about the dev
mailing list