| Message ID | 20200915014455.1232507-7-jesse.brandeburg@intel.com |
|---|---|
| State | New |
| Headers | show |
| Series | make drivers/net/ethernet W=1 clean | expand |
On Mon, 2020-09-14 at 18:44 -0700, Jesse Brandeburg wrote: > While fixing the W=1 builds, this warning came up because the > developers used a very tricky way to get structures initialized > to a non-zero value, but this causes GCC to warn about an > override. In this case the override was intentional, so just > disable the warning for this code with a macro that results > in disabling the warning for compiles on GCC versions after 8. > > NOTE: the __diag_ignore macro currently only accepts a second > argument of 8 (version 80000) > > Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com> > --- > drivers/net/ethernet/renesas/sh_eth.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/drivers/net/ethernet/renesas/sh_eth.c > b/drivers/net/ethernet/renesas/sh_eth.c > index 586642c33d2b..c63304632935 100644 > --- a/drivers/net/ethernet/renesas/sh_eth.c > +++ b/drivers/net/ethernet/renesas/sh_eth.c > @@ -45,6 +45,15 @@ > #define SH_ETH_OFFSET_DEFAULTS \ > [0 ... SH_ETH_MAX_REGISTER_OFFSET - 1] = SH_ETH_OFFSET_INVALID > > +/* use some intentionally tricky logic here to initialize the whole > struct to > + * 0xffff, but then override certain fields, requiring us to > indicate that we > + * "know" that there are overrides in this structure, and we'll need > to disable > + * that warning from W=1 builds. GCC has supported this option since > 4.2.X, but > + * the macros available to do this only define GCC 8. > + */ > +__diag_push(); > +__diag_ignore(GCC, 8, "-Woverride-init", > + "logic to initialize all and then override some is OK"); > static const u16 sh_eth_offset_gigabit[SH_ETH_MAX_REGISTER_OFFSET] = > { > SH_ETH_OFFSET_DEFAULTS, > > @@ -332,6 +341,7 @@ static const u16 > sh_eth_offset_fast_sh3_sh2[SH_ETH_MAX_REGISTER_OFFSET] = { > > [TSU_ADRH0] = 0x0100, > }; > +__diag_pop(); > I don't have any strong feeling against disabling compiler warnings, but maybe the right thing to do here is to initialize the gaps to the invalid value instead of pre-initializing the whole thing first and then setting up the valid values on the 2nd pass. I don't think there are too many gaps to fill, it is doable, so maybe add this as a comment to this driver maintainer so they could pickup the work from here.
Saeed Mahameed wrote: > On Mon, 2020-09-14 at 18:44 -0700, Jesse Brandeburg wrote: > > While fixing the W=1 builds, this warning came up because the > > developers used a very tricky way to get structures initialized > > to a non-zero value, but this causes GCC to warn about an > > override. In this case the override was intentional, so just > > disable the warning for this code with a macro that results > > in disabling the warning for compiles on GCC versions after 8. > > > > NOTE: the __diag_ignore macro currently only accepts a second > > argument of 8 (version 80000) > > > > Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com> > > --- > > drivers/net/ethernet/renesas/sh_eth.c | 10 ++++++++++ > > 1 file changed, 10 insertions(+) > > > > diff --git a/drivers/net/ethernet/renesas/sh_eth.c > > b/drivers/net/ethernet/renesas/sh_eth.c > > index 586642c33d2b..c63304632935 100644 > > --- a/drivers/net/ethernet/renesas/sh_eth.c > > +++ b/drivers/net/ethernet/renesas/sh_eth.c > > @@ -45,6 +45,15 @@ > > #define SH_ETH_OFFSET_DEFAULTS \ > > [0 ... SH_ETH_MAX_REGISTER_OFFSET - 1] = SH_ETH_OFFSET_INVALID > > > > +/* use some intentionally tricky logic here to initialize the whole > > struct to > > + * 0xffff, but then override certain fields, requiring us to > > indicate that we > > + * "know" that there are overrides in this structure, and we'll need > > to disable > > + * that warning from W=1 builds. GCC has supported this option since > > 4.2.X, but > > + * the macros available to do this only define GCC 8. > > + */ > > +__diag_push(); > > +__diag_ignore(GCC, 8, "-Woverride-init", > > + "logic to initialize all and then override some is OK"); > > static const u16 sh_eth_offset_gigabit[SH_ETH_MAX_REGISTER_OFFSET] = > > { > > SH_ETH_OFFSET_DEFAULTS, > > > > @@ -332,6 +341,7 @@ static const u16 > > sh_eth_offset_fast_sh3_sh2[SH_ETH_MAX_REGISTER_OFFSET] = { > > > > [TSU_ADRH0] = 0x0100, > > }; > > +__diag_pop(); > > > > I don't have any strong feeling against disabling compiler warnings, > but maybe the right thing to do here is to initialize the gaps to the > invalid value instead of pre-initializing the whole thing first and > then setting up the valid values on the 2nd pass. > > I don't think there are too many gaps to fill, it is doable, so maybe > add this as a comment to this driver maintainer so they could pickup > the work from here. added linux-renesas-soc list. @list, any comments on Saeed's comment above? Thanks, Jesse
at 8:25 PM, Saeed Mahameed <saeed@kernel.org> wrote: > I don't have any strong feeling against disabling compiler warnings, > but maybe the right thing to do here is to initialize the gaps to the > invalid value instead of pre-initializing the whole thing first and > then setting up the valid values on the 2nd pass. > > I don't think there are too many gaps to fill, it is doable, so maybe > add this as a comment to this driver maintainer so they could pickup > the work from here. The problem is that filling in the gaps creates a maintenance hazard. As another example, the kernel does exactly this kind of thing to initialize the system call table. Using the override in this way is the clearest, easiest to maintain way of expressing this. Disabling the warning not only gets rid of the warning, but should also be seen as a "something special is happening here" flag in the code. With the warning disabled, there is a hazard of now missing the possible introduction of an override here, but I think that risk is less than the risk of failing to fill a gap when changes are made. I completely support doing this in exactly this way. -- Mark Rustad, Ethernet Products Group, Intel Corporation
diff --git a/drivers/net/ethernet/renesas/sh_eth.c b/drivers/net/ethernet/renesas/sh_eth.c index 586642c33d2b..c63304632935 100644 --- a/drivers/net/ethernet/renesas/sh_eth.c +++ b/drivers/net/ethernet/renesas/sh_eth.c @@ -45,6 +45,15 @@ #define SH_ETH_OFFSET_DEFAULTS \ [0 ... SH_ETH_MAX_REGISTER_OFFSET - 1] = SH_ETH_OFFSET_INVALID +/* use some intentionally tricky logic here to initialize the whole struct to + * 0xffff, but then override certain fields, requiring us to indicate that we + * "know" that there are overrides in this structure, and we'll need to disable + * that warning from W=1 builds. GCC has supported this option since 4.2.X, but + * the macros available to do this only define GCC 8. + */ +__diag_push(); +__diag_ignore(GCC, 8, "-Woverride-init", + "logic to initialize all and then override some is OK"); static const u16 sh_eth_offset_gigabit[SH_ETH_MAX_REGISTER_OFFSET] = { SH_ETH_OFFSET_DEFAULTS, @@ -332,6 +341,7 @@ static const u16 sh_eth_offset_fast_sh3_sh2[SH_ETH_MAX_REGISTER_OFFSET] = { [TSU_ADRH0] = 0x0100, }; +__diag_pop(); static void sh_eth_rcv_snd_disable(struct net_device *ndev); static struct net_device_stats *sh_eth_get_stats(struct net_device *ndev);
While fixing the W=1 builds, this warning came up because the developers used a very tricky way to get structures initialized to a non-zero value, but this causes GCC to warn about an override. In this case the override was intentional, so just disable the warning for this code with a macro that results in disabling the warning for compiles on GCC versions after 8. NOTE: the __diag_ignore macro currently only accepts a second argument of 8 (version 80000) Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com> --- drivers/net/ethernet/renesas/sh_eth.c | 10 ++++++++++ 1 file changed, 10 insertions(+)