| Message ID | 20230217095650.2305559-1-arnd@kernel.org (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | a59f832a71c938779b8abdb022d6d39903e635c3 |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | sfc: use IS_ENABLED() checks for CONFIG_SFC_SRIOV | expand |
On 17/02/2023 09:56, Arnd Bergmann wrote: > From: Arnd Bergmann <arnd@arndb.de> > > One local variable has become unused after a recent change: > > drivers/net/ethernet/sfc/ef100_nic.c: In function 'ef100_probe_netdev_pf': > drivers/net/ethernet/sfc/ef100_nic.c:1155:21: error: unused variable 'net_dev' [-Werror=unused-variable] > struct net_device *net_dev = efx->net_dev; > ^~~~~~~ > > The variable is still used in an #ifdef. Replace the #ifdef with > an if(IS_ENABLED()) check that lets the compiler see where it is > used, rather than adding another #ifdef. So we've had Leon telling us[1] to use __maybe_unused, and you're saying to use IS_ENABLED() instead. Which is right? (And does it make any difference to build time? I'm assuming the compiler is smart enough that this change doesn't affect text size...?) -ed [1]: https://lore.kernel.org/netdev/cac3fa89-50a3-6de0-796c-a215400f3710@intel.com/T/#md2ecc82f18c200391dc6581ff68ff08eee9a65cf
On Fri, Feb 17, 2023, at 17:13, Edward Cree wrote: > On 17/02/2023 09:56, Arnd Bergmann wrote: >> From: Arnd Bergmann <arnd@arndb.de> >> >> One local variable has become unused after a recent change: >> >> drivers/net/ethernet/sfc/ef100_nic.c: In function 'ef100_probe_netdev_pf': >> drivers/net/ethernet/sfc/ef100_nic.c:1155:21: error: unused variable 'net_dev' [-Werror=unused-variable] >> struct net_device *net_dev = efx->net_dev; >> ^~~~~~~ >> >> The variable is still used in an #ifdef. Replace the #ifdef with >> an if(IS_ENABLED()) check that lets the compiler see where it is >> used, rather than adding another #ifdef. > > So we've had Leon telling us[1] to use __maybe_unused, and you're > saying to use IS_ENABLED() instead. Which is right? > (And does it make any difference to build time? I'm assuming the > compiler is smart enough that this change doesn't affect text > size...?) > -ed Both are correct, but I prefer the IS_ENABLED() change because it improves build coverage. The resulting object code should be the same, as the dead-code-elimination in gcc takes care of removing it the same way. If you use the __maybe_uninitialized annotation, you still need an extra fix to initialize the ef100_probe_netdev_pf() return code. Arnd
On Fri, Feb 17, 2023 at 04:13:59PM +0000, Edward Cree wrote: > On 17/02/2023 09:56, Arnd Bergmann wrote: > > From: Arnd Bergmann <arnd@arndb.de> > > > > One local variable has become unused after a recent change: > > > > drivers/net/ethernet/sfc/ef100_nic.c: In function 'ef100_probe_netdev_pf': > > drivers/net/ethernet/sfc/ef100_nic.c:1155:21: error: unused variable 'net_dev' [-Werror=unused-variable] > > struct net_device *net_dev = efx->net_dev; > > ^~~~~~~ > > > > The variable is still used in an #ifdef. Replace the #ifdef with > > an if(IS_ENABLED()) check that lets the compiler see where it is > > used, rather than adding another #ifdef. > > So we've had Leon telling us[1] to use __maybe_unused, and you're > saying to use IS_ENABLED() instead. Which is right? > (And does it make any difference to build time? I'm assuming the > compiler is smart enough that this change doesn't affect text > size...?) You are mixing answers, __maybe_unused is for variables. For functions, it will be much saner to create empty declarations for relevant functions in tc.h, for !CONFIG_SFC_SRIOV flow. It will be much cleaner than spaghetti code in .c files which you have now. Thanks > -ed > > [1]: https://lore.kernel.org/netdev/cac3fa89-50a3-6de0-796c-a215400f3710@intel.com/T/#md2ecc82f18c200391dc6581ff68ff08eee9a65cf
On Fri, Feb 17, 2023 at 05:19:26PM +0100, Arnd Bergmann wrote: > On Fri, Feb 17, 2023, at 17:13, Edward Cree wrote: > > On 17/02/2023 09:56, Arnd Bergmann wrote: > >> From: Arnd Bergmann <arnd@arndb.de> > >> > >> One local variable has become unused after a recent change: > >> > >> drivers/net/ethernet/sfc/ef100_nic.c: In function 'ef100_probe_netdev_pf': > >> drivers/net/ethernet/sfc/ef100_nic.c:1155:21: error: unused variable 'net_dev' [-Werror=unused-variable] > >> struct net_device *net_dev = efx->net_dev; > >> ^~~~~~~ > >> > >> The variable is still used in an #ifdef. Replace the #ifdef with > >> an if(IS_ENABLED()) check that lets the compiler see where it is > >> used, rather than adding another #ifdef. > > > > So we've had Leon telling us[1] to use __maybe_unused, and you're > > saying to use IS_ENABLED() instead. Which is right? > > (And does it make any difference to build time? I'm assuming the > > compiler is smart enough that this change doesn't affect text > > size...?) > > -ed > > Both are correct, but I prefer the IS_ENABLED() change because it > improves build coverage. The resulting object code should be the > same, as the dead-code-elimination in gcc takes care of removing > it the same way. > > If you use the __maybe_uninitialized annotation, you still need > an extra fix to initialize the ef100_probe_netdev_pf() return > code. FWIIW, IS_ENABLED() is the approach that is more familiar to me. Though I have nothing in particular against other approaches. Questions of consistency aside, this patch does look good to me and does appear to address the build problem in question - on x86_64. Reviewed-by: Simon Horman <simon.horman@corigine.com>
Hello: This patch was applied to netdev/net-next.git (master) by David S. Miller <davem@davemloft.net>: On Fri, 17 Feb 2023 10:56:39 +0100 you wrote: > From: Arnd Bergmann <arnd@arndb.de> > > One local variable has become unused after a recent change: > > drivers/net/ethernet/sfc/ef100_nic.c: In function 'ef100_probe_netdev_pf': > drivers/net/ethernet/sfc/ef100_nic.c:1155:21: error: unused variable 'net_dev' [-Werror=unused-variable] > struct net_device *net_dev = efx->net_dev; > ^~~~~~~ > > [...] Here is the summary with links: - sfc: use IS_ENABLED() checks for CONFIG_SFC_SRIOV https://git.kernel.org/netdev/net-next/c/a59f832a71c9 You are awesome, thank you!
diff --git a/drivers/net/ethernet/sfc/ef100_nic.c b/drivers/net/ethernet/sfc/ef100_nic.c index becd21c2325d..4dc643b0d2db 100644 --- a/drivers/net/ethernet/sfc/ef100_nic.c +++ b/drivers/net/ethernet/sfc/ef100_nic.c @@ -399,14 +399,14 @@ static int ef100_filter_table_up(struct efx_nic *efx) * filter insertion will need to take the lock for read. */ up_write(&efx->filter_sem); -#ifdef CONFIG_SFC_SRIOV - rc = efx_tc_insert_rep_filters(efx); + if (IS_ENABLED(CONFIG_SFC_SRIOV)) + rc = efx_tc_insert_rep_filters(efx); + /* Rep filter failure is nonfatal */ if (rc) netif_warn(efx, drv, efx->net_dev, "Failed to insert representor filters, rc %d\n", rc); -#endif return 0; fail_vlan0: @@ -419,9 +419,8 @@ static int ef100_filter_table_up(struct efx_nic *efx) static void ef100_filter_table_down(struct efx_nic *efx) { -#ifdef CONFIG_SFC_SRIOV - efx_tc_remove_rep_filters(efx); -#endif + if (IS_ENABLED(CONFIG_SFC_SRIOV)) + efx_tc_remove_rep_filters(efx); down_write(&efx->filter_sem); efx_mcdi_filter_del_vlan(efx, 0); efx_mcdi_filter_del_vlan(efx, EFX_FILTER_VID_UNSPEC); @@ -737,7 +736,6 @@ static unsigned int efx_ef100_recycle_ring_size(const struct efx_nic *efx) return 10 * EFX_RECYCLE_RING_SIZE_10G; } -#ifdef CONFIG_SFC_SRIOV static int efx_ef100_get_base_mport(struct efx_nic *efx) { struct ef100_nic_data *nic_data = efx->nic_data; @@ -773,7 +771,6 @@ static int efx_ef100_get_base_mport(struct efx_nic *efx) return 0; } -#endif static int compare_versions(const char *a, const char *b) { @@ -1155,10 +1152,9 @@ int ef100_probe_netdev_pf(struct efx_nic *efx) struct net_device *net_dev = efx->net_dev; int rc; - if (!nic_data->grp_mae) + if (!IS_ENABLED(CONFIG_SFC_SRIOV) || !nic_data->grp_mae) return 0; -#ifdef CONFIG_SFC_SRIOV rc = efx_init_struct_tc(efx); if (rc) return rc; @@ -1193,7 +1189,6 @@ int ef100_probe_netdev_pf(struct efx_nic *efx) net_dev->features |= NETIF_F_HW_TC; efx->fixed_features |= NETIF_F_HW_TC; } -#endif return rc; } @@ -1206,12 +1201,11 @@ void ef100_remove(struct efx_nic *efx) { struct ef100_nic_data *nic_data = efx->nic_data; -#ifdef CONFIG_SFC_SRIOV - if (efx->mae) { + if (IS_ENABLED(CONFIG_SFC_SRIOV) && efx->mae) { efx_ef100_fini_reps(efx); efx_fini_mae(efx); } -#endif + efx_mcdi_detach(efx); efx_mcdi_fini(efx); if (nic_data) @@ -1304,9 +1298,8 @@ const struct efx_nic_type ef100_pf_nic_type = { .update_stats = ef100_update_stats, .pull_stats = efx_mcdi_mac_pull_stats, .stop_stats = efx_mcdi_mac_stop_stats, -#ifdef CONFIG_SFC_SRIOV - .sriov_configure = efx_ef100_sriov_configure, -#endif + .sriov_configure = IS_ENABLED(CONFIG_SFC_SRIOV) ? + efx_ef100_sriov_configure : NULL, /* Per-type bar/size configuration not used on ef100. Location of * registers is defined by extended capabilities.