From: Ben Hutchings Subject: [PATCH 0/8] New driver "sfc" for Solarstorm SFC4000 controller (try #9) Date: Wed, 26 Mar 2008 12:12:12 +0000 Message-ID: <20080326121210.GE24160@solarflare.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-net-drivers@solarflare.com, Jeff Garzik , David Miller To: netdev@vger.kernel.org Return-path: Received: from 82-69-137-158.dsl.in-addr.zen.co.uk ([82.69.137.158]:59107 "EHLO uklogin.uk.level5networks.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751554AbYCZMM2 (ORCPT ); Wed, 26 Mar 2008 08:12:28 -0400 Content-Disposition: inline Sender: netdev-owner@vger.kernel.org List-ID: This is a resubmission of a new driver for Solarflare network controllers. The driver supports the 10Xpress PHY and XFP modules on our reference designs SFE4001 and SFE4002 and the SMC models SMC10GPCIe-XFP and SMC10GPCIe-10BT. The previous patches began with: http://marc.info/?l=linux-netdev&m=120528862826282&w=2 Some explanation of the driver structure was posted in: http://marc.info/?l=linux-netdev&m=119999015817920&w=2 Since the last submission we have addressed the review comments: - Remove local unused flags variables. - Removed bogus uses of __attribute__ ((unused)). - Removed conditional use of topology_core_siblings(). - Moved net-dev allocation and freeing from efx_{init,fini}_napi() to efx_pci_{probe,remove}() since we only have one port per PCI function. Moved net-dev initialisation and association from efx_init_napi() to appropriate functions. Removed now-obsolete test that efx->net_dev is non-null. - Simplified Makefile. We have also made a number of bug fixes: - Always reconfigure MAC in efx_start_port(); this avoids potential missed signalling of net carrier. - Moved call to falcon_update_nic_stats() into efx_net_stats() since it now requires NIC stats as well. - Remove useless call to efx_reconfigure_port() from efx_ioctl(). - When fini'ing the tx queue, don't complain about completing empty descriptors. Whilst here make skipping tx descriptors a proper error, and perform an invisible reset. - Don't propagate link state changes whilst the port is down. - Fixed failure path for efx_probe_all() to call efx_remove_nic() if necessary. - Fixed efx_ethtool_set_pauseparam() to propagate result of efx_mac_operations::set_pause. We also did some cleanup and simplification: - Removed declaration of removed function falcon_debugfs_read_hardware_desc(). - Removed unused definitions of EFX_WORD, EFX_DWORD, EFX_OWORD. Removed definition of EFX_BYTE and replaced with literal 1 in sfe4001.c. - Removed unused inline function efx_is_over_4gb(). - Changed return type of efx_probe_interrupts() and efx_select_used() to void, as they can never fail. Simplified efx_probe_nic() accordingly. - Renamed FALCON_BUF_ALIGN to FALCON_BUF_SIZE (it is both size and alignment). Replaced literal 4096 in falcon_alloc_special_buffer() with FALCON_BUF_SIZE. Made use of kernel ALIGN() macro in falcon_alloc_special_buffer(). - Corrected first comment in i2c-direct.c - we don't use I2C for EEPROM access on Falcon. - Removed name format macros only used by ethtool self-tests and debugfs. - Removed SPI-related members from efx_nic, only needed for MTD driver. - Remove the net_dev_registered flag, replacing it with checks on the net_dev->flags field. netif_running() is implicitly false when NET_DEV_REGISTERED()==0, so remove some spurious checks. - Since we always call efx_reconfigure_port() in efx_start_port(), then always drain the port in efx_stop_all(). This removes an unneeded abstraction. We hope the driver is now ready to be merged. Ben. -- Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job.