-
Notifications
You must be signed in to change notification settings - Fork 79
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
43222e9396ac6002481b7a97aca46e6f openonload-201210.tgz
- Loading branch information
Showing
264 changed files
with
14,987 additions
and
8,512 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,130 +1,132 @@ | ||
|
|
||
| OpenOnload 201205-u1 Release Notes | ||
| ================================== | ||
| OpenOnload 201210 Release Notes | ||
| =============================== | ||
|
|
||
| This is an update release of OpenOnload that includes fixes for a number | ||
| of bugs found during testing. Please see the ChangeLog for the full list | ||
| of changes. | ||
| This is a beta release that adds new features to OpenOnload. We would be | ||
| very grateful if you could test this release and report any issues you | ||
| find to support@solarflare.com. | ||
|
|
||
| There are no major new features in this release, so the notes below detail | ||
| features added in the openonload-201205 release. | ||
| Below is a brief summary of the major new features and limitations that we | ||
| are aware of. See the ChangeLog for further details and bugs fixed. | ||
|
|
||
|
|
||
| ---------------------------------------- | ||
| TCP loopback acceleration | ||
| ------------------------- | ||
| New features | ||
| ------------ | ||
|
|
||
| - Use huge pages for Onload packet buffers (EF_USE_HUGE_PAGES). | ||
|
|
||
| - Improvements to onload_stackdump default output and new commands. | ||
|
|
||
| This release adds acceleration of TCP loopback connections. This feature | ||
| is disabled by default; To enable loopback acceleration, set the following | ||
| environment variables: | ||
| - Performance improvements in epoll_wait() and poll(). | ||
|
|
||
| export EF_TCP_SERVER_LOOPBACK=2 | ||
| export EF_TCP_CLIENT_LOOPBACK=2 | ||
| - Physical addressing mode (EF_PACKET_BUFFER_MODE=2 or 3). | ||
|
|
||
| The sockets at both ends of an accelerated TCP loopback connection have to | ||
| reside in the same Onload stack. In the configuration given above, they | ||
| will both reside in the stack of the connecting socket. It is also | ||
| possible to have the connecting socket moved into the stack of the | ||
| listening socket, by setting EF_TCP_CLIENT_LOOPBACK=3. | ||
| - Option to put TCP loopback connections in a separate stack. | ||
|
|
||
| See the documentation for details of other options. | ||
| - sendmmsg() support for accelerated UDP sockets, whether or not GLIBC | ||
| and/or the kernel support this call. | ||
|
|
||
| NOTE: If spinning is enabled, then loopback acceleration may cause threads | ||
| to spin that didn't in previous releases. | ||
| - Net driver: VLAN insertion and stripping support. | ||
|
|
||
| - Firewalling including Iptables integration (BETA). | ||
|
|
||
|
|
||
| ---------------------------------------- | ||
| New configuration options | ||
| ------------------------- | ||
| Configuration options | ||
| --------------------- | ||
|
|
||
| In addition to those mentioned elsewhere, the following configuration | ||
| options have been added. See the manual for more details: | ||
| The following new configuration options have been added: | ||
|
|
||
| EF_TCP_RST_DELAYED_CONN | ||
| - Set this option to prevent data being delivered late by TCP due to | ||
| retransmits. Useful for trading applications where delivering the data | ||
| late is worse than not delivering it at all. | ||
| EF_PACKET_BUFFER_MODE | ||
| - New options 2 and 3 enable physical addressing mode. Physical | ||
| addressing mode is insecure in that the user-level application can | ||
| provoke the network adapter to read and write arbitrary addresses in | ||
| memory. See also phys_mode_gid below. | ||
|
|
||
| EF_DYNAMIC_ACK_THRESH | ||
| - This option reduces the ACK rate when streaming TCP data, which | ||
| increases throughput and reduces CPU utilisation. This option is | ||
| enabled by default; set to 0 to get the old behaviour. | ||
| EF_TCP_CLIENT_LOOPBACK=4 | ||
| - New option 4 causes TCP loopback connections to be placed in a separate | ||
| Onload stack. This can improve performance by reducing contention | ||
| between resources. | ||
|
|
||
| EF_USE_HUGE_PAGES | ||
| - Enable use of huge pages for packet buffers. | ||
|
|
||
| ---------------------------------------- | ||
| Control plane table sizes | ||
| ------------------------- | ||
| EF_VALIDATE_ENV | ||
| - Warn about any obsolete or misspelled options in the environment. | ||
|
|
||
| We've added the following new module parameters to the 'onload' kernel | ||
| module. | ||
|
|
||
| max_layer2_interfaces | ||
| - Sets the maximum number of network interfaces (including physical | ||
| interfaces, vlan interfaces and bonds) supported in Onload's control | ||
| plane. The default is 50. | ||
| ---------------------------------------- | ||
| Module options | ||
| -------------- | ||
|
|
||
| max_routes | ||
| - Sets the maximum number of entries in Onload's route table. The | ||
| default is 256. | ||
| The following module options have been added: | ||
|
|
||
| max_neighs | ||
| - Sets the maximum number of entries in Onload's ARP/neighbour table. | ||
| This is rounded up internally to a power of two. The default is 1024. | ||
| epoll2_max_stacks (onload module) | ||
| - Controls the maximum number of stacks that an epoll fd can handle when | ||
| EF_UL_EPOLL=2. | ||
|
|
||
| phys_mode_gid (onload module) | ||
| - Set this option to enable physical addressing mode and restrict which | ||
| users can use it. | ||
|
|
||
| ---------------------------------------- | ||
| ef_vi Improvements | ||
| ------------------ | ||
| shared_buffer_table (sfc_resource module) | ||
| - This module option needs to be set to enable legacy ef_vi applications | ||
| that use the ef_iobufset API. For example, setting | ||
| shared_buffer_table=10000 makes 10000 buffer table entries available for | ||
| use with ef_iobufset. | ||
|
|
||
| ef_vi has been extended with two new types: | ||
|
|
||
| - ef_pd: A protection domain essentially defines an address space. | ||
| The following module options have been removed: | ||
|
|
||
| - ef_memreg: Allows you to register ordinary program memory for DMA. | ||
| unsafe_sriov_without_iommu (sfc_resource module) | ||
| - Obsoleted by the new physical addressing modes and phys_mode_gid. | ||
|
|
||
| Together these increase flexibility significantly. Multiple VIs can be | ||
| allocated within a protection domain, and any VI in the domain can access | ||
| any memory registered in the domain. | ||
| buffer_table_min (sfc_resource module) | ||
| buffer_table_max | ||
| - These options existed because previous versions of the sfc_resource | ||
| driver required uniform resources across all network adapters. That | ||
| caused problems when different types of adapter were placed together in | ||
| one host. That limitation is no longer present, so these module options | ||
| are not needed. | ||
|
|
||
| We've also added an option (EF_PD_VF) to allocate VIs within SR-IOV VFs. | ||
| The advantages of this are reduced latency and no limit on the amount of | ||
| memory that can be registered. | ||
|
|
||
| Another option (EF_PD_PHYS_MODE) allows a VI to use physical addresses | ||
| rather than protected I/O addresses. This option is only available to | ||
| processes running as root, as it allows the application to cause the | ||
| adapter to read and write arbitrary addresses in memory. | ||
| ---------------------------------------- | ||
| Iptables and firewalling | ||
| ------------------------ | ||
|
|
||
| The interface between the ef_vi user-space library and the driver is | ||
| backwards compatible with the following earlier releases: | ||
| This release adds a BETA of support for iptables integration. This | ||
| feature allows the administrator to place limits on which flows are | ||
| permitted to bypass the kernel stack. Flows that are not permitted to | ||
| bypass the kernel stack are therefore subject to iptables rules (if | ||
| enabled) in the normal way. | ||
|
|
||
| openonload-201109-u2 | ||
| enterpriseonload-2.0.0.x | ||
| This feature does not perform iptables rules checking on the adapter | ||
| itself. | ||
|
|
||
| ef_vi applications built against those releases will continue to work with | ||
| the drivers in this release. | ||
| The onload_iptables script provides a way to configure the Onload firewall | ||
| automatically so that a subset of iptables rules will be respected when | ||
| using Onload. The script should tell you about any iptables rules that | ||
| are incompatible with Onload acceleration. | ||
|
|
||
|
|
||
| ---------------------------------------- | ||
| Scalable packet buffer mode | ||
| --------------------------- | ||
| ef_iobufset is deprecated | ||
| ------------------------- | ||
|
|
||
| Support for SR-IOV is disabled on 32-bit kernels. This means the | ||
| following features are not available on 32-bit kernels: | ||
| The ef_iobufset interface in ef_vi is deprecated and will be removed in a | ||
| future release. We recommend you migrate ef_vi applications to the | ||
| ef_memreg interface instead. | ||
|
|
||
| - Scalable packet buffer mode (EF_PACKET_BUFFER_MODE=1) | ||
| - ef_vi with VFs | ||
|
|
||
| ---------------------------------------- | ||
| Known bugs | ||
| ---------- | ||
|
|
||
| On some recent kernel versions, configuring the adapter to have a large | ||
| number of VFs (via sfboot) can cause kernel panics. This problem affects | ||
| kernel versions in the range 3.0 to 3.3 inclusive, and is due to netlink | ||
| messages that include information about network interfaces growing too | ||
| large. | ||
| - bug31498: sendmmsg() returns error code incorrectly. | ||
|
|
||
| This problem can be avoided by ensuring that the total number of physical | ||
| network interfaces, including VFs, is no more than 30. | ||
| - bug31280: Build failure with gcc 4.7.1 due to string concatenation. | ||
|
|
||
|
|
||
| David Riddoch | ||
| 2012/07/05 | ||
| 2012/10/09 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.