Karsten Graul says:
====================
More patches to address abnormal termination processing of
sockets and link groups.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
With the introduction of the link group termination worker there is
no longer a need to postpone smc_close_active_abort() to a worker.
To protect socket destruction due to normal and abnormal socket
closing, the socket refcount is increased.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Use a worker for link group termination to guarantee process context.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
If a link group and its connections must be terminated,
* wake up socket waiters
* do not enable buffer reuse
A linkgroup might be terminated while normal connection closing
is running. Avoid buffer reuse and its related LLC DELETE RKEY
call, if linkgroup termination has started. And use the earliest
indication of linkgroup termination possible, namely the removal
from the linkgroup list.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
There are lots of link group termination scenarios. Most of them
still allow to inform the peer of the terminating sockets about aborting.
This patch tries to call smc_close_abort() for terminating sockets.
And the internal TCP socket is reset with tcp_abort().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Usually link groups are freed delayed to enable quick connection
creation for a follow-on SMC socket. Terminated link groups are
freed faster. This patch makes sure, fast schedule of link group
freeing is not rescheduled by a delayed schedule. And it makes sure
link group freeing is not rescheduled, if the real freeing is already
running.
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Locking hierarchy requires that the link group conns_lock can be
taken if the socket lock is held, but not vice versa. Nevertheless
socket termination during abnormal link group termination should
be protected by the socket lock.
This patch reduces the time segments the link group conns_lock is
held to enable usage of lock_sock in smc_lgr_terminate().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
When a link group is to be terminated, it is sufficient to hold
the lgr lock when unlinking the link group from its list.
Move the lock-protected link group unlinking into smc_lgr_terminate().
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
The resources for a terminated socket are being cleaned up.
This patch makes sure
* no more data is received for an actively terminated socket
* no more data is sent for an actively or passively terminated socket
Signed-off-by: Ursula Braun <ubraun@linux.ibm.com>
Signed-off-by: Karsten Graul <kgraul@linux.ibm.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Extend the size of QSFP EEPROM for the cable types SSF8436 and SFF8636
from 256 to 640 bytes in order to expose all the EEPROM pages by
ethtool.
For SFF-8636 and SFF-8436 specifications, the driver exposes 256 bytes
of data for ethtool's get_module_eeprom() callback. This is because the
driver uses the below defines to specify SFF module length in ethtool's
get_module_info() callback:
'ETH_MODULE_SFF_8636_LEN' and 'ETH_MODULE_SFF_8436_LEN' (both are 256).
As a result of exposing 256 bytes only, ethtool shows wrong "zero" info
for pages 1, 2, 3.
The patch changes the length returned by callback for get_module_info()
to the values from the next defines: 'ETH_MODULE_SFF_8636_MAX_LEN' and
'ETH_MODULE_SFF_8436_MAX_LEN' (both are 640) to allow exposing of upper
page 1, 2 and 3.
Signed-off-by: Vadim Pasternak <vadimp@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Provide a macro for getting QSFP module EEPROM page number from the
optional upper page number row offset, specified in request.
Signed-off-by: Vadim Pasternak <vadimp@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Do some cleanup of the netback init and deinit code:
- add an omnipotent queue deinit function usable from
xenvif_disconnect_data() and the error path of xenvif_connect_data()
- only install the irq handlers after initializing all relevant items
(especially the kthreads related to the queue)
- there is no need to use get_task_struct() after creating a kthread
and using put_task_struct() again after having stopped it.
- use kthread_run() instead of kthread_create() to spare the call of
wake_up_process().
Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Paul Durrant <pdurrant@gmail.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Hayes Wang says:
====================
Support loading the firmware of the PHY with the type of RTL_FW_PHY_NC.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Support the firmware of PHY NC which is used to fix the issue found
for PHY. Currently, only RTL_VER_04, RTL_VER_05, and RTL_VER_06 need
it.
The order of loading PHY firmware would be
RTL_FW_PHY_START
RTL_FW_PHY_NC
RTL_FW_PHY_STOP
The RTL_FW_PHY_START/RTL_FW_PHY_STOP are used to lock/unlock the PHY,
and set/clear the patch key from the firmware file.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Move r8153_patch_request() forward for later patch.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Make sure @fw_offset field of struct fw_mac is more than the size
of struct fw_mac.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
The struct fw_type_1 is used by MAC only, so rename it to a meaningful one.
Besides, adjust two messages. Replace "load xxx fail" with "check xxx fail"
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Jeff Kirsher says:
====================
1GbE Intel Wired LAN Driver Updates 2019-10-21
This series contains updates to e1000e and igc only.
Sasha adds stream control transmission protocol (SCTP) CRC checksum
support for igc. Also added S0ix support to the e1000e driver. Then
added multicast support by adding the address list to the MTA table and
providing the option for IPv6 address for igc. In addition, added
receive checksum support to igc as well. Lastly, cleaned up some code
that was not fully implemented yet for the VLAN filter table array.
v2: Dropped patch 1 & 2 from the original series. Patch 1 is being sent
to 'net' tree as a fix and patch 2 implementation needs to be
re-worked. Updated the patch to add support for S0ix to fix the
reverse Xmas tree issues and made the entry/exit functions void
since they constantly returned success. All based on community
feedback.
v3: Cleaned up patch 4 of the series based on feedback from the
community. Cleaned up a stray comma in a code comment and removed
the 'inline' of a function that would be inlined by the compiler
anyways.
====================
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Use tcf_tm_dump(), instead of an open coded variant (no functional change
in this patch).
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Heiner Kallweit says:
====================
net: phy: marvell: support downshift as PHY tunable
So far downshift is implemented for one small use case only and can't
be controlled from userspace. So let's implement this feature properly
as a PHY tunable so that it can be controlled via ethtool.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of superseded function marvell_set_downshift() we can use new
function m88e1111_set_downshift() in m88e1116r_config_init().
For this m88e1116r_config_init() has to be moved in the code.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So far downshift is implemented for one small use case only and can't
be controlled from userspace. So let's implement this feature properly
as a PHY tunable so that it can be controlled via ethtool.
More Marvell PHY's may support downshift, but I restricted it for now
to the ones where I have the datasheet.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Added test cases for IP header operations:
- set tos/precedence
- add value to tos/precedence
- clear tos/precedence
- invert tos/precedence
Signed-off-by: Roman Mashak <mrv@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lorenzo Bianconi says:
====================
add XDP support to mvneta driver
Add XDP support to mvneta driver for devices that rely on software
buffer management. Supported verdicts are:
- XDP_DROP
- XDP_PASS
- XDP_REDIRECT
- XDP_TX
Moreover set ndo_xdp_xmit net_device_ops function pointer in order
to support redirecting from other device (e.g. virtio-net).
Convert mvneta driver to page_pool API.
This series is based on previous work done by Jesper and Ilias.
We will send follow-up patches to reduce DMA-sync operations.
Changes since v4:
- reset page_pool pointer to NULL in mvneta_rxq_drop_pkts and in
mvneta_create_page_pool error path
- move dma sync in mvneta_rx_refill() in patch 2/7
- verify bpf prog pointer in mvneta_xdp_setup to double-check if
stop/start is really necessary
- coding style fixes
Changes since v3:
- rename MVNETA_XDP_CONSUMED in MVNETA_XDP_DROPPED
- squash patch 4/8 and patch 3/8
- fix dma sync for XDP_TX verdict
- fix queue_index in xdp_rxq_info_reg
- cosmetics
Changes since v2:
- rely on page_pool_recycle_direct instead of xdp_return_buff for XDP_DROP
- define xdp buffer in mvneta_rx_swbm and avoid default initializations
- use dma_sync_single_for_cpu instead of dma_sync_single_range_for_cpu
- run page_pool_release_page in mvneta_swbm_add_rx_fragment even if
the buffer contains just ETH_FCS
Changes since v1:
- sync dma buffers before refilling hw queues
- fix stats accounting
Changes since RFC:
- implement XDP_TX
- make tx pending buffer list agnostic
- code refactoring
- check if device is running in mvneta_xdp_setup
====================
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement XDP_TX verdict and ndo_xdp_xmit net_device_ops function
pointer
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow tx buffer array to contain both skb and xdp buffers in order to
enable xdp frame recycling adding XDP_TX verdict support
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move data buffer prefetch in mvneta_swbm_rx_frame after
dma_sync_single_range_for_cpu
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Refactor mvneta_rx_swbm code introducing mvneta_swbm_rx_frame and
mvneta_swbm_add_rx_fragment routines. Rely on build_skb in oreder to
allocate skb since the previous patch introduced buffer recycling using
the page_pool API.
This patch fixes even an issue in the original driver where dma buffers
are accessed before dma sync.
mvneta driver can run on not cache coherent devices so it is
necessary to sync DMA buffers before sending them to the device
in order to avoid memory corruptions. Running perf analysis we can
see a performance cost associated with this DMA-sync (anyway it is
already there in the original driver code). In follow up patches we
will add more logic to reduce DMA-sync as much as possible.
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the page_pool api for allocations and DMA handling instead of
__dev_alloc_page()/dma_map_page() and free_page()/dma_unmap_page().
Pages are unmapped using page_pool_release_page before packets
go into the network stack.
The page_pool API offers buffer recycling capabilities for XDP but
allocates one page per packet, unless the driver splits and manages
the allocated page.
This is a preliminary patch to add XDP support to mvneta driver
Signed-off-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce mvneta_update_stats routine to collect {rx/tx} statistics
(packets and bytes). This is a preliminary patch to add XDP support to
mvneta driver
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
VLAN filter table array not implemented yet and shadow_vfta pointer
not used. Clean up the code and remove the unused shadow_vfta pointer.
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Extend the socket buffer field process and add Rx checksum functionality
Minor: fix indentation with tab instead of spaces.
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add multicast addresses list to the MTA table.
Implement basic Rx mode support.
Add option for IPv6 address settings.
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Implement flow for S0ix support. Modern SoCs support S0ix low power
states during idle periods, which are sub-states of ACPI S0 that increase
power saving while supporting an instant-on experience for providing
lower latency that ACPI S0. The S0ix states shut off parts of the SoC
when they are not in use, while still maintaning optimal performance.
This patch add support for S0ix started from an Ice Lake platform.
Suggested-by: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>
Signed-off-by: Vitaly Lifshits <vitaly.lifshits@intel.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@linux.intel.com>
Signed-off-by: Sasha Neftin <sasha.neftin@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Huazhong Tan says:
====================
net: hns3: add some cleanups & optimizations
This patchset includes some cleanups and optimizations for the HNS3
ethernet driver.
[patch 1/8] removes unused and unnecessary structures.
[patch 2/8] uses a ETH_ALEN u8 array to replace two mac_addr_*
field in struct hclge_mac_mgr_tbl_entry_cmd.
[patch 3/8] optimizes the barrier used in the IO path.
[patch 4/8] introduces macro ring_to_netdev() to get netdevive
from struct hns3_enet_ring variable.
[patch 5/8] makes struct hns3_enet_ring to be cacheline aligned
[patch 6/8] adds a minor cleanup for hns3_handle_rx_bd().
[patch 7/8] removes linear data allocating for fraglist SKB.
[patch 8/8] clears hardware error when resetting.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When device is resetting, the CMDQ service may be stopped until
reset completed. If a new RAS error occurs at this moment, it
will no be able to clear the RAS source. This patch fixes it
by clear the RAS source after reset complete.
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, napi_alloc_skb() is used to allocate skb for fraglist
when the head skb is not enough to hold the remaining data, and
the remaining data is added to the frags part of the fraglist skb,
leaving the linear part unused.
So this patch passes length of 0 to allocate fraglist skb with
zero size of linear data.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit e559709505 ("net: hns3: Add handling of GRO Pkts
not fully RX'ed in NAPI poll"), ring->skb is used to record the
current SKB when processing the RX BD in hns3_handle_rx_bd(),
so the parameter out_skb is unnecessary.
This patch also adjusts the err checking to reduce duplication
in hns3_handle_rx_bd(), and "err == -ENXIO" is rare case, so put
it in the unlikely annotation.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since struct hns3_enet_ring is a frequently used in critical data
path, so make it cacheline aligned as struct hns3_enet_tqp_vector.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are a few places that need to access the netdev of a ring
through ring->tqp->handle->kinfo.netdev, and ring->tqp is a struct
which both in enet and hclge modules, it is better to use the
struct that is only used in enet module.
This patch adds the ring_to_netdev() to access the netdev of ring
through ring->tqp_vector->napi.dev.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the TX and RX ring in a queue is bounded to the
same IRQ, there may be unnecessary barrier op when only one of
the ring need to be processed.
This patch adjusts the location of rmb() in hns3_clean_tx_ring()
and adds a checking in hns3_clean_rx_ring() to avoid unnecessary
barrier op when there is nothing to do for the ring.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
mac_addr_hi32 and mac_addr_lo16 are used to store the MAC address
for management table. But using array of mac_addr[ETH_ALEN] would
be more general and not need to care about the big-endian mode of
the CPU.
Signed-off-by: Guojia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Only the queue_index field in struct hns3_nic_ring_data is
used, other field is unused and unnecessary for hns3 driver,
so this patch removes it and move the queue_index field to
hns3_enet_ring.
This patch also removes an unused struct hns_queue declaration.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking fixes from David Miller:
"I was battling a cold after some recent trips, so quite a bit piled up
meanwhile, sorry about that.
Highlights:
1) Fix fd leak in various bpf selftests, from Brian Vazquez.
2) Fix crash in xsk when device doesn't support some methods, from
Magnus Karlsson.
3) Fix various leaks and use-after-free in rxrpc, from David Howells.
4) Fix several SKB leaks due to confusion of who owns an SKB and who
should release it in the llc code. From Eric Biggers.
5) Kill a bunc of KCSAN warnings in TCP, from Eric Dumazet.
6) Jumbo packets don't work after resume on r8169, as the BIOS resets
the chip into non-jumbo mode during suspend. From Heiner Kallweit.
7) Corrupt L2 header during MPLS push, from Davide Caratti.
8) Prevent possible infinite loop in tc_ctl_action, from Eric
Dumazet.
9) Get register bits right in bcmgenet driver, based upon chip
version. From Florian Fainelli.
10) Fix mutex problems in microchip DSA driver, from Marek Vasut.
11) Cure race between route lookup and invalidation in ipv4, from Wei
Wang.
12) Fix performance regression due to false sharing in 'net'
structure, from Eric Dumazet"
* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (145 commits)
net: reorder 'struct net' fields to avoid false sharing
net: dsa: fix switch tree list
net: ethernet: dwmac-sun8i: show message only when switching to promisc
net: aquantia: add an error handling in aq_nic_set_multicast_list
net: netem: correct the parent's backlog when corrupted packet was dropped
net: netem: fix error path for corrupted GSO frames
macb: propagate errors when getting optional clocks
xen/netback: fix error path of xenvif_connect_data()
net: hns3: fix mis-counting IRQ vector numbers issue
net: usb: lan78xx: Connect PHY before registering MAC
vsock/virtio: discard packets if credit is not respected
vsock/virtio: send a credit update when buffer size is changed
mlxsw: spectrum_trap: Push Ethernet header before reporting trap
net: ensure correct skb->tstamp in various fragmenters
net: bcmgenet: reset 40nm EPHY on energy detect
net: bcmgenet: soft reset 40nm EPHYs before MAC init
net: phy: bcm7xxx: define soft_reset for 40nm EPHY
net: bcmgenet: don't set phydev->link from MAC
net: Update address for MediaTek ethernet driver in MAINTAINERS
ipv4: fix race condition between route lookup and invalidation
...
Intel test robot reported a ~7% regression on TCP_CRR tests
that they bisected to the cited commit.
Indeed, every time a new TCP socket is created or deleted,
the atomic counter net->count is touched (via get_net(net)
and put_net(net) calls)
So cpus might have to reload a contended cache line in
net_hash_mix(net) calls.
We need to reorder 'struct net' fields to move @hash_mix
in a read mostly cache line.
We move in the first cache line fields that can be
dirtied often.
We probably will have to address in a followup patch
the __randomize_layout that was added in linux-4.13,
since this might break our placement choices.
Fixes: 355b985537 ("netns: provide pure entropy for net_hash_mix()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If there are multiple switch trees on the device, only the last one
will be listed, because the arguments of list_add_tail are swapped.
Fixes: 83c0afaec7 ("net: dsa: Add new binding implementation")
Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
