This patch add the support of proxy multicast, ie being able to build a static
multicast tree. It adds the support of (*,*) and (*,G) entries.
The user should define an (*,*) entry which is not used for real forwarding.
This entry defines the upstream in iif and contains all interfaces from the
static tree in its oifs. It will be used to forward packet upstream when they
come from an interface belonging to the static tree.
Hence, the user should define (*,G) entries to build its static tree. Note that
upstream interface must be part of oifs: packets are sent to all oifs
interfaces except the input interface. This ensures to always join the whole
static tree, even if the packet is not coming from the upstream interface.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This will ease further addition of new MRT[6]_* values and avoid to update
in6.h each time.
Note that we reduce the maximum value from 210 to 209, but 210 does not match
any known value in ip[6]_mroute_setsockopt().
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The original intent of this file was to list limitations in
drivers/hardware relating to multicast use, back when some
modest hardware from the early 1990s did not support things
we might take for granted today.
I was intending to delete some now-gone MCA/token ring entries
in this file, but once I opened it, I found it only contained
information on the earliest (pre-2000) linux networking drivers.
Checking the git history shows that the file hasn't been touched
since 2005. Clearly nobody is actively consulting this file
as a meaningful reference.
Rather than add a "YES YES YES" line for all of the drivers we
currently have, lets just take advantage of the fact that nobody
is using the file to delete it.
This has the side benefit of not having to do a line-by-line
deletion of the file content as each older driver is expired.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With the previous patch the state transition handling of the reset
code itself is now (hopefully) race free and solid. But that still
leaves out everyone else - with the various lock-free wait paths
we have there's the possibility that the reset happens between the
point where we read the seqno we should wait on and the actual wait.
And if __wait_seqno then never sees the RESET_IN_PROGRESS state, we'll
happily wait for a seqno which will in all likelyhood never signal.
In practice this is not a big problem since the X server gets
constantly interrupted, and can then submit more work (hopefully) to
unblock everyone else: As soon as a new seqno write lands, all waiters
will unblock. But running the i-g-t reset testcase ZZ_hangman can
expose this race, especially on slower hw with fewer cpu cores.
Now looking forward to ARB_robustness and friends that's not the best
possible behaviour, hence this patch adds a reset_counter to be able
to detect any reset, even if a given thread never observed the
in-progress state.
The important part is to correctly order things:
- The write side needs to increment the counter after any seqno gets
reset. Hence we need to do that at the end of the reset work, and
again wake everyone up. We also need to place a barrier in between
any possible seqno changes and the counter increment, since any
unlock operations only guarantee that nothing leaks out, but not
that at later load operation gets moved ahead.
- On the read side we need to ensure that no reset can sneak in and
invalidate the seqno. In all cases we can use the one-sided barrier
that unlock operations guarantee (of the lock protecting the
respective seqno/ring pair) to ensure correct ordering. Hence it is
sufficient to place the atomic read before the mutex/spin_unlock and
no additional barriers are required.
The end-result of all this is that we need to wake up everyone twice
in a reset operation:
- First, before the reset starts, to get any lockholders of the locks,
so that the reset can proceed.
- Second, after the reset is completed, to allow waiters to properly
and reliably detect the reset condition and bail out.
I admit that this entire reset_counter thing smells a bit like
overkill, but I think it's justified since it makes it really explicit
what the bail-out condition is. And we need a reset counter anyway to
implement ARB_robustness, and imo with finer-grained locking on the
horizont this is the most resilient scheme I could think of.
v2: Drop spurious change in the wait_for_error EXIT_COND - we only
need to wait until we leave the reset-in-progress wedged state.
v3: Don't play tricks with barriers in the throttle ioctl, the
spin_unlock is barrier enough.
I've also considered using a little helper to grab the current
reset_counter, but then decided that hiding the atomic_read isn't a
great idea, since having it explicitly show up in the code is a nice
remainder to reviews to check the memory barriers.
v4: Add a comment to explain why we need to fall through in
__wait_seqno in the end variable assignments.
v5: Review from Damien:
- s/smb/smp/ in a comment
- don't increment the reset counter after we've set it to WEDGED. Now
we (again) properly wedge the gpu when the reset fails.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/usb/gadget/u_serial.c:1291:5: sparse: symbol \
'userial_init' was not declared. Should it be static?
drivers/usb/gadget/zero.c:66:25: sparse: symbol \
'gzero_options' was not declared. Should it be static?
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Use usb_gstrings_attach() to assign strings in f_acm to assign strings
ids.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The USB strings don't (yet) fully work in multiple configs/gadget
environment. The string id is assigned to the descriptor and the struct
usb_strings. We create a copy of the individual descriptor so we don't
clash if we use a function more than once. However, we have only one
struct usb_string for each string.
Currently each function which is used multiple times checks for
"id != 0" and only assigns string ids if it did not happen yet. This
works well if we use the same function multiple times as long as we do
it within the "one" gadget we have. Trouble starts once we use the same
function in a second gadget.
In order to solve this I introduce usb_gstrings_attach(). This function will
crate a copy all structs except for the strings which are not copied.
After the copy it will assign USB ids and attach it to cdev.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The configfs can't use all of composite's hooks because ->bind() and
->unbind() has to be done a little differently. ->disconnect() and
->setup() on the hand can be recycled.
This patch exports them both so configfs can use them.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch factors out two helper functions from composite_bind()
that is composite_dev_prepare() and its counterpart
composite_dev_cleanup().
This will be used by the configfs which requries a slightly different
bind/setup code because part of its configurations (i.e. config
descripts, cdev, …) are setup in advance and VID/PID and so one should
not be overwritten. Also the setup of ep0 endpoint can be delayed until
the UDC is assigned.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch adds udc_attach_driver() which allows to bind an UDC which is
specified by name to a driver. The name of available UDCs can be
obtained from /sys/class/udc. This interface is intended for configfs
interface.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
We can have multiple usb_functions which origin is the same "instance".
Within one USB configuration there should be only one function of an
instance. This back pointer helps configfs to recoginze to which
instance a given usb_function belongs.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch converts the acm_ms gadget to make use of the function
framework to request the ACM function.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch converts the acm_ms gadget to make use of the function
framework to request the ACM function.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch converts the acm_ms gadget to make use of the function
framework to request the ACM function.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch converts f_acm into a module which uses the new function
interface. It also converts one of its users that is g_serial to make
use of it. The other users of it (g_nokia for instance) are still using
the old include file system and should not notice the change at all. So
they can be converter later independently.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch removes gserial_setup() and gserial_cleanup() and adds
gserial_alloc_line() and gserial_free_line() to replace them.
The initial setup of u_serial happens now on module load time. A
maximum of four TTY ports can be requested which is the current limit.
In theory we could extend this limit, the hard limit is the number of
available endpoints.
alloc_tty_driver() is now called at module init time with the max
available ports. The per-line footprint here is on 32bit is 3 * size of
pointer + 60 bytes (for cdevs).
The remaining memory (struct gs_port) is allocated once a port is
requested.
With this change it is possible to load g_multi and g_serial at the same
time. GS0 receives the module that is loaded first, GS1 is received by
the next module and so on. With the configfs interface the port number
can be exported and the device node is more predictable. Nothing changes
for g_serial and friends as long as one module is used.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This will be used to remove a single function from a given config. Right
now "ignore" that an error at ->bind() time and cleanup later during
composite_unbind() / remove_config().
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Every user of u_serial has now to select the U_SERIAL symbol instead of
including the file.
There is one limition with this: ports and and gs_tty_driver are global
variables in u_serial. Since all users share them, there can be only one
user loaded at a time i.e. either g_serial or g_nokia.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch factors out the three possible functions into three possible
bind functions which are passed as an argument to usb_add_config(). This
will ease the step by step converting of the individual functions to the
new function registration method.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The significant part of this function was removed in 90f7976 ("USB:
Remove unsupported usb gadget drivers"). I would move this to function
bind time but I don't see the point in moving an empty function.
Therefore bye bye.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch converts the f_sourcesink and f_loopback file to the USB-function
module. Both functions shares a few common utility functions which are
currently implemented in g_zero.c itself. This patch moves the common
code into the sourcesink file and creates one module out of the the two
functions (source sink and loop back).
The g_zero gadget is function specific to source sink and loop back to
set a few options. This Symbol dependency enforces a modul load right
now.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This patch provides an infrastructure to register & unregister a USB
function. This allows to turn a function into a module and avoid the
'#include "f_.*.c"' magic and we get a clear API / cut between the bare
gadget and its functions.
The concept is simple:
Each function defines the DECLARE_USB_FUNCTION_INIT macro whith an unique
name of the function and two allocation functions.
- one to create an "instance". The instance holds the current configuration
set. In case there are two usb_configudations with one function there will
be one instance and two usb_functions
- one to create an "function" from the instance.
The name of the instance is used to automaticaly load the module if it the
instance is not yet available.
The usb_function callbacks are slightly modified and extended:
- usb_get_function()
creates a struct usb_function inclunding all pointers (bind,
unbind,…). It uses the "instance" to map its configuration. So we can
have _two_ struct usb_function, one for each usb_configuration.
- ->unbind()
Since the struct usb_function was not allocated in ->bind() it should
not kfree()d here. This function should only reverse what happens in
->bind() that is request cleanup and the cleanup of allocated
descriptors.
- ->free_func()
a simple kfree() of the struct usb_function
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
f_loopback should only include the bare function but it also includes
the config descriptor. This patch moves the config descriptor into
zero.c, the only user of this function.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
f_sourcesink should only include the bare function but it also includes
the config descriptor. This patch moves the config descriptor into
zero.c, the only user of this function.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The setup request can be sent to an interface/endpoint or to the device
itself. If it is sent to an interface / endpoint then we forward it to
the function that is mapped to that interface / endpoint.
If the device is the target of the setup request then we forward it to the
->setup() callback of the currently active configuration.
In case of the sourcesink function the requests are function specific
but are sent to the device.
This patch introduces a setup wrapper at configuration level which
forwards the request to the function. By using this wrapper we can keep
the function specific code within the function file and we need just a
hint at config level to forward the request.
The here introduced global variable will be moved into the gadget (which
combines the two functions) in a later patch.
SourceSink is currently the only function using ->setup() at config level.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Removes a couple of:
|drivers/usb/gadget/s3c-hsudc.c: In function 's3c_hsudc_epin_intr':
|drivers/usb/gadget/s3c-hsudc.c:438:2: warning: passing argument 1 of '__raw_readl' makes pointer from integer without a cast
|arch/arm/include/asm/io.h:104:19: note: expected 'const volatile void *' but argument is of type 'unsigned int'
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Lets assume nokia_bind() starts with "return -EINVAL". After loading the
gadget we end up with:
|udc dummy_udc.0: registering UDC driver [g_nokia]
|BUG: unable to handle kernel NULL pointer dereference at 00000040
|IP: [<c11f9555>] __list_add+0x25/0xf0
|Call Trace:
| [<c12d4e21>] rollback_registered+0x21/0x40
| [<c12d513f>] unregister_netdevice_queue+0x4f/0xa0
| [<c12d5259>] unregister_netdev+0x19/0x30
| [<f81335b2>] gphonet_cleanup+0x32/0x50 [g_nokia]
| [<f8133f1c>] nokia_unbind+0x1c/0x2a [g_nokia]
| [<f802509f>] __composite_unbind.constprop.10+0x4f/0xb0 [libcomposite]
| [<f80255be>] composite_bind+0x1ae/0x230 [libcomposite]
| [<c129e576>] usb_gadget_probe_driver+0xc6/0x1b0
| [<f8024aba>] usb_composite_probe+0x7a/0xa0 [libcomposite]
That is crash from nokia_unbind() invoked via nokia_bind(). This crash
will look different we if make it until usb_string_ids_tab() before we
enter an error condition in the probe function.
nokia_bind_config() tries to clean up which is IMHO the right thing to
do. Leaving things as-is and hoping that its unbind() will clean it up
is kinda backwards. Especially since the bind function never succeeded so
it can't know how much it needs to clean up.
This fixes the behaviour by not calling the driver's unbind function if
its bind function failed.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Commit 46d3ceab "tcp: TCP Small Queues" has severly degraded
performance for single connection RR workloads on HiperSockets with
MTU >=16K due to a conflict of the TCP Small Queues approach with our
buffer scan threshold which releases buffers not frequently enough yet.
This fix restores performance to the same level as before cited commit.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As of now, s390dbf entries for the cards are discarded as soon as the
device is removed. However, this will also bar us of all chances of
getting valuable debug information after a device has been removed.
This patch will keep the s390dbf entries around until the qeth module
is removed.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Reviewed-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Refer to virtual NICs instead of GuestLANs.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing port isolation mode 'forward' will now verify that the adjacent
switch port supports the required reflective relay (RR) mode. This patch adds
the required error handling for the cases where enabling port isolation mode
'forward' can now fail.
Furthermore, once established, we never fall back from one of the port
isolation modes to a non-isolated mode without further user-interaction.
This includes cases where the isolation mode was enabled successfully, but
ceases to work e.g. due to configuration changes at the switch port.
Finally, configuring an isolation mode with the device being offline
will make onlining the device fail permanently upon errors encountered until
either errors are resolved or the isolation mode is changed by the user to a
different mode.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Reviewed-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove exports that are not used anywhere else.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The previous code did never retry any idx setup unless retries were done
for device offline/online at the beginning of the function.
Signed-off-by: Stefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Reviewed-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
YOSHIFUJI Hideaki says:
====================
This series of changes basically clean up NDISC logic,
especially on sender side.
We originally do For NS/NA/RS:
1) build temporary ICMPv6 header
2) ndisc_build_skb() with temporary ICMPv6 header and rather
criptic arguments.
- Calculate total length and allocate sk_buff
- Build IPv6 header.
- copy ICMPv6 header, additional data and ND options.
- Fill-in ICMPv6 checksum.
Here, structures defined for message format was not used
at all, it is difficult to understand what is being sent,
and it was not generic.
3) __ndisc_send()
- Allocate temporary dst.
- Send it.
Several issues:
- We could not defer decision if we should/can send some ND
option.
- It is hard to see the packet format at a glance.
- ICMPv6 header was built as temporary variable, and then
copied to the buffer.
- Some code path for Redirect was not shared.
With these patches, we do:
1) Calculate (or estimate) message length and option length.
2) Allocate skb (via new ndisc_skb_alloc()).
3) Fill-in ICMPv6 message directly using compound literals.
4) Fill-in ICMPv6 checksum
5) Build IPv6 header (including length)
6) Send the packet (via ndisc_send_skb()).
- allocate temporary dst and send it.
- We can defer calculating real length of the packet.
For example, we can give up filling some option at when
filling in.
- Message is built directly without temporary buffer.
- Structures defined for message format is easier to understand
what is being built.
- NS/NA/RS/Redirect share same logic.
- Reduced code/data size:
text data bss dec hex filename
265407 14133 3488 283028 45194 old/net/ipv6/ipv6.o
264955 14109 3488 282552 44fb8 new/net/ipv6/ipv6.o
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the UART2 muxing data to the board file (this used to be,
erroneously, done in the bootloader).
Cc: stable <stable@vger.kernel.org> [3.7]
Signed-off-by: Luciano Coelho <coelho@ti.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
The iterator correctly handles of_node_put() calls.
Remove it before continue'ing the loop.
Without this patch you get the following with
CONFIG_OF_DYNAMIC set:
ERROR: Bad of_node_put() on /ocp/timer@44e31000!
[<c001329c>] (unwind_backtrace+0x0/0xe0) from [<c03dd8f0>] (of_node_release+0x2c/0xa0)!
[<c03dd8f0>] (of_node_release+0x2c/0xa0) from [<c03ddea0>] (of_find_matching_node_and_match+0x78/0x90)!
[<c03ddea0>] (of_find_matching_node_and_match+0x78/0x90) from [<c06d349c>] (omap_get_timer_dt+0x78/0x90)!
[<c06d349c>] (omap_get_timer_dt+0x78/0x90) from [<c06d3664>] (omap_dm_timer_init_one.clone.2+0x34/0x2bc)!
[<c06d3664>] (omap_dm_timer_init_one.clone.2+0x34/0x2bc) from [<c06d3a2c>] (omap2_gptimer_clocksource_init.clone.4+0x24/0xa8)!
[<c06d3a2c>] (omap2_gptimer_clocksource_init.clone.4+0x24/0xa8) from [<c06cca58>] (time_init+0x20/0x30)!
[<c06cca58>] (time_init+0x20/0x30) from [<c06c9690>] (start_kernel+0x1a8/0x2fc)!
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Acked-by: Jon Hunter <jon-hunter@ti.com>
[tony@atomide.com: updated description per Jon]
Signed-off-by: Tony Lindgren <tony@atomide.com>
Otherwise we will get:
WARNING: vmlinux.o(.text+0x1d4f0): Section mismatch in reference from the
function omap_init_ocp2scp() to the function .init.text:omap_device_build()
The function omap_init_ocp2scp() references
the function __init omap_device_build().
This is often because omap_init_ocp2scp lacks a __init
annotation or the annotation of omap_device_build is wrong.
Signed-off-by: Tony Lindgren <tony@atomide.com>
Construct NS/NA/RS message directly using C99 compound literals.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
skb_transport_header() (thus icmp6_hdr()) is available here,
use it.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Build ICMPv6 message first and make buffer management easier;
we can use skb->len when filling checksum in ICMPv6 header,
and then build IP header with length field.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
