mainly to quiet a warning emitted by GCC 3.3 about comparing
a variable to a value which is larger than the former can hold.
The value was checked to make sure the `np->squeue' array is
not accessed behind its boundary.
This worked due to possibly accidental truncation when
(np->squeueput + 1) was larger than or equal to MAX_START (256)
when it was assigned to `qidx'.
`qidx' is used to hold the next position in the start queue
for an insertion. The new type was chosen because some other
code in the function ncr_freeze_devq() also uses plain integers
to hold those indices.
Wrapped the line after the closing parenthesis of an `if'
condition.
sis_ioctl() was called, so one had to use ifconfig each time the cable got
plugged in to be able to use the connection.
Do it a better way now, add a "in_tick" field in the softc structure,
call timeout() in sis_tick() and don't call it in sis_init() if in_tick is
non-zero.
Reported by: Landmark Networks
Pointy hat to: cognet
Some of the calls to bus_dmamap_sync() were syncing the DMA descriptor
ring maps using the mbuf tag, when they should have been using the
descriptor ring tag instead.
series, the 8139C+ has a descriptor-based DMA mechanism, and its
performance is actually pretty respectable. Note: the 8139D chip does
not support C+ mode. Only the 8139C+ and 8169 gigE chips support C+ mode.
Supported features:
- RX and TX checksum offload
- hardware VLAN tag insertion/extraction
- TX interrupt moderation using the 8139's on-board timer
Everything should be properly busdma'ed and endian-independent, so
things should work ok on non-x86 platforms. Unfortunately, my call
for testers on this code was met with deafening silence, and I don't
have access to any non-x86 FreeBSD boxes at the moment, so this is
speculation.
The device detection code has been cleaned up a little as well
(thanks to Michal Mertl) for the patches.
There are also updates to the rl(4) man page (which I accidentally
checked in before when I updated the dc(4) man page. Oops.)
Todo: finish support for the 8169 gigabit ethernet chip. This
mainly requires writing an rlgphy driver to handle the 8169's built-in
PHY. This will have to wait until I actually get my hands on an 8169
card for testing though. (I still can't find a source for one in the
U.S. Suggestions/pointers welcome.)
- MN-110 10/100 USB ethernet (ADMtek Pegasus II, if_aue)
- MN-120 10/100 cardbus (ADMtek Centaur-C, if_dc)
- MN-130 10/100 PCI (ADMtek Centaur-P, if_dc)
Also update dc(4) man page to mention support for MN-120 and MN-130.
register, present only on 3c90xB and later NICs. This meant that you could
not use a 1500 byte MTU with VLANs on original 3c905/3c900 cards (boomerang
chipset). The boomerang chip does support large frames though, just not
in the same way: you can set the 'allow large frames' bit in the MAC
control register to receive frames up to 4K in size.
Changes:
- Set the 'allow large frames' bit for boomerang chips and increase
the packet size register for cyclone and later chips. This allows
us to use IFCAP_VLAN_MTU on all supported xl(4) NICs.
- Actually set the IFCAP_VLAN_MTU flag in the capabilities word
in xl_attach().
- Change the method used to detect older boomerang chips. My 3c575C
cardbus NIC was being incorrectly identified as 3c90x chip instead
of 3c90xB because the capabilities word in its EEPROM reports
a bizzare value. In addition to checking for the supportsNoTxLength
bit, also check for the absence of the supportsLargePackets bit.
Both of these cases denote a 3c90xB chip.
- Make RX and TX checksums configurable via the SIOCSIFCAP ioctl.
- Avoid an unecessary le32toh() in xl_rxeof(): we already have the
received frame size in the lower 16 bits of rxstat, no need to
read it again.
Tested with 3c905-TX, 3c900-TPO, 3c980C and 3c575C NICs.
to have this driver working on sparc64. It still needs to be made
endian-clean before it can work there.
Special thanks to dragonk@evilcode.net for sending me a dc(4) card so
that I was able to do this work.
Many cheers to all the people that tested this change, thanks to them,
this change shouldn't break anything :-).
Tested by: marcel (i386 and ia64), ru (i386), wilko (alpha),
mbr (i386), wpaul (i386) and
Will Saxon <WillS@housing.ufl.edu> (i386)
BUS_DMA_NOWAIT flag, since the code can't handle this.
- Use NULL, NULL for the lockfunc and lockfuncarg parameters of
bus_dma_tag_create() since deferred loads can't happen now.
forced to do slightly bogus power state manipulation. However, this
is one of those features that is preventing further progress, so mark
them as BURN_BIRDGES like I did for the drivers in sys/dev/...
This, like the other change, are a no-op unless you have BURN_BRIDGES
in your kernel.
Add two new arguments to bus_dma_tag_create(): lockfunc and lockfuncarg.
Lockfunc allows a driver to provide a function for managing its locking
semantics while using busdma. At the moment, this is used for the
asynchronous busdma_swi and callback mechanism. Two lockfunc implementations
are provided: busdma_lock_mutex() performs standard mutex operations on the
mutex that is specified from lockfuncarg. dftl_lock() is a panic
implementation and is defaulted to when NULL, NULL are passed to
bus_dma_tag_create(). The only time that NULL, NULL should ever be used is
when the driver ensures that bus_dmamap_load() will not be deferred.
Drivers that do not provide their own locking can pass
busdma_lock_mutex,&Giant args in order to preserve the former behaviour.
sparc64 and powerpc do not provide real busdma_swi functions, so this is
largely a noop on those platforms. The busdma_swi on is64 is not properly
locked yet, so warnings will be emitted on this platform when busdma
callback deferrals happen.
If anyone gets panics or warnings from dflt_lock() being called, please
let me know right away.
Reviewed by: tmm, gibbs
bit in the EEPROM mode register on. Also, the address must be written
in two 32-bit register accesses instead of 6 8-bit accesses.
Tested with my 8139B cardbus NIC.
PR: kern/35900
Submitted by: Mark Kettenis <kettenis@chello.nl>
mapped I/O mode, we pause for .1 seconds after issuing the reset command
before trying to poll the 'command busy' bit in the status register.
With my 3c575C cardbus NIC, my Sony Picturebook locks up when it tries
to read the status register immediately after the reset. This appears
to be a problem only with certain NICs on certain hardware, but the
added delay should not hurt cards that already work.
This bug seems to have been brought to light by the fact that the xl
driver now defaults to memory mapped I/O mode instead of programmed
I/O mode like it used to. With PIO mode, the delay isn't needed and
everything works (which is why this NIC worked with 5.0-RELEASE but
not 5.1). I suspect that what's happening is that when the chip is
reset, it takes a little while for the memory-mapped decoding logic
to recover. Trying to access the chip's registers during this period
causes an error condition of some kind that wedges the system.
Devices below may experience a change in geometry.
* Due to a bug, aic(4) never used extended geometry. Changes all drives
>1G to now use extended translation.
* sbp(4) drives exactly 1 GB in size now no longer use extended geometry.
* umass(4) drives exactly 1 GB in size now no longer use extended geometry.
For all other controllers in this commit, this should be a no-op.
Looked over by: scottl
are the same that those of the kernel in the KLD_MODULE case. If
we ever want to detect that kind of problems, this is not the right
place to do this since every network driver would be affected by
such desynchronisation.
toggle several media options (sonet/sdh, for example) with ifconfig and
to see the carrier state in ifconfig's output. It gives also read/write
access (given the right privilegs) to the S/Uni registers to user space
programs.
about the driver version in case of an error report. It conflicts with
some other variable of the same name that has been added to the kernel
just recently and there haven't been any bug reports for quite some
time now, anyway ...
bzero(ptr, sizeof(DC_RXLEN * 5));
which should obviously be:
bzero(ptr, DC_RXLEN * 5);
Looks like this bug may have reduced the effectiveness of the
workaround for the hardware bug in the PNIC chips.
MFC after: 1 week
o Remove register keyword
o ANSIfy prototypes
o Remove "return;" at the end of void functions
o Remove trailing spaces
o Don't align local variables with tabs and reorder them
o Don't use /* FOO */ at the end of a #ifdef FOO block if
it's a small block
- Other non-functional changes :
o 6 -> ETHER_ADDR_LEN
o Don't initialize if_output; ether_ifattach() does it for us
865. The APSIZE register has a variable-sized field of enabled bits.
To figure out how many bits a specific host bridge supports, write the
maximum width and see how many bits are set in the hardware. We then
use this mask for setting and getting the aperture size. Prior to this,
the agp(4) driver would treat an aperture size of 256 MB as 128 MB and
would not allocate enough physical memory for the GART as a result.
MFC after: 3 days
Sponsored by: The Weather Channel
Approved by: re (rwatson)
The submitter of PR 32118 told me that this patch also fixes autoselecting
for znyx 4 port cards (10baseT, 100baseTX did work already).
PR: 32118
Reviewed by: imp
Approved by: rwatson (re)
leads to a panic at unload time, as we own 2 instances of callout and
untimeout() only one.
Will I'm there, remove a call to callout_handler_init(), one is enough.
Reviewed by: wpaul
ILMI daemons. Factor out common softc fields for all ATM interfaces that
need to be externally visible into an ifatm structure and make the midway
driver using this structure and fill the MIB.
(currently) only consumer (en).
Add a sysctl node hw.atm where the atm drivers will hook on their hardware
sysctl sub-trees.
Make atm_ifattach call if_attach and remove the corresponding call to if_attach
from en. Create atm_ifdetach and use that in en.
While the last change actually changes the interface this is not a problem in
practice because the only other consumer of this API is an older LANAI driver
on the net, that is not ready for current anyway.
Reviewed by: -atm
following changes have been done:
- stylify. The original code was too hard to read.
- get rid of a number of compilation options (Adaptec-only, Eni-only, no-DMA).
- more debugging features.
- locking. This is not correct yet in the absence of interface layer locking,
but is correct enough to not to cause lock order reversals.
- remove RAW mode. There are no users of this in the tree and I doubt that
there are any.
- remove NetBSD compatibility code. There was no way to keep NetBSD non-busdma
and FreeBSD busdma code together.
- if_en now buildable as a module.
This has been actively tested on sparc64 and i386 with ENI server and
client cards and an Adaptec card (thanks to kjc).
Reviewed by: mdodd, arr
if attach succeeded. device_is_alive just tells us that probe
succeeded. Since we were using it to do things like detach net
interfaces, this caused problems when there were errors in the attach
routine.
Symptoms of problem reported by: martin blapp
- Unconditionally call *_stop() if device is in the tree. This is to
prevent callouts from happening after the device is gone. Checks for
bus_child_present() should be added in the future to keep from touching
potentially non-existent hardware in *_detach(). Found by iedowse@.
- Always check for and free miibus children, even if the device is not in
the tree since some failure cases could have gotten here.
- Call ether_ifdetach() in the irq setup failure case
- ti(4), xl(4): move ifmedia_init() calls to the beginning of attach so
that ifmedia_removeall() can be unconditionally called on detach. There
is no way to detect whether ifmedia has been initialized without using
a separate variable (as tl(4) does).
- Add comments to indicate assumptions of code path
in dc_detach() instead of only calling it if the hardware is preset.
This is a workaround for page faults in softclock() after a `dc'
device was detached, caused by not disabling a timer before freeing
its memory. The bus_child_present() checks should probably be
re-added later, but only to avoid the hardware accesses and not the
other resource cleanups in dc_stop().
Approved by: njl
network layer (ether).
- Don't abuse module names to facilitate ifconfig module loading;
such abuse isn't really needed. (And if we do need type information
associated with a module then we should make it explicit and not
use hacks.)
unencapsulated packet back into the IFQ. Unfortunately, the only reason
rl_encap would fail was due to m_defrag failing, which should only happen
when we're low on mbufs. Hence, it was possible for us to end up with
an IFQ full of packets which could never clear the queue because they could
never be defragmented because they were themselves taking up all the mbufs.
To solve this, take if_xl's approach to the problem of encapsulation failure:
drop the packet.
MFC after: 3 days
- Don't bother setting OACTIVE when the descriptors are all full
or there's a vr_encap failure, it doesn't help anything.
- Correctly roll back on the descriptor list after a failure
so as not to corrupt the list.
- Add a missing VR_UNLOCK().
Without these changes, vr_encap failure (which is assured during
a low mbuf situation) would result in the card locking until
the watchdog could fire.
MFC after: 1 week
properly (likely due to mbuf exhaustion.) Previously, the driver
got somewhat wedged.
Also, remove the annoying messages printed every time xl_encap
couldn't allocate a mbuf; they served no useful purpose, and just made
an mbuf exhaustion situation more annoying.
MFC after: 1 week
RX part of this driver too. It's better since the code wasn't
dealing with bus_dmamap_load() returning EINPROGRESS, and this
can't happen with bus_dmamap_load_mbuf().
Submitted by: jake
- Remove locking of the softc in the attach method, instead depending on
bus_setup_intr being at the end of attach (delaying interrupt enable until
after ether_ifattach is called)
- Call *_detach directly in the error case of attach, depending on checking
in detach to only free resources that were allocated. This puts all
resource freeing in one place, avoiding thinkos that lead to memory leaks.
- Add bus_child_present check to calls to *_stop in the detach method to
be sure hw is present before touching its registers.
- Remove bzero softc calls since device_t should do this for us.
- dc: move interrupt allocation back where it was before. It was unnecessary
to move it. This reverts part of 1.88
- rl: move irq allocation before ether_ifattach. Problems might have been
caused by allocating the irq after enabling interrupts on the card.
- rl: call rl_stop before ether_ifdetach
- sf: call sf_stop before ether_ifdetach
- sis: add missed free of sis_tag
- sis: check errors from tag creation
- sis: move dmamem_alloc and dmamap_load to happen at same time as tag creation
- sk: remove duplicate initialization of sk_dev
- ste: add missed bus_generic_detach
- ti: call ti_stop before ether_ifdetach
- ti: add missed error setting in ti_rdata alloc failure
- vr: add missed error setting in I/O, memory mapping cases
- xl: add missed error setting in I/O, memory mapping cases
- xl: remove multi-level goto on attach failure
- xl: move dmamem_alloc and dmamap_load to happen at same time as tag creation
- Calls to free(9) are unconditional because it is valid to call free with a
null pointer.
Reviewed by: imp, mdodd
function.
Also, use m_defrag where appropriate to defrag long mbuf chains
in the same fashion as was done in if_sis.c. Before this change,
if_dc would blow up and take down the interface if fed a really long
mbuf chain.
MFC after: 2 weeks
1. The chain passed in is > 31 fragments long
or
2. The chain will not fit in the remaining descriptors without
defragmentation.
This is slightly less clear than other network drivers because the sis
chips share one descriptor list for all packets, it seems.
Before this change, a > 127 fragment chain would get stuck in the IFQUEUE
permanently, bringing all network traffic to a halt.
MFC after: 2 weeks
code messed up on B & C chipsets because it lost the packet header
and therefore the flag indicating the need for hardware checksums.
MFC after: 2 weeks
where physical addresses larger than virtual addresses, such as i386s
with PAE.
- Use this to represent physical addresses in the MI vm system and in the
i386 pmap code. This also changes the paddr parameter to d_mmap_t.
- Fix printf formats to handle physical addresses >4G in the i386 memory
detection code, and due to kvtop returning vm_paddr_t instead of u_long.
Note that this is a name change only; vm_paddr_t is still the same as
vm_offset_t on all currently supported platforms.
Sponsored by: DARPA, Network Associates Laboratories
Discussed with: re, phk (cdevsw change)
driver should use port or memory based IO, determine it dynamically
at runtime, preferring MMIO where possible. This helps us support newer
arches which dislike port based access better.
Tested on i386 & sparc64, with 3c900, 905, 905b, and 905C cards.
(in varying combinations by both jake and myself)
branches:
Initialize struct cdevsw using C99 sparse initializtion and remove
all initializations to default values.
This patch is automatically generated and has been tested by compiling
LINT with all the fields in struct cdevsw in reverse order on alpha,
sparc64 and i386.
Approved by: re(scottl)
- Get rid of the useless atop() / pmap_phys_address() detour. The
device mmap handlers must now give back the physical address
without atop()'ing it.
- Don't borrow the physical address of the mapping in the returned
int. Now we properly pass a vm_offset_t * and expect it to be
filled by the mmap handler when the mapping was successful. The
mmap handler must now return 0 when successful, any other value
is considered as an error. Previously, returning -1 was the only
way to fail. This change thus accidentally fixes some devices
which were bogusly returning errno constants which would have been
considered as addresses by the device pager.
- Garbage collect the poorly named pmap_phys_address() now that it's
no longer used.
- Convert all the d_mmap_t consumers to the new API.
I'm still not sure wheter we need a __FreeBSD_version bump for this,
since and we didn't guarantee API/ABI stability until 5.1-RELEASE.
Discussed with: alc, phk, jake
Reviewed by: peter
Compile-tested on: LINT (i386), GENERIC (alpha and sparc64)
Runtime-tested on: i386
time and there's no indication that it will improve anytime soon.
By removing support for SimOS it is possible to build LINT on
Alpha, which is considered more important at the moment.
Not objected to on: alpha@
as opposed to one after the other. This is faster in both -CURRENT
and -STABLE. Additionally, there is less code duplication for
error-checking.
One thing to note is that this code seems to return(1) when no buffers
are available; perhaps ENOBUFS should be the correct return value?
Partially submitted & tested by: Hiten Pandya <hiten@unixdaemons.com>
MFC after: 1 week
wish the busdma APIs were more consistent accross architectures.
We should probably move all the other DMA map creations in
xl_attach() where we can really handle them failing, since
xl_init() is void and shouldn't fail.
Pointy hat to: mux
Tested by: Anders Andersson <anders@hack.org>
bus_dmamap_load() it.
- Make it so reusing mbufs when we can't allocate (or map) new ones
actually works. We were previously trying to reuse a mbuf which
was already bus_dmamap_unload()'ed.
Reviewed by: silby
for the agp module, and add agp to the list of modules to compile for alpha.
Add an alpha_mb() to agp_flush_cache for alpha -- it's not correct but may
improve the situation, and it's what linux and NetBSD do.
using the Rhine's internal shift registers which are designed
for the job. This reduces the amount of time we wait around shifting
bits, and seems to work better with some chips.
Also, provide a workaround for some newer cards which report fake PHYs
at multiple addresses. (As more cards are ID'd, I'm sure this part
of the code will have to be expanded to cover more cases.)
Submitted by: Thomas Nystrom <thn@saeab.se>
MFC after: 1 week
under load.
This patch has been tested by Thomas and other for more than a month now,
and all (known) hangs seem to be solved.
Thomas's explanation of the patch:
* Fix the problem with the printing of the RX-error.
* Code from if_fet do better deal with the RX-recovery including a
timeout of the RX-turnoff.
* The call to vr_rxeof before vr_rxeoc have been moved to a point
where the RX-part of the chip is turned off. Otherwise there is a
window where new data could have been written to the buffer chain
before the RX-part is turned off. If this happens the chip will see
a busy rx-buffer. I have no evidence that this have occured but
god knows what the chip will do in this case!
* I have added a timeout of the TX-turnoff. I have checked and in
my 900 MHz system the flags for turnoff (both RX & TX) is seen at
the first check in the loop.
* I could see that I got the VR_ISR_DROPPED interrupt sometimes and
started to thinking about this. I then realized that no recovery is
needed for this case and therefore I only count it as an rxerror
(which was not done before).
* Finally I have changed the FIFO RX threshhold to 128 bytes. When I
did this the VR_ISR_DROPPED interrupt went away. Theory: The chip
will receive a complete frame before it tries to write it out to
memory then the RX threshold is set to store'n'forward. IF the frame
is large AND the next rx frame also is large AND the bus is busy
transfering a TX frame to the TX fifo THEN the second received
frame wont fit in the FIFO and is then dropped. By having the RX
threshold set to 128 the RX fifo is emptied faster.
MFC after: 5 days
on 900B and 635(A).
Re-add the enhanced PHY access register method again for older chipsets,
they do not seem to work with all old chips.
Reviewed by: phk
MFC after: 7 days
- Add conversions to/from little endian for fields that the NIC accesses
by DMA as required.
- Add some bus_dmamap_sync() calls, and correct some existing ones.
- Read the receiver information from the EEPROM in an endian-neutral
manner.
- Load all RX and TX descriptors in a single DMA map up front, and
get the bus addresses of individual descriptors by address arithmetic;
this fixes multiple use of the descriptor tags, which would have
undesired effects.
It seems that xl still does not work on e250 boxen, for reasons which
are not clear yet.
Reviewed by: mux
and ether_ifattach() to end. This fixes a "could sleep" case and
simplifies error exit cases as well. Also be sure to set errno
and clean up resources in !mac error case.
Tested by: Ryan Beasley
Allow to read EEPROM from LAN. It is shared
between a 1394 controller and the NIC and each
time we access it, we need to set SIS_EECMD_REQ.
Idea from: linux driver source
Reviewed by: luoqi
Obtained from: linux driver source (idea)
register, and phy has to be directly accessed via mdio.
Patch converted to CURRENT from STABLE.
Submitted by: luoqi
Reviewed by: luoqi (again)
MFC after: 2 weeks
the PCI bus. When this bit is set, the Max DMA Burst Size
for TX/RX DMA should be no larger than 64 bytes.
Reviewed by: luoqi
Obtained from: (idea from linux driver source)
MFC after: 2 weeks
is read one clock edge too late. This bit is driven low by
slave (as any other input data bits from slave) when the clock
is LOW. The current code did read the bit after the clock was
driven high again.
Reviewed by: luoqi
MFC after: 2 weeks
bus_dmamap_load_mbuf() returned EFBIG.
o Fix mbuf leaks in an error (rare) code path.
o Reuse the TX descriptor if xl_encap() failed instead of
just picking the next one.
o Better error messages.
it possible to use this driver under ia64, sparc64 (though
there may be endianness issues with this one) and other archs.
Tested on: i386, alpha (gallatin)
pci_get_ether accesor, which gets it from the CIS for cardbus cards
(and from other pci-like buses via whatever mechanism is used there).
Submitted by: sam
Approved by: re (blanket)
1. Detect the revision of the Rhine chip we're using.
2. Use the force reset command on revisions which support
it whenever the normal reset command fails.
This should solve a wide range of "my vr0 locks up with reset
failed messages" problems. (Although the root causes should
be eventually tracked down.)
Tested by: grenville armitage <garmitage@swin.edu.au>
Obtained from: Via's if_fet driver
MFC after: 3 days
Approved by: re
o use if_input for input packet processing
o don't strip the Ethernet header for input packets
o use BPF_* macros bpf tapping
o call ether_ioctl to handle default ioctl case
o track vlan changes
Reviewed by: many
Approved by: re
OpenBSD who got the code (or the idea) from the NetBSD tlp driver.
This gets some cardbus dc cards working (either completely or nearly
so). It also appears to get additional pci cards working, without
breaking working ones.
# Maybe some additional work is needed here. Also, the cardbus attachment
# might need to match on the CIS rather than on the vendor/device so we have
# a finer level of detail as to what the card is. Technically, the
# vendor/device fields are undefined for CardBus (even though most cards are
# using common silicon with pci models).
According to the MII specification, the delay produced by our
reads alone are sufficient for correct operation.
This reduces the time mii_tick takes from 10ms to ~1ms here. That's
still a lot, but much better than before.
Submitted by: Harti Brandt <brandt@fokus.gmd.de>
MFC after: 3 weeks
RXSTATE to STOPPED, but to WAIT. This should fix hangs which
could only be solved by replugging the cable.
Submitted by: jhb
Reviewed by: phk
MFC after: 2 weeks
The 550 version is location at address 1 but since it works right we
let the code find whatever PHY it can.
Fix a fragment issue on TX. If the number of frags are more then the
driver has allocated then bring all the frags together into one packet
and send it out. Code derived from the fxp driver.
Tested and found by: Francois Tigeot <francois.tigeot@nic.fr>
Hellmuth Michaelis <hm@kts.org>
MFC after: 1 week
pmap_zero_page() and pmap_zero_page_area() were modified to accept
a struct vm_page * instead of a physical address, vm_page_zero_fill()
and vm_page_zero_fill_area() have served no purpose.
when this is needed. Work around bogus second PHY in the DFE-580 card
via a change in the if_ste.c driver.
Suggested by: jdp
Reviewed by: jdp
MFC after: 3 days
elimiates the driver lockup problem reported by many.
Concepts used were taken from Via's if_fet driver. Verification
and implementation were done by Thomas Nystrom.
Submitted by: Thomas Nystrom <thn@saeab.se>
MFC after: 3 days
This is pretty much fixes any issue I can find:
- Watchdog timeouts were due to starting the TX DMA engine
before we had a packet ready for it. So the first packet
sent never got out only if we sent more then one packet
at a time did the others make it out and not blow up.
Of course reseting the chip then caused us not to transmit
the first packet again ie. catch-22. This required logic changes.
- Combine interrupts on TX packets being queued up.
- Don't keep running around the RX ring since we might get
out of sync so only go around once per receive
- Let the RX engine recover via the poll interface which is
similar to the TX interface. This way the chip wakes
up with no effort when we read enough packets.
- Do better hand-shaking on RX & TX packets so they don't
start of to soon.
- Force a duplex setting when the link comes up after
an ste_init or it will default to half-duplex and be
really slow. This only happens on subsequent ste_init.
The first one worked.
- Don't call stat_update for every overflow. We only monitor
the collisions so the tick interval is good enough for that.
Just read in the collision stats to minimize bus reads.
- Don't read the miibus every tick since it uses delays and
delays are not good for performance.
- Tie link events directly to the miibus code so the port
gets set correctly if someone changes the port settings.
- Reduce the extreme number of {R,T}FD's. They would consume
130K of kernel memory for each NIC.
- Set the TX_THRESH to wait for the DMA engine to complete
before running the TX FIFO. This hurts peak TX performance
but under bi-directional load the DMA engine can't keep up
with the FIFO. Testing shows that we end up in the case
anyways (a la dc(4) issues but worse since the RX engine hogs
everything).
- When stopping the card do a reset since the reset verifies the
card has stopped. Otherwise on heavy RX load the RX DMA engine
is still stuffing packets into memory. If that happens after
we free the DMA area memory bits get scribled in memory and
bad things happen.
This card still has seemingly unfixable issues under heavy RX load in
which the card takes over the PCI bus.
Sponsored by: Vernier Networks
MFC after: 1 week
Remove some unnecessary assignments to mbuf fields in sis_newbuf(),
the "length" fields are of no use while the mbuf is in the receive ring.
MFC after: 3 days
sizes. Previously, the end result was at the mercy of the card's default
setting. This change will reduce the number of buffer underruns for
some users.
PR: kern/37929
Submitted by: Thomas Nystrom <thn@saeab.se>
MFC after: 7 days
has gone away instead of spinning in the interrupt handler. This stops
my machine from hanging when I eject a rl(4)-based cardbus card.
Reviewed by: imp
The value we use is still questionable for 440BX chipsets.
- When flushing the TLB just toggle the bit in question instead of writing
a magic value that could trash other unrelated bits.
o Assert that the page queues lock is held in vm_page_unwire().
o Make vm_page_lock_queues() and vm_page_unlock_queues() visible
to kernel loadable modules.
still queued for transmission. This should solve the problem of
the device stalling on transmissions if some link event prevents
transmission.
There are other drivers which have the same problem and need to be
fixed in the same way.
MFC after: 3 days
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
allow recovery from transmission lockups which occur in the middle
of the descriptor list, rather than just at the beginning.
For some unknown reason, Rhine II chips have a tendency to stop
transmitting while under heavy load, possibly due to collisions.
Whether this behavior is due to a hardware bug or a driver glitch
is unknown as of now.
In either case, this change allows the driver to gracefully recover
from such situations.
Special thanks go to The Anarcat <anarcat@anarcat.dyndns.org>, who
bugged me into looking at this and to
Dominic Marks <dominic_marks@btinternet.com>, who performed a great
deal of testing to help characterize this problem.
MFC after: 3 days
This is temporary hack, better and generalized solution probably
should be implemented at lower layer(MII or PCI?).
Tested by: shoko.araki@soliton.co.jp
MFC after: 1 week
the same thing as the SMC 1211, but with their own vendor ID.
Update the device list to support this NIC. (Discovered these
cards lying around the lab at work.)
doesn't support NWAY, the RealTek PHY (both the integrated ones on 8139
chips and the RTL8201L 10/100 PHY) will not report the link speed via
the ANLPAR or BMSR registers. For the 8201L, we need to look in magic
vendor-specific PHY register 0x19. For the 8139 MAC+PHY combo, we have
to be able to test the RL_MEDIASTAT register.
The changes to rlphy.c are based largely on the patch from PR 30836,
however I tried to eliminate some magic numbers by creating an entry
for the 8201 PHY in miidevs.
Also updated if_rl.c to allow the rlphy driver to read the RL_MEDIASTAT
register via the rl_miibus_readreg() routine.
most cases NULL is passed, but in some cases such as network driver locks
(which use the MTX_NETWORK_LOCK macro) and UMA zone locks, a name is used.
Tested on: i386, alpha, sparc64
One to notify the system that the MTU for VLAN can be 1500 so the vlan
will automatically be configured with a 1500 MTU the other is to ignore
the error case if the received frame is to long.
The frame size notification came from code in the SIS driver, and
the support for long frames derived from the NetBSD Tulip driver.
Tested on: 4 port D-Link adapter DFE-570TX 4 Intel 21143
Netgear card with 82c169 PNIC 10/100BaseTX
Reviewed by: ru (manpage), wpaul (not objected to), archie
Approved by: imp
Obtained from: NetBSD
motherboard chipsets. We need to force the chip to reload its MAC address
into the receive filter, and enable software access mode for the PHY.
PR: kern/33294
to Phil Kernick:
"The problem is that in full duplex mode, the Conexant chip always reports a
carrier lost error, even when the frame is successfully sent. So, if we
have a Conexant chip, then ignore carrier lost when in full duplex
mode."
Since the Xircom chips seem to have the same issue and since we already
have a workaround for this, just expand the workaround test to also
check for DC_IS_CONEXANT().
to a label is inside an #ifdef block, then the label should *also* be
inside an #ifdef block. Hide the "done:" label which is only used if
DEVICE_POLLING is enabled under #ifdef DEVICE_POLLING.
a major slowdown, and re-enable stats overflow interrupts.
For future reference, the bug was in our code, and not
some bug in the 3com chips.
Reviewed by: wpaul
MFC after: 2 days
Non-SMP, i386-only, no polling in the idle loop at the moment.
To use this code you must compile a kernel with
options DEVICE_POLLING
and at runtime enable polling with
sysctl kern.polling.enable=1
The percentage of CPU reserved to userland can be set with
sysctl kern.polling.user_frac=NN (default is 50)
while the remainder is used by polling device drivers and netisr's.
These are the only two variables that you should need to touch. There
are a few more parameters in kern.polling but the default values
are adequate for all purposes. See the code in kern_poll.c for
more details on them.
Polling in the idle loop will be implemented shortly by introducing
a kernel thread which does the job. Until then, the amount of CPU
dedicated to polling will never exceed (100-user_frac).
The equivalent (actually, better) code for -stable is at
http://info.iet.unipi.it/~luigi/polling/
and also supports polling in the idle loop.
NOTE to Alpha developers:
There is really nothing in this code that is i386-specific.
If you move the 2 lines supporting the new option from
sys/conf/{files,options}.i386 to sys/conf/{files,options} I am
pretty sure that this should work on the Alpha as well, just that
I do not have a suitable test box to try it. If someone feels like
trying it, I would appreciate it.
NOTE to other developers:
sure some things could be done better, and as always I am open to
constructive criticism, which a few of you have already given and
I greatly appreciated.
However, before proposing radical architectural changes, please
take some time to possibly try out this code, or at the very least
read the comments in kern_poll.c, especially re. the reason why I
am using a soft netisr and cannot (I believe) replace it with a
simple timeout.
Quick description of files touched by this commit:
sys/conf/files.i386
new file kern/kern_poll.c
sys/conf/options.i386
new option
sys/i386/i386/trap.c
poll in trap (disabled by default)
sys/kern/kern_clock.c
initialization and hardclock hooks.
sys/kern/kern_intr.c
minor swi_net changes
sys/kern/kern_poll.c
the bulk of the code.
sys/net/if.h
new flag
sys/net/if_var.h
declaration for functions used in device drivers.
sys/net/netisr.h
NETISR_POLL
sys/dev/fxp/if_fxp.c
sys/dev/fxp/if_fxpvar.h
sys/pci/if_dc.c
sys/pci/if_dcreg.h
sys/pci/if_sis.c
sys/pci/if_sisreg.h
device driver modifications
Introduce an additional device flag for those NICs which require the
transmit buffers to be aligned to 32-bit boundaries.
(the equivalen fix for STABLE is slightly simpler because there are
no supported chips which require this alignment there.)
controllers. There still seems to be some issues with the DRI copying code
for some adapters, at least it doesn't hang the system now. Input would be
appreciated.
PR: 32301
Obtained from: Eric Anhlot <eanholt@gladstone.uoregon.edu>, Joe <joeo@nks.net>
The reason we are required to commit to -current first is so that later
MFC's do not risk the loss of existing bug fixes. Even if this was not
strictly required in -current, it should still be fixed there too.
The reason we are required to commit to -current first is so that later
MFC's do not risk the loss of existing bug fixes. Even if this was not
strictly required in -current, it should still be fixed there too.
mbuf allocation fails, and fix (i hope) a couple of style bugs.
I believe these printf() are extremely dangerous because now they can
occur on every incoming packet and are not rate limited. They were
meant to warn the sysadmin about lack of resources, but now they
can become a nice way to panic your system under load.
Other drivers (e.g. the fxp driver) have nothing like this.
There is a pending discussion on putting this kind of warnings
elsewhere, and I hope we can fix this soon.
underlying unaligned bcopy) on incoming packets that are already
available (albeit unaligned) in a buffer.
The performance improvement varies, depending on CPU and memory
speed, but can be quite large especially on slow CPUs. I have seen
over 50% increase on forwarding speed on the sis driver for the
486/133 (embedded systems), which does exactly the same thing.
The behaviour is controlled by a sysctl variable, hw.dc_quick which
defaults to 1. Set it to 0 to restore the old behaviour.
After running a few experiments (in userland, though) I am convinced
that doing the m_devget() is detrimental to performance in almost
all cases.
Even if your CPU has degraded performance with misaligned data,
the bcopy() in the driver has the same overhead due to misaligment
as the one that you save in the uiomove(), plus you do one extra
copy and pollute the cache.
But more often than not, you do not even have to touch the payload,
e.g. when you are forwarding packets, and even in the often-cited
case of NFS, you often end up passing a pointer to the payload to
the disk controller.
In any case, you can play with the sysctl variable to toggle between
the two behaviours, and see if it makes a difference.
MFC-after: 3 days
have alignment problems.
On small boxes (e.g. the net4501 from Soekris, featuring a 486/133)
this provides huge performance benefits: the peak forwarding rate
with avg.sized packets goes up by 50-70% because of this change
alone. Faster CPUs might benefit less from this change, but in any
case the CPU has better things to do than waste time on useless
memory-to-memory copies.
Several drivers (for Tulip-like cards) might benefit from a similar
change.
Right now the new behaviour is controlled by a sysctl variable,
hw.sis_quick which defaults to 1 (on), you can set it to 0 to
reintroduce the old behaviour (and compare the results). The
variable is only there to show how much you can gain with this
change, it will go away soon.
Also, slightly simplify the code to initialize the ring buffers,
and remove a couple of dangerous printf's which could trigger on
any packet in case of mbuf shortage.
MFC-after: 3 days
idle and the driver would not detect the event, requiring userland
to cycle the interface to bring it up again.
The fix consists in adding SIS_IMR_RX_IDLE to the interrupt mask and
add a command in sis_intr() to restart the receiver when this happens.
While at it, make the test of status bits more efficient.
in the 21143, instead of giving priority to the receive unit.
This gives a 10-15% performance improvement in the forwarding rate
under heavy load.
Reviewed-by: Bill Paul
Jonathon Lemon's driver (gx) is at least as fast and has more features
and is likely to be better supported.
It is also possible that Intel might support this chipset in FreeBSD
with their own driver. Somewhat secretive and furtive rumblings from
certain Yahoo employees have indicated that this might happen soon.
I'm a little unhappy at the lack of discussion on the net list about
this, or on developers, or on hackers, or the lack of mention on
audit. This then leaves me to try and figure out the right thing
to do.
I've concluded that the right thing to do is to remove wx from FreeBSD,
as this is probably best for FreeBSD.
What the heck, the OpenBSD version will benefit.
1. Add wx_txint_delay as a tunable (defaults to 5000 now, or ~5ms) and switch
to using delayed TXDW interrupts. Since the chip continues to reload the
TIDV with this value for each descriptor written back, this allows continued
deferral of the actual interrupt until the last packet completes (assuming
that 5ms between multiple packets transmitting is reasonable).
2. Add two other SYSCTL entities:
hw.wx.dump_stats
hw.wx.clear_stats
to be used, hackey hackey, to get the watchdog routine to dump/clear
the current softc statistics.
Usage would be:
sysctl -w hw.wx.dump_stats=UNIT
to cause the current stats to be dumped for UNIT.
3. Attempt to clean up wx_detach routine so we don't panic. Well, things
still panic, but given that the code is just like other NIC drivers,
I suspect it's actually something elsewhere, like e1000phy, that's actually
blowing up.
4. Skip the entire test for runt packets- after doing somet thinking
and experimenting, I believe that the chip only doesn't like it if
the whole frame to xmit is < 16 bytes- each TFD can be some fragment
of that. This should improve performance a chunk because of all of the
(14 byte ETHERHEADER + DATA) mbuf chains.
5. Keep track of total frame length. Try not to xmit an odd byte frame-
this is supposed to get around some dumb Cisco switch problems.
6. On the last packet, also set Interrupt Delay && Report Packet Sent
(see #1 above)
7. Attempt to do xmit garbage collection *first* in order to avoid setting
IFF_OACTIVE if at all possible.
MFC after: 1 week
We only support the size of frame we are currently allocating, which
is MCLBYTES - sizeof (struct ether_header) usable, so don't set an
MTU that would go over this.
