- Don't use a single big DMA block for all rings. Create separate
DMA area for each ring instead. Currently the following DMA
areas are created:
Event ring, standard RX ring, jumbo RX ring, RX return ring,
hardware MAC statistics and producer/consumer status area.
For Tigon II, mini RX ring and TX ring are additionally created.
- Added missing bus_dmamap_sync(9) in various TX/RX paths.
- TX ring is no longer created for Tigon 1 such that it saves more
resources on Tigon 1.
- Data sheet is not clear about alignment requirement of each ring
so use 32 bytes alignment for normal DMA area but use 64 bytes
alignment for jumbo RX ring where the extended RX descriptor
size is 64 bytes.
- For each TX/RX buffers use separate DMA tag(e.g. the size of a
DMA segment, total size of DMA segments etc).
- Tigon allows separate DMA area for event producer, RX return
producer and TX consumer which is really cool feature. This
means TX and RX path could be independently run in parallel.
However ti(4) uses a single driver lock so it's meaningless
to have separate DMA area for these producer/consumer such that
this change creates a single status DMA area.
- It seems Tigon has no limits on DMA address space and I also
don't see any problem with that but old comments in driver
indicates there could be issues on descriptors being located in
64bit region. Introduce a tunable, dev.ti.%d.dac, to disable
using 64bit DMA in driver. The default is 0 which means it would
use full 64bit DMA. If there are DMA issues, users can disable
it by setting the tunable to 0.
- Do not increase watchdog timer in ti_txeof(). Previously driver
increased the watchdog timer whenever there are queued TX frames.
- When stat ticks is set to 0, skip processing ti_stats_update(),
avoiding bus_dmamap_sync(9) and updating if_collisions counter.
- MTU does not include FCS bytes, replace it with
ETHER_VLAN_ENCAP_LEN.
With these changes, ti(4) should work on PAE environments.
Many thanks to Jay Borkenhagen for remote hardware access.
Pre-allocate the memory in device attach time. While I'm here
remove unnecessary reassignment of error variable as it was already
initialized. Also added a missing driver lock in TIIOCSETTRACE
handler.
allocator with UMA backed jumbo allocator by default. Previously
ti(4) used sf_buf(9) interface for jumbo buffers but it was broken
at this moment such that enabling jumbo frame caused instant panic.
Due to the nature of sf_buf(9) it heavily relies on VM changes but
it seems ti(4) was not received much blessing from VM gurus. I
don't understand VM magic and implications used in driver either.
Switching to UMA backed jumbo allocator like other network drivers
will make jumbo frame work on ti(4).
While I'm here, fully allocate all RX buffers. This means ti(4) now
uses 512 RX buffer and 1024 mini RX buffers.
To use sf_buf(9) interface for jumbo buffers, introduce a new
'options TI_SF_BUF_JUMBO'. If it is proven that sf_buf(9) is better
for jumbo buffers, interesting developers can fix the issue in
future.
ti(4) still needs more bus_dma(9) cleanups and should use separate
DMA tag/map for each ring(standard, jumbo, mini, command, event
etc) but it should work on all platforms except PAE.
Special thanks to Jay[1] who provided complete remote debugging
access.
Tested by: Jay Borkenhagen <jayb <> braeburn dot org > [1]
if_watchdog and if_timer.
- Fix some issues in detach for sn(4), ste(4), and ti(4). Primarily this
means calling ether_ifdetach() before anything else.
requiried to keep consistent softc state before/after callback function
invocation and supposed to be sligntly faster than previous one as it
wouldn't incur callback overhead. With this change callback function
was gone.
- Decrease TI_MAXTXSEGS to 32 from 128. It seems that most mbuf chain
length is less than 32 and it would be re-packed with m_defrag(9) if
its chain length is larger than TI_MAXTXSEGS. This would protect ti(4)
against possible kernel stack overflow when txsegs[] is put on stack.
Alternatively, we can embed the txsegs[] into softc. However, that
would waste memory and make Tx/Rx speration hard when we want to
sperate Tx/Rx handlers to optimize locking.
- Fix dma map tracking used in Tx path. Previously it used the dma map
of the last mbuf chain in ti_txeof() which was incorrect as ti(4)
used dma map of the first mbuf chain when it loads a mbuf chain with
bus_dmamap_load_mbuf(9). Correct the bug by introducing queues that
keep track of active/inactive dma maps/mbuf chain.
- Use ti_txcnt to check whether driver need to set watchdog timer instead
of blidnly clearing the timer in ti_txeof().
- Remove the 3rd arg. of ti_encap(). Since ti(4) now caches the last
descriptor index(ti_tx_saved_prodidx) used in Tx there is no need to
pass it as a fuction arg.
- Change data type of producer/consumer index to int from u_int16_t in
order to remove implicit type conversions in Tx/Rx handlers.
- Check interface queue before getting a mbuf chain to reduce locking
overhead.
- Check number of available Tx descriptores to be 16 or higher in
ti_start(). This wouldn't protect Tx descriptor shortage but it would
reduce number of bus_dmamap_unload(9) calls in ti_encap() when we are
about to running out of Tx descriptors.
- Command NIC to send packets ony when the driver really has packets
enqueued. Previously it always set TI_MB_SENDPROD_IDX which would
command NIC to DMA Tx descriptors into NIC local memory regardless
of Tx descriptor changes.
Reviewed by: scottl
. remove unnecessay header files after Scott's bus_dma(9) commit.
. remove global variable tis which was introduced at the time of
zero_copy(9) changes. The variable tis was not used at all. The
same applyes to ti_links in softc so axe it.
. deregister variables.
. axe ti_vhandle and switch to use explicit register access for
accessing NIC local memory. Creates three variants of ti_mem to
read/write NIC local memory(ti_mem_read, ti_mem_write) and clearing
NIC local memory(ti_mem_zero). This greatly enhances code
readability and have ti(4) drop using shared memory scheme for
Tigon 1. As Tigon 1 switched to use explicit register access for Tx,
axe ti_tx_ring_nic/ti_cmd_ring in softc.(Tigon 2 used to host ring
scheme which means there is no need to access NIC local memory via
register access for Tx and NIC would DMA the modified Tx rings into
its local memory.) [1]
. introduce new macro TI_EVENT_*/TI_CMD_* to handle NIC envent/command.
Instead of using bit fields assginment for accessing the event, use
shift operations to set/get it. [1]
. add additional check for valid DMA tags in ti_free_dmamaps().
. add missing bus_dmamap_sync/bus_dmamap_unload in ti_free_*_ring_*.
. fix locking nits(MTX_RECURSE mutex) and make ti(4) MPSAFE.
. change data type of ti_rdata_phys to bus_addr_t and don't blindly
cast to uint32_t.
. rearrange detach path and make ti(4) survive during device detach.
. for Tigon 1, use explicit register access for checking Tx descriptors
in ti_encap()/ti_txeof(). [1]
. properly call bus_dmamap_sync(9) for updating statistics.
. remove extra semicolon in ti_encap()
. rewrite loading MAC address to work on strict-alignment architectures.
. move TI_RD_OFF macro to if_tireg.h
. axe ETHER_ALIGN as it's already defined in <net/ethernet.h>.
. make macros immuine from expansion by adding parenthesis and do-while.
. remove alpha specific hack as vtophys(9) is no longer used in ti(4)
after Scott's bus_dma(9) fix.
Reviewed by: scottl
Obtained from: OpenBSD [1]
to use busdma. Unlike most of the other drivers, but similar to the
if_em driver, pre-allocate the dmamaps at init time instead of allocating
them on the fly when descriptors need to be filled. This isn't ideal right
now because a map is allocated for every descriptor slot in the tx, rx, mini,
and jumbo rings (which is a lot!) in order to simplify the bookkeeping, even
though the driver might support filling only a subset of those slots.
Luckily, maps are typically NULL on i386 and amd64, so the cost isn't
very high. It could be an issue with sparc64, but the driver isn't endian
clean either, and that is a much bigger problem to solve first.
Note that jumbo frame support is under-tested, and I'm not even sure if
it till really works correctly given the evil VM magic that is does.
The changes here attempt to preserve the existing semanitcs.
Thanks to Martin Nillson for contributing the Netgear card for this work.
MFC-After: 3 weeks
struct ifnet or the layer 2 common structure it was embedded in have
been replaced with a struct ifnet pointer to be filled by a call to the
new function, if_alloc(). The layer 2 common structure is also allocated
via if_alloc() based on the interface type. It is hung off the new
struct ifnet member, if_l2com.
This change removes the size of these structures from the kernel ABI and
will allow us to better manage them as interfaces come and go.
Other changes of note:
- Struct arpcom is no longer referenced in normal interface code.
Instead the Ethernet address is accessed via the IFP2ENADDR() macro.
To enforce this ac_enaddr has been renamed to _ac_enaddr.
- The second argument to ether_ifattach is now always the mac address
from driver private storage rather than sometimes being ac_enaddr.
Reviewed by: sobomax, sam
the packets are immediately returned for sending (e.g. when bridging
or packet forwarding). There are more efficient ways to do this
but for now use the least intrusive approach.
Reviewed by: imp, rwatson
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.
The constant I was using was correct, but I mislabeled it as 256K when
it should have been 512K. This doesn't actually change the code, but
it clarifies things somewhat.
Submitted by: Chuck Cranor <chuck@research.att.com>
mtx_enter(lock, type) becomes:
mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks)
mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized)
similarily, for releasing a lock, we now have:
mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN.
We change the caller interface for the two different types of locks
because the semantics are entirely different for each case, and this
makes it explicitly clear and, at the same time, it rids us of the
extra `type' argument.
The enter->lock and exit->unlock change has been made with the idea
that we're "locking data" and not "entering locked code" in mind.
Further, remove all additional "flags" previously passed to the
lock acquire/release routines with the exception of two:
MTX_QUIET and MTX_NOSWITCH
The functionality of these flags is preserved and they can be passed
to the lock/unlock routines by calling the corresponding wrappers:
mtx_{lock, unlock}_flags(lock, flag(s)) and
mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN
locks, respectively.
Re-inline some lock acq/rel code; in the sleep lock case, we only
inline the _obtain_lock()s in order to ensure that the inlined code
fits into a cache line. In the spin lock case, we inline recursion and
actually only perform a function call if we need to spin. This change
has been made with the idea that we generally tend to avoid spin locks
and that also the spin locks that we do have and are heavily used
(i.e. sched_lock) do recurse, and therefore in an effort to reduce
function call overhead for some architectures (such as alpha), we
inline recursion for this case.
Create a new malloc type for the witness code and retire from using
the M_DEV type. The new type is called M_WITNESS and is only declared
if WITNESS is enabled.
Begin cleaning up some machdep/mutex.h code - specifically updated the
"optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN
and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently
need those.
Finally, caught up to the interface changes in all sys code.
Contributors: jake, jhb, jasone (in no particular order)
Have if_ti stop "hiding" the softc pointer in the buffer region. Rather,
use the available void * passed to the free routine and pass the softc
pointer through there.
To note: in MEXTADD(), TI_JUMBO_FRAMELEN should probably be TI_JLEN. I left it
unchanged, because this way I'm sure to not damage anything in this respect...
takes care of all the 10/100 and gigE PCI drivers that I've done.
Next will be the wireless drivers, then the USB ones. I may pick up
some stragglers along the way. I'm sort of playing this by ear: if
anyone spots any places where I've screwed up horribly, please let me
know.
that should be better.
The old code counted references to mbuf clusters by using the offset
of the cluster from the start of memory allocated for mbufs and
clusters as an index into an array of chars, which did the reference
counting. If the external storage was not a cluster then reference
counting had to be done by the code using that external storage.
NetBSD's system of linked lists of mbufs was cosidered, but Alfred
felt it would have locking issues when the kernel was made more
SMP friendly.
The system implimented uses a pool of unions to track external
storage. The union contains an int for counting the references and
a pointer for forming a free list. The reference counts are
incremented and decremented atomically and so should be SMP friendly.
This system can track reference counts for any sort of external
storage.
Access to the reference counting stuff is now through macros defined
in mbuf.h, so it should be easier to make changes to the system in
the future.
The possibility of storing the reference count in one of the
referencing mbufs was considered, but was rejected 'cos it would
often leave extra mbufs allocated. Storing the reference count in
the cluster was also considered, but because the external storage
may not be a cluster this isn't an option.
The size of the pool of reference counters is available in the
stats provided by "netstat -m".
PR: 19866
Submitted by: Bosko Milekic <bmilekic@dsuper.net>
Reviewed by: alfred (glanced at by others on -net)
cards. This basically involves switching to the 12.4.13 firmware, plus
a couple of minor tweaks to the driver.
Also changed the jumbo buffer allocation scheme just a little to avoid
'failed to allocate jumbo buffer' conditions in certain cases.
a PCI memory mapped region, rman_get_bushandle() returns what happens
to be a kernel virtual address pointing to the base of the PCI shared
memory window. However this is not the behavior on all platforms:
the only thing you should do with the bushandle is pass it to the
bus_spare_read()/bus_space_write() routines. If you actually do want
the kernel virtual address of the base of the PCI memory window, you
need to use rman_get_virtual().
The problem is that at the moment, rman_get_virtual() returns a physical
address, which is bad. In order to get the kernel virtual address we
need, we have to play with it a little.
Presumeably this behavior will be changed, but in the meantime the
Tigon driver won't work. So for the moment, I'm adding a kludge to
make things happy on the alpha: the correct kernel virtual address
is calculated from the value returned by rman_get_virtual(). This
should be removed once rman_get_virtual() starts doing the right
thing.
This should make the Tigon actuall work on the alpha now.
critical mbuf fields to sane values. Simplify the use of ETHER_ALIGN to
enforce payload alignment, and turn it on on the x86 as well as alpha
since it helps with NFS which wants the payload to be longword aligned
even though the hardware doesn't require it.
This fixes a problem with the ti driver causing an unaligned access trap
on the Alpha due to m_adj() sometimes not setting the alignment correctly
because of incomplete mbuf initialization.
in ti_rxeof() instead. This doesn't really seem to provide much in the
way of a performance boost, and I'm pretty sure it can cause mbuf leakage
in some extreme cases.
positively not let ti_encap() fill up the TX ring all the way and wrap
around. This fixes a potential transmit lockup where a really fast
machine (or particular TX traffic pattern) can overrun the end of the
ring.
Reported by: John Plevyak <jplevyak@inktomi.com>
preparation for tsunami support. Previous chipsets' direct-mapped DMA
mask was always 1024*1024*1024. The Tsunami chipset needs it to be
2*1024*1024*1024
These changes should not affect the i386 port
Reviewed by: Doug Rabson <dfr@nlsystems.com>
bug in the stats accounting (nicSendBDs counter was bogus when TX ring was
configured to be in host memory).
Update if_tireg.h to look for new firmware fix level.
Networks Tigon 1 and Tigon 2 chipsets. There are a _lot_ of OEM'ed
gigabit ethernet adapters out there which use the Alteon chipset so
this driver covers a fair amount of hardware. I know that it works with
the Alteon AceNIC, 3Com 3c985 and Netgear GA620, however it should also
work with the DEC/Compaq EtherWORKS 1000, Silicon Graphics Gigabit
ethernet board, NEC Gigabit Ethernet board and maybe even the IBM and
and Sun boards. The Netgear board is the cheapest (~$350US) but still
yields fairly good performance.
Support is provided for jumbo frames with all adapters (just set the
MTU to something larger than 1500 bytes), as well as hardware multicast
filtering and vlan tagging (in conjunction with the vlan support in
-current, which I should merge into -stable soon). There are some hooks
for checksum offload support, but they're turned off for now since
FreeBSD doesn't have an officially sanctioned way to support checksum
offloading (yet).
I have not added the 'device ti0' entry to GENERIC since the driver
with all the firmware compiled in is quite large, and it doesn't really
fit into the category of generic hardware.
