Implemented a more sophisticated mechanism for handling transmitter
interrupts which now defers them until the transmit queue if filled
up with completed buffers. This has two advantages: first, it reduces
the number of transmitter interrupts to just 1/120th of the rate
that they occured previously, and two, running down many buffers
at once has much improved cache effects.
Defer rundown (m_freem) of completed transmit buffers for no longer
than 1 second.
This brings us up-to-date with the most recent "fxp" driver in FreeBSD.
directly. That would require that we map the scsipi_xfer into DMA
space. Instead, copy to/from the NCR CCB, which the script already
has to DMA to/from. These copies are small, and don't seem to affect
performance.
Separate the ncb (i.e. softc) members that are accessed by the script into
a separate structure. Allocate one of these structures in DMA safe memory
using bus_dma, and change RELOC_SOFTC to use the DMA address of this
structure.
"If a channel has no drives, do *not* unmap its I/O regions.
It's not really safe to use them for anything else, and in legacy mode it
will just cause us to probe the channel again as an ISA device."
- keep the modes supported by the drive in struct ata_drive_datas (will be
later used for downgrading the DMA/PIO mode on error)
- use config flags to force/disable PIO/DMA/UDMA modes
- For the CMD PCI0643/6 setup DMA mode to DMA Read multiple.
was stuck in the while (!(cbp->cb_status & FXP_CB_STATUS_C)); loop,
I think because the relevent CPU cache entry was never updated.
This was on a compaq deskpro system.
eg the Cypress ISA bridge:
-put channel mapping into the chip specific part, unify with
channel_probe() into channel_map()
-use pointer to channel data as function call argument wherever possible
instead of the channel number
-allow the "compat" channel number to differ from the per-controller
channel number - for mapping and interrupt functions
-add support for SiS5598 and Cypress 82C693 chips
Mostly done by Manuel, I only contributed to the first 2 items.
- returned EOPNOTSUPP rather than -1.
- no check for negative offset.
many of these fix potential security problems in these drivers.
XXX XXX XXX
the d_mmap cdev routine should be changed to have a prototype like:
paddr_t (*d_mmap) __P((dev_t, off_t, int));
by someone!
when this driver was changed to use the MAC-independent MII code. This
"PHY" does not have a programming interface, but simply senses the duplex
mode from the link partner. Since it is not possible to read the 80c24
to determine the link type or status, we consider this to be "manual".
Don't claim DMA support if we don't have explicit support for this chip.
They're to may way to loose when trying to use DMA without configuring the
controller and disks.
in the printing of DMA mode (piix3/4 only)
others: set the debug_mask to 0, so that debug messages are turned off by
default but can be easily turned on.
Reset drive_flags to 0 for unconfigured devices, so that they are ignored
later. For configured devices, reset state to 0 after probe/attach.
struct scsipi_adapter; they were not used.
Add a scsipi_ioctl entry point to struct scsipi_adapter. This will be
used to issue ioctl commands to the host adapters.
Inspired by PR #6090, from Matt Jacob.
Add vendor ID for ESS Technology Inc.
Add product ID for ESS Technology ES1938 SOLO-1 AudioDrive
Add product ID for Fujitsu K56flex LTMODEM
Add product ID for TI PCI1250 PC Card bridge
Add product ID for Trident Cyber 9397 Video controller.
descriptor count gets miscounted on txprobes. The second (and more
important) is that transmit stalls should now be fixed. The problem
was not due to lack of ring resources but dmamaps. When Jason changed
the driver to use pre-allocated maps (instead of the dynamic ones I
used), and when there were no more maps, the driver just gave up instead
of calling tx_intr to free any transmitted but unreclaimed dma maps.
Since there was nothing being transmitted, no transmit interrupts
would fire to restore things (and OACTIVE prevented other transmits
from happenning). So it stayed starved until another interrupt cause
(like a received packet) "woke" it up.
-use of uninitialized variable in attach (mostly harmless)
-removal of dead code in attach()
-fix of interpretation of SIA blocks on >=21142 (serious, caused strange
things with 10Mbit)
egcs won't attempt to optimize inline copies of them into a single
load/store (the RFA is not 4-byte aligned in the DMA area).
Just use memcpy() to copy those fields, now, not our own home-grown.
cursor code. These are unneeded on the Bt485, since the RAMDAC handles
the hardware cursor. They are needed for the Bt463, however, since the
RAMDAC only provides cursor color.
* Improve the midisyn layer a little.
* Add a driver for the Yamaha OPL[23] FM synths.
The opl driver is not finished yet; it sounds pretty awful.
For some strange reason I cannot get any FM sound from my SB64 cards,
but a regular SB16 works fine.
if i drive is responding):
-if the reset succeeds, check some registers to make sure there is really
a drive (and not a chipset which echoes back the last written value)
-explicitely select the master before trying to read the master's status
the 3Com 3c905-TX and 3c905-T4. Fix handling of the MII on the
3c595-MII and 3c597-MII (can't talk to the PHYs on these cards; just
use "manual" for the external MII port).
Fixes kern/4782 (Chris Demetriou).
if_tlvar.h, since the ThunderLAN PHY driver needs to know some things
about it's parent (and since it can only attach to a ThunderLAN chip,
this is the easiest way).
associated with any observed lossage...it was just noticed while reading
ncr_start().
g/c some now-unreachable code produced by the earlier race condition fix.
* Use the new trigger_{in,out}put interface.
* Allow the play and record channels to have different encodings (but the
same sample rate).
* Play u-law and a-law as 16-bit data. (This may not work yet...)
>We're not Linux. If we still want driver version strings to be displayed
>at boot, then I suppose DEBUG is a reasonable compromise.
Makes the whole concept useless. This is for default printouts. If you
can build a debug kernel, you know what version you have. This was under
the concept of 'RAS' so that hapless users could tell you microversion
things. But I guess this isn't the right way according to our local
Jesuits. Oh, well. I'll think up something different and hopefully
less objectionable. And yes, NetBSD isn't linux. The developers seem
to be equally bad tempered, but linux is more successful.
Part B:
Field interrupts in OS layer so that (in this OS) bus_dmamap_sync(POSTREAD)
can (formally) ensure that the result queue is stable wrt to buffering
and that for sending a command a bus_dmamap_sync(PREWRITE) is done to
ensure that the device gets a good view of what the mailbox contents
should be.
which contain 'standard' com- and lpt-type ports. Some of these present
as PCI simple-communications/serial or simple-communications/parallel
devices, but many do not. (Additionally, there is no document that I can
find that describes the "specific well-konwn register-level" description
of how the 'standard' devices' config space headers shold work.) Eventually,
some of the devices driven by this code should become simple pci attachments
for the 'lpt' and 'com' drivers, but that requires solid documentation.
the BAR-printing function to print a name for the register, factor out
a common register-bits function which can handle the fact that type 2
headers have a different size than is usual, and actually do something
useful with the rest of the bits in the type 2 header.
as with user-land programs, include files are installed by each directory
in the tree that has includes to install. (This allows more flexibility
as to what gets installed, makes 'partial installs' easier, and gives us
more options as to which machines' includes get installed at any given
time.) The old SYS_INCLUDES={symlinks,copies} behaviours are _both_
still supported, though at least one bug in the 'symlinks' case is
fixed by this change. Include files can't be build before installation,
so directories that have includes as targets (e.g. dev/pci) have to move
those targets into a different Makefile.
used in the SMC EtherPower II.
Media control isn't yet supported, due to some MII infrastructure
problems which I hope to address soon. This isn't a huge deal, since
the PHY defaults to auto-negotiate mode.
Also, the device just programs the multicast hash table to accept all
multicast, to avoid a hardware bug that causes the multicast address
filter to lose in 10Mb/s mode. This bug will be fixed in a more sane
way once the media control issues are dealt with.
of last resort when trying to communicate information about
bogus behaviour of PCI devices to the MI autoconfiguration code.
In general, bogus behaviour should be handled by drivers, but there
are some types of bogons which can't be addressed that way. The
only quirk currently defined is one which indicates that the device
is multi-function even though the device's header says otherwise.
(Mmm, Intel 82371FB PCI-to-ISA Bridge (PIIX); you'd think that at least
Intel would have gotten it right...)
of functions on a given device. Also, clean up the #if 0'd
major-debugging-spew code so that it's all one piece, so that
it's a bit prettier, and so that it prints out quirk information.
of last resort when trying to communicate information about
bogus behaviour of PCI devices to the MI autoconfiguration code.
In general, bogus behaviour should be handled by drivers, but there
are some types of bogons which can't be addressed that way. The
only quirk currently defined is one which indicates that the device
is multi-function even though the device's header says otherwise.
(Mmm, Intel 82371FB PCI-to-ISA Bridge (PIIX); you'd think that at least
Intel would have gotten it right...)
printing into a function, add a bit more pretty-printing of existing stuff.
Implement pretty-printers for type 1 and type 2 headers. (Right now,
these are just quick stabs based on some on-line bridge docs that I have
handy on my laptop. Mmmm, meetings. I'll check the bits when I get
back within reach of my official docs.)
lots of data, e.g. ~18k on a PCI system with few add-in devices; use
with MSGBUFSIZE=...). Useful to have here so that people who want as
much data about the PCI configuration in a machine can get it without
having to craft their own code. Also, clean up a few of the other
#if 0'd printfs.
* print all configuration space registers. Then, where possible,
interpret them. (That is, PRESENT ALL THE DATA, then interpret it --
don't hide data behind interpretation. Also, when interpreting
fields, try to print out the specific value that's being interpreted.)
* handle different header types.
* allow caller to specify a function which can interpret the
device-dependent header and is responsible for pretty-printing it.
It spews (use 'options MSGBUFSIZE=...' 8-), but when you want the data,
you really want _all_ of it.
Still needs some cleanup and additional code (e.g. interepretation
of PCI-PCI (type 1) and PCI-Cardbus (type 2(?)) bridge headers).
pci_mapreg_info() call. pci_mapreg_map() implies this check,
but code which calls pci_mapreg_info() has to check it explicitly.
Otherwise, if memory space is disabled, the driver does the wrong
thing, and tries to use memory space anyway, potentially resulting
incorrect driver operation and no useful error message.
The graphics device driver passes a "default attribute" for normal text
output to the wscons framework. If the emulation module needs more
attributes (for different "renditions") it can allocate them via a
callback.
For now, only the "sun" emulation makes use of it.
needs some testing, but it seems to produce sound. The driver was written
by me, but since I don't have the hardware the debugging and testing was
done by Andreas Gustafsson <gson@araneus.fi>, Chuck Cranor
<chuck@maria.wustl.edu>, and Phil Nelson <phil@cs.wwu.edu>. Thanks.
ID) when determining if the Vendor ID is invalid. The spec says that
Vendor ID of 0xffff is invalid, so, it doesn't _matter_ what the product
ID is in that case. Treat Vendor ID 0 as invalid because we always have.
and see if anything responds. if nothing (that's attributable to the
PCI IDE controller) responds, then that channel either has no devices on
it or has been disabled (via a non-standard mechanism) by the BIOS. If nothing
responds, don't map the compat.-mode interrupt or attach the wdc to that
channel, because the BIOS is likely to assign that IRQ to a different PCI
device. If that happens, the kernel will panic because that device will
try to map the IRQ level-triggered, but the compat interrupt will have been
mapped edge-triggered. (One possible way around this is to map the compat
interrupt edge-triggered, but it's not clear reading the spec that this
is correct or desirable.)
insert a check to see whether a channel appears to be enabled. Shouldn't
be necessary, according to the spec, but some PC chipsets allow individual
compatibility channels to be disabled. "I hate PCs."
controller driver. These are commented out here until the wdc
declaration mess is resolved, and until then need to go into MD
files files in places where they play nice with the wdc declaration.
entry (because the same product id is used for the 640B, as well). Note
that a few of the entries (PCI0642, PCI0650A) no longer have data to
be found on the CMD web site, and note that PCI0650A should probably have
its "A" trimmed as well.
in pci/if_tl.c, call config_found() with a print function, instead of
printing ourself a message in if_tl.c if no miibus was found. The print
function is in mii/mii.c (mii_adapter_print()) so that it can be used by any
adapter (idea from the scsi system).
it simple, and allocate one for each transmit and receive descriptor.
In addition to being simpler and faster, this fixes a serious memory leak
in the transmit path.
tested on any other platform other than i386. Use of bus_dma(9) can be
turned by defining TULIP_BUS_DMA_NOTX and TULIP_BUS_DMA_NORX. These allow
one to determine if the problem is in the transmit or receive path.
A problem reported by mycroft is also fixed.
- Fall back on i/o space if mem space isn't available.
- Card reports mem space as prefetchable, but mapping the card into dense
space fails in nasty ways on the Alpha. Force mapping into sparse
space by clearing the BUS_SPACE_MAP_CACHEABLE bit (XXX!).
nullbuf (used to pad packets < ETHER_MIN_SIZE) is used for all tl
interfaces. Allocates only once, and never deallocate it (as we can't say
if another instance of the driver is interface is using it).
i/o-mapped space to always be used), we discover that at least one
ThunderLAN interface can't read the EEPROM properly if memory-mapped
access is used. Kludge around this for now by "prefering" i/o space.
- Do PCI space configuration like the other drivers. In particular,
don't _disable_ the space we're not using because some lame firmware
implementations might not reenable it on warm boot. Also, prefer
memory space always.
- Make match and info-gathering in attach table-driven.
- Rearrange things a bit to be a bit more visually pleasing during boot.
Also, fixup some #include problems.
- Work around a bug in the 82557 that causes the receiver to lock up
in certain conditions by kicking the multicast address filter if we
haven't heard anything come down the wire for some period of time.
- Fix a bug that could cause TxCB descriptor chains to cross page boundaries
on the Alpha.
- Remove some unneeded register masking.
- Fix a bug where too much data was copied from the config template, causing
memory corruption.
- Fix handing of if_timer (it was be cleared too early in some cases).
- Attempt to reduce the chances of receiver overrun by doubling the
number of receive DMA segments, and processing receive interrupts
before transmit interrupts.
- Remove a gratuitous assignment.
- Fix a bug where incoming packets were counted twice.
pseudo-device rnd # /dev/random and in-kernel generator
in config files.
o Add declaration to all architectures.
o Clean up copyright message in rnd.c, rnd.h, and rndpool.c to include
that this code is derived in part from Ted Tyso's linux code.
Add newer Intel / VIA i386 chipsets.
Correct one S3 chip, add one.
Mostly from Carl Shapiro <css@samsara.dialup.access.net> per PR kern/4169
and kern/4170 (identical).
memory-mapped or i/o-mapped access to the device registers, and always
choose memory-mapped if it is enabled. In particular, do _not_ explictly
disable the space we decide to not use, as this confuses some versions
of Alpha console software (which are arguably buggy because of this
problem).
Also, fix a logic error pointed out by Ross Harvey <ross@teraflop.com>
that would cause memory-mapped access to never be enabled.
pci_map_mem(), with appropriate changes for bus_space.
* Add extra arguments for use by pci_mapreg_*(), and make the functions
static.
* Rewrite pci_mapreg_info() and pci_mapreg_map() as wrappers around
pci_*_find(), using the documented interface.
(currently only CD-ROM drives on i386). The sys/dev/scsipi system provides 2
busses to which devices can attach (scsibus and atapibus). This needed to
change some include files and structure names in the low level scsi drivers.
able to read SXP registers (instead of panicing). Probably have done a bit
of setup overkill- we now alloc some scratch memory for the 2100 that I thought
I would need for the Port database, but so far I haven't needed to retrieve
the port database. Well, early days yet.
call for the board's memory space to be PCI_MAPREG_MEM_TYPE_32BIT_1M or
PCI_MAPREG_MEM_TYPE_32BIT depending on the board ID. Also, remove a
bogus extra argument to an interrupt-establishment-error printf. Problems
pointed out by Jarkko Torppa <torppa@cute.fi> in PR 3753, but fixed slightly
differently than he suggested.
- Add NetBSD autoconfiguration support.
- Rearrange code slightly to minimize the number of #ifdefs.
- Don't use a structure to access CSRs. Use macros that DTRT for
the NetBSD and FreeBSD cases.
- Deal with alignment contraint on Alpha - add 2-byte padding at the
beginning of the RFA, so that the data will be 4-byte aligned, after
the 14-byte Ethernet header.
Thanks to Matthias Drochner for the testing, and David Greenman for
the feedback on the changes.
by default if it's usable, but falling back to I/O space if mem isn't usable.
If NCR_IOMAPPED is defined (default on the x86), prefer I/O space
then fall back to mem. Also, clean up the various memory consistency checks
so that they can deal with run-time determination of whether or not the
device is to be memory- or I/O-mapped.
mjacob