support machines having multiple independently numbered PCI domains
and don't support reenumeration without ambiguity amongst the
devices as seen by the OS and represented by PCI location strings.
This includes introducing a function pci_find_dbsf(9) which works
like pci_find_bsf(9) but additionally takes a domain number argument
and limiting pci_find_bsf(9) to only search devices in domain 0 (the
only domain in single-domain systems). Bge(4) and ofw_pcibus(4) are
changed to use pci_find_dbsf(9) instead of pci_find_bsf(9) in order
to no longer report false positives when searching for siblings and
dupe devices in the same domain respectively.
Along with this change the sole host-PCI bridge driver converted to
actually make use of PCI domain support is uninorth(4), the others
continue to use domain 0 only for now and need to be converted as
appropriate later on.
Note that this means that the format of the location strings as used
by pciconf(8) has been changed and that consumers of <sys/pciio.h>
potentially need to be recompiled.
Suggested by: jhb
Reviewed by: grehan, jhb, marcel
Approved by: re (kensmith), jhb (PCI maintainer hat)
to gem_attach() as the former access softc members not yet initialized
at that time and gem_reset() actually is enough to stop the chip. [1]
o Revise the use of gem_bitwait(); add bus_barrier() calls before calling
gem_bitwait() to ensure the respective bit has been written before we
starting polling on it and poll for the right bits to change, f.e. even
though we only reset RX we have to actually wait for both GEM_RESET_RX
and GEM_RESET_TX to clear. Add some additional gem_bitwait() calls in
places we've been missing them according to the GEM documentation.
Along with this some excessive DELAYs, which probably only were added
because of bugs in gem_bitwait() and its use in the first place, as
well as as have of an gem_bitwait() reimplementation in gem_reset_tx()
were removed.
o Add gem_reset_rxdma() and use it to deal with GEM_MAC_RX_OVERFLOW errors
more gracefully as unlike gem_init_locked() it resets the RX DMA engine
only, causing no link loss and the FIFOs not to be cleared. Also use it
deal with GEM_INTR_RX_TAG_ERR errors, with previously were unhandled.
This was based on information obtained from the Linux GEM and OpenSolaris
ERI drivers.
o Turn on workarounds for silicon bugs in the Apple GMAC variants.
This was based on information obtained from the Darwin GMAC and Linux GEM
drivers.
o Turn on "infinite" (i.e. maximum 31 * 64 bytes in length) DMA bursts.
This greatly improves especially RX performance.
o Optimize the RX path, this consists of:
- kicking the receiver as soon as we've a spare descriptor in gem_rint()
again instead of just once after all the ready ones have been handled;
- kicking the receiver the right way, i.e. as outlined in the GEM
documentation in batches of 4 and by pointing it to the descriptor
after the last valid one;
- calling gem_rint() before gem_tint() in gem_intr() as gem_tint() may
take quite a while;
- doubling the size of the RX ring to 256 descriptors.
Overall the RX performance of a GEM in a 1GHz Sun Fire V210 was improved
from ~100Mbit/s to ~850Mbit/s.
o In gem_add_rxbuf() don't assign the newly allocated mbuf to rxs_mbuf
before calling bus_dmamap_load_mbuf_sg(), if bus_dmamap_load_mbuf_sg()
fails we'll free the newly allocated mbuf, unable to recycle the
previous one but a NULL pointer dereference instead.
o In gem_init_locked() honor the return value of gem_meminit().
o Simplify gem_ringsize() and dont' return garbage in the default case.
Based on OpenBSD.
o Don't turn on MAC control, MIF and PCS interrupts unless GEM_DEBUG is
defined as we don't need/use these interrupts for operation.
o In gem_start_locked() sync the DMA maps of the descriptor rings before
every kick of the transmitter and not just once after enqueuing all
packets as the NIC might instantly start transmitting after we kicked
it the first time.
o Keep state of the link state and use it to enable or disable the MAC
in gem_mii_statchg() accordingly as well as to return early from
gem_start_locked() in case the link is down. [3]
o Initialize the maximum frame size to a sane value.
o In gem_mii_statchg() enable carrier extension if appropriate.
o Increment if_ierrors in case of an GEM_MAC_RX_OVERFLOW error and in
gem_eint(). [3]
o Handle IFF_ALLMULTI correctly; don't set it if we've turned promiscuous
group mode on and don't clear the flag if we've disabled promiscuous
group mode (these were mostly NOPs though). [2]
o Let gem_eint() also report GEM_INTR_PERR errors.
o Move setting sc_variant from gem_pci_probe() to gem_pci_attach() as
device probe methods are not supposed to touch the softc.
o Collapse sc_inited and sc_pci into bits for sc_flags.
o Add CTASSERTs ensuring that GEM_NRXDESC and GEM_NTXDESC are set to
legal values.
o Correctly set up for 802.3x flow control, though #ifdef out the code
that actually enables it as this needs more testing and mainly a proper
framework to support it.
o Correct and add some conversions from hard-coded functions names to
__func__ which were borked or forgotten in if_gem.c rev. 1.42.
o Use PCIR_BAR instead of a homegrown macro.
o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN].
o In gem_pci_attach() in case attaching fails release the resources in
the opposite order they were allocated.
o Make gem_reset() static to if_gem.c as it's not needed outside that
module.
o Remove the GEM_GIGABIT flag and the associated code; GEM_GIGABIT was
never set and the associated code was in the wrong place.
o Remove sc_mif_config; it was only used to cache the contents of the
respective register within gem_attach().
o Remove the #ifdef'ed out NetBSD/OpenBSD code for establishing a suspend
hook as it will never be used on FreeBSD.
o Also probe Apple Intrepid 2 GMAC and Apple Shasta GMAC, add support for
Apple K2 GMAC. Based on OpenBSD.
o Add support for Sun GBE/P cards, or in other words actually add support
for cards based on GEM to gem(4). This mainly consists of adding support
for the TBI of these chips. Along with this the PHY selection code was
rewritten to hardcode the PHY number for certain configurations as for
example the PHY of the on-board ERI of Blade 1000 shows up twice causing
no link as the second incarnation is isolated.
These changes were ported from OpenBSD with some additional improvements
and modulo some bugs.
o Add code to if_gem_pci.c allowing to read the MAC-address from the VPD on
systems without Open Firmware.
This is an improved version of my variant of the respective code in
if_hme_pci.c
o Now that gem(4) is MI enable it for all archs.
Pointed out by: yongari [1]
Suggested by: rwatson [2], yongari [3]
Tested on: i386 (GEM), powerpc (GMACs by marcel and yongari),
sparc64 (ERI and GEM)
Reviewed by: yongari
Approved by: re (kensmith)
ways:
(1) Cached pages are no longer kept in the object's resident page
splay tree and memq. Instead, they are kept in a separate per-object
splay tree of cached pages. However, access to this new per-object
splay tree is synchronized by the _free_ page queues lock, not to be
confused with the heavily contended page queues lock. Consequently, a
cached page can be reclaimed by vm_page_alloc(9) without acquiring the
object's lock or the page queues lock.
This solves a problem independently reported by tegge@ and Isilon.
Specifically, they observed the page daemon consuming a great deal of
CPU time because of pages bouncing back and forth between the cache
queue (PQ_CACHE) and the inactive queue (PQ_INACTIVE). The source of
this problem turned out to be a deadlock avoidance strategy employed
when selecting a cached page to reclaim in vm_page_select_cache().
However, the root cause was really that reclaiming a cached page
required the acquisition of an object lock while the page queues lock
was already held. Thus, this change addresses the problem at its
root, by eliminating the need to acquire the object's lock.
Moreover, keeping cached pages in the object's primary splay tree and
memq was, in effect, optimizing for the uncommon case. Cached pages
are reclaimed far, far more often than they are reactivated. Instead,
this change makes reclamation cheaper, especially in terms of
synchronization overhead, and reactivation more expensive, because
reactivated pages will have to be reentered into the object's primary
splay tree and memq.
(2) Cached pages are now stored alongside free pages in the physical
memory allocator's buddy queues, increasing the likelihood that large
allocations of contiguous physical memory (i.e., superpages) will
succeed.
Finally, as a result of this change long-standing restrictions on when
and where a cached page can be reclaimed and returned by
vm_page_alloc(9) are eliminated. Specifically, calls to
vm_page_alloc(9) specifying VM_ALLOC_INTERRUPT can now reclaim and
return a formerly cached page. Consequently, a call to malloc(9)
specifying M_NOWAIT is less likely to fail.
Discussed with: many over the course of the summer, including jeff@,
Justin Husted @ Isilon, peter@, tegge@
Tested by: an earlier version by kris@
Approved by: re (kensmith)
of pages don't sum to anywhere near the total number of pages on amd64.
This is for the most part because uma_small_alloc() pages have never been
counted as wired pages, like their kmem_malloc() brethren. They should
be. This changes fixes that.
It is no longer necessary for the page queues lock to be held to free
pages allocated by uma_small_alloc(). I removed the acquisition and
release of the page queues lock from uma_small_free() on amd64 and ia64
weeks ago. This patch updates the other architectures that have
uma_small_alloc() and uma_small_free().
Approved by: re (kensmith)
o Revamp the PIC I/F to only abstract the PIC hardware. The
resource handling has been moved to nexus, where it belongs.
o Include EOI and MASK+EOI methods to the PIC I/F in support of
INTR_FILTER.
o With the allocation of interrupt resources and setup of
interrupt handlers in the common platform code we can delay
talking to the PIC hardware after enumeration of all devices.
Introduce a call to powerpc_intr_enable() in configure_final()
to achieve that and have powerpc_setup_intr() only program the
PIC when !cold.
o As a consequence of the above, remove all early_attach() glue
from the OpenPIC and Heathrow PIC drivers and have them
register themselves when they're found during enumeration.
o Decouple the interrupt vector from the interrupt request line.
Allocate vectors increasingly so that they can be used for
the intrcnt index as well. Extend the Heathrow PIC driver to
translate between IRQ and vector. The OpenPIC driver already
has the support for vectors in hardware.
Approved by: re (blanket)
syscall. It was broken when a new lseek syscall was introduced.
The problem is that we need to swap the 32-bit td_retval values
for the __syscall indirect syscall when the actual syscall has
a 32-bit return value. Hence, we need to exclude lseek(2). And
this means the "old" lseek(2) as well -- which we didn't.
Based on a patch from: grehan@
Approved by: re (rwatson)
Previously it didn't honor parent dma tag's restrictions such that
an invalid dma segment could be passed to device. The driver for the
device may panic in sanity check routine for the dma segment or may
produce unexpected results. I have no idea how it could ever have
worked before.
Reviewed by: grehan
Tested by: gad
Approved by: re (hrs)
used to return PAGE_SIZE without respect to restrictions of a DMA tag.
This affected all of the busdma load functions that use
_bus_dmamap_loader_buffer() as their back-end.
Reviewed by: scottl (long a ago)
Approved by: re (hrs)
This allocator uses a binary buddy system with a twist. First and
foremost, this allocator is required to support the implementation of
superpages. As a side effect, it enables a more robust implementation
of contigmalloc(9). Moreover, this reimplementation of
contigmalloc(9) eliminates the acquisition of Giant by
contigmalloc(..., M_NOWAIT, ...).
The twist is that this allocator tries to reduce the number of TLB
misses incurred by accesses through a direct map to small, UMA-managed
objects and page table pages. Roughly speaking, the physical pages
that are allocated for such purposes are clustered together in the
physical address space. The performance benefits vary. In the most
extreme case, a uniprocessor kernel running on an Opteron, I measured
an 18% reduction in system time during a buildworld.
This allocator does not implement page coloring. The reason is that
superpages have much the same effect. The contiguous physical memory
allocation necessary for a superpage is inherently colored.
Finally, the one caveat is that this allocator does not effectively
support prezeroed pages. I hope this is temporary. On i386, this is
a slight pessimization. However, on amd64, the beneficial effects of
the direct-map optimization outweigh the ill effects. I speculate
that this is true in general of machines with a direct map.
Approved by: re
more exposure. The current state of SCTP implementation is
considered to be ready for 32-bit platforms, but still need some
work/testing on 64-bit platforms.
Approved by: re (kensmith)
Discussed with: rrs
caches with data caches after writing to memory. This typically
is required to make breakpoints work on ia64 and powerpc. For
those architectures the function is implemented.
- Rename PCPU_LAZY_INC into PCPU_INC
- Add the PCPU_ADD interface which just does an add on the pcpu member
given a specific value.
Note that for most architectures PCPU_INC and PCPU_ADD are not safe.
This is a point that needs some discussions/work in the next days.
Reviewed by: alc, bde
Approved by: jeff (mentor)
Probabilly, a general approach is not the better solution here, so we should
solve the sched_lock protection problems separately.
Requested by: alc
Approved by: jeff (mentor)
handler is wrapped in a couple of functions - a filter wrapper and an
ithread wrapper. In this case (and just in this case), the filter
wrapper could ask the system to schedule the ithread and mask the
interrupt source if the wrapped handler is composed of just an ithread
handler: modify the "old" interrupt code to make it support
this situation, while the "new" interrupt code is already ok.
Discussed with: jhb
same way it was enabled for Linux binares in linuxulator.
This allows binaries built with -pie. Many ports auto-detect -fPIE support
in GCC 4.2 and build binaries FreeBSD was unable to run.
vmcnts. This can be used to abstract away pcpu details but also changes
to use atomics for all counters now. This means sched lock is no longer
responsible for protecting counts in the switch routines.
Contributed by: Attilio Rao <attilio@FreeBSD.org>
VM_PHYSSEG_SPARSE depending on whether the physical address space is
densely or sparsely populated with memory. The effect of this
definition is to determine which of two implementations of
vm_page_array and PHYS_TO_VM_PAGE() is used. The legacy
implementation is obtained by defining VM_PHYSSEG_DENSE, and a new
implementation that trades off time for space is obtained by defining
VM_PHYSSEG_SPARSE. For now, all architectures except for ia64 and
sparc64 define VM_PHYSSEG_DENSE. Defining VM_PHYSSEG_SPARSE on ia64
allows the entirety of my Itanium 2's memory to be used. Previously,
only the first 1 GB could be used. Defining VM_PHYSSEG_SPARSE on
sparc64 allows USIIIi-based systems to boot without crashing.
This change is a combination of Nathan Whitehorn's patch and my own
work in perforce.
Discussed with: kmacy, marius, Nathan Whitehorn
PR: 112194
read the same register back. It can cause hangs or machine
checks in certain cases. One particular case is with bge(4)
when a reset is initiated for the controller.
MFC after: 1 month
GETATTRs being generated - one from lookup()/namei() and the other
from nfs_open() (for cto consistency). This change eliminates the
GETATTR in nfs_open() if an otw GETATTR was done from the namei()
path. Instead of extending the vop interface, we timestamp each attr
load, and use this to detect whether a GETATTR was done from namei()
for this syscall. Introduces a thread-local variable that counts the
syscalls made by the thread and uses <pid, tid, thread syscalls> as
the attrload timestamp. Thanks to jhb@ and peter@ for a discussion on
thread state that could be used as the timestamp with minimal overhead.
partitioning class that supports multiple schemes. Current
schemes supported are APM (Apple Partition Map) and GPT.
Change all GEOM_APPLE anf GEOM_GPT options into GEOM_PART_APM
and GEOM_PART_GPT (resp).
The ctlreq interface supports verbs to create and destroy
partitioning schemes on a disk; to add, delete and modify
partitions; and to commit or undo changes made.
