Provide support in loader for booting 64-bit PowerPC kernels. Like amd64,
64-bit PowerPC kernels are loaded by a 32-bit loader, since nearly all
powerpc64 firmwares execute in 32-bit mode.
Convert Freescale PowerPC platforms to FDT convention.
The following systems are affected:
- MPC8555CDS
- MPC8572DS
This overhaul covers the following major changes:
- All integrated peripherals drivers for Freescale MPC85XX SoC, which are
currently in the FreeBSD source tree are reworked and adjusted so they
derive config data out of the device tree blob (instead of hard coded /
tabelarized values).
Change the implementation of vm_hold_free_pages() so that it performs at
most one call to pmap_qremove(), and thus one TLB shootdown, instead of one
call and TLB shootdown per page.
Remove interval validation from cpu_tick_calibrate(). As I found, check
was needed at preliminary version of the patch, where number of CPU ticks
was divided strictly on 16 seconds. Final code instead uses real interval
duration, so precise interval should not be important. Same time aliasing
issues around second boundary causes false positives, periodically logging
useless "t_delta ... too long/short" messages when HZ set below 256.
Fixes a bug for LP PHY that some frames have 2 padding bytes at the
start so we should adjust the mbuf if the driver is running in PIO mode.
Now it should work well with WPA authentication and association for LP
PHY devices.
Reduce the number of global TLB shootdowns generated by pmap_qenter().
Specifically, teach pmap_qenter() to recognize the case when it is being
asked to replace a mapping with the very same mapping and not generate
a shootdown. Unfortunately, the buffer cache commonly passes an entire
buffer to pmap_qenter() when only a subset of the mappings are changing.
For the extension of buffers in allocbuf() this was resulting in
unnecessary shootdowns. The addition of new pages to the end of the
buffer need not and did not trigger a shootdown, but overwriting the
initial mappings with the very same mappings was seen as a change that
necessitated a shootdown. With this change, that is no longer so.
For a "buildworld" on amd64, this change eliminates 14-15% of the
pmap_invalidate_range() shootdowns, and about 4% of the overall
shootdowns.
Since powerpc and powerpc64 share an instruction set, bind can and should
use the 32-bit atomic operations unmodified. Accomplish this by switching
some MACHINE_ARCH values to MACHINE_CPUARCH.
RTLD support for powerpc64. A few small modifications to the Makefile
and symbol map are required to support various consequences of the dot
symbol scheme:
- Symbols beginning with a dot are reserved, so start private symbols with
an underscore.
- In order to set RTLD breakpoints, gdb must be able to locate the text
entry point, not the data section function descriptor, so add
.r_debug_state to the symbol map on powerpc64.
If ata_sata_phy_reset() failed and ata_generic_reset() is not called, mark
channel as having no devices connected. This improves hot-unplug operation
on legacy-emulating SATA controllers.
On attach, grab channel lock before setting up interrupt. This fixes crash
in ATA_CAM mode if phy connect event arrive before CAM bus initialization
completed.
Teach libstand what to do for powerpc64: the same thing as powerpc. Like
amd64, libstand apps for powerpc64 systems are 32-bit, due to 32-bit Open
Firmware implementations.
The 64-bit PowerPC ABI implemented in binutils 2.15 requires some special
quirks for weak-symbol handling. Text symbols require also marking weak
the special dot-symbol associated with the function, and data symbols
require that you not do that. To fix this, provide a hacked
__weak_reference for powerpc64, and define a new __weak_reference_data
for the single weak data symbol in base.
Add definitions for powerpc64 ELF relocation types. Some 64-bit relocations
are identical to 32-bit relocations in meaning, name, and number, and I
have chosen not to duplicate those here.
Fix reading of empty fifolog files. When we don't have anything
to inflate, skip the rest of the fifolog reader code to avoid
hitting the assert about Z_OK a bit further down.
Teach our toolchain how to generate 64-bit PowerPC binaries. This fixes
a variety of bugs in binutils related to handling of 64-bit PPC ELF,
provides a GCC configuration for 64-bit PowerPC on FreeBSD, and
associated build systems tweaks.
Some revision of Yukon controller generates corrupted frame when TX
checksum offloading is enabled. The frame has a valid checksum
value so payload might be modified during TX checksum calculation.
Disable TX checksum offloading but give users chance to enable it
when they know their controller works without problems with TX
checksum offloading.
Reported by: Andrzej Tobola <ato <> iem dot pw dot edu dot pl>
For both i386 and amd64 pmap,
- change the type of pm_active to cpumask_t, which it is;
- in pmap_remove_pages(), compare with PCPU(curpmap), instead of
dereferencing the long chain of pointers [1].
For amd64 pmap, remove the unneeded checks for validity of curpmap
in pmap_activate(), since curpmap should be always valid after
r209789.
Add support for the VM_ALLOC_COUNT() hint to vm_page_alloc(). Consequently,
the maintenance of vm_pageout_deficit can be localized to just two places:
vm_page_alloc() and vm_pageout_scan().
This change also corrects an off-by-one error in the maintenance of
vm_pageout_deficit. Historically, the buffer cache functions, allocbuf()
and vm_hold_load_pages(), have not taken into account that vm_page_alloc()
already increments vm_pageout_deficit by one.
Improve cxgb(4)'s behaviour when faced with temporarily "bouncy" links:
- Run the adapter's tick at 1Hz and remove link state checks from it.
Instead, have each port check its link state. Delay the check so that
it takes place slightly after the driver is notified of a change in
link state. This is a cheap way to debounce these notifications if
many are received in rapid succession. POLL_LINK_1ST_TIME flag can
also be eliminated as a side effect of these changes.
- Do not reset the PHY when link goes down.
- Clear port's link_fault flag if the PHY indicates link is down.
- get_link_status_r should leave speed and duplex alone when link is down.
Eliminate ext_intr_task. The "slow" interrupt handler is already
running on the adapter's task queue. Just do what the task does
instead of enqueueing it.
Sync. printf() of libstand(3) with sys/kern/subr_prf.c.
CVS r1.94 jhb:
Cast the integer read as the first argument for %b to an unsigned integer
so it's value is not sign extended when assigned to the uintmax_t variable
used internally by printf. For example, if bit 31 is set in the cpuid
feature word, then %b would print out the initial value as a 16 character
hexadecimal value. Now it only prints out an 8 character value.
CVS r1.109 njl:
Add support for 'h' and 'hh' modifiers for printf(9).
CVS r1.117 phk:
If we ignore an unknown % sequence, we must stop interpreting the remaining
% arguments because the varargs are now out of sync and there is a risk that
we might for instance dereference an integer in a %s argument.
SVN r209836 jkim:
Implement optional 'precision' for numbers. Previously, it was parsed but
ignored. Some third-party modules (e.g., APCICA) prefer this format over
zero padding flag '0'.
Implement optional 'precision' for numbers. Previously, it was parsed but
ignored. Some third-party modules (e.g., APCICA) prefer this format over
zero padding flag '0'.
Remove enabling RX checksum offloading in RX filter setup. RX
checksum is enabled in sge_init_locked().
While I'm here do not set RX checksum bits in RX descriptor
initialization. It is controller's job to set these bits.
Replace the existing PowerPC busdma implementation with the one from
amd64 (with slight modifications). This provides support for bounce
buffers, which are required on systems with RAM above 4 GB.
Enable KX bit, which is needed for 64 bit access, in status register
for XLR. Update exception handlers and other functions which set/change
status registers to preserve this.
64 bit compilation support XLR platform code.
Mostly changes to make casting between int and pointer and printing
64bit values safe for 32 and 64 bit compile.
Merge jmallett@'s n64 work into HEAD - changeset 8
Updated PTE/PDE macros from http://svn.freebsd.org/base/user/jmallett/octeon
Introduce pmap_segshift() macro, use pmap_segmap() in place of pmap_pde, and
remove pmap_pde().
Fix several bugs in the real-mode Open Firmware implementation and provide
a virtual-mode version for use on 64-bit systems, which have 32-bit
firmware implementations and require similar constraints on addressing
to the real-mode implementation.
Make VM_ALLOC_RETRY flag mandatory for vm_page_grab(). Assert that the
flag is always provided, and unconditionally retry after sleep for the
busy page or failed allocation.
The intent is to remove VM_ALLOC_RETRY eventually.
Correctly maintain the per-cpu field "curpmap" on amd64 just like we
do on i386. The consequences of not doing so on amd64 became apparent
with the introduction of the COUNT_IPIS and COUNT_XINVLTLB_HITS
options. Specifically, single-threaded applications were generating
unnecessary IPIs to shoot-down the TLB on other processors. However,
this is clearly nonsensical because a single-threaded application is
only running on the current processor. The reason that this happens
is that pmap_activate() is unable to properly update the old pmap's
field "pm_active" without the correct "curpmap". So, in effect, stale
bits in "pm_active" were leading pmap_protect(), pmap_remove(),
pmap_remove_pages(), etc. to flush the TLB contents on some arbitrary
processor that wasn't even running the same application.
Remove the ability to upgrade via sysinstall. This feature has been flaky
and broken for a number of years. freebsd-update now provides a reasonable
way to handle upgrades, so keeping this functionality in sysinstall only
serves to mislead users.
Use the kernel's start address to determine what to map. This allows
us to link the kernel at different addresses without needing to build
a corresponding loader.
Fix an error in the EXAMPLES section of getopt(1), which is based on
the same fix present in NetBSD.
Note: the getopt man page contains more antique information like this.
An overhaul of the man page and/or sync with NetBSD would be the right
thing to do. But since this is out of the scope of the PR, I'll leave
it as it is for now.
PR: docs/133118
Submitted by: Oleg A. Mamontov (oleg at mamontov dot net)
Discussed with: jilles@
MFC after: 2 weeks
Increase the default and minimum partition sizes.
Bump / to 1GB, /var to 4GB and /tmp to 1GB.
A 512MB root partition is now too small to hold two kernels, so to
prevent installkernel failing 1GB is a better default. Likewise,
applications are storing more data in /var so 4GB is more reasonable
on a modern machine.
The previous minimum partition sizes were too small to hold a minimal
installation, and have been adjusted accordingly.
- Simplify logic in handling ticks wrap-up
- Fix a bug where thread may be in sleeping state but the wchan won't
be set, leading to an empty container for sleepq_type(). [0]
Sponsored by: Sandvine Incorporated
[0] Submitted by: Bryan Venteicher
<bryanv at daemoninthecloset dot org>
MFC after: 3 days
X-MFC: 209577
projects / FreeBSD / FreeBSD.git / log