It turns out the kernel relies on barrier() to force a reload of the
percpu offset value. Since we can't easily modify the definition of
barrier() to include "tp" as an output register, we instead provide a
definition of __my_cpu_offset as extended assembly that includes a fake
stack read to hazard against barrier(), forcing gcc to know that it
must reread "tp" and recompute anything based on "tp" after a barrier.
This fixes observed hangs in the slub allocator when we are looping
on a percpu cmpxchg_double.
A similar fix for ARMv7 was made in June in change 509eb76ebf.
Cc: stable@vger.kernel.org
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
atomic* value is signed value, and atomic* functions need also process
signed value (parameter value, and return value), so use 'long long'
instead of 'u64'.
After replacement, it will also fix a bug for atomic64_add_negative():
"u64 is never less than 0".
The modifications are:
in vim, use "1,% s/\<u64\>/long long/g" command.
remove redundant '__aligned(8)'.
be sure of 80 (and macro '\') columns limitation after replacement.
Signed-off-by: Chen Gang <gang.chen@asianux.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com> [re-instated const cast]
The hardware architecture descriptor headers have been updated, in
particular to reflect some larger MMIO fields on the mPIPE shims for
controlling the network hardware, from the recent Gx72 release.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
A config option to allow a variant vmap() using huge pages that was never
upstreamed had some bits of code related to it scattered around the tile
architecture; the config option was removed downstream and this commit
cleans up the scattered evidence of it from the upstream as well.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Pull Tile arch updates from Chris Metcalf:
"These changes bring in a bunch of new functionality that has been
maintained internally at Tilera over the last year, plus other stray
bits of work that I've taken into the tile tree from other folks.
The changes include some PCI root complex work, interrupt-driven
console support, support for performing fast-path unaligned data
fixups by kernel-based JIT code generation, CONFIG_PREEMPT support,
vDSO support for gettimeofday(), a serial driver for the tilegx
on-chip UART, KGDB support, more optimized string routines, support
for ftrace and kprobes, improved ASLR, and many bug fixes.
We also remove support for the old TILE64 chip, which is no longer
buildable"
* git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile: (85 commits)
tile: refresh tile defconfig files
tile: rework <asm/cmpxchg.h>
tile PCI RC: make default consistent DMA mask 32-bit
tile: add null check for kzalloc in tile/kernel/setup.c
tile: make __write_once a synonym for __read_mostly
tile: remove support for TILE64
tile: use asm-generic/bitops/builtin-*.h
tile: eliminate no-op "noatomichash" boot argument
tile: use standard tile_bundle_bits type in traps.c
tile: simplify code referencing hypervisor API addresses
tile: change <asm/system.h> to <asm/switch_to.h> in comments
tile: mark pcibios_init() as __init
tile: check for correct compiler earlier in asm-offsets.c
tile: use standard 'generic-y' model for <asm/hw_irq.h>
tile: use asm-generic version of <asm/local64.h>
tile PCI RC: add comment about "PCI hole" problem
tile: remove DEBUG_EXTRA_FLAGS kernel config option
tile: add virt_to_kpte() API and clean up and document behavior
tile: support FRAME_POINTER
tile: support reporting Tilera hypervisor statistics
...
The macrology in cmpxchg.h was designed to allow arbitrary pointer
and integer values to be passed through the routines. To support
cmpxchg() on 64-bit values on the 32-bit tilepro architecture, we
used the idiom "(typeof(val))(typeof(val-val))". This way, in the
"size 8" branch of the switch, when the underlying cmpxchg routine
returns a 64-bit quantity, we cast it first to a typeof(val-val)
quantity (i.e. size_t if "val" is a pointer) with no warnings about
casting between pointers and integers of different sizes, then cast
onwards to typeof(val), again with no warnings. If val is not a
pointer type, the additional cast is a no-op. We can't replace the
typeof(val-val) cast with (for example) unsigned long, since then if
"val" is really a 64-bit type, we cast away the high bits.
HOWEVER, this fails with current gcc (through 4.7 at least) if "val"
is a pointer to an incomplete type. Unfortunately gcc isn't smart
enough to realize that "val - val" will always be a size_t type
even if it's an incomplete type pointer.
Accordingly, I've reworked the way we handle the casting. We have
given up the ability to use cmpxchg() on 64-bit values on tilepro,
which is OK in the kernel since we should use cmpxchg64() explicitly
on such values anyway. As a result, I can just use simple "unsigned
long" casts internally.
As I reworked it, I realized it would be cleaner to move the
architecture-specific conditionals for cmpxchg and xchg out of the
atomic.h headers and into cmpxchg, and then use the cmpxchg() and
xchg() primitives directly in atomic.h and elsewhere. This allowed
the cmpxchg.h header to stand on its own without relying on the
implicit include of it that is performed by <asm/atomic.h>.
It also allowed collapsing the atomic_xchg/atomic_cmpxchg routines
from atomic_{32,64}.h into atomic.h.
I improved the tests that guard the allowed size of the arguments
to the routines to use a __compiletime_error() test. (By avoiding
the use of BUILD_BUG, I could include cmpxchg.h into bitops.h as
well and use the macros there, which is otherwise impossible due
to include order dependency issues.)
The tilepro _atomic_xxx internal methods were previously set up to
take atomic_t and atomic64_t arguments, which isn't as convenient
with the new model, so I modified them to take int or u64 arguments,
which is consistent with how they used the arguments internally
anyway, so provided some nice simplification there too.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Pull networking changes from David Miller:
"Noteworthy changes this time around:
1) Multicast rejoin support for team driver, from Jiri Pirko.
2) Centralize and simplify TCP RTT measurement handling in order to
reduce the impact of bad RTO seeding from SYN/ACKs. Also, when
both timestamps and local RTT measurements are available prefer
the later because there are broken middleware devices which
scramble the timestamp.
From Yuchung Cheng.
3) Add TCP_NOTSENT_LOWAT socket option to limit the amount of kernel
memory consumed to queue up unsend user data. From Eric Dumazet.
4) Add a "physical port ID" abstraction for network devices, from
Jiri Pirko.
5) Add a "suppress" operation to influence fib_rules lookups, from
Stefan Tomanek.
6) Add a networking development FAQ, from Paul Gortmaker.
7) Extend the information provided by tcp_probe and add ipv6 support,
from Daniel Borkmann.
8) Use RCU locking more extensively in openvswitch data paths, from
Pravin B Shelar.
9) Add SCTP support to openvswitch, from Joe Stringer.
10) Add EF10 chip support to SFC driver, from Ben Hutchings.
11) Add new SYNPROXY netfilter target, from Patrick McHardy.
12) Compute a rate approximation for sending in TCP sockets, and use
this to more intelligently coalesce TSO frames. Furthermore, add
a new packet scheduler which takes advantage of this estimate when
available. From Eric Dumazet.
13) Allow AF_PACKET fanouts with random selection, from Daniel
Borkmann.
14) Add ipv6 support to vxlan driver, from Cong Wang"
Resolved conflicts as per discussion.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1218 commits)
openvswitch: Fix alignment of struct sw_flow_key.
netfilter: Fix build errors with xt_socket.c
tcp: Add missing braces to do_tcp_setsockopt
caif: Add missing braces to multiline if in cfctrl_linkup_request
bnx2x: Add missing braces in bnx2x:bnx2x_link_initialize
vxlan: Fix kernel panic on device delete.
net: mvneta: implement ->ndo_do_ioctl() to support PHY ioctls
net: mvneta: properly disable HW PHY polling and ensure adjust_link() works
icplus: Use netif_running to determine device state
ethernet/arc/arc_emac: Fix huge delays in large file copies
tuntap: orphan frags before trying to set tx timestamp
tuntap: purge socket error queue on detach
qlcnic: use standard NAPI weights
ipv6:introduce function to find route for redirect
bnx2x: VF RSS support - VF side
bnx2x: VF RSS support - PF side
vxlan: Notify drivers for listening UDP port changes
net: usbnet: update addr_assign_type if appropriate
driver/net: enic: update enic maintainers and driver
driver/net: enic: Exposing symbols for Cisco's low latency driver
...
This change sets the PCI devices' initial DMA capabilities
conservatively and promotes them at the request of the driver,
as opposed to assuming advanced DMA capabilities. The old design
runs the risk of breaking drivers that assume default capabilities.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This was really only useful for TILE64 when we mapped the
kernel data with small pages. Now we use a huge page and we
really don't want to map different parts of the kernel
data in different ways.
We retain the __write_once name in case we want to bring
it back to life at some point in the future.
Note that this change uncovered a latent bug where the
"smp_topology" variable happened to always be aligned mod 8
so we could store two "int" values at once, but when we
eliminated __write_once it ended up only aligned mod 4.
Fix with an explicit annotation.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This chip is no longer being actively developed for (it was superceded
by the TILEPro64 in 2008), and in any case the existing compiler and
toolchain in the community do not support it. It's unlikely that the
kernel works with TILE64 at this point as the configuration has not been
tested in years. The support is also awkward as it requires maintaining
a significant number of ifdefs. So, just remove it altogether.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The definisions of __ffs(), __fls(), and ffs() for tile are almost same
as asm-generic/bitops-*.h. The only difference is that it is defined
as __always_inline or inline. So this switches to use those headers.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com> [moved #includes to end]
We use virt_to_pte(NULL, va) a lot, which isn't very obvious.
I added virt_to_kpte(va) as a more obvious wrapper function,
that also validates the va as being a kernel adddress.
And, I fixed the semantics of virt_to_pte() so that we handle
the pud and pmd the same way, and we now document the fact that
we handle the final pte level differently.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Newer hypervisors have an API for reporting per-cpu statistics
information. This change allows seeing that information via
/sys/devices/system/cpu/cpuN/hv_stats file for each core.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Enter kernel debugger at boot with:
--hvd UART_1=1 --hvx kgdbwait --hvx kgdboc=ttyS1,115200
or at runtime with:
echo ttyS1,115200 > /sys/module/kgdboc/parameters/kgdboc
echo g > /proc/sysrq-trigger
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The TILE-Gx chip includes an on-chip UART. This change adds support
for using the UART from within the kernel. The UART shim has more
functionality than is exposed here, but to keep the kernel code and
binary simpler, this is a subset of the full API designed to enable
a standard Linux tty serial driver only.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The existing code relied on the hardware definition (<arch/chip.h>)
to specify how much VA and PA space was available. It's convenient
to allow customizing this for some configurations, so provide symbols
MAX_PA_WIDTH and MAX_VA_WIDTH in <asm/page.h> that can be modified
if desired.
Additionally, move away from the MEM_XX_INTRPT nomenclature to
define the start of various regions within the VA space. In fact
the cleaner symbol is, for example, MEM_SV_START, to indicate the
start of the area used for supervisor code; the actual address of the
interrupt vectors is not as important, and can be changed if desired.
As part of this change, convert from "intrpt1" nomenclature (which
built in the old privilege-level 1 model) to a simple "intrpt".
Also strip out some tilepro-specific code supporting modifying the
PL the kernel could run at, since we don't actually support using
different PLs in tilepro, only tilegx.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Technically, user privilege is anything less than kernel
privilege. We modify the existing user_mode() macro to have
this semantic (and use it in a couple of places it wasn't being
used before), and add an IS_KERNEL_EX1() macro to the assembly
code as well.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This tile-specific API had a minor bug, in that if a super huge (>4GB)
page mapped a particular address range, we wouldn't handle it correctly.
As part of fixing that bug, I also cleaned up some of the pud and pmd
accessors to make them more consistent.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Previously, we used a special-purpose register (SPR_SYSTEM_SAVE_K_0)
to hold the CPU number and the top of the current kernel stack
by using the low bits to hold the CPU number, and using the high
bits to hold the address of the page just above where we'd want
the kernel stack to be. That way we could initialize a new SP
when first entering the kernel by just masking the SPR value and
subtracting a couple of words.
However, it's actually more useful to be able to place an arbitrary
kernel-top value in the SPR. This allows us to create a new stack
context (e.g. for virtualization) with an arbitrary top-of-stack VA.
To make this work, we now store the CPU number in the high bits,
above the highest legal VA bit (42 bits in the current tilegx
microarchitecture). The full 42 bits are thus available to store the
top of stack value. Getting the current cpu (a relatively common
operation) is still fast; it's now a shift rather than a mask.
We make this change only for tilegx, since tilepro has too few SPR
bits to do this, and we don't need this support on tilepro anyway.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Normally the build doesn't include these warnings, but at one
point I built with -Wsign-compare, and noticed a few things that
are technically bugs. This change fixes those things.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Nothing in the codebase was using them, and as written they took
"unsigned long" as the physical address rather than "phys_addr_t",
which is wrong on tilepro anyway. Rather than fixing stale APIs,
just remove them; if there's ever demand for them on this platform,
we can put them back.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
With this change, tile Linux now supports address-space layout
randomization for shared objects, stack, heap and vdso.
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Tony Lu <zlu@tilera.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This may fix a reported bug where an R_TILEGX_64 in a module was not
pointing to an aligned address.
Reported-by: Simon Marchi <simon.marchi@polymtl.ca>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change includes support for Kprobes, Jprobes and Return Probes.
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Tony Lu <zlu@tilera.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This commit adds support for static ftrace, graph function support,
and dynamic tracer support.
Signed-off-by: Tony Lu <zlu@tilera.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change creates the framework for vDSO calls, makes the existing
rt_sigreturn() mechanism use it, and adds a fast gettimeofday().
Now that we need to expose the vDSO address to userspace, we add
AT_SYSINFO_EHDR to the set of aux entries provided to userspace.
(You can disable any extra vDSO support by booting with vdso=0,
but the rt_sigreturn vDSO page will still be provided.)
Note that glibc has supported the tile vDSO since release 2.17.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
First, fix a bug in asm/unaligned.h; we need to just use the asm-generic
unaligned.h so we properly choose endian-correct flavors.
Second, keep the hv/hypervisor.h ABI fully "native" in the sense that
we don't have __BIG_ENDIAN__ ifdefs there. Instead, we use macros in
the head_NN.S assembly code to properly extract two 32-bit structure
members from a 64-bit register holding the structure.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change adds support for CONFIG_PREEMPT (full kernel preemption).
In addition to the core support, this change includes a number
of places where we fix up uses of smp_processor_id() and per-cpu
variables. I also eliminate the PAGE_HOME_HERE and PAGE_HOME_UNKNOWN
values for page homing, as it turns out they weren't being used.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change adds support for avoiding recursive backtracer crashes;
we haven't seen this in practice other than when things are seriously
corrupt, but it may help avoid losing the root cause of a crash.
Also, don't abort kernel backtracers for invalid userspace PC's.
If we do, we lose the ability to backtrace through a userspace
call to a bad address above PAGE_OFFSET, even though that it can
be perfectly reasonable to continue the backtrace in such a case.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change enables unaligned userspace memory access via a kernel
fast path on tilegx. The kernel tracks user PC/instruction pairs
per-thread using a direct-mapped cache in userspace. The cache
maps those PC/instruction pairs to JIT'ed instruction sequences that
load or store using byte-wide load store intructions and then
synthesize 2-, 4- or 8-byte load or store results. Once an
instruction has been seen to generate an unaligned access once,
subsequent hits on that instruction typically require overhead
of only around 50 cycles if cache and TLB is hot.
We support the prctl() PR_GET_UNALIGN / PR_SET_UNALIGN sys call to
enable or disable unaligned fixups on a per-process basis.
To do this we pull some of the tilepro unaligned support out of the
single_step.c file; tilepro uses instruction disassembly for both
single-step and unaligned access support. Since tilegx actually has
hardware singlestep support, though, it's cleaner to keep the tilegx
unaligned access code in a separate file. While we're at it,
properly rename the tilepro-specific types, etc., to have tilepro
suffixes instead of generic tile suffixes.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
This change improves and cleans up the tile console.
- We enable HVC_IRQ support on tilegx, with the addition of a new
Tilera hypervisor API for tilegx to allow a console IPI. If IPI
support is not available we fall back to the previous polling mode.
- We simplify the earlyprintk code to use CON_BOOT and eliminate some
of the other supporting earlyprintk code.
- A new tile_console_write() primitive is used to send output to
the console and is factored out of the hvc_tile driver.
This lets us support a "sim_console" boot argument to allow using
simulator hooks to send output to the "console" as a slightly
faster alternative to emulating the hardware more directly.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On Tilera Gx72 systems, the logic for figuring out whether
a given port is root complex is slightly different.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The standard kernel function dma_get_required_mask() uses the
highest DRAM address to determine if 32-bit or 64-bit DMA addressing
is needed. This only works on architectures that have direct mapping
between the PA and the PCI address space, i.e. those that don't have
I/O TLBs or have I/O TLB but choose to use direct mapping. Neither
of these are true for tilegx. Whether to use 64-bit DMA should depend
on the PCI device's capability only, not on the amount of DRAM
installeds, so we now advertise a 64-bit DMA mask unconditionally.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The .mem_resources[] field in the pci_controller struct
is now obsoleted by the .mem_space and .io_space fields.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The TRIO shim initialization is shared with other kernel drivers
such as the endpoint and StreamIO drivers, so reorganize the
initialization flow to ensure that the root complex driver properly
initializes TRIO state regardless of what kind of TRIO driver will
end up using the shim.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
To enable this functionality, configure CONFIG_TILE_PCI_IO. Without
this flag, the kernel still assigns I/O address ranges to the
devices, but no TRIO resource and mapping support is provided.
We assign disjoint I/O address ranges to separate PCIe domains.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Besides using pr_info() to print the linkdown status for a plug-in
slot, add extra indication that this is expected if the slot is empty.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
To support PCIe devices with higher number of MSI-X interrupt vectors,
e.g. 16 for the LSI RAID card, enhance the Gx RC stack to provide more
MSI-X vectors by using the TRIO Scatter Queues, which provide 8 more
vectors in addition to ~10 from the Map Mem regions.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
The LSI MEGARAID SAS HBA suffers from the problem where it can do
64-bit DMA to streaming buffers but not to consistent buffers.
In other words, 64-bit DMA is used for disk data transfers and 32-bit
DMA must be used for control message transfers. According to LSI,
the firmware is not fully functional yet. This change implements a
kind of hybrid dma_ops to support this.
Note that on most other platforms, the 64-bit DMA addressing space is the
same as the 32-bit DMA space and they overlap the physical memory space.
No special arrangement is needed to support this kind of mixed DMA
capability. On TILE-Gx, the 64-bit DMA space is completely separate
from the 32-bit DMA space. Due to the use of the IOMMU, the 64-bit DMA
space doesn't overlap the physical memory space. On the other hand,
the 32-bit DMA space overlaps the physical memory space under 4GB.
The separate address spaces make it necessary to have separate dma_ops.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
- remove unneeded <linux/bootmem.h> include in pci.c
- eliminate unused pci_controller.first_busno field
- prefer msleep to mdelay
- remove stale comment about pci_scan_bus_parented()
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Using strlen as a model, add length checking to create strnlen.
Signed-off-by: Ken Steele <ken@tilera.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The initial driver support was for a single mPIPE shim on the chip
(as is the case for the Gx36 hardware). The Gx72 chip has two mPIPE
shims, so we extend the driver to handle that case.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The "inv" (invalidate) instruction is generally less safe than "finv"
(flush and invalidate), as it will drop dirty data from the cache.
It turns out we have almost no need for "inv" (other than for the
older 32-bit architecture in some limited cases), so convert to
"finv" where possible and delete the extra "inv" infrastructure.
Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
