Commit:
a33d331761 ("x86/CPU/AMD: Fix Bulldozer topology")
restored the initial approach we had with the Fam15h topology of
enumerating CU (Compute Unit) threads as cores. And this is still
correct - they're beefier than HT threads but still have some
shared functionality.
Our current approach has a problem with the Mad Max Steam game, for
example. Yves Dionne reported a certain "choppiness" while playing on
v4.9.5.
That problem stems most likely from the fact that the CU threads share
resources within one CU and when we schedule to a thread of a different
compute unit, this incurs latency due to migrating the working set to a
different CU through the caches.
When the thread siblings mask mirrors that aspect of the CUs and
threads, the scheduler pays attention to it and tries to schedule within
one CU first. Which takes care of the latency, of course.
Reported-by: Yves Dionne <yves.dionne@gmail.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: <stable@vger.kernel.org> # 4.9
Cc: Brice Goglin <Brice.Goglin@inria.fr>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yazen Ghannam <yazen.ghannam@amd.com>
Link: http://lkml.kernel.org/r/20170205105022.8705-1-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull irq fixes from Thomas Gleixner:
- Prevent double activation of interrupt lines, which causes problems
on certain interrupt controllers
- Handle the fallout of the above because x86 (ab)uses the activation
function to reconfigure interrupts under the hood.
* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/irq: Make irq activate operations symmetric
irqdomain: Avoid activating interrupts more than once
commit 8fd524b355 ("x86: Kill bad_dma_address variable") has killed
bad_dma_address variable and used instead of macro DMA_ERROR_CODE
which is always zero. Since dma_addr is unsigned, the statement
dma_addr >= DMA_ERROR_CODE
is always true, and not needed.
arch/x86/kernel/pci-calgary_64.c: In function ‘iommu_free’:
arch/x86/kernel/pci-calgary_64.c:299:2: warning: comparison of unsigned expression >= 0 is always true [-Wtype-limits]
if (unlikely((dma_addr >= DMA_ERROR_CODE) && (dma_addr < badend))) {
Fixes: 8fd524b355 ("x86: Kill bad_dma_address variable")
Signed-off-by: Nikola Pajkovsky <npajkovsky@suse.cz>
Cc: iommu@lists.linux-foundation.org
Cc: Jon Mason <jdmason@kudzu.us>
Cc: Muli Ben-Yehuda <mulix@mulix.org>
Link: http://lkml.kernel.org/r/7612c0f9dd7c1290407dbf8e809def922006920b.1479161177.git.npajkovsky@suse.cz
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Introduce ELF_HWCAP2 variable for x86 and reserve its bit 0 to expose the
ring 3 MONITOR/MWAIT.
HWCAP variables contain bitmasks which can be used by userspace
applications to detect which instruction sets are supported by CPU. On x86
architecture information about CPU capabilities can be checked via CPUID
instructions, unfortunately presence of ring 3 MONITOR/MWAIT feature cannot
be checked this way. ELF_HWCAP cannot be used as well, because on x86 it is
set to CPUID[1].EDX which means that all bits are reserved there.
HWCAP2 approach was chosen because it reuses existing solution present
in other architectures, so only minor modifications are required to the
kernel and userspace applications. When ELF_HWCAP2 is defined
kernel maps it to AT_HWCAP2 during the start of the application.
This way the ring 3 MONITOR/MWAIT feature can be detected using getauxval()
API in a simple and fast manner. ELF_HWCAP2 type is u32 to be consistent
with x86 ELF_HWCAP type.
Signed-off-by: Grzegorz Andrejczuk <grzegorz.andrejczuk@intel.com>
Cc: Piotr.Luc@intel.com
Cc: dave.hansen@linux.intel.com
Link: http://lkml.kernel.org/r/1484918557-15481-3-git-send-email-grzegorz.andrejczuk@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Move the check to whether this is a UV system that needs initialization
from is_uv_system() to the internal uv_system_init() function. This is
because on a UV system without a HUB the is_uv_system() returns false.
But we still need some specific UV system initialization. See the
uv_system_init() for change to a quick check if UV is applicable. This
change should not increase overhead since is_uv_system() also called
into this same area.
Signed-off-by: Mike Travis <travis@sgi.com>
Reviewed-by: Russ Anderson <rja@hpe.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Dimitri Sivanich <sivanich@hpe.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170125163518.256403963@asylum.americas.sgi.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Erik reported that on a preproduction hardware a CMCI storm triggers the
BUG_ON in add_timer_on(). The reason is that the per CPU MCE timer is
started by the CMCI logic before the MCE CPU hotplug callback starts the
timer with add_timer_on(). So the timer is already queued which triggers
the BUG.
Using add_timer_on() is pretty pointless in this code because the timer is
strictlty per CPU, initialized as pinned and all operations which arm the
timer happen on the CPU to which the timer belongs.
Simplify the whole machinery by using mod_timer() instead of add_timer_on()
which avoids the problem because mod_timer() can handle already queued
timers. Use __start_timer() everywhere so the earliest armed expiry time is
preserved.
Reported-by: Erik Veijola <erik.veijola@intel.com>
Tested-by: Borislav Petkov <bp@alien8.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Borislav Petkov <bp@alien8.de>
Cc: Tony Luck <tony.luck@intel.com>
Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1701310936080.3457@nanos
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
CONFIG_DEBUG_NX_TEST has been broken since CONFIG_DEBUG_SET_MODULE_RONX=y
was added in v2.6.37 via:
84e1c6bb38 ("x86: Add RO/NX protection for loadable kernel modules")
since the exception table was then made read-only.
Additionally, the manually constructed extables were never fixed when
relative extables were introduced in v3.5 via:
706276543b ("x86, extable: Switch to relative exception table entries")
However, relative extables won't work for test_nx.c, since test instruction
memory areas may be more than INT_MAX away from an executable fixup
(e.g. stack and heap too far away from executable memory with the fixup).
Since clearly no one has been using this code for a while now, and similar
tests exist in LKDTM, this should just be removed entirely.
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jinbum Park <jinb.park7@gmail.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170131003711.GA74048@beast
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add the TSC value to the MCE record only when the MCE being logged is
precise, i.e., it is logged as an exception or an MCE-related interrupt.
So it doesn't look particularly easy to do without touching/changing a
bunch of places. That's why I'm trying tricks first.
For example, the mce-apei.c case I'm addressing by setting ->tsc only
for errors of panic severity. The idea there is, that, panic errors will
have raised an #MC and not polled.
And then instead of propagating a flag to mce_setup(), it seems
easier/less code to set ->tsc depending on the call sites, i.e.,
are we polling or are we preparing an MCE record in an exception
handler/thresholding interrupt.
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yazen Ghannam <Yazen.Ghannam@amd.com>
Cc: linux-edac <linux-edac@vger.kernel.org>
Link: http://lkml.kernel.org/r/20170123183514.13356-5-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The compacted-format XSAVES area is determined at boot time and
never changed after. The field xsave.header.xcomp_bv indicates
which components are in the fixed XSAVES format.
In fpstate_init() we did not set xcomp_bv to reflect the XSAVES
format since at the time there is no valid data.
However, after we do copy_init_fpstate_to_fpregs() in fpu__clear(),
as in commit:
b22cbe404a x86/fpu: Fix invalid FPU ptrace state after execve()
and when __fpu_restore_sig() does fpu__restore() for a COMPAT-mode
app, a #GP occurs. This can be easily triggered by doing valgrind on
a COMPAT-mode "Hello World," as reported by Joakim Tjernlund and
others:
https://bugzilla.kernel.org/show_bug.cgi?id=190061
Fix it by setting xcomp_bv correctly.
This patch also moves the xcomp_bv initialization to the proper
place, which was in copyin_to_xsaves() as of:
4c833368f0 x86/fpu: Set the xcomp_bv when we fake up a XSAVES area
which fixed the bug too, but it's more efficient and cleaner to
initialize things once per boot, not for every signal handling
operation.
Reported-by: Kevin Hao <haokexin@gmail.com>
Reported-by: Joakim Tjernlund <Joakim.Tjernlund@infinera.com>
Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@suse.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi V. Shankar <ravi.v.shankar@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: haokexin@gmail.com
Link: http://lkml.kernel.org/r/1485212084-4418-1-git-send-email-yu-cheng.yu@intel.com
[ Combined it with 4c833368f0. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The equivalence ID was needed outside of the container scanning logic
but now, after this has been cleaned up, not anymore. Now, cont_desc.mc
is used to denote whether the container we're looking at has the proper
microcode patch for this CPU or not.
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/20170120202955.4091-17-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
The idea was to not scan the microcode blob on each AP (Application
Processor) during boot and thus save us some milliseconds. However, on
architectures where the microcode engine is shared between threads, this
doesn't work. Here's why:
The microcode on CPU0, i.e., the first thread, gets updated. The second
thread, i.e., CPU1, i.e., the first AP walks into load_ucode_amd_ap(),
sees that there's no container cached and goes and scans for the proper
blob.
It finds it and as a last step of apply_microcode_early_amd(), it tries
to apply the patch but that core has already the updated microcode
revision which it has received through CPU0's update. So it returns
false and we do desc->size = -1 to prevent other APs from scanning.
However, the next AP, CPU2, has a different microcode engine which
hasn't been updated yet. The desc->size == -1 test prevents it from
scanning the blob anew and we fail to update it.
The fix is much more straight-forward than it looks: the BSP
(BootStrapping Processor), i.e., CPU0, caches the microcode patch
in amd_ucode_patch. We use that on the AP and try to apply it.
In the 99.9999% of cases where we have homogeneous cores - *not*
mixed-steppings - the application will be successful and we're good to
go.
In the remaining small set of systems, we will simply rescan the blob
and find (or not, if none present) the proper patch and apply it then.
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170120202955.4091-16-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
It was pretty clumsy before and the whole work of parsing the microcode
containers was spread around the functions wrongly.
Clean it up so that there's a main scan_containers() function which
iterates over the microcode blob and picks apart the containers glued
together. For each container, it calls a parse_container() helper which
concentrates on one container only: sanity-checking, parsing, counting
microcode patches in there, etc.
It makes much more sense now and it is actually very readable. Oh, and
we luvz a diffstat removing more crap than adding.
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/20170120202955.4091-8-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
