[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20240325064023.2997-1-adrian.hunter@intel.com>
Date: Mon, 25 Mar 2024 08:40:04 +0200
From: Adrian Hunter <adrian.hunter@...el.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: Michael Ellerman <mpe@...erman.id.au>,
Nicholas Piggin <npiggin@...il.com>,
Christophe Leroy <christophe.leroy@...roup.eu>,
"Aneesh Kumar K.V" <aneesh.kumar@...nel.org>,
"Naveen N. Rao" <naveen.n.rao@...ux.ibm.com>,
Heiko Carstens <hca@...ux.ibm.com>,
Vasily Gorbik <gor@...ux.ibm.com>,
Alexander Gordeev <agordeev@...ux.ibm.com>,
Christian Borntraeger <borntraeger@...ux.ibm.com>,
Sven Schnelle <svens@...ux.ibm.com>,
Ingo Molnar <mingo@...hat.com>,
Borislav Petkov <bp@...en8.de>,
Dave Hansen <dave.hansen@...ux.intel.com>,
x86@...nel.org,
"H. Peter Anvin" <hpa@...or.com>,
Andy Lutomirski <luto@...nel.org>,
Vincenzo Frascino <vincenzo.frascino@....com>,
John Stultz <jstultz@...gle.com>,
Stephen Boyd <sboyd@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
Randy Dunlap <rdunlap@...radead.org>,
Bjorn Helgaas <bhelgaas@...gle.com>,
Arnd Bergmann <arnd@...db.de>,
Anna-Maria Behnsen <anna-maria@...utronix.de>,
linuxppc-dev@...ts.ozlabs.org,
linux-kernel@...r.kernel.org,
linux-s390@...r.kernel.org
Subject: [PATCH V2 00/19] timekeeping: Handle potential multiplication overflow
Hi
Kernel timekeeping calculates a clock value by keeping a base value and
adding the number of nanoseconds since that time. Those nanoseconds are
calculated from the clocksource delta. Then periodically, the base value is
moved forwards (refer timekeeping_advance()) which is done by the local
timer interrupt handler. It is designed such that there will always be a
timer interrupt before the delta becomes big enough to overflow the 64-bit
multiplication used in the conversion of delta to nanoseconds (refer
timekeeping_delta_to_ns()). Obviously if timer interrupts are stopped, then
the multiplication does eventually overflow.
Timekeeping multiplication overflow results in a "time loop", typically
cycling about every 15 minutes with x86 TSC, for example starting at 10:00:
10:00, 10:01, 10:02 ... 10:15, 10:00, 10:01, ... 10:15, 10:00, 10:01 ...
Because a VMM can deliberately stop timer interrupts for a guest, a virtual
machine can be exposed to this issue.
TDX maintains a monotonically increasing virtual TSC for a TDX guest, so
the overflow is allowing a backwards movement of timekeeping that would not
happen otherwise.
It is considered this could break security of cryptographic protocols that
rely on the timestamps for freshness / replay protection, and consequently
the kernel should prevent such a time loop.
Handle multiplication overflows by falling back to higher precision
calculation when the possibility of an overflow is detected.
Extend the facility also to VDSO, dependent on new config option
GENERIC_VDSO_OVERFLOW_PROTECT which is selected by x86 only, so other
architectures are not affected. The result is a calculation that has
similar performance as before. Most machines showed performance benefit,
except Skylake-based hardware such as Intel Kaby Lake which was seen <1%
worse.
Changes in V2:
vdso: Consolidate vdso_calc_delta()
Keep powerpc comment about mask
Move ifdef out of function
vdso: Consolidate nanoseconds calculation
Adjusted due to changes in "vdso: Consolidate vdso_calc_delta()"
Adrian Hunter (19):
vdso: Consolidate vdso_calc_delta()
vdso: Consolidate nanoseconds calculation
vdso: Add CONFIG_GENERIC_VDSO_OVERFLOW_PROTECT
math64: Tidy mul_u64_u32_shr()
vdso: math64: Provide mul_u64_u32_add_u64_shr()
vdso: Add vdso_data::max_cycles
vdso: Make delta calculation overflow safe
x86/vdso: Make delta calculation overflow safe
timekeeping: Move timekeeping helper functions
timekeeping: Rename fast_tk_get_delta_ns() to __timekeeping_get_ns()
timekeeping: Tidy timekeeping_cycles_to_ns() slightly
timekeeping: Reuse timekeeping_cycles_to_ns()
timekeeping: Refactor timekeeping helpers
timekeeping: Consolidate timekeeping helpers
timekeeping: Fold in timekeeping_delta_to_ns()
timekeeping: Prepare timekeeping_cycles_to_ns() for overflow safety
timekeeping: Make delta calculation overflow safe
timekeeping: Let timekeeping_cycles_to_ns() handle both under and overflow
clocksource: Make watchdog and suspend-timing multiplication overflow safe
arch/powerpc/include/asm/vdso/gettimeofday.h | 26 +++----
arch/s390/include/asm/vdso/gettimeofday.h | 7 +-
arch/x86/Kconfig | 1 +
arch/x86/include/asm/vdso/gettimeofday.h | 42 +++++++----
include/linux/math64.h | 8 +-
include/vdso/datapage.h | 4 +
include/vdso/math64.h | 38 ++++++++++
kernel/time/clocksource.c | 42 +++++------
kernel/time/timekeeping.c | 106 ++++++++++++++-------------
kernel/time/vsyscall.c | 6 ++
lib/vdso/Kconfig | 7 ++
lib/vdso/gettimeofday.c | 55 +++++++++-----
12 files changed, 208 insertions(+), 134 deletions(-)
Regards
Adrian
Powered by blists - more mailing lists