Rather than deferring unaligned access speed determinations to a vendor
function, let's probe them and find out how fast they are. If we
determine that an unaligned word access is faster than N byte accesses,
mark the hardware's unaligned access as "fast". Otherwise, we mark
accesses as slow.
The algorithm itself runs for a fixed amount of jiffies. Within each
iteration it attempts to time a single loop, and then keeps only the best
(fastest) loop it saw. This algorithm was found to have lower variance from
run to run than my first attempt, which counted the total number of
iterations that could be done in that fixed amount of jiffies. By taking
only the best iteration in the loop, assuming at least one loop wasn't
perturbed by an interrupt, we eliminate the effects of interrupts and
other "warm up" factors like branch prediction. The only downside is it
depends on having an rdtime granular and accurate enough to measure a
single copy. If we ever manage to complete a loop in 0 rdtime ticks, we
leave the unaligned setting at UNKNOWN.
There is a slight change in user-visible behavior here. Previously, all
boards except the THead C906 reported misaligned access speed of
UNKNOWN. C906 reported FAST. With this change, since we're now measuring
misaligned access speed on each hart, all RISC-V systems will have this
key set as either FAST or SLOW.
Currently, we don't have a way to confidently measure the difference between
SLOW and EMULATED, so we label anything not fast as SLOW. This will
mislabel some systems that are actually EMULATED as SLOW. When we get
support for delegating misaligned access traps to the kernel (as opposed
to the firmware quietly handling it), we can explicitly test in Linux to
see if unaligned accesses trap. Those systems will start to report
EMULATED, though older (today's) systems without that new SBI mechanism
will continue to report SLOW.
I've updated the documentation for those hwprobe values to reflect
this, specifically: SLOW may or may not be emulated by software, and FAST
represents means being faster than equivalent byte accesses. The change
in documentation is accurate with respect to both the former and current
behavior.
Signed-off-by: Evan Green <evan@rivosinc.com>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
Link: https://lore.kernel.org/r/20230818194136.4084400-2-evan@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
The kernel maintains a mask of ISA extensions ANDed together across all
harts. Let's also keep a bitmap of ISA extensions for each CPU. Although
the kernel is currently unlikely to enable a feature that exists only on
some CPUs, we want the ability to report asymmetric CPU extensions
accurately to usermode.
Note that riscv_fill_hwcaps() runs before the per_cpu_offsets are built,
which is why I've used a [NR_CPUS] array rather than per_cpu() data.
Signed-off-by: Evan Green <evan@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Palmer Dabbelt <palmer@rivosinc.com>
Link: https://lore.kernel.org/r/20230509182504.2997252-3-evan@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
This allows userspace to select various routines to use based on the
performance of misaligned access on the target hardware.
Rather than adding DT bindings, this change taps into the alternatives
mechanism used to probe CPU errata. Add a new function pointer alongside
the vendor-specific errata_patch_func() that probes for desirable errata
(otherwise known as "features"). Unlike the errata_patch_func(), this
function is called on each CPU as it comes up, so it can save
feature information per-CPU.
The T-head C906 has fast unaligned access, both as defined by GCC [1],
and in performing a basic benchmark, which determined that byte copies
are >50% slower than a misaligned word copy of the same data size (source
for this test at [2]):
bytecopy size f000 count 50000 offset 0 took 31664899 us
wordcopy size f000 count 50000 offset 0 took 5180919 us
wordcopy size f000 count 50000 offset 1 took 13416949 us
[1] https://github.com/gcc-mirror/gcc/blob/master/gcc/config/riscv/riscv.cc#L353
[2] https://pastebin.com/EPXvDHSW
Co-developed-by: Palmer Dabbelt <palmer@rivosinc.com>
Signed-off-by: Evan Green <evan@rivosinc.com>
Reviewed-by: Heiko Stuebner <heiko.stuebner@vrull.eu>
Tested-by: Heiko Stuebner <heiko.stuebner@vrull.eu>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Paul Walmsley <paul.walmsley@sifive.com>
Link: https://lore.kernel.org/r/20230407231103.2622178-5-evan@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
In preparation for tracking and exposing microarchitectural details to
userspace (like whether or not unaligned accesses are fast), move the
riscv_cpuinfo struct out to its own new cpufeatures.h header. It will
need to be used by more than just cpu.c.
Signed-off-by: Evan Green <evan@rivosinc.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Heiko Stuebner <heiko.stuebner@vrull.eu>
Tested-by: Heiko Stuebner <heiko.stuebner@vrull.eu>
Reviewed-by: Paul Walmsley <paul.walmsley@sifive.com>
Link: https://lore.kernel.org/r/20230407231103.2622178-2-evan@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
