Detect Power8 in-core crypto through getauxval?

341 views Asked by At

I'm on GCC112, which is a little-endian Power8 machine running Linux. Power8 has in-core crypto providing AES, SHA and a few other useful features. I'm trying to determine the availability of the features at runtime using getauxval. The use case is distros building for a "minimum" capable machine, and we need to swap-in a faster function at runtime.

The dump of hwcaps.h is shown below, but it lacks specific bits for Power8, AES, SHA and others. However, I believe Power8 is ISA 2.07, and ISA 2.07 has the bit PPC_FEATURE2_ARCH_2_07.

The thing I am not clear on is, is Power8 in-core crypto optional like ARM's crypto under ARMv8. I can't find a document that clearly states the requirement, and I don't have a membership to OpenPower to access ISA documents. (Another possibility is, it is stated but I missed it in the docs).

Is it possible to use getauxval to query the runtime environment for the features? If not, then how do we determine feature availability at runtime? Is CPU probing the only alternative available?

Maybe more generally, how do we determine Power6, Power7 and Power8 runtime environments?


auxv.h is mostly empty. The header file includes hwcaps.h.

$ cat /usr/include/bits/hwcap.h
...

/* The bit numbers must match those in the kernel's asm/cputable.h.  */

/* Feature definitions in AT_HWCAP.  */
#define PPC_FEATURE_32              0x80000000 /* 32-bit mode. */
#define PPC_FEATURE_64              0x40000000 /* 64-bit mode. */
#define PPC_FEATURE_601_INSTR       0x20000000 /* 601 chip, Old POWER ISA.  */
#define PPC_FEATURE_HAS_ALTIVEC     0x10000000 /* SIMD/Vector Unit.  */
#define PPC_FEATURE_HAS_FPU         0x08000000 /* Floating Point Unit.  */
#define PPC_FEATURE_HAS_MMU         0x04000000 /* Memory Management Unit.  */
#define PPC_FEATURE_HAS_4xxMAC      0x02000000 /* 4xx Multiply Accumulator.  */
#define PPC_FEATURE_UNIFIED_CACHE   0x01000000 /* Unified I/D cache.  */
#define PPC_FEATURE_HAS_SPE         0x00800000 /* Signal Processing ext.  */
#define PPC_FEATURE_HAS_EFP_SINGLE  0x00400000 /* SPE Float.  */
#define PPC_FEATURE_HAS_EFP_DOUBLE  0x00200000 /* SPE Double.  */
#define PPC_FEATURE_NO_TB           0x00100000 /* 601/403gx have no timebase */
#define PPC_FEATURE_POWER4          0x00080000 /* POWER4 ISA 2.00 */
#define PPC_FEATURE_POWER5          0x00040000 /* POWER5 ISA 2.02 */
#define PPC_FEATURE_POWER5_PLUS     0x00020000 /* POWER5+ ISA 2.03 */
#define PPC_FEATURE_CELL_BE         0x00010000 /* CELL Broadband Engine */
#define PPC_FEATURE_BOOKE           0x00008000 /* ISA Category Embedded */
#define PPC_FEATURE_SMT             0x00004000 /* Simultaneous
                                                  Multi-Threading */
#define PPC_FEATURE_ICACHE_SNOOP    0x00002000
#define PPC_FEATURE_ARCH_2_05       0x00001000 /* ISA 2.05 */
#define PPC_FEATURE_PA6T            0x00000800 /* PA Semi 6T Core */
#define PPC_FEATURE_HAS_DFP         0x00000400 /* Decimal FP Unit */
#define PPC_FEATURE_POWER6_EXT      0x00000200 /* P6 + mffgpr/mftgpr */
#define PPC_FEATURE_ARCH_2_06       0x00000100 /* ISA 2.06 */
#define PPC_FEATURE_HAS_VSX         0x00000080 /* P7 Vector Extension.  */
#define PPC_FEATURE_PSERIES_PERFMON_COMPAT  0x00000040
#define PPC_FEATURE_TRUE_LE         0x00000002
#define PPC_FEATURE_PPC_LE          0x00000001

/* Feature definitions in AT_HWCAP2.  */
#define PPC_FEATURE2_ARCH_2_07     0x80000000 /* ISA 2.07 */
#define PPC_FEATURE2_HAS_HTM       0x40000000 /* Hardware Transactional
                                                 Memory */
#define PPC_FEATURE2_HAS_DSCR      0x20000000 /* Data Stream Control
                                                 Register */
#define PPC_FEATURE2_HAS_EBB       0x10000000 /* Event Base Branching */
#define PPC_FEATURE2_HAS_ISEL      0x08000000 /* Integer Select */
#define PPC_FEATURE2_HAS_TAR       0x04000000 /* Target Address Register */
1

There are 1 answers

3
Jeremy Kerr On BEST ANSWER

I'd say that getauxval() would be the best way to do this; the HWCAP & HWCAP2 values are exactly for determining hardware features. Missing from your list is the PPC_FEATURE2_VEC_CRYPTO, which indicates the presence of the vector crypto instructions, which sounds like the one you need.

As a side note: you probably don't want to detect processor implementations, but processor features. Specifically, check for the individual feature, rather than trying to check for a process that provides that feature. (eg., detect VEC_CRYPTO directly, rather than trying to check for POWER8, and assume that that implies crypto functionality).

As a bit of detail, Linux's cputable entries specify the HWCAP/HWCAP2 values. Using POWER8 as an example:

#define COMMON_USER2_POWER8 (PPC_FEATURE2_ARCH_2_07 | \
                 PPC_FEATURE2_HTM_COMP | \
                 PPC_FEATURE2_HTM_NOSC_COMP | \
                 PPC_FEATURE2_DSCR | \
                 PPC_FEATURE2_ISEL | PPC_FEATURE2_TAR | \
                 PPC_FEATURE2_VEC_CRYPTO)

That's from arch/powerpc/include/asm/cputable.h in the kernel (which also provides the actual hwcap bits that can be set in the aux vector).

Finally, I'm fairly sure you don't need to be an OpenPOWER foundation member to download the ISA (latest is 3.0B) - you just need an account on the website.