cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

fpu.h (4630B)


      1/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
      2#ifndef _UAPI__ASM_ALPHA_FPU_H
      3#define _UAPI__ASM_ALPHA_FPU_H
      4
      5
      6/*
      7 * Alpha floating-point control register defines:
      8 */
      9#define FPCR_DNOD	(1UL<<47)	/* denorm INV trap disable */
     10#define FPCR_DNZ	(1UL<<48)	/* denorms to zero */
     11#define FPCR_INVD	(1UL<<49)	/* invalid op disable (opt.) */
     12#define FPCR_DZED	(1UL<<50)	/* division by zero disable (opt.) */
     13#define FPCR_OVFD	(1UL<<51)	/* overflow disable (optional) */
     14#define FPCR_INV	(1UL<<52)	/* invalid operation */
     15#define FPCR_DZE	(1UL<<53)	/* division by zero */
     16#define FPCR_OVF	(1UL<<54)	/* overflow */
     17#define FPCR_UNF	(1UL<<55)	/* underflow */
     18#define FPCR_INE	(1UL<<56)	/* inexact */
     19#define FPCR_IOV	(1UL<<57)	/* integer overflow */
     20#define FPCR_UNDZ	(1UL<<60)	/* underflow to zero (opt.) */
     21#define FPCR_UNFD	(1UL<<61)	/* underflow disable (opt.) */
     22#define FPCR_INED	(1UL<<62)	/* inexact disable (opt.) */
     23#define FPCR_SUM	(1UL<<63)	/* summary bit */
     24
     25#define FPCR_DYN_SHIFT	58		/* first dynamic rounding mode bit */
     26#define FPCR_DYN_CHOPPED (0x0UL << FPCR_DYN_SHIFT)	/* towards 0 */
     27#define FPCR_DYN_MINUS	 (0x1UL << FPCR_DYN_SHIFT)	/* towards -INF */
     28#define FPCR_DYN_NORMAL	 (0x2UL << FPCR_DYN_SHIFT)	/* towards nearest */
     29#define FPCR_DYN_PLUS	 (0x3UL << FPCR_DYN_SHIFT)	/* towards +INF */
     30#define FPCR_DYN_MASK	 (0x3UL << FPCR_DYN_SHIFT)
     31
     32#define FPCR_MASK	0xffff800000000000L
     33
     34/*
     35 * IEEE trap enables are implemented in software.  These per-thread
     36 * bits are stored in the "ieee_state" field of "struct thread_info".
     37 * Thus, the bits are defined so as not to conflict with the
     38 * floating-point enable bit (which is architected).  On top of that,
     39 * we want to make these bits compatible with OSF/1 so
     40 * ieee_set_fp_control() etc. can be implemented easily and
     41 * compatibly.  The corresponding definitions are in
     42 * /usr/include/machine/fpu.h under OSF/1.
     43 */
     44#define IEEE_TRAP_ENABLE_INV	(1UL<<1)	/* invalid op */
     45#define IEEE_TRAP_ENABLE_DZE	(1UL<<2)	/* division by zero */
     46#define IEEE_TRAP_ENABLE_OVF	(1UL<<3)	/* overflow */
     47#define IEEE_TRAP_ENABLE_UNF	(1UL<<4)	/* underflow */
     48#define IEEE_TRAP_ENABLE_INE	(1UL<<5)	/* inexact */
     49#define IEEE_TRAP_ENABLE_DNO	(1UL<<6)	/* denorm */
     50#define IEEE_TRAP_ENABLE_MASK	(IEEE_TRAP_ENABLE_INV | IEEE_TRAP_ENABLE_DZE |\
     51				 IEEE_TRAP_ENABLE_OVF | IEEE_TRAP_ENABLE_UNF |\
     52				 IEEE_TRAP_ENABLE_INE | IEEE_TRAP_ENABLE_DNO)
     53
     54/* Denorm and Underflow flushing */
     55#define IEEE_MAP_DMZ		(1UL<<12)	/* Map denorm inputs to zero */
     56#define IEEE_MAP_UMZ		(1UL<<13)	/* Map underflowed outputs to zero */
     57
     58#define IEEE_MAP_MASK		(IEEE_MAP_DMZ | IEEE_MAP_UMZ)
     59
     60/* status bits coming from fpcr: */
     61#define IEEE_STATUS_INV		(1UL<<17)
     62#define IEEE_STATUS_DZE		(1UL<<18)
     63#define IEEE_STATUS_OVF		(1UL<<19)
     64#define IEEE_STATUS_UNF		(1UL<<20)
     65#define IEEE_STATUS_INE		(1UL<<21)
     66#define IEEE_STATUS_DNO		(1UL<<22)
     67
     68#define IEEE_STATUS_MASK	(IEEE_STATUS_INV | IEEE_STATUS_DZE |	\
     69				 IEEE_STATUS_OVF | IEEE_STATUS_UNF |	\
     70				 IEEE_STATUS_INE | IEEE_STATUS_DNO)
     71
     72#define IEEE_SW_MASK		(IEEE_TRAP_ENABLE_MASK |		\
     73				 IEEE_STATUS_MASK | IEEE_MAP_MASK)
     74
     75#define IEEE_CURRENT_RM_SHIFT	32
     76#define IEEE_CURRENT_RM_MASK	(3UL<<IEEE_CURRENT_RM_SHIFT)
     77
     78#define IEEE_STATUS_TO_EXCSUM_SHIFT	16
     79
     80#define IEEE_INHERIT    (1UL<<63)	/* inherit on thread create? */
     81
     82/*
     83 * Convert the software IEEE trap enable and status bits into the
     84 * hardware fpcr format. 
     85 *
     86 * Digital Unix engineers receive my thanks for not defining the
     87 * software bits identical to the hardware bits.  The chip designers
     88 * receive my thanks for making all the not-implemented fpcr bits
     89 * RAZ forcing us to use system calls to read/write this value.
     90 */
     91
     92static inline unsigned long
     93ieee_swcr_to_fpcr(unsigned long sw)
     94{
     95	unsigned long fp;
     96	fp = (sw & IEEE_STATUS_MASK) << 35;
     97	fp |= (sw & IEEE_MAP_DMZ) << 36;
     98	fp |= (sw & IEEE_STATUS_MASK ? FPCR_SUM : 0);
     99	fp |= (~sw & (IEEE_TRAP_ENABLE_INV
    100		      | IEEE_TRAP_ENABLE_DZE
    101		      | IEEE_TRAP_ENABLE_OVF)) << 48;
    102	fp |= (~sw & (IEEE_TRAP_ENABLE_UNF | IEEE_TRAP_ENABLE_INE)) << 57;
    103	fp |= (sw & IEEE_MAP_UMZ ? FPCR_UNDZ | FPCR_UNFD : 0);
    104	fp |= (~sw & IEEE_TRAP_ENABLE_DNO) << 41;
    105	return fp;
    106}
    107
    108static inline unsigned long
    109ieee_fpcr_to_swcr(unsigned long fp)
    110{
    111	unsigned long sw;
    112	sw = (fp >> 35) & IEEE_STATUS_MASK;
    113	sw |= (fp >> 36) & IEEE_MAP_DMZ;
    114	sw |= (~fp >> 48) & (IEEE_TRAP_ENABLE_INV
    115			     | IEEE_TRAP_ENABLE_DZE
    116			     | IEEE_TRAP_ENABLE_OVF);
    117	sw |= (~fp >> 57) & (IEEE_TRAP_ENABLE_UNF | IEEE_TRAP_ENABLE_INE);
    118	sw |= (fp >> 47) & IEEE_MAP_UMZ;
    119	sw |= (~fp >> 41) & IEEE_TRAP_ENABLE_DNO;
    120	return sw;
    121}
    122
    123
    124#endif /* _UAPI__ASM_ALPHA_FPU_H */