/* IBM_PROLOG_BEGIN_TAG */
/* This is an automatically generated prolog. */
/* */
/* bos72Q src/bos/kernel/sys/xmem.h 1.55.2.10 */
/* */
/* Licensed Materials - Property of IBM */
/* */
/* COPYRIGHT International Business Machines Corp. 1988,2019 */
/* All Rights Reserved */
/* */
/* US Government Users Restricted Rights - Use, duplication or */
/* disclosure restricted by GSA ADP Schedule Contract with IBM Corp. */
/* */
/* IBM_PROLOG_END_TAG */
/* @(#)61 1.55.2.10 src/bos/kernel/sys/xmem.h, sysvmm, bos72Q, q2019_13A4 2/6/19 00:42:17 */
/*
* COMPONENT_NAME: (SYSVMM) Virtual Memory Manager
*
* FUNCTIONS:
*
* ORIGINS: 27
*
* (C) COPYRIGHT International Business Machines Corp. 1988, 1997
* All Rights Reserved
* Licensed Materials - Property of IBM
*
* US Government Users Restricted Rights - Use, duplication or
* disclosure restricted by GSA ADP Schedule Contract with IBM Corp.
*/
#ifndef _H_XMEM
#define _H_XMEM
#include <sys/types.h>
#include <sys/vm_types.h>
#if defined (__64BIT_KERNEL) && defined (_KERNEL)
#include <sys/kerrno.h>
#include <sys/skeys.h>
#endif /* __64BIT_KERNEL && _KERNEL */
#ifdef __cplusplus
extern "C" {
#endif
/*
* Memory Address Space Definitions
*
* These values define the address space for kernel services requiring
* address space information (usually via the segflag parameter).
*
* value of 3 should be reserved for use by UIO_XMEM in uio.h.
*
* LW_XMATTACH can only be used as a segflag value when calling xmattach(),
* not in uio code. An xmattach() with LW_XMATTACH implies
* USER_ADSPACE; calling xmattach() with LW_XMATTACH to attach from system
* address space is illegal.
*/
#define USER_ADSPACE 0 /* user address space */
#define SYS_ADSPACE 1 /* system address space */
#define USERI_ADSPACE 2 /* user instruction address space */
/* #define UIO_XMEM 3 <sys/uio.h> */
#define LW_XMATTACH 4 /* lw attach in user address space */
#define SYS_ADSPACE_ASSIGN_KEYSET 5 /* use current keyset */
/*
* Return Codes
*/
#define XMEM_SUCC (0) /* successful completion */
#define XMEM_FAIL (-1) /* unsuccessful completion */
#define XMEM_AGAIN (-2) /* attempt the operation again */
/*
* Cross Memory Address Space Definitions
*/
#define XMEM_PROC ( 0) /* process address space */
#define XMEM_GLOBAL (-1) /* global address space */
#define XMEM_INVAL (-2) /* access removed */
#ifndef __64BIT_KERNEL /* These aren't used in 64-bit kernel */
#define XMEM_PROC2 (-3) /* process aspace, split portion */
#define XMEM_XLATE (-4) /* Special pre-translated */
#endif /* ! __64BIT_KERNEL */
#define XMEM_REMIO (-5) /* remote i/o address space */
/*
* Structure for managing pre-translated memory buffers
* within the xmem descriptor
*/
#ifdef __64BIT_KERNEL
struct xmem_phys {
rpn_t total;
rpn_t used;
vpn_t s_vpn;
rpn_t *rpn;
};
#else
struct xmem_phys {
int total;
int used;
unsigned int s_vpn;
rpn_t *rpn;
};
#endif
/*
* Pretranslation Definitions
*/
#define XLATE_PERSISTENT 0x1 /* prex data structures should */
/* be persistent across calls */
#define XLATE_PRESENT 0x2 /* pre-translated buffer */
#define XLATE_ALLOC 0x4 /* only allocate prex data */
/* structures ....xlate later */
#define XLATE_SPARSE 0x8 /* sparse translations */
#define XLATE_DR_SAFE 0x10 /* dynamic reconfig safe prex data */
#define XLATE_TRANSIENT 0x20 /* transient pretranslation */
#define XLATE_FAST 0x40 /* fast-path pre-xlation used */
#ifdef __64BIT_KERNEL
/*
* xmemflags definitions.
*/
#define XMEM_AUTHORITY 0x0001 /* unprivileged or priviledged */
#define XMEM_TEMP_ATT 0x0002 /* xmem from a temp attach */
#define XMEM_LW 0x0004 /* xmem from a lightweight attach */
#define XMEM_SEG_XLATE 0x0008 /* segments are already pre-xlated */
#define XMEM_SUB_PTR 0x0010 /* use subspace_ptr not _id */
#define XMEM_SEG_1T 0x0020 /* segments are 1T, not 256M */
#define XMEM_AMR_SAFE 0x0040 /* AMR in xmemd somewhere */
/*
* xm_amr definitions
*/
#define XM_ALL_HKEYS 0 /* access to all h/w storage keys */
#ifdef _KERNSYS
/* Last value in vmhandle array of xmem descriptor is really a
* timestamp.
*/
#define LW_TIME(_dp) ((_dp)->subspace_ptr[(_dp)->num_sids])
/* 64-bit AMR is at end of subspace extent list if there is one.
*/
#define XM_AMRX(_dp) ((_dp)->subspace_ptr[(_dp)->num_sids])
#endif /* _KERNSYS */
#endif /* __64BIT_KERNEL */
/*
* Cross Memory Descriptor.
* Describes what portions of memory are accessable.
*/
#ifdef __64BIT_KERNEL
#if defined(__64BIT__) || defined(_LONG_LONG)
/*
* This is the 64-bit kernel's definition of a cross memory descriptor
* There is no limit to the number of segment crossings allowed.
*/
struct xmem
{
int aspace_id; /* describe address space */
int num_sids; /* number of segment IDs */
/* If !(xmemflags & XMEM_SUB_PTR), this field is the Segment ID.
*
* If (xmemflags & XMEM_SUB_PTR), this field is a pointer to a
* list of sids, supporting multi-segment crossings.
*
* For XMEM_GLOBAL descriptors, this field contains the AMR, if
* XMEM_AMR_SAFE is set.
*
* NOTE: it is not valid to statically construct a global xmemd
* with an AMR in it. Users must use xmsethkeyset() or xmattach()
* to get an AMR value in a cross-memory descriptor.
*
* The hkeyset_t type is only available with _KERNEL defined so
* we must be careful not to use it if thats not the case.
*/
union
{
vmhandle_t _subspace_id;
vmhandle_t *_subspace_ptr;
#ifdef _KERNEL
hkeyset_t _xm_globamr;
#else
unsigned long _xm_globamr;
#endif /* _KERNEL */
} _u;
#define subspace_id _u._subspace_id
#define subspace_ptr _u._subspace_ptr
#define xm_globamr _u._xm_globamr
char *vaddr; /* Attached start vaddr */
ushort xmemflags; /* xmem flags */
ushort prexflags; /* pre-translation flags */
ushort xm_amr; /* partial AMR, 8 keys (0-7) */
pshift_t l2psize; /* log2 of region's page size */
struct xmem_phys xp;
/* XMEM_REMIO overloaded fields:
* vaddr - remote liobn
* subspace_id - remote i/o address
* xm_len - length in bytes (overloaded on 32bk)
*/
#define rem_liobn vaddr
#define rem_addr _u._subspace_id
#define rem_size num_sids
};
#endif /* defined(__64BIT__) || defined(_LONG_LONG) */
#else /* __64BIT_KERNEL */
struct xmem {
union {
struct {
short version;
short flag; /* address space of buffer */
} s;
int _aspace_id;
} _u;
#define aspace_id _u._aspace_id
#define xm_flag _u.s.flag
#define xm_version _u.s.version
union {
vmhandle_t _subspace_id; /* portion of aspace for this */
uint _prexflags; /* pre-translation flag word */
} u1;
#define subspace_id u1._subspace_id
#define prexflags u1._prexflags
union {
vmhandle_t _subspace_id2; /* 2nd portion of aspace */
struct xmem *_orig_xmem; /* ptr to real original xmem */
} u2;
#define subspace_id2 u2._subspace_id2
#define orig_xmem u2._orig_xmem
char *uaddr; /* user buffer address */
/* pre-translation overloaded fields:
* uaddr - pointer to the xmem_phys struct describing the
* pre-xlated information
*/
#define rlist uaddr
/* XMEM_REMIO overloaded fields:
* uaddr - remote liobn
* subspace_id - remote i/o address
* subspace_id2 - remote length in bytes
*/
#define rem_liobn uaddr
#define rem_addr u1._subspace_id
#define rem_size u2._subspace_id2
};
#endif /* __64BIT_KERNEL */
/* This is the maximum size of a remote TCE table. Bytes per TCE is defined
* as TCE_MAP_SIZE in <sys/dma.h>
*/
#define MAX_TCE_MAP_SIZE 0x100000000ULL
/*
* For VIO support
*/
#define IS_REMOTE(xp) (xp->aspace_id == XMEM_REMIO)
#define GET_LIOBN(xp) ((liobn_t)xp->rem_liobn)
extern struct xmem *xmem_global;
/*
* Macros for cross memory attach/detach operations.
*/
#define xmemat(vaddr,count,dp) xmattach(vaddr,count,dp,SYS_ADSPACE)
#define xmemdt(dp) xmdetach(dp)
/*
* Entry points for cross memory operations.
*/
/*cross memory atomic operations*/
int xmem_compare_and_swap_32(int *location,int *original,int newval,
struct xmem *dp);
#ifdef _LONG_LONG /*ensure that compiler knows about long long*/
int xmem_compare_and_swap_64(long long *location,long long *original,
long long newval,struct xmem *dp);
#endif /*_LONG_LONG*/
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
int xmattach(caddr_t, int32long64_t, struct xmem *, int);
#else
int xmattach(); /* setup for cross memory */
#endif
/* char *uaddr; */ /* user buffer address */
/* int32long64_t count;*/ /* user buffer length */
/* struct xmem *dp; */ /* cross memory descriptor*/
/* int segflag; */ /* address space of buffer */
#ifndef __64BIT_KERNEL
/* 64bit prototype
*/
#if defined(_LONG_LONG)
int xmattach64(unsigned long long vaddr, int count, struct xmem *dp,
int segflag);
#endif
#endif
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
int xmbsterile(caddr_t, size_t, struct xmem*, ulong);
#else
int xmbsterile();
#endif
/* caddr_t bufaddr; */ /* starting address of buffer */
/* size_t buflen; */ /* buffer size in bytes */
/* struct xmem *dp; */ /* cross memory descriptor */
/* ulong flags; */ /* flags */
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
/*
* xm_ue_query
* Syntax:
* #include <sys/xmem.h>
* long xm_ue_query(bufaddr, buflen, dp, ue_addr, flags)
* caddr_t bufaddr;
* size_t buflen;
* struct xmem *dp;
* caddr_t *ue_addr;
* ulong flags;
*
* Input Parameters:
* bufaddr Starting address of buffer
* buflen Buffer size in bytes
* dp Cross-memory descriptor
* flags Flags :
* 0x1 - Use an exception handler.
*
* Output Parameters:
* ue_addr Address of the first cache line containing the first UE or the
* address of the buffer if the first UE is in the first cache line
* and the buffer is not cache aligned.
*
* Description: The xm_ue_query kernel service determines if there are any
* uncorrectable errors present in the memory consumed by the
* specified buffer. Upon finding an uncorrectable error, the
* xm_ue_query kernel service returns the address of the first
* uncorrectable error. If the buffer is not cache aligned and the
* uncorrectable error is in the first cache line consumed by the
* buffer, the xm_ue_query subroutine will return the address of
* the start of the buffer.
*
* The xm_ue_query kernel service should be used where the buffer
* might be in either user or kernel space.
*
* The cross-memory descriptor must have been filled in correctly
* prior to the xm_ue_query call (for example, by calling the
* xmattach kernel service).
*
* Execution Environment: The xm_ue_query subroutine can be called from either
* the process or interrupt environment.
*
* Return Values:
* 0 - No UE detected.
* 1 - UE detected. The address of the UE is returned in the ue_addr
* parameter.
* < 0 - Indicates one of the following errors:
* 1. flags parameter is non-zero
* 2. The xmem descriptor is in use, i.e. (aspace_id != XMEM_INVAL)
* 3. The range of the buffer is outside of the xmem descriptor
* 4. The buffer wraps.
* 5. Bad buffer address.
*/
long xm_ue_query(caddr_t, size_t, struct xmem*, caddr_t*, ulong);
#else
long xm_ue_query();
#endif
/* caddr_t bufaddr; */ /* starting address of buffer */
/* size_t buflen; */ /* buffer size in bytes */
/* struct xmem *dp; */ /* cross memory descriptor */
/* caddr_t *ue_addr */ /* address of cache line */
/* containing the first UE */
/* ulong flags; */ /* flags */
/* flags definitions */
#define XM_UE_XCP 0x0001 /* Use exception handler. */
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
long xmattach_remio(liobn_t, ulong, ulong, struct xmem*);
#else
long xmattach_remio();
#endif
/* liobn_t rliobn; */ /* remote LIOBN */
/* ulong raddr; */ /* remote starting address */
/* ulong count; */ /* count in bytes */
/* struct xmem *dp; */ /* remote xmem (XMEM_REMIO)*/
/* cross mem copy in */
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
int xmemin(caddr_t, caddr_t, int32long64_t, struct xmem *);
#else
int xmemin(); /* cross memory copy in */
#endif
/* char *uaddr; */ /* address in other address space */
/* char *kaddr; */ /* address in global memory */
/* int32long64_t count; */ /* amount of data to copy */
/* struct xmem *dp; */ /* cross memory descriptor */
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
int xmemout(caddr_t, caddr_t, int32long64_t, struct xmem *); /* xmem copy out */
#else
int xmemout(); /* cross memory copy out */
#endif
/* char *kaddr; */ /* address in global memory */
/* char *uaddr; */ /* address in other address space*/
/* int32long64_t count; */ /* amount of data to copy */
/* struct xmem *dp; */ /* cross memory descriptor */
#if defined(__64BIT_KERNEL) || defined(__FULL_PROTO)
int xmdetach(struct xmem *); /* cross memory detach */
#else
int xmdetach(); /* cross memory detach */
#endif
/* struct xmem *dp; */ /* cross memory descriptor */
/*
* Entry points for kernel only cross memory operations.
*/
#define XMEM_HIDE 0x00000000 /* hide the page */
#define XMEM_UNHIDE 0x00000001 /* unhide the page */
#define XMEM_ACC_CHK 0x00000002 /* perform access checking */
#define XMEM_WRITE_ONLY 0x00000004 /* intended transfer is outbound only*/
#define XMEM_DR_SAFE 0x00000008 /* caller is DR safe/aware */
#define XMEM_DR_RETRY 0x00000010 /* try xmemdma64() again. only valid
with XMEM_DR_SAFE */
/* cross memory DMA setup */
#ifndef __64BIT_KERNEL
#if defined(__FULL_PROTO)
int xmemdma(struct xmem *, caddr_t, int);
int xmemacc(struct xmem *, caddr_t, int);
#else
int xmemdma(); /* cross memory DMA setup */
/* struct xmem *dp; */ /* cross memory descriptor */
/* caddr_t uaddr; */ /* address in other address space */
/* int flags; */ /* operation flags */
/* (page can be read-only) */
/* returns the real address */
int xmemacc(); /* cross memory access */
/* struct xmem *dp; */ /* cross memory descriptor */
/* caddr_t raddr; */ /* real address */
/* int modify; */ /* non-zero if page modified */
#endif /* __FULL_PROTO */
#endif /* ! __64BIT_KERNEL */
#define XMEM_RDONLY 0 /* readonly access allowed */
#define XMEM_RDWR 1 /* read/write access allowed */
#define XMEM_NOACC 2 /* no access allowed */
#ifdef _KERNSYS
int xlate_create(struct xmem *, caddr_t baddr, int32long64_t, uint);
int xlate_remove(struct xmem *);
int xlate_pin(struct xmem *, caddr_t baddr, int32long64_t, int);
int xlate_unpin(struct xmem *, caddr_t, int32long64_t);
int xlate_addr(struct xmem *, caddr_t, ulong_t, ulong_t *);
int xlate_addr_update(struct xmem *, caddr_t, ulong_t, ulong_t);
int xmemdma64_list(struct xmem *, caddr_t, int32long64_t, void *, int, int);
int xmemdma64_list_ext(struct xmem *, caddr_t, int32long64_t, void *, int, int, uint);
int xmemzero(struct xmem *, caddr_t, long);
#endif /* _KERNSYS */
#if defined(__64BIT_KERNEL)
void *xmemdma64(struct xmem *, caddr_t, int); /* cross memory DMA setup */
#else
#if defined(_LONG_LONG)
#if defined(__FULL_PROTO)
unsigned long long xmemdma64(struct xmem *, caddr_t, int);
#else
unsigned long long xmemdma64(); /* cross memory DMA setup */
#endif /* __FULL_PROTO */
#endif /* _LONG_LONG */
#endif /* __64BIT_KERNEL */
/* struct xmem *dp; */ /* cross memory descriptor */
/* char *uaddr; */ /* address in other address space */
/* int flags; */ /* operation flags, same as */
/* xmemdma() above */
/* (page can be read-only) */
/* returns the 64-bit real address */
#ifdef __64BIT_KERNEL
extern int xmem_getlocation(struct xmem *, caddr_t, size_t, int);
long long xmempsize(struct xmem *, void *, size_t);
#ifdef _KERNEL
kerrno_t xmsethkeyset(struct xmem *, hkeyset_t, long);
/* xmsethkeyset returns */
#define EINVAL_XMSETHKEYSET KERROR(EINVAL, sysvmm_BLOCK_01, 1)
kerrno_t xmgethkeyset(struct xmem *, hkeyset_t *, long);
/* xmgethkeyset returns */
#define EINVAL_XMGETHKEYSET KERROR(EINVAL, sysvmm_BLOCK_01, 2)
#endif /* _KERNEL */
#endif /* __64BIT_KERNEL */
/* RPN allocator structures and prototypes */
#ifdef _KERNEL
#define RPN_INVAL -1
#define RPN_DR_INVAL -2
#endif /* _KERNEL */
#ifdef __cplusplus
}
#endif
#ifdef _KERNSYS
#include <sys/processor.h>
#include <sys/lock_def.h>
#include <sys/syspest.h>
#define CACHELINESIZE 128
#ifdef __cplusplus
extern "C" {
#endif
struct xmfree_s {
struct xmfree_s *next;
char mem_type; /* type of memory */
};
struct rpn_header {
struct rpn_list *__rpn_next; /* next allocated list */
struct rpn_list *__rpn_prev; /* previous allocated list */
int __rpn_length; /* length of array */
cpu_t __rpn_cpu;
char __pad;
char __l2psize; /* log2 of rpn-list psize */
} rpn_header_t;
typedef struct rpn_lock {
Simple_lock lock;
/* pad to next cache line */
char fill[CACHELINESIZE - (sizeof(Simple_lock))];
} rpn_lock_t;
typedef struct rpn_list {
struct rpn_header __rpn_header;
rpn_t rpn_trans[1]; /* pre-translation of first page */
} rpn_list_t;
typedef struct rpn_anchor {
struct rpn_list list;
/* pad to next cache line */
char fill[CACHELINESIZE - (sizeof(struct rpn_list))];
} rpn_anchor_t;
#define rpn_next __rpn_header.__rpn_next
#define rpn_prev __rpn_header.__rpn_prev
#define rpn_length __rpn_header.__rpn_length
#define rpn_cpu __rpn_header.__rpn_cpu
/* Given a pointer to the start of a field in a rpn_list_t, determine the
* address of the start of the rpn_list_t containing the field.
*/
#define RPNFLD_TO_LISTP(_ptr, _fld) \
((rpn_list_t *)(((char *)(_ptr)) - ((char *)&(((rpn_list_t *)0)->_fld))))
#define RPN_CPU_LOCK(_cpunum) simple_lock(&rpn_list_locks[_cpunum].lock)
#define RPN_CPU_UNLOCK(_cpunum) simple_unlock(&rpn_list_locks[_cpunum].lock)
#define RPN_REG(_rpnl, _rpnl_length, _cpunum) { \
rpn_list_t *_rpnl_h = RPNFLD_TO_LISTP(_rpnl, rpn_trans); \
ASSERT(lock_mine(&rpn_list_locks[_cpunum].lock)); \
if ((_rpnl_h->rpn_next = rpn_anchor[_cpunum].list.rpn_next) != NULL) \
_rpnl_h->rpn_next->rpn_prev = _rpnl_h; \
rpn_anchor[_cpunum].list.rpn_next = _rpnl_h; \
_rpnl_h->rpn_prev = &rpn_anchor[_cpunum].list; \
_rpnl_h->rpn_length = _rpnl_length; \
_rpnl_h->rpn_cpu = _cpunum; \
}
#define RPN_UNREG(_rpnl) { \
rpn_list_t *_rpnl_h = RPNFLD_TO_LISTP(_rpnl, rpn_trans); \
cpu_t _cpunum = _rpnl_h->rpn_cpu; \
ASSERT(lock_mine(&rpn_list_locks[_cpunum].lock)); \
if ((_rpnl_h->rpn_prev->rpn_next = _rpnl_h->rpn_next) != NULL) \
_rpnl_h->rpn_next->rpn_prev = _rpnl_h->rpn_prev; \
}
extern rpn_anchor_t *rpn_anchor;
extern rpn_lock_t *rpn_list_locks;
void rpn_init(void);
void rpn_inval(rpn_t, rpn_t);
rpn_t *rpn_alloc(int, ushort);
void rpn_free(rpn_t *);
int rpn_free_interrupt(rpn_t *);
#ifdef __cplusplus
}
#endif
#endif /* _KERNSYS */
#endif /* _H_XMEM */