162306a36Sopenharmony_ci// SPDX-License-Identifier: GPL-2.0-or-later 262306a36Sopenharmony_ci 362306a36Sopenharmony_ci/* 462306a36Sopenharmony_ci * Copyright 2020 IBM Corp. 562306a36Sopenharmony_ci * 662306a36Sopenharmony_ci * Author: Bulent Abali <abali@us.ibm.com> 762306a36Sopenharmony_ci * 862306a36Sopenharmony_ci */ 962306a36Sopenharmony_ci#include <stdio.h> 1062306a36Sopenharmony_ci#include <stdlib.h> 1162306a36Sopenharmony_ci#include <string.h> 1262306a36Sopenharmony_ci#include <unistd.h> 1362306a36Sopenharmony_ci#include <stdint.h> 1462306a36Sopenharmony_ci#include <sys/types.h> 1562306a36Sopenharmony_ci#include <sys/stat.h> 1662306a36Sopenharmony_ci#include <sys/time.h> 1762306a36Sopenharmony_ci#include <sys/fcntl.h> 1862306a36Sopenharmony_ci#include <sys/mman.h> 1962306a36Sopenharmony_ci#include <endian.h> 2062306a36Sopenharmony_ci#include <bits/endian.h> 2162306a36Sopenharmony_ci#include <sys/ioctl.h> 2262306a36Sopenharmony_ci#include <assert.h> 2362306a36Sopenharmony_ci#include <errno.h> 2462306a36Sopenharmony_ci#include <signal.h> 2562306a36Sopenharmony_ci#include "vas-api.h" 2662306a36Sopenharmony_ci#include "nx.h" 2762306a36Sopenharmony_ci#include "copy-paste.h" 2862306a36Sopenharmony_ci#include "nxu.h" 2962306a36Sopenharmony_ci#include "nx_dbg.h" 3062306a36Sopenharmony_ci#include <sys/platform/ppc.h> 3162306a36Sopenharmony_ci 3262306a36Sopenharmony_ci#define barrier() 3362306a36Sopenharmony_ci#define hwsync() ({ asm volatile("sync" ::: "memory"); }) 3462306a36Sopenharmony_ci 3562306a36Sopenharmony_ci#ifndef NX_NO_CPU_PRI 3662306a36Sopenharmony_ci#define cpu_pri_default() ({ asm volatile ("or 2, 2, 2"); }) 3762306a36Sopenharmony_ci#define cpu_pri_low() ({ asm volatile ("or 31, 31, 31"); }) 3862306a36Sopenharmony_ci#else 3962306a36Sopenharmony_ci#define cpu_pri_default() 4062306a36Sopenharmony_ci#define cpu_pri_low() 4162306a36Sopenharmony_ci#endif 4262306a36Sopenharmony_ci 4362306a36Sopenharmony_civoid *nx_fault_storage_address; 4462306a36Sopenharmony_ci 4562306a36Sopenharmony_cistruct nx_handle { 4662306a36Sopenharmony_ci int fd; 4762306a36Sopenharmony_ci int function; 4862306a36Sopenharmony_ci void *paste_addr; 4962306a36Sopenharmony_ci}; 5062306a36Sopenharmony_ci 5162306a36Sopenharmony_cistatic int open_device_nodes(char *devname, int pri, struct nx_handle *handle) 5262306a36Sopenharmony_ci{ 5362306a36Sopenharmony_ci int rc, fd; 5462306a36Sopenharmony_ci void *addr; 5562306a36Sopenharmony_ci struct vas_tx_win_open_attr txattr; 5662306a36Sopenharmony_ci 5762306a36Sopenharmony_ci fd = open(devname, O_RDWR); 5862306a36Sopenharmony_ci if (fd < 0) { 5962306a36Sopenharmony_ci fprintf(stderr, " open device name %s\n", devname); 6062306a36Sopenharmony_ci return -errno; 6162306a36Sopenharmony_ci } 6262306a36Sopenharmony_ci 6362306a36Sopenharmony_ci memset(&txattr, 0, sizeof(txattr)); 6462306a36Sopenharmony_ci txattr.version = 1; 6562306a36Sopenharmony_ci txattr.vas_id = pri; 6662306a36Sopenharmony_ci rc = ioctl(fd, VAS_TX_WIN_OPEN, (unsigned long)&txattr); 6762306a36Sopenharmony_ci if (rc < 0) { 6862306a36Sopenharmony_ci fprintf(stderr, "ioctl() n %d, error %d\n", rc, errno); 6962306a36Sopenharmony_ci rc = -errno; 7062306a36Sopenharmony_ci goto out; 7162306a36Sopenharmony_ci } 7262306a36Sopenharmony_ci 7362306a36Sopenharmony_ci addr = mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0ULL); 7462306a36Sopenharmony_ci if (addr == MAP_FAILED) { 7562306a36Sopenharmony_ci fprintf(stderr, "mmap() failed, errno %d\n", errno); 7662306a36Sopenharmony_ci rc = -errno; 7762306a36Sopenharmony_ci goto out; 7862306a36Sopenharmony_ci } 7962306a36Sopenharmony_ci handle->fd = fd; 8062306a36Sopenharmony_ci handle->paste_addr = (void *)((char *)addr + 0x400); 8162306a36Sopenharmony_ci 8262306a36Sopenharmony_ci rc = 0; 8362306a36Sopenharmony_ciout: 8462306a36Sopenharmony_ci close(fd); 8562306a36Sopenharmony_ci return rc; 8662306a36Sopenharmony_ci} 8762306a36Sopenharmony_ci 8862306a36Sopenharmony_civoid *nx_function_begin(int function, int pri) 8962306a36Sopenharmony_ci{ 9062306a36Sopenharmony_ci int rc; 9162306a36Sopenharmony_ci char *devname = "/dev/crypto/nx-gzip"; 9262306a36Sopenharmony_ci struct nx_handle *nxhandle; 9362306a36Sopenharmony_ci 9462306a36Sopenharmony_ci if (function != NX_FUNC_COMP_GZIP) { 9562306a36Sopenharmony_ci errno = EINVAL; 9662306a36Sopenharmony_ci fprintf(stderr, " NX_FUNC_COMP_GZIP not found\n"); 9762306a36Sopenharmony_ci return NULL; 9862306a36Sopenharmony_ci } 9962306a36Sopenharmony_ci 10062306a36Sopenharmony_ci 10162306a36Sopenharmony_ci nxhandle = malloc(sizeof(*nxhandle)); 10262306a36Sopenharmony_ci if (!nxhandle) { 10362306a36Sopenharmony_ci errno = ENOMEM; 10462306a36Sopenharmony_ci fprintf(stderr, " No memory\n"); 10562306a36Sopenharmony_ci return NULL; 10662306a36Sopenharmony_ci } 10762306a36Sopenharmony_ci 10862306a36Sopenharmony_ci nxhandle->function = function; 10962306a36Sopenharmony_ci rc = open_device_nodes(devname, pri, nxhandle); 11062306a36Sopenharmony_ci if (rc < 0) { 11162306a36Sopenharmony_ci errno = -rc; 11262306a36Sopenharmony_ci fprintf(stderr, " open_device_nodes failed\n"); 11362306a36Sopenharmony_ci return NULL; 11462306a36Sopenharmony_ci } 11562306a36Sopenharmony_ci 11662306a36Sopenharmony_ci return nxhandle; 11762306a36Sopenharmony_ci} 11862306a36Sopenharmony_ci 11962306a36Sopenharmony_ciint nx_function_end(void *handle) 12062306a36Sopenharmony_ci{ 12162306a36Sopenharmony_ci int rc = 0; 12262306a36Sopenharmony_ci struct nx_handle *nxhandle = handle; 12362306a36Sopenharmony_ci 12462306a36Sopenharmony_ci rc = munmap(nxhandle->paste_addr - 0x400, 4096); 12562306a36Sopenharmony_ci if (rc < 0) { 12662306a36Sopenharmony_ci fprintf(stderr, "munmap() failed, errno %d\n", errno); 12762306a36Sopenharmony_ci return rc; 12862306a36Sopenharmony_ci } 12962306a36Sopenharmony_ci close(nxhandle->fd); 13062306a36Sopenharmony_ci free(nxhandle); 13162306a36Sopenharmony_ci 13262306a36Sopenharmony_ci return rc; 13362306a36Sopenharmony_ci} 13462306a36Sopenharmony_ci 13562306a36Sopenharmony_cistatic int nx_wait_for_csb(struct nx_gzip_crb_cpb_t *cmdp) 13662306a36Sopenharmony_ci{ 13762306a36Sopenharmony_ci long poll = 0; 13862306a36Sopenharmony_ci uint64_t t; 13962306a36Sopenharmony_ci 14062306a36Sopenharmony_ci /* Save power and let other threads use the h/w. top may show 14162306a36Sopenharmony_ci * 100% but only because OS doesn't know we slowed the this 14262306a36Sopenharmony_ci * h/w thread while polling. We're letting other threads have 14362306a36Sopenharmony_ci * higher throughput on the core. 14462306a36Sopenharmony_ci */ 14562306a36Sopenharmony_ci cpu_pri_low(); 14662306a36Sopenharmony_ci 14762306a36Sopenharmony_ci#define CSB_MAX_POLL 200000000UL 14862306a36Sopenharmony_ci#define USLEEP_TH 300000UL 14962306a36Sopenharmony_ci 15062306a36Sopenharmony_ci t = __ppc_get_timebase(); 15162306a36Sopenharmony_ci 15262306a36Sopenharmony_ci while (getnn(cmdp->crb.csb, csb_v) == 0) { 15362306a36Sopenharmony_ci ++poll; 15462306a36Sopenharmony_ci hwsync(); 15562306a36Sopenharmony_ci 15662306a36Sopenharmony_ci cpu_pri_low(); 15762306a36Sopenharmony_ci 15862306a36Sopenharmony_ci /* usleep(0) takes around 29000 ticks ~60 us. 15962306a36Sopenharmony_ci * 300000 is spinning for about 600 us then 16062306a36Sopenharmony_ci * start sleeping. 16162306a36Sopenharmony_ci */ 16262306a36Sopenharmony_ci if ((__ppc_get_timebase() - t) > USLEEP_TH) { 16362306a36Sopenharmony_ci cpu_pri_default(); 16462306a36Sopenharmony_ci usleep(1); 16562306a36Sopenharmony_ci } 16662306a36Sopenharmony_ci 16762306a36Sopenharmony_ci if (poll > CSB_MAX_POLL) 16862306a36Sopenharmony_ci break; 16962306a36Sopenharmony_ci 17062306a36Sopenharmony_ci /* Fault address from signal handler */ 17162306a36Sopenharmony_ci if (nx_fault_storage_address) { 17262306a36Sopenharmony_ci cpu_pri_default(); 17362306a36Sopenharmony_ci return -EAGAIN; 17462306a36Sopenharmony_ci } 17562306a36Sopenharmony_ci 17662306a36Sopenharmony_ci } 17762306a36Sopenharmony_ci 17862306a36Sopenharmony_ci cpu_pri_default(); 17962306a36Sopenharmony_ci 18062306a36Sopenharmony_ci /* hw has updated csb and output buffer */ 18162306a36Sopenharmony_ci hwsync(); 18262306a36Sopenharmony_ci 18362306a36Sopenharmony_ci /* Check CSB flags. */ 18462306a36Sopenharmony_ci if (getnn(cmdp->crb.csb, csb_v) == 0) { 18562306a36Sopenharmony_ci fprintf(stderr, "CSB still not valid after %d polls.\n", 18662306a36Sopenharmony_ci (int) poll); 18762306a36Sopenharmony_ci prt_err("CSB still not valid after %d polls, giving up.\n", 18862306a36Sopenharmony_ci (int) poll); 18962306a36Sopenharmony_ci return -ETIMEDOUT; 19062306a36Sopenharmony_ci } 19162306a36Sopenharmony_ci 19262306a36Sopenharmony_ci return 0; 19362306a36Sopenharmony_ci} 19462306a36Sopenharmony_ci 19562306a36Sopenharmony_cistatic int nxu_run_job(struct nx_gzip_crb_cpb_t *cmdp, void *handle) 19662306a36Sopenharmony_ci{ 19762306a36Sopenharmony_ci int i, ret, retries; 19862306a36Sopenharmony_ci struct nx_handle *nxhandle = handle; 19962306a36Sopenharmony_ci 20062306a36Sopenharmony_ci assert(handle != NULL); 20162306a36Sopenharmony_ci i = 0; 20262306a36Sopenharmony_ci retries = 5000; 20362306a36Sopenharmony_ci while (i++ < retries) { 20462306a36Sopenharmony_ci hwsync(); 20562306a36Sopenharmony_ci vas_copy(&cmdp->crb, 0); 20662306a36Sopenharmony_ci ret = vas_paste(nxhandle->paste_addr, 0); 20762306a36Sopenharmony_ci hwsync(); 20862306a36Sopenharmony_ci 20962306a36Sopenharmony_ci NXPRT(fprintf(stderr, "Paste attempt %d/%d returns 0x%x\n", 21062306a36Sopenharmony_ci i, retries, ret)); 21162306a36Sopenharmony_ci 21262306a36Sopenharmony_ci if ((ret == 2) || (ret == 3)) { 21362306a36Sopenharmony_ci 21462306a36Sopenharmony_ci ret = nx_wait_for_csb(cmdp); 21562306a36Sopenharmony_ci if (!ret) { 21662306a36Sopenharmony_ci goto out; 21762306a36Sopenharmony_ci } else if (ret == -EAGAIN) { 21862306a36Sopenharmony_ci long x; 21962306a36Sopenharmony_ci 22062306a36Sopenharmony_ci prt_err("Touching address %p, 0x%lx\n", 22162306a36Sopenharmony_ci nx_fault_storage_address, 22262306a36Sopenharmony_ci *(long *) nx_fault_storage_address); 22362306a36Sopenharmony_ci x = *(long *) nx_fault_storage_address; 22462306a36Sopenharmony_ci *(long *) nx_fault_storage_address = x; 22562306a36Sopenharmony_ci nx_fault_storage_address = 0; 22662306a36Sopenharmony_ci continue; 22762306a36Sopenharmony_ci } else { 22862306a36Sopenharmony_ci prt_err("wait_for_csb() returns %d\n", ret); 22962306a36Sopenharmony_ci break; 23062306a36Sopenharmony_ci } 23162306a36Sopenharmony_ci } else { 23262306a36Sopenharmony_ci if (i < 10) { 23362306a36Sopenharmony_ci /* spin for few ticks */ 23462306a36Sopenharmony_ci#define SPIN_TH 500UL 23562306a36Sopenharmony_ci uint64_t fail_spin; 23662306a36Sopenharmony_ci 23762306a36Sopenharmony_ci fail_spin = __ppc_get_timebase(); 23862306a36Sopenharmony_ci while ((__ppc_get_timebase() - fail_spin) < 23962306a36Sopenharmony_ci SPIN_TH) 24062306a36Sopenharmony_ci ; 24162306a36Sopenharmony_ci } else { 24262306a36Sopenharmony_ci /* sleep */ 24362306a36Sopenharmony_ci unsigned int pr = 0; 24462306a36Sopenharmony_ci 24562306a36Sopenharmony_ci if (pr++ % 100 == 0) { 24662306a36Sopenharmony_ci prt_err("Paste attempt %d/", i); 24762306a36Sopenharmony_ci prt_err("%d, failed pid= %d\n", retries, 24862306a36Sopenharmony_ci getpid()); 24962306a36Sopenharmony_ci } 25062306a36Sopenharmony_ci usleep(1); 25162306a36Sopenharmony_ci } 25262306a36Sopenharmony_ci continue; 25362306a36Sopenharmony_ci } 25462306a36Sopenharmony_ci } 25562306a36Sopenharmony_ci 25662306a36Sopenharmony_ciout: 25762306a36Sopenharmony_ci cpu_pri_default(); 25862306a36Sopenharmony_ci 25962306a36Sopenharmony_ci return ret; 26062306a36Sopenharmony_ci} 26162306a36Sopenharmony_ci 26262306a36Sopenharmony_ciint nxu_submit_job(struct nx_gzip_crb_cpb_t *cmdp, void *handle) 26362306a36Sopenharmony_ci{ 26462306a36Sopenharmony_ci int cc; 26562306a36Sopenharmony_ci 26662306a36Sopenharmony_ci cc = nxu_run_job(cmdp, handle); 26762306a36Sopenharmony_ci 26862306a36Sopenharmony_ci if (!cc) 26962306a36Sopenharmony_ci cc = getnn(cmdp->crb.csb, csb_cc); /* CC Table 6-8 */ 27062306a36Sopenharmony_ci 27162306a36Sopenharmony_ci return cc; 27262306a36Sopenharmony_ci} 27362306a36Sopenharmony_ci 27462306a36Sopenharmony_ci 27562306a36Sopenharmony_civoid nxu_sigsegv_handler(int sig, siginfo_t *info, void *ctx) 27662306a36Sopenharmony_ci{ 27762306a36Sopenharmony_ci fprintf(stderr, "%d: Got signal %d si_code %d, si_addr %p\n", getpid(), 27862306a36Sopenharmony_ci sig, info->si_code, info->si_addr); 27962306a36Sopenharmony_ci 28062306a36Sopenharmony_ci nx_fault_storage_address = info->si_addr; 28162306a36Sopenharmony_ci} 28262306a36Sopenharmony_ci 28362306a36Sopenharmony_ci/* 28462306a36Sopenharmony_ci * Fault in pages prior to NX job submission. wr=1 may be required to 28562306a36Sopenharmony_ci * touch writeable pages. System zero pages do not fault-in the page as 28662306a36Sopenharmony_ci * intended. Typically set wr=1 for NX target pages and set wr=0 for NX 28762306a36Sopenharmony_ci * source pages. 28862306a36Sopenharmony_ci */ 28962306a36Sopenharmony_ciint nxu_touch_pages(void *buf, long buf_len, long page_len, int wr) 29062306a36Sopenharmony_ci{ 29162306a36Sopenharmony_ci char *begin = buf; 29262306a36Sopenharmony_ci char *end = (char *) buf + buf_len - 1; 29362306a36Sopenharmony_ci volatile char t; 29462306a36Sopenharmony_ci 29562306a36Sopenharmony_ci assert(buf_len >= 0 && !!buf); 29662306a36Sopenharmony_ci 29762306a36Sopenharmony_ci NXPRT(fprintf(stderr, "touch %p %p len 0x%lx wr=%d\n", buf, 29862306a36Sopenharmony_ci (buf + buf_len), buf_len, wr)); 29962306a36Sopenharmony_ci 30062306a36Sopenharmony_ci if (buf_len <= 0 || buf == NULL) 30162306a36Sopenharmony_ci return -1; 30262306a36Sopenharmony_ci 30362306a36Sopenharmony_ci do { 30462306a36Sopenharmony_ci t = *begin; 30562306a36Sopenharmony_ci if (wr) 30662306a36Sopenharmony_ci *begin = t; 30762306a36Sopenharmony_ci begin = begin + page_len; 30862306a36Sopenharmony_ci } while (begin < end); 30962306a36Sopenharmony_ci 31062306a36Sopenharmony_ci /* When buf_sz is small or buf tail is in another page */ 31162306a36Sopenharmony_ci t = *end; 31262306a36Sopenharmony_ci if (wr) 31362306a36Sopenharmony_ci *end = t; 31462306a36Sopenharmony_ci 31562306a36Sopenharmony_ci return 0; 31662306a36Sopenharmony_ci} 317