static void find_capp_checkstop_reason(int flat_chip_id,
struct OpalHMIEvent *hmi_evt,
uint64_t *out_flags)
{
struct capp_info info;
struct phb *phb;
uint64_t capp_fir;
uint64_t capp_fir_mask;
uint64_t capp_fir_action0;
uint64_t capp_fir_action1;
uint64_t reg;
int64_t rc;
/* Find the CAPP on the chip associated with the HMI. */
for_each_phb(phb) {
/* get the CAPP info */
rc = capp_get_info(flat_chip_id, phb, &info);
if (rc == OPAL_PARAMETER)
continue;
if (xscom_read(flat_chip_id, info.capp_fir_reg, &capp_fir) ||
xscom_read(flat_chip_id, info.capp_fir_mask_reg,
&capp_fir_mask) ||
xscom_read(flat_chip_id, info.capp_fir_action0_reg,
&capp_fir_action0) ||
xscom_read(flat_chip_id, info.capp_fir_action1_reg,
&capp_fir_action1)) {
prerror("CAPP: Couldn't read CAPP#%d (PHB:#%x) FIR registers by XSCOM!\n",
info.capp_index, info.phb_index);
continue;
}
if (!(capp_fir & ~capp_fir_mask))
continue;
prlog(PR_DEBUG, "CAPP#%d (PHB:#%x): FIR 0x%016llx mask 0x%016llx\n",
info.capp_index, info.phb_index, capp_fir,
capp_fir_mask);
prlog(PR_DEBUG, "CAPP#%d (PHB:#%x): ACTION0 0x%016llx, ACTION1 0x%016llx\n",
info.capp_index, info.phb_index, capp_fir_action0,
capp_fir_action1);
/*
* If this bit is set (=1) a Recoverable Error has been
* detected
*/
xscom_read(flat_chip_id, info.capp_err_status_ctrl_reg, ®);
if ((reg & PPC_BIT(0)) != 0) {
phb_lock(phb);
phb->ops->set_capp_recovery(phb);
phb_unlock(phb);
hmi_evt->severity = OpalHMI_SEV_NO_ERROR;
hmi_evt->type = OpalHMI_ERROR_CAPP_RECOVERY;
queue_hmi_event(hmi_evt, 1, out_flags);
return;
}
}
}
static void show_all_regs(struct npu2 *npu, int brick_index)
{
int i, stack, stack_min, stack_max;
uint64_t fir_val, mask_val, fir_addr, mask_addr;
struct npu2_dev *dev;
npu2_scom_dump_t scom_reg;
if (brick_index != -1) {
stack_min = stack_max = NPU2_STACK_STCK_0 + brick_index / 2;
} else {
stack_min = NPU2_STACK_STCK_0;
stack_max = NPU2_STACK_STCK_2;
/* Avoid dumping unused stacks for opencapi on Lagrange */
if (npu->total_devices == 2)
stack_min = stack_max = NPU2_STACK_STCK_1;
}
/* NPU FIRs */
for (i = 0; i < NPU2_TOTAL_FIR_REGISTERS; i++) {
fir_addr = NPU2_FIR_REGISTER_0 + i * NPU2_FIR_OFFSET;
mask_addr = fir_addr + NPU2_FIR_MASK_OFFSET;
xscom_read(npu->chip_id, fir_addr, &fir_val);
xscom_read(npu->chip_id, mask_addr, &mask_val);
prlog(PR_ERR, "NPU[%d] FIR%d = 0x%016llx (mask 0x%016llx => 0x%016llx)\n",
npu->chip_id, i, fir_val, mask_val, fir_val & ~mask_val);
}
/* NPU global, per-stack registers */
for (i = 0; i < ARRAY_SIZE(npu2_scom_dump_global); i++) {
for (stack = stack_min; stack <= stack_max; stack++)
print_one_npu_reg(npu, &npu2_scom_dump_global[i], stack);
}
/*
* NPU global registers, stack independent
* We have only one for now, so dump it directly
*/
scom_reg.name = "XTS.REG.ERR_HOLD";
scom_reg.block = NPU2_BLOCK_XTS;
scom_reg.offset = 0;
print_one_npu_reg(npu, &scom_reg, NPU2_STACK_MISC);
/* nvlink- or opencapi-specific registers */
for (i = 0; i < npu->total_devices; i++) {
dev = &npu->devices[i];
if (brick_index == -1 || dev->brick_index == brick_index) {
if (dev->type == NPU2_DEV_TYPE_NVLINK)
show_nvlink_regs(npu, dev->brick_index);
else if (dev->type == NPU2_DEV_TYPE_OPENCAPI)
show_opencapi_regs(npu, dev->brick_index);
}
}
}
void p8_sbe_init_timer(void)
{
struct dt_node *np;
int64_t rc;
uint32_t tick_us;
np = dt_find_compatible_node(dt_root, NULL, "ibm,power8-sbe-timer");
if (!np)
return;
sbe_timer_chip = dt_get_chip_id(np);
tick_us = dt_prop_get_u32(np, "tick-time-us");
sbe_timer_inc = usecs_to_tb(tick_us);
sbe_timer_target = ~0ull;
rc = xscom_read(sbe_timer_chip, 0xE0006, &sbe_last_gen);
if (rc) {
prerror("SLW: Error %lld reading tmr gen count\n", rc);
return;
}
sbe_last_gen_stamp = mftb();
prlog(PR_INFO, "SLW: Timer facility on chip %d, resolution %dus\n",
sbe_timer_chip, tick_us);
sbe_has_timer = true;
}
static int64_t opb_write(struct proc_chip *chip, uint32_t addr, uint32_t data,
uint32_t sz)
{
uint64_t ctl = ECCB_CTL_MAGIC, stat;
int64_t rc, tout;
uint64_t data_reg;
switch(sz) {
case 1:
data_reg = ((uint64_t)data) << 56;
break;
case 2:
data_reg = ((uint64_t)data) << 48;
break;
case 4:
data_reg = ((uint64_t)data) << 32;
break;
default:
prerror("LPC: Invalid data size %d\n", sz);
return OPAL_PARAMETER;
}
rc = xscom_write(chip->id, chip->lpc_xbase + ECCB_DATA, data_reg);
if (rc) {
log_simple_error(&e_info(OPAL_RC_LPC_WRITE),
"LPC: XSCOM write to ECCB DATA error %lld\n", rc);
return rc;
}
ctl = SETFIELD(ECCB_CTL_DATASZ, ctl, sz);
ctl = SETFIELD(ECCB_CTL_ADDRLEN, ctl, ECCB_ADDRLEN_4B);
ctl = SETFIELD(ECCB_CTL_ADDR, ctl, addr);
rc = xscom_write(chip->id, chip->lpc_xbase + ECCB_CTL, ctl);
if (rc) {
log_simple_error(&e_info(OPAL_RC_LPC_WRITE),
"LPC: XSCOM write to ECCB CTL error %lld\n", rc);
return rc;
}
for (tout = 0; tout < ECCB_TIMEOUT; tout++) {
rc = xscom_read(chip->id, chip->lpc_xbase + ECCB_STAT, &stat);
if (rc) {
log_simple_error(&e_info(OPAL_RC_LPC_WRITE),
"LPC: XSCOM read from ECCB STAT err %lld\n",
rc);
return rc;
}
if (stat & ECCB_STAT_OP_DONE) {
if (stat & ECCB_STAT_ERR_MASK) {
log_simple_error(&e_info(OPAL_RC_LPC_WRITE),
"LPC: Error status: 0x%llx\n", stat);
return OPAL_HARDWARE;
}
return OPAL_SUCCESS;
}
time_wait(100);
}
log_simple_error(&e_info(OPAL_RC_LPC_WRITE), "LPC: Write timeout !\n");
return OPAL_HARDWARE;
}
static int read_core_fir(uint32_t chip_id, uint32_t core_id, uint64_t *core_fir)
{
int rc;
switch (proc_gen) {
case proc_gen_p8:
rc = xscom_read(chip_id,
XSCOM_ADDR_P8_EX(core_id, P8_CORE_FIR), core_fir);
break;
case proc_gen_p9:
rc = xscom_read(chip_id,
XSCOM_ADDR_P9_EC(core_id, P9_CORE_FIR), core_fir);
break;
default:
rc = OPAL_HARDWARE;
}
return rc;
}
/* same as above, but for direct access registers */
static void print_one_reg(int chip_id, int brick_index,
uint64_t reg_addr, const char *reg_name)
{
uint64_t val;
xscom_read(chip_id, reg_addr, &val);
prlog(PR_ERR, "NPU[%d] %s brick %d 0x%llx = 0x%016llx\n",
chip_id, reg_name, brick_index, reg_addr, val);
}
static uint64_t __unused dl_read(struct npu_dev *npu_dev, uint32_t addr)
{
uint64_t val;
xscom_write(npu_dev->npu->chip_id,
npu_dev->xscom + NX_DL_REG_ADDR, addr);
xscom_read(npu_dev->npu->chip_id,
npu_dev->xscom + NX_DL_REG_DATA, &val);
return val;
}
uint64_t npu2_scom_read(uint64_t gcid, uint64_t scom_base,
uint64_t reg, uint64_t size)
{
uint64_t val;
npu2_scom_set_addr(gcid, scom_base, reg, size);
xscom_read(gcid, scom_base + NPU2_MISC_SCOM_IND_SCOM_DATA, &val);
return val;
}
static int64_t opb_read(struct lpcm *lpc, uint32_t addr, uint32_t *data,
uint32_t sz)
{
uint64_t ctl = ECCB_CTL_MAGIC | ECCB_CTL_READ, stat;
int64_t rc, tout;
if (lpc->mbase)
return opb_mmio_read(lpc, addr, data, sz);
if (sz != 1 && sz != 2 && sz != 4) {
prerror("Invalid data size %d\n", sz);
return OPAL_PARAMETER;
}
ctl = SETFIELD(ECCB_CTL_DATASZ, ctl, sz);
ctl = SETFIELD(ECCB_CTL_ADDRLEN, ctl, ECCB_ADDRLEN_4B);
ctl = SETFIELD(ECCB_CTL_ADDR, ctl, addr);
rc = xscom_write(lpc->chip_id, lpc->xbase + ECCB_CTL, ctl);
if (rc) {
log_simple_error(&e_info(OPAL_RC_LPC_READ),
"LPC: XSCOM write to ECCB CTL error %lld\n", rc);
return rc;
}
for (tout = 0; tout < ECCB_TIMEOUT; tout++) {
rc = xscom_read(lpc->chip_id, lpc->xbase + ECCB_STAT,
&stat);
if (rc) {
log_simple_error(&e_info(OPAL_RC_LPC_READ),
"LPC: XSCOM read from ECCB STAT err %lld\n",
rc);
return rc;
}
if (stat & ECCB_STAT_OP_DONE) {
uint32_t rdata = GETFIELD(ECCB_STAT_RD_DATA, stat);
if (stat & ECCB_STAT_ERR_MASK) {
log_simple_error(&e_info(OPAL_RC_LPC_READ),
"LPC: Error status: 0x%llx\n", stat);
return OPAL_HARDWARE;
}
switch(sz) {
case 1:
*data = rdata >> 24;
break;
case 2:
*data = rdata >> 16;
break;
default:
*data = rdata;
break;
}
return 0;
}
time_wait_nopoll(100);
}
static uint16_t phy_read(struct npu_dev *npu_dev, uint64_t addr)
{
uint64_t val;
if (pl_use_scom)
xscom_read(npu_dev->npu->chip_id, npu_dev->pl_xscom_base + addr, &val);
else
val = in_be16((void *) npu_dev->pl_base + PL_MMIO_ADDR(addr));
return val & 0xffff;
}
static int read_core_wof(uint32_t chip_id, uint32_t core_id, uint64_t *core_wof)
{
int rc;
switch (proc_gen) {
case proc_gen_p9:
rc = xscom_read(chip_id,
XSCOM_ADDR_P9_EC(core_id, P9_CORE_WOF), core_wof);
break;
default:
rc = OPAL_HARDWARE;
}
return rc;
}
static bool read_pba_bar(struct proc_chip *chip, unsigned int bar_no,
uint64_t *base, uint64_t *size)
{
uint64_t bar, mask;
int rc;
rc = xscom_read(chip->id, pba_bar0 + bar_no, &bar);
if (rc) {
prerror("SLW: Error %d reading PBA BAR%d on chip %d\n",
rc, bar_no, chip->id);
return false;
}
rc = xscom_read(chip->id, pba_barmask0 + bar_no, &mask);
if (rc) {
prerror("SLW: Error %d reading PBA BAR MASK%d on chip %d\n",
rc, bar_no, chip->id);
return false;
}
prlog(PR_DEBUG, " PBA BAR%d : 0x%016llx\n", bar_no, bar);
prlog(PR_DEBUG, " PBA MASK%d: 0x%016llx\n", bar_no, mask);
if (mask == PBA_MASK_ALL_BITS) {
/*
* This could happen if all HOMER users are not enabled during
* early system bringup. Skip using the PBA BAR.
*/
mask = 0;
bar = 0;
prerror(" PBA MASK%d uninitalized skipping BAR\n", bar_no);
}
*base = bar & 0x0ffffffffffffffful;
*size = (mask | 0xfffff) + 1;
return (*base) != 0;
}
void fsi_recovery()
{
int32_t rc = SUCCESS;
/* Clear out OPB error */
uint64_t scom_data = 0;
scom_data = 0x8000000000000000; /*0=Unit Reset*/
rc |= xscom_write( OPB_REG_RES, scom_data );
rc |= xscom_write( OPB_REG_STAT, scom_data );
/* Check if we have any errors left */
rc |= xscom_read( OPB_REG_STAT, &scom_data );
TRACFCOMP( "PIB2OPB Status after cleanup = %08X%08X (rc=%d)",
(uint32_t)(scom_data >> 32), (uint32_t)scom_data, rc );
}
static int __ipoll_update_mask(uint32_t proc, bool set, uint64_t bits)
{
uint64_t mask;
int rc;
rc = xscom_read(proc, PRD_IPOLL_REG_MASK, &mask);
if (rc)
return rc;
if (set)
mask |= bits;
else
mask &= ~bits;
return xscom_write(proc, PRD_IPOLL_REG_MASK, mask);
}
static int ipoll_record_and_mask_pending(uint32_t proc)
{
uint64_t status;
int rc;
lock(&ipoll_lock);
rc = xscom_read(proc, PRD_IPOLL_REG_STATUS, &status);
if (!rc)
__ipoll_update_mask(proc, true, status);
unlock(&ipoll_lock);
if (!rc)
ipoll_status[proc] |= (status & PRD_IPOLL_MASK);
return rc;
}
static void decode_malfunction(struct OpalHMIEvent *hmi_evt, uint64_t *out_flags)
{
int i;
uint64_t malf_alert, flags;
flags = 0;
if (!setup_scom_addresses()) {
prerror("Failed to setup scom addresses\n");
/* Send an unknown HMI event. */
hmi_evt->u.xstop_error.xstop_type = CHECKSTOP_TYPE_UNKNOWN;
hmi_evt->u.xstop_error.xstop_reason = 0;
queue_hmi_event(hmi_evt, false, out_flags);
return;
}
xscom_read(this_cpu()->chip_id, malf_alert_scom, &malf_alert);
if (!malf_alert)
return;
for (i = 0; i < 64; i++) {
if (malf_alert & PPC_BIT(i)) {
xscom_write(this_cpu()->chip_id, malf_alert_scom,
~PPC_BIT(i));
find_capp_checkstop_reason(i, hmi_evt, &flags);
find_nx_checkstop_reason(i, hmi_evt, &flags);
find_npu_checkstop_reason(i, hmi_evt, &flags);
}
}
find_core_checkstop_reason(hmi_evt, &flags);
/*
* If we fail to find checkstop reason, send an unknown HMI event.
*/
if (!(flags & OPAL_HMI_FLAGS_NEW_EVENT)) {
hmi_evt->u.xstop_error.xstop_type = CHECKSTOP_TYPE_UNKNOWN;
hmi_evt->u.xstop_error.xstop_reason = 0;
queue_hmi_event(hmi_evt, false, &flags);
}
*out_flags |= flags;
}
static void print_chip_info(uint32_t chip_id)
{
uint64_t f000f, cfam_id;
const char *name;
char uname_buf[64];
int rc;
rc = xscom_read(chip_id, 0xf000f, &f000f);
if (rc)
return;
cfam_id = f000f >> 44;
switch(cfam_id & 0xff) {
case 0xf9:
name = "P7 processor";
break;
case 0xe8:
name = "P7+ processor";
break;
case 0xef:
name = "P8E (Murano) processor";
break;
case 0xea:
name = "P8 (Venice) processor";
break;
case 0xd3:
name = "P8NVL (Naples) processor";
break;
case 0xe9:
name = "Centaur memory buffer";
break;
default:
snprintf(uname_buf, sizeof(uname_buf), "Unknown ID 0x%02lx",
cfam_id & 0xff);
name = uname_buf;
}
printf("%08x | DD%lx.%lx | %s\n",
chip_id, (cfam_id >> 16) & 0xf, (cfam_id >> 8) & 0xf, name);
}
static int populate_ipoll_msg(struct opal_prd_msg *msg, uint32_t proc)
{
uint64_t ipoll_mask;
int rc;
lock(&ipoll_lock);
rc = xscom_read(proc, PRD_IPOLL_REG_MASK, &ipoll_mask);
unlock(&ipoll_lock);
if (rc) {
prlog(PR_ERR, "PRD: Unable to read ipoll status (chip %d)!\n",
proc);
return -1;
}
msg->attn.proc = proc;
msg->attn.ipoll_status = ipoll_status[proc];
msg->attn.ipoll_mask = ipoll_mask;
return 0;
}
static int nx_cfg_umac_status_ctrl(u32 gcid, u64 xcfg)
{
u64 uctrl;
int rc;
#define CRB_ENABLE 1
rc = xscom_read(gcid, xcfg, &uctrl);
if (rc)
return rc;
uctrl = SETFIELD(NX_P9_UMAC_STATUS_CTRL_CRB_ENABLE, uctrl, CRB_ENABLE);
rc = xscom_write(gcid, xcfg, uctrl);
if (rc)
prerror("NX%d: ERROR: Setting UMAC Status Control failure %d\n",
gcid, rc);
else
prlog(PR_DEBUG, "NX%d: Setting UMAC Status Control 0x%016lx\n",
gcid, (unsigned long)uctrl);
return rc;
}
static void p8_sbe_dump_timer_ffdc(void)
{
uint64_t i, val;
int64_t rc;
static const uint32_t dump_regs[] = {
0xe0000, 0xe0001, 0xe0002, 0xe0003,
0xe0004, 0xe0005, 0xe0006, 0xe0007,
0xe0008, 0xe0009, 0xe000a, 0xe000b,
0xe000c, 0xe000d, 0xe000e, 0xe000f,
0xe0010, 0xe0011, 0xe0012, 0xe0013,
0xe0014, 0xe0015, 0xe0016, 0xe0017,
0xe0018, 0xe0019,
0x5001c,
0x50038, 0x50039, 0x5003a, 0x5003b
};
/**
* @fwts-label SLWRegisterDump
* @fwts-advice An error condition occurred in sleep/winkle
* engines timer state machine. Dumping debug information to
* root-cause. OPAL/skiboot may be stuck on some operation that
* requires SLW timer state machine (e.g. core powersaving)
*/
prlog(PR_DEBUG, "SLW: Register state:\n");
for (i = 0; i < ARRAY_SIZE(dump_regs); i++) {
uint32_t reg = dump_regs[i];
rc = xscom_read(sbe_timer_chip, reg, &val);
if (rc) {
prlog(PR_DEBUG, "SLW: XSCOM error %lld reading"
" reg 0x%x\n", rc, reg);
break;
}
prlog(PR_DEBUG, "SLW: %5x = %016llx\n", reg, val);
}
}
/**
* @brief Poll for completion of a FSI operation, return data on read
*/
int32_t poll_for_complete( uint32_t * o_val )
{
int32_t rc = SUCCESS;
enum { MAX_OPB_TIMEOUT_NS = 10*NS_PER_MSEC }; /*=10ms */
*o_val = 0;
uint64_t read_data = 0;
uint64_t elapsed_time_ns = 0;
do
{
rc = xscom_read( OPB_REG_STAT, &read_data );
if ( SUCCESS != rc )
{
fsi_recovery(); /* Try to recover the engine. */
return rc;
}
/* Check for completion. Note: not checking for FSI errors. */
if ( (read_data & OPB_STAT_BUSY) == 0 ) break; /* Not busy */
sleep( 10000 ); /* sleep for 10,000 ns */
elapsed_time_ns += 10000;
} while ( elapsed_time_ns <= MAX_OPB_TIMEOUT_NS );
if ( MAX_OPB_TIMEOUT_NS < elapsed_time_ns )
{
TRAC_ERR( "[poll_for_complete] FSI request timed out." );
return FAIL;
}
*o_val = (uint32_t)read_data; /* Data in the bottom half. */
return rc;
}
int main(int argc, char *argv[])
{
uint64_t val, addr = -1ull;
uint32_t def_chip, chip_id = 0xffffffff;
bool list_chips = false;
bool no_work = false;
int rc;
while(1) {
static struct option long_opts[] = {
{"chip", required_argument, NULL, 'c'},
{"list-chips", no_argument, NULL, 'l'},
{"help", no_argument, NULL, 'h'},
{"version", no_argument, NULL, 'v'},
};
int c, oidx = 0;
c = getopt_long(argc, argv, "-c:hlv", long_opts, &oidx);
if (c == EOF)
break;
switch(c) {
case 1:
addr = strtoull(optarg, NULL, 16);
break;
case 'c':
chip_id = strtoul(optarg, NULL, 0);
break;
case 'h':
print_usage(0);
break;
case 'l':
list_chips = true;
break;
case 'v':
printf("xscom utils version %s\n", version);
exit(0);
default:
exit(1);
}
}
if (addr == -1ull)
no_work = true;
if (no_work && !list_chips) {
fprintf(stderr, "Invalid or missing address\n");
print_usage(1);
}
def_chip = xscom_init();
if (def_chip == 0xffffffff) {
fprintf(stderr, "No valid XSCOM chip found\n");
exit(1);
}
if (list_chips) {
printf("Chip ID | Rev | Chip type\n");
printf("---------|-------|--------\n");
xscom_for_each_chip(print_chip_info);
}
if (no_work)
return 0;
if (chip_id == 0xffffffff)
chip_id = def_chip;
rc = xscom_read(chip_id, addr, &val);
if (rc) {
fprintf(stderr,"Error %d reading XSCOM\n", rc);
exit(1);
}
printf("%016" PRIx64 "\n", val);
return 0;
}
static void find_nx_checkstop_reason(int flat_chip_id,
struct OpalHMIEvent *hmi_evt,
uint64_t *out_flags)
{
uint64_t nx_status;
uint64_t nx_dma_fir;
uint64_t nx_pbi_fir_val;
int i;
/* Get NX status register value. */
if (xscom_read(flat_chip_id, nx_status_reg, &nx_status) != 0) {
prerror("XSCOM error reading NX_STATUS_REG\n");
return;
}
/* Check if NX has driven an HMI interrupt. */
if (!(nx_status & NX_HMI_ACTIVE))
return;
/* Initialize HMI event */
hmi_evt->severity = OpalHMI_SEV_FATAL;
hmi_evt->type = OpalHMI_ERROR_MALFUNC_ALERT;
hmi_evt->u.xstop_error.xstop_type = CHECKSTOP_TYPE_NX;
hmi_evt->u.xstop_error.u.chip_id = flat_chip_id;
/* Get DMA & Engine FIR data register value. */
if (xscom_read(flat_chip_id, nx_dma_engine_fir, &nx_dma_fir) != 0) {
prerror("XSCOM error reading NX_DMA_ENGINE_FIR\n");
return;
}
/* Get PowerBus Interface FIR data register value. */
if (xscom_read(flat_chip_id, nx_pbi_fir, &nx_pbi_fir_val) != 0) {
prerror("XSCOM error reading NX_PBI_FIR\n");
return;
}
/* Find NX checkstop reason and populate HMI event with error info. */
for (i = 0; i < ARRAY_SIZE(nx_dma_xstop_bits); i++)
if (nx_dma_fir & PPC_BIT(nx_dma_xstop_bits[i].bit))
hmi_evt->u.xstop_error.xstop_reason
|= nx_dma_xstop_bits[i].reason;
for (i = 0; i < ARRAY_SIZE(nx_pbi_xstop_bits); i++)
if (nx_pbi_fir_val & PPC_BIT(nx_pbi_xstop_bits[i].bit))
hmi_evt->u.xstop_error.xstop_reason
|= nx_pbi_xstop_bits[i].reason;
/*
* Set NXDMAENGFIR[38] to signal PRD that service action is required.
* Without this inject, PRD will not be able to do NX unit checkstop
* error analysis. NXDMAENGFIR[38] is a spare bit and used to report
* a software initiated attention.
*
* The behavior of this bit and all FIR bits are documented in
* RAS spreadsheet.
*/
xscom_write(flat_chip_id, nx_dma_engine_fir, PPC_BIT(38));
/* Send an HMI event. */
queue_hmi_event(hmi_evt, 0, out_flags);
}
static void find_npu2_checkstop_reason(int flat_chip_id,
struct OpalHMIEvent *hmi_evt,
uint64_t *out_flags)
{
struct phb *phb;
int i;
bool npu2_hmi_verbose = false, found = false;
uint64_t npu2_fir;
uint64_t npu2_fir_mask;
uint64_t npu2_fir_action0;
uint64_t npu2_fir_action1;
uint64_t npu2_fir_addr;
uint64_t npu2_fir_mask_addr;
uint64_t npu2_fir_action0_addr;
uint64_t npu2_fir_action1_addr;
uint64_t fatal_errors;
int total_errors = 0;
const char *loc;
/* Find the NPU on the chip associated with the HMI. */
for_each_phb(phb) {
/* NOTE: if a chip ever has >1 NPU this will need adjusting */
if (phb_is_npu2(phb->dt_node) &&
(dt_get_chip_id(phb->dt_node) == flat_chip_id)) {
found = true;
break;
}
}
/* If we didn't find a NPU on the chip, it's not our checkstop. */
if (!found)
return;
npu2_fir_addr = NPU2_FIR_REGISTER_0;
npu2_fir_mask_addr = NPU2_FIR_REGISTER_0 + NPU2_FIR_MASK_OFFSET;
npu2_fir_action0_addr = NPU2_FIR_REGISTER_0 + NPU2_FIR_ACTION0_OFFSET;
npu2_fir_action1_addr = NPU2_FIR_REGISTER_0 + NPU2_FIR_ACTION1_OFFSET;
for (i = 0; i < NPU2_TOTAL_FIR_REGISTERS; i++) {
/* Read all the registers necessary to find a checkstop condition. */
if (xscom_read(flat_chip_id, npu2_fir_addr, &npu2_fir) ||
xscom_read(flat_chip_id, npu2_fir_mask_addr, &npu2_fir_mask) ||
xscom_read(flat_chip_id, npu2_fir_action0_addr, &npu2_fir_action0) ||
xscom_read(flat_chip_id, npu2_fir_action1_addr, &npu2_fir_action1)) {
prerror("HMI: Couldn't read NPU FIR register%d with XSCOM\n", i);
continue;
}
fatal_errors = npu2_fir & ~npu2_fir_mask & npu2_fir_action0 & npu2_fir_action1;
if (fatal_errors) {
loc = chip_loc_code(flat_chip_id);
if (!loc)
loc = "Not Available";
prlog(PR_ERR, "NPU: [Loc: %s] P:%d FIR#%d FIR 0x%016llx mask 0x%016llx\n",
loc, flat_chip_id, i, npu2_fir, npu2_fir_mask);
prlog(PR_ERR, "NPU: [Loc: %s] P:%d ACTION0 0x%016llx, ACTION1 0x%016llx\n",
loc, flat_chip_id, npu2_fir_action0, npu2_fir_action1);
total_errors++;
}
/* Can't do a fence yet, we are just logging fir information for now */
npu2_fir_addr += NPU2_FIR_OFFSET;
npu2_fir_mask_addr += NPU2_FIR_OFFSET;
npu2_fir_action0_addr += NPU2_FIR_OFFSET;
npu2_fir_action1_addr += NPU2_FIR_OFFSET;
}
if (!total_errors)
return;
npu2_hmi_verbose = nvram_query_eq_safe("npu2-hmi-verbose", "true");
/* Force this for now until we sort out something better */
npu2_hmi_verbose = true;
if (npu2_hmi_verbose) {
npu2_dump_scoms(flat_chip_id);
prlog(PR_ERR, " _________________________ \n");
prlog(PR_ERR, "< It's Debug time! >\n");
prlog(PR_ERR, " ------------------------- \n");
prlog(PR_ERR, " \\ ,__, \n");
prlog(PR_ERR, " \\ (oo)____ \n");
prlog(PR_ERR, " (__) )\\ \n");
prlog(PR_ERR, " ||--|| * \n");
}
/* Set up the HMI event */
hmi_evt->severity = OpalHMI_SEV_WARNING;
hmi_evt->type = OpalHMI_ERROR_MALFUNC_ALERT;
hmi_evt->u.xstop_error.xstop_type = CHECKSTOP_TYPE_NPU;
hmi_evt->u.xstop_error.u.chip_id = flat_chip_id;
/* Marking the event as recoverable so that we don't crash */
queue_hmi_event(hmi_evt, 1, out_flags);
}
int nx_cfg_rx_fifo(struct dt_node *node, const char *compat,
const char *priority, u32 gcid, u32 pid, u32 tid,
u64 umac_bar, u64 umac_notify)
{
u64 cfg;
int rc, size;
uint64_t fifo;
u32 lpid = 0xfff; /* All 1's for 12 bits in UMAC notify match reg */
#define MATCH_ENABLE 1
fifo = (uint64_t) local_alloc(gcid, RX_FIFO_SIZE, RX_FIFO_SIZE);
assert(fifo);
/*
* When configuring the address of the Rx FIFO into the Receive FIFO
* BAR, we should _NOT_ shift the address into bits 8:53. Instead we
* should copy the address as is and VAS/NX will extract relevant bits.
*/
/*
* Section 5.21 of P9 NX Workbook Version 2.42 shows Receive FIFO BAR
* 54:56 represents FIFO size
* 000 = 1KB, 8 CRBs
* 001 = 2KB, 16 CRBs
* 010 = 4KB, 32 CRBs
* 011 = 8KB, 64 CRBs
* 100 = 16KB, 128 CRBs
* 101 = 32KB, 256 CRBs
* 110 = 111 reserved
*/
size = RX_FIFO_SIZE / 1024;
cfg = SETFIELD(NX_P9_RX_FIFO_BAR_SIZE, fifo, ilog2(size));
rc = xscom_write(gcid, umac_bar, cfg);
if (rc) {
prerror("NX%d: ERROR: Setting UMAC FIFO bar failure %d\n",
gcid, rc);
return rc;
} else
prlog(PR_DEBUG, "NX%d: Setting UMAC FIFO bar 0x%016lx\n",
gcid, (unsigned long)cfg);
rc = xscom_read(gcid, umac_notify, &cfg);
if (rc)
return rc;
/*
* VAS issues asb_notify with the unique ID to identify the target
* co-processor/engine. Logical partition ID (lpid), process ID (pid),
* and thread ID (tid) combination is used to define the unique ID
* in the system. Export these values in device-tree such that the
* driver configure RxFIFO with VAS. Set these values in RxFIFO notify
* match register for each engine which compares the ID with each
* request.
* To define unique indentification, 0xfff (1's for 12 bits),
* co-processor type, and counter within coprocessor type are used
* for lpid, pid, and tid respectively.
*/
cfg = SETFIELD(NX_P9_RX_FIFO_NOTIFY_MATCH_LPID, cfg, lpid);
cfg = SETFIELD(NX_P9_RX_FIFO_NOTIFY_MATCH_PID, cfg, pid);
cfg = SETFIELD(NX_P9_RX_FIFO_NOTIFY_MATCH_TID, cfg, tid);
cfg = SETFIELD(NX_P9_RX_FIFO_NOTIFY_MATCH_MATCH_ENABLE, cfg,
MATCH_ENABLE);
rc = xscom_write(gcid, umac_notify, cfg);
if (rc) {
prerror("NX%d: ERROR: Setting UMAC notify match failure %d\n",
gcid, rc);
return rc;
} else
prlog(PR_DEBUG, "NX%d: Setting UMAC notify match 0x%016lx\n",
gcid, (unsigned long)cfg);
dt_add_property_string(node, "compatible", compat);
dt_add_property_string(node, "priority", priority);
dt_add_property_u64(node, "rx-fifo-address", fifo);
dt_add_property_cells(node, "rx-fifo-size", RX_FIFO_SIZE);
dt_add_property_cells(node, "lpid", lpid);
dt_add_property_cells(node, "pid", pid);
dt_add_property_cells(node, "tid", tid);
return 0;
}
static void find_npu_checkstop_reason(int flat_chip_id,
struct OpalHMIEvent *hmi_evt,
uint64_t *out_flags)
{
struct phb *phb;
struct npu *p = NULL;
uint64_t npu_fir;
uint64_t npu_fir_mask;
uint64_t npu_fir_action0;
uint64_t npu_fir_action1;
uint64_t fatal_errors;
/* Only check for NPU errors if the chip has a NPU */
if (PVR_TYPE(mfspr(SPR_PVR)) != PVR_TYPE_P8NVL)
return find_npu2_checkstop_reason(flat_chip_id, hmi_evt, out_flags);
/* Find the NPU on the chip associated with the HMI. */
for_each_phb(phb) {
/* NOTE: if a chip ever has >1 NPU this will need adjusting */
if (dt_node_is_compatible(phb->dt_node, "ibm,power8-npu-pciex") &&
(dt_get_chip_id(phb->dt_node) == flat_chip_id)) {
p = phb_to_npu(phb);
break;
}
}
/* If we didn't find a NPU on the chip, it's not our checkstop. */
if (p == NULL)
return;
/* Read all the registers necessary to find a checkstop condition. */
if (xscom_read(flat_chip_id,
p->at_xscom + NX_FIR, &npu_fir) ||
xscom_read(flat_chip_id,
p->at_xscom + NX_FIR_MASK, &npu_fir_mask) ||
xscom_read(flat_chip_id,
p->at_xscom + NX_FIR_ACTION0, &npu_fir_action0) ||
xscom_read(flat_chip_id,
p->at_xscom + NX_FIR_ACTION1, &npu_fir_action1)) {
prerror("Couldn't read NPU registers with XSCOM\n");
return;
}
fatal_errors = npu_fir & ~npu_fir_mask & npu_fir_action0 & npu_fir_action1;
/* If there's no errors, we don't need to do anything. */
if (!fatal_errors)
return;
prlog(PR_DEBUG, "NPU: FIR 0x%016llx mask 0x%016llx\n",
npu_fir, npu_fir_mask);
prlog(PR_DEBUG, "NPU: ACTION0 0x%016llx, ACTION1 0x%016llx\n",
npu_fir_action0, npu_fir_action1);
/* Set the NPU to fenced since it can't recover. */
npu_set_fence_state(p, true);
/* Set up the HMI event */
hmi_evt->severity = OpalHMI_SEV_WARNING;
hmi_evt->type = OpalHMI_ERROR_MALFUNC_ALERT;
hmi_evt->u.xstop_error.xstop_type = CHECKSTOP_TYPE_NPU;
hmi_evt->u.xstop_error.u.chip_id = flat_chip_id;
/* The HMI is "recoverable" because it shouldn't crash the system */
queue_hmi_event(hmi_evt, 1, out_flags);
}
