| Message ID | 1429720465-2551-1-git-send-email-shilpa.bhat@linux.vnet.ibm.com |
|---|---|
| State | Superseded |
| Headers | show |
Hi Shilpa, On 04/22/2015 10:04 PM, Shilpasri G Bhat wrote: > Add a new class of message definition OPAL_MSG_OCC to > opal_message_type to notify the following OCC events to host: > 1) OCC Reset > 2) OCC Load > 3) OCC Throttle Status Change > > Add an opal poller to periodically read throttle status updated by OCC > for each chip and notify any change in throttle status to host. The > throttle status indicates the reason why OCC may have limited the max > Pstate of the chip. > > Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com> > --- > hw/occ.c | 85 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ > include/chip.h | 1 + > include/opal-api.h | 7 +++++ > 3 files changed, 93 insertions(+) > > diff --git a/hw/occ.c b/hw/occ.c > index 34d6de5..d3711b2 100644 > --- a/hw/occ.c > +++ b/hw/occ.c > @@ -24,6 +24,8 @@ > #include <timebase.h> > #include <hostservices.h> > #include <errorlog.h> > +#include <opal-api.h> > +#include <opal-msg.h> > > /* OCC Communication Area for PStates */ > > @@ -31,6 +33,16 @@ > > #define MAX_PSTATES 256 > > +#define OCC_RESET 0 > +#define OCC_LOAD 1 > +#define OCC_THROTTLE 2 > + > +#define chip_occ_data(chip) \ > + ((struct occ_pstate_table *)(chip->homer_base + \ > + P8_HOMER_SAPPHIRE_DATA_OFFSET)) > + > +static bool occ_reset; > + > struct occ_pstate_entry { > s8 id; > u8 flags; > @@ -302,6 +314,58 @@ static bool cpu_pstates_prepare_core(struct proc_chip *chip, struct cpu_thread * > return true; > } > > +/* occ_throttle_poll: This function will queue a meassage of type > + * OPAL_MSG_OCC to notify any change in the throttle status of the > + * chip. Throttle status indicates the reason why OCC may have limited > + * the max Pstate of the chip. > + * 0x00 = No throttle > + * 0x01 = Power Cap > + * 0x02 = Processor Over Temperature > + * 0x03 = Power Supply Failure (currently not used) > + * 0x04 = Over current (currently not used) > + * 0x05 = OCC Reset (not reliable as some failures will not allow for > + * OCC to update throttle status, so use 'occ_reset') > + */ > +static void occ_throttle_poll(void *data __unused) > +{ > + struct proc_chip *chip; > + struct occ_pstate_table *occ_data; > + > + if (occ_reset) { > + int inactive = 0; > + > + for_each_chip(chip) { > + occ_data = chip_occ_data(chip); > + if (occ_data->valid != 1) { > + inactive = 1; > + break; > + } else if (!occ_data->throttle) { > + chip->prev_throttle = occ_data->throttle; Why do you do this ? ^^^^ Why cant we initialize prev_throttle to 0x00 - No throttle at boot and on each OCC reset ? I did not see the point of introducing another value to initialize prev_throttle. > + } > + } > + if (!inactive) { > + occ_reset = false; > + /* > + * Queue OCC_THROTTLE with throttle status as 0 to > + * indicate all OCCs are active after a reset. > + */ > + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, > + OCC_THROTTLE, 0, 0); Note that opal_queue_msg() is expected to fail. So we should handle it. Refer to the other call sites to get an idea. Should we do something like cancel the poller if the queuing of messages fail ? Regards Preeti U Murthy
Hi Preeti, On 04/23/2015 04:40 PM, Preeti U Murthy wrote: > Hi Shilpa, > > On 04/22/2015 10:04 PM, Shilpasri G Bhat wrote: >> + chip->prev_throttle = occ_data->throttle; > > Why do you do this ? ^^^^ > > Why cant we initialize prev_throttle to 0x00 - No throttle at boot and > on each OCC reset ? I did not see the point of introducing another value > to initialize prev_throttle. Yes agree. I can initialize prev_throttle to '0' and skip the above assignment. > >> + } >> + } >> + if (!inactive) { >> + occ_reset = false; >> + /* >> + * Queue OCC_THROTTLE with throttle status as 0 to >> + * indicate all OCCs are active after a reset. >> + */ >> + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, >> + OCC_THROTTLE, 0, 0); > > Note that opal_queue_msg() is expected to fail. So we should handle it. > Refer to the other call sites to get an idea. Should we do something > like cancel the poller if the queuing of messages fail ? I think adding a pr_log would suffice if any of the opal_queue_msg fails except throttle as we are just reporting the status of OCC and not taking any action based on the event. For throttle/unthrottle message we can skip the prev_throttle/occ_reset update if we fail to queue the message so that we can retry queuing it the next time occ_throttle_poll() is called, Thanks and Regards, Shilpa
On 04/23/2015 05:28 PM, Shilpasri G Bhat wrote: > Hi Preeti, > > On 04/23/2015 04:40 PM, Preeti U Murthy wrote: >> Hi Shilpa, >> >> On 04/22/2015 10:04 PM, Shilpasri G Bhat wrote: > >>> + chip->prev_throttle = occ_data->throttle; >> >> Why do you do this ? ^^^^ >> >> Why cant we initialize prev_throttle to 0x00 - No throttle at boot and >> on each OCC reset ? I did not see the point of introducing another value >> to initialize prev_throttle. > > Yes agree. I can initialize prev_throttle to '0' and skip the above assignment. >> >>> + } >>> + } >>> + if (!inactive) { >>> + occ_reset = false; >>> + /* >>> + * Queue OCC_THROTTLE with throttle status as 0 to >>> + * indicate all OCCs are active after a reset. >>> + */ >>> + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, >>> + OCC_THROTTLE, 0, 0); >> >> Note that opal_queue_msg() is expected to fail. So we should handle it. >> Refer to the other call sites to get an idea. Should we do something >> like cancel the poller if the queuing of messages fail ? > > I think adding a pr_log would suffice if any of the opal_queue_msg fails except > throttle as we are just reporting the status of OCC and not taking any action > based on the event. For throttle/unthrottle message we can skip the > prev_throttle/occ_reset update if we fail to queue the message so that we can > retry queuing it the next time occ_throttle_poll() is called, This sounds good. Regards Preeti U Murthy > > Thanks and Regards, > Shilpa > > >
diff --git a/hw/occ.c b/hw/occ.c index 34d6de5..d3711b2 100644 --- a/hw/occ.c +++ b/hw/occ.c @@ -24,6 +24,8 @@ #include <timebase.h> #include <hostservices.h> #include <errorlog.h> +#include <opal-api.h> +#include <opal-msg.h> /* OCC Communication Area for PStates */ @@ -31,6 +33,16 @@ #define MAX_PSTATES 256 +#define OCC_RESET 0 +#define OCC_LOAD 1 +#define OCC_THROTTLE 2 + +#define chip_occ_data(chip) \ + ((struct occ_pstate_table *)(chip->homer_base + \ + P8_HOMER_SAPPHIRE_DATA_OFFSET)) + +static bool occ_reset; + struct occ_pstate_entry { s8 id; u8 flags; @@ -302,6 +314,58 @@ static bool cpu_pstates_prepare_core(struct proc_chip *chip, struct cpu_thread * return true; } +/* occ_throttle_poll: This function will queue a meassage of type + * OPAL_MSG_OCC to notify any change in the throttle status of the + * chip. Throttle status indicates the reason why OCC may have limited + * the max Pstate of the chip. + * 0x00 = No throttle + * 0x01 = Power Cap + * 0x02 = Processor Over Temperature + * 0x03 = Power Supply Failure (currently not used) + * 0x04 = Over current (currently not used) + * 0x05 = OCC Reset (not reliable as some failures will not allow for + * OCC to update throttle status, so use 'occ_reset') + */ +static void occ_throttle_poll(void *data __unused) +{ + struct proc_chip *chip; + struct occ_pstate_table *occ_data; + + if (occ_reset) { + int inactive = 0; + + for_each_chip(chip) { + occ_data = chip_occ_data(chip); + if (occ_data->valid != 1) { + inactive = 1; + break; + } else if (!occ_data->throttle) { + chip->prev_throttle = occ_data->throttle; + } + } + if (!inactive) { + occ_reset = false; + /* + * Queue OCC_THROTTLE with throttle status as 0 to + * indicate all OCCs are active after a reset. + */ + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, + OCC_THROTTLE, 0, 0); + } + } else { + for_each_chip(chip) { + occ_data = chip_occ_data(chip); + if (occ_data->valid == 1 && + (chip->prev_throttle != occ_data->throttle)) { + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, + OCC_THROTTLE, chip->id, + occ_data->throttle); + chip->prev_throttle = occ_data->throttle; + } + } + } +} + /* CPU-OCC PState init */ /* Called after OCC init on P8 */ void occ_pstates_init(void) @@ -345,6 +409,11 @@ void occ_pstates_init(void) cpu_pstates_prepare_core(chip, c, pstate_nom); } } + + /* Add opal_poller to poll OCC throttle status of each chip */ + for_each_chip(chip) + chip->prev_throttle = 0xF; + opal_add_poller(occ_throttle_poll, NULL); } struct occ_load_req { @@ -386,6 +455,7 @@ static void __occ_do_load(u8 scope, u32 dbob_id __unused, u32 seq_id) prlog(PR_INFO, "OCC: Load: Fallback to preloaded image\n"); rc = 0; } else if (!rc) { + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, OCC_LOAD); /* Success, start OCC */ rc = host_services_occ_start(); } @@ -509,6 +579,21 @@ static void occ_do_reset(u8 scope, u32 dbob_id, u32 seq_id) rc = 0; } if (!rc) { + opal_queue_msg(OPAL_MSG_OCC, NULL, NULL, OCC_RESET); + /* + * Set 'valid' byte of chip_occ_data to 0 since OCC + * may not clear this byte on a reset. + * OCC will set the 'valid' byte to 1 when it becomes + * active again. + */ + for_each_chip(chip) { + struct occ_pstate_table *occ_data; + + occ_data = chip_occ_data(chip); + occ_data->valid = 0; + chip->prev_throttle = 0xF; + } + occ_reset = true; /* Send a single success response for all chips */ stat = fsp_mkmsg(FSP_CMD_RESET_OCC_STAT, 2, 0, seq_id); if (stat) diff --git a/include/chip.h b/include/chip.h index 0547902..340fdfc 100644 --- a/include/chip.h +++ b/include/chip.h @@ -147,6 +147,7 @@ struct proc_chip { uint64_t homer_size; uint64_t occ_common_base; uint64_t occ_common_size; + u8 prev_throttle; /* Must hold capi_lock to change */ u8 capp_phb3_attached_mask; diff --git a/include/opal-api.h b/include/opal-api.h index 1698311..abe798e 100644 --- a/include/opal-api.h +++ b/include/opal-api.h @@ -417,6 +417,13 @@ enum opal_msg_type { OPAL_MSG_HMI_EVT, OPAL_MSG_DPO, OPAL_MSG_PRD, + OPAL_MSG_OCC, /* + * params[0] = 0 reset, + * 1 load, + * 2 throttle + * params[1] = chip_id, + * params[2] = throttle_status + */ OPAL_MSG_TYPE_MAX, };
Add a new class of message definition OPAL_MSG_OCC to opal_message_type to notify the following OCC events to host: 1) OCC Reset 2) OCC Load 3) OCC Throttle Status Change Add an opal poller to periodically read throttle status updated by OCC for each chip and notify any change in throttle status to host. The throttle status indicates the reason why OCC may have limited the max Pstate of the chip. Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com> --- hw/occ.c | 85 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ include/chip.h | 1 + include/opal-api.h | 7 +++++ 3 files changed, 93 insertions(+)