@@ -424,6 +424,7 @@ void xive_router_end_notify(XiveRouter *xrtr, XiveEAS *eas);
typedef struct XiveTCTXMatch {
XiveTCTX *tctx;
uint8_t ring;
+ bool precluded;
} XiveTCTXMatch;
#define TYPE_XIVE_PRESENTER "xive-presenter"
@@ -452,7 +453,9 @@ int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
uint8_t nvt_blk, uint32_t nvt_idx,
bool cam_ignore, uint8_t priority,
- uint32_t logic_serv);
+ uint32_t logic_serv, bool *precluded);
+
+uint32_t xive_get_vpgroup_size(uint32_t nvp_index);
/*
* XIVE Fabric (Interface between Interrupt Controller and Machine)
@@ -121,6 +121,7 @@ uint64_t xive2_tm_pull_os_ctx(XivePresenter *xptr, XiveTCTX *tctx,
hwaddr offset, unsigned size);
void xive2_tm_pull_os_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
hwaddr offset, uint64_t value, unsigned size);
+bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority);
void xive2_tm_set_hv_target(XivePresenter *xptr, XiveTCTX *tctx,
hwaddr offset, uint64_t value, unsigned size);
void xive2_tm_pull_phys_ctx_ol(XivePresenter *xptr, XiveTCTX *tctx,
@@ -660,21 +660,34 @@ static int pnv_xive2_match_nvt(XivePresenter *xptr, uint8_t format,
logic_serv);
}
- /*
- * Save the context and follow on to catch duplicates,
- * that we don't support yet.
- */
if (ring != -1) {
- if (match->tctx) {
+ /*
+ * For VP-specific match, finding more than one is a
+ * problem. For group notification, it's possible.
+ */
+ if (!cam_ignore && match->tctx) {
qemu_log_mask(LOG_GUEST_ERROR, "XIVE: already found a "
"thread context NVT %x/%x\n",
nvt_blk, nvt_idx);
- return false;
+ /* Should set a FIR if we ever model it */
+ return -1;
+ }
+ /*
+ * For a group notification, we need to know if the
+ * match is precluded first by checking the current
+ * thread priority. If the interrupt can be delivered,
+ * we always notify the first match (for now).
+ */
+ if (cam_ignore &&
+ xive2_tm_irq_precluded(tctx, ring, priority)) {
+ match->precluded = true;
+ } else {
+ if (!match->tctx) {
+ match->ring = ring;
+ match->tctx = tctx;
+ }
+ count++;
}
-
- match->ring = ring;
- match->tctx = tctx;
- count++;
}
}
}
@@ -1671,6 +1671,16 @@ static uint32_t xive_tctx_hw_cam_line(XivePresenter *xptr, XiveTCTX *tctx)
return xive_nvt_cam_line(blk, 1 << 7 | (pir & 0x7f));
}
+uint32_t xive_get_vpgroup_size(uint32_t nvp_index)
+{
+ /*
+ * Group size is a power of 2. The position of the first 0
+ * (starting with the least significant bits) in the NVP index
+ * gives the size of the group.
+ */
+ return 1 << (ctz32(~nvp_index) + 1);
+}
+
static uint8_t xive_get_group_level(uint32_t nvp_index)
{
/* FIXME add crowd encoding */
@@ -1743,30 +1753,39 @@ int xive_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
/*
* This is our simple Xive Presenter Engine model. It is merged in the
* Router as it does not require an extra object.
- *
- * It receives notification requests sent by the IVRE to find one
- * matching NVT (or more) dispatched on the processor threads. In case
- * of a single NVT notification, the process is abbreviated and the
- * thread is signaled if a match is found. In case of a logical server
- * notification (bits ignored at the end of the NVT identifier), the
- * IVPE and IVRE select a winning thread using different filters. This
- * involves 2 or 3 exchanges on the PowerBus that the model does not
- * support.
- *
- * The parameters represent what is sent on the PowerBus
*/
bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
uint8_t nvt_blk, uint32_t nvt_idx,
bool cam_ignore, uint8_t priority,
- uint32_t logic_serv)
+ uint32_t logic_serv, bool *precluded)
{
XiveFabricClass *xfc = XIVE_FABRIC_GET_CLASS(xfb);
- XiveTCTXMatch match = { .tctx = NULL, .ring = 0 };
+ XiveTCTXMatch match = { .tctx = NULL, .ring = 0, .precluded = false };
uint8_t group_level;
int count;
/*
- * Ask the machine to scan the interrupt controllers for a match
+ * Ask the machine to scan the interrupt controllers for a match.
+ *
+ * For VP-specific notification, we expect at most one match and
+ * one call to the presenters is all we need (abbreviated notify
+ * sequence documented by the architecture).
+ *
+ * For VP-group notification, match_nvt() is the equivalent of the
+ * "histogram" and "poll" commands sent to the power bus to the
+ * presenters. 'count' could be more than one, but we always
+ * select the first match for now. 'precluded' tells if (at least)
+ * one thread matches but can't take the interrupt now because
+ * it's running at a more favored priority. We return the
+ * information to the router so that it can take appropriate
+ * actions (backlog, escalation, broadcast, etc...)
+ *
+ * If we were to implement a better way of dispatching the
+ * interrupt in case of multiple matches (instead of the first
+ * match), we would need a heuristic to elect a thread (for
+ * example, the hardware keeps track of an 'age' in the TIMA) and
+ * a new command to the presenters (the equivalent of the "assign"
+ * power bus command in the documented full notify sequence.
*/
count = xfc->match_nvt(xfb, format, nvt_blk, nvt_idx, cam_ignore,
priority, logic_serv, &match);
@@ -1779,6 +1798,8 @@ bool xive_presenter_notify(XiveFabric *xfb, uint8_t format,
group_level = cam_ignore ? xive_get_group_level(nvt_idx) : 0;
trace_xive_presenter_notify(nvt_blk, nvt_idx, match.ring, group_level);
xive_tctx_pipr_update(match.tctx, match.ring, priority, group_level);
+ } else {
+ *precluded = match.precluded;
}
return !!count;
@@ -1818,7 +1839,7 @@ void xive_router_end_notify(XiveRouter *xrtr, XiveEAS *eas)
uint8_t nvt_blk;
uint32_t nvt_idx;
XiveNVT nvt;
- bool found;
+ bool found, precluded;
uint8_t end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
uint32_t end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
@@ -1901,8 +1922,9 @@ void xive_router_end_notify(XiveRouter *xrtr, XiveEAS *eas)
found = xive_presenter_notify(xrtr->xfb, format, nvt_blk, nvt_idx,
xive_get_field32(END_W7_F0_IGNORE, end.w7),
priority,
- xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7));
-
+ xive_get_field32(END_W7_F1_LOG_SERVER_ID, end.w7),
+ &precluded);
+ /* we don't support VP-group notification on P9, so precluded is not used */
/* TODO: Auto EOI. */
if (found) {
@@ -739,6 +739,12 @@ int xive2_router_write_nvgc(Xive2Router *xrtr, bool crowd,
return xrc->write_nvgc(xrtr, crowd, nvgc_blk, nvgc_idx, nvgc);
}
+static bool xive2_vp_match_mask(uint32_t cam1, uint32_t cam2,
+ uint32_t vp_mask)
+{
+ return (cam1 & vp_mask) == (cam2 & vp_mask);
+}
+
/*
* The thread context register words are in big-endian format.
*/
@@ -753,44 +759,50 @@ int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
uint32_t qw1w2 = xive_tctx_word2(&tctx->regs[TM_QW1_OS]);
uint32_t qw0w2 = xive_tctx_word2(&tctx->regs[TM_QW0_USER]);
- /*
- * TODO (PowerNV): ignore mode. The low order bits of the NVT
- * identifier are ignored in the "CAM" match.
- */
+ uint32_t vp_mask = 0xFFFFFFFF;
if (format == 0) {
- if (cam_ignore == true) {
- /*
- * F=0 & i=1: Logical server notification (bits ignored at
- * the end of the NVT identifier)
- */
- qemu_log_mask(LOG_UNIMP, "XIVE: no support for LS NVT %x/%x\n",
- nvt_blk, nvt_idx);
- return -1;
+ /*
+ * i=0: Specific NVT notification
+ * i=1: VP-group notification (bits ignored at the end of the
+ * NVT identifier)
+ */
+ if (cam_ignore) {
+ vp_mask = ~(xive_get_vpgroup_size(nvt_idx) - 1);
}
- /* F=0 & i=0: Specific NVT notification */
+ /* For VP-group notifications, threads with LGS=0 are excluded */
/* PHYS ring */
if ((be32_to_cpu(qw3w2) & TM2_QW3W2_VT) &&
- cam == xive2_tctx_hw_cam_line(xptr, tctx)) {
+ !(cam_ignore && tctx->regs[TM_QW3_HV_PHYS + TM_LGS] == 0) &&
+ xive2_vp_match_mask(cam,
+ xive2_tctx_hw_cam_line(xptr, tctx),
+ vp_mask)) {
return TM_QW3_HV_PHYS;
}
/* HV POOL ring */
if ((be32_to_cpu(qw2w2) & TM2_QW2W2_VP) &&
- cam == xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2)) {
+ !(cam_ignore && tctx->regs[TM_QW2_HV_POOL + TM_LGS] == 0) &&
+ xive2_vp_match_mask(cam,
+ xive_get_field32(TM2_QW2W2_POOL_CAM, qw2w2),
+ vp_mask)) {
return TM_QW2_HV_POOL;
}
/* OS ring */
if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
- cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) {
+ !(cam_ignore && tctx->regs[TM_QW1_OS + TM_LGS] == 0) &&
+ xive2_vp_match_mask(cam,
+ xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2),
+ vp_mask)) {
return TM_QW1_OS;
}
} else {
/* F=1 : User level Event-Based Branch (EBB) notification */
+ /* FIXME: what if cam_ignore and LGS = 0 ? */
/* USER ring */
if ((be32_to_cpu(qw1w2) & TM2_QW1W2_VO) &&
(cam == xive_get_field32(TM2_QW1W2_OS_CAM, qw1w2)) &&
@@ -802,6 +814,22 @@ int xive2_presenter_tctx_match(XivePresenter *xptr, XiveTCTX *tctx,
return -1;
}
+bool xive2_tm_irq_precluded(XiveTCTX *tctx, int ring, uint8_t priority)
+{
+ uint8_t *regs = &tctx->regs[ring];
+
+ /*
+ * The xive2_presenter_tctx_match() above tells if there's a match
+ * but for VP-group notification, we still need to look at the
+ * priority to know if the thread can take the interrupt now or if
+ * it is precluded.
+ */
+ if (priority < regs[TM_CPPR]) {
+ return false;
+ }
+ return true;
+}
+
static void xive2_router_realize(DeviceState *dev, Error **errp)
{
Xive2Router *xrtr = XIVE2_ROUTER(dev);
@@ -841,7 +869,7 @@ static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
Xive2End end;
uint8_t priority;
uint8_t format;
- bool found;
+ bool found, precluded;
Xive2Nvp nvp;
uint8_t nvp_blk;
uint32_t nvp_idx;
@@ -922,7 +950,8 @@ static void xive2_router_end_notify(Xive2Router *xrtr, uint8_t end_blk,
found = xive_presenter_notify(xrtr->xfb, format, nvp_blk, nvp_idx,
xive2_end_is_ignore(&end),
priority,
- xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7));
+ xive_get_field32(END2_W7_F1_LOG_SERVER_ID, end.w7),
+ &precluded);
/* TODO: Auto EOI. */