mem-ruby: Fix MOESI_CMP_directory blocked line handling
Using recycle in the L2 controllers to put messages back into the buffer
may lead to starvation when there are many L1 requests for the same line.
This can easily trigger the deadlock detection mechanism in configurations
with many cores (16+). Replacing recycle by stall_and_wait for L1
requests avoids this issue. wakeUpBuffers calls were added to all
transitions from transient to stable states.
Change-Id: I28b8aeacc48919ccf38e69653cd9205a4153514b
Signed-off-by: Tiago Muck <tiago.muck@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/17568
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
Reviewed-by: Matthew Poremba <matthew.poremba@amd.com>
Maintainer: Jason Lowe-Power <jason@lowepower.com>
Tested-by: kokoro <noreply+kokoro@google.com>
diff --git a/src/mem/protocol/MOESI_CMP_directory-L2cache.sm b/src/mem/protocol/MOESI_CMP_directory-L2cache.sm
index 0c00bd9..6252219 100644
--- a/src/mem/protocol/MOESI_CMP_directory-L2cache.sm
+++ b/src/mem/protocol/MOESI_CMP_directory-L2cache.sm
@@ -234,6 +234,7 @@
void set_tbe(TBE b);
void unset_tbe();
MachineID mapAddressToMachine(Addr addr, MachineType mtype);
+ void wakeUpAllBuffers(Addr a);
Entry getCacheEntry(Addr address), return_by_pointer="yes" {
return static_cast(Entry, "pointer", L2cache[address]);
@@ -1537,13 +1538,6 @@
localDirectory.deallocate(address);
}
- action(zz_recycleL1RequestQueue, "zz", desc="Send the head of the mandatory queue to the back of the queue.") {
- peek(L1requestNetwork_in, RequestMsg) {
- APPEND_TRANSITION_COMMENT(in_msg.Requestor);
- }
- L1requestNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency));
- }
-
action(zz_recycleRequestQueue, "\zz", desc="Send the head of the mandatory queue to the back of the queue.") {
peek(requestNetwork_in, RequestMsg) {
APPEND_TRANSITION_COMMENT(in_msg.Requestor);
@@ -1558,6 +1552,14 @@
responseNetwork_in.recycle(clockEdge(), cyclesToTicks(recycle_latency));
}
+ action(st_stallAndWaitL1RequestQueue, "st", desc="Stall and wait on the address") {
+ stall_and_wait(L1requestNetwork_in, address);
+ }
+
+ action(wa_wakeUpDependents, "wa", desc="Wake up any requests waiting for this address") {
+ wakeUpAllBuffers(address);
+ }
+
action(da_sendDmaAckUnblock, "da", desc="Send dma ack to global directory") {
enqueue(responseNetwork_out, ResponseMsg, response_latency) {
out_msg.addr := address;
@@ -1576,11 +1578,11 @@
//*****************************************************
transition({II, IFGX, IFGS, ISFGS, IFGXX, IFLXO, OFGX, ILOW, ILOXW, ILOSW, ILOSXW, SLSW, OLSW, ILSW, IW, OW, SW, OXW, OLSXW, ILXW, IFLS, IFLO, IFLOX, IFLOXX, IFLOSX, OLSXS, IGS, IGM, IGMLS, IGMO, IGMIO, OGMIO, IGMIOF, OGMIOF, MM, SS, OO, OI, MI, MII, OLSI, ILSI, SLSS, OLSS, OLSF, IGMIOFS, ILOSD, ILOSXD, ILOD, ILXD, ILOXD}, {L1_PUTO, L1_PUTS, L1_PUTS_only, L1_PUTX}) {
- zz_recycleL1RequestQueue;
+ st_stallAndWaitL1RequestQueue;
}
transition({II, IFGX, IFGS, ISFGS, IFGXX, IFLXO, OFGX, ILOW, ILOXW, ILOSW, ILOSXW, SLSW, OLSW, ILSW, IW, OW, SW, OXW, OLSXW, ILXW, IFLS, IFLO, IFLOX, IFLOXX, IFLOSX, OLSXS, IGS, IGM, IGMLS, IGMO, IGMIO, OGMIO, IGMIOF, OGMIOF, MM, SS, OO, OI, MI, MII, OLSI, ILSI, SLSS, OLSS, OLSF, IGMIOFS, ILOSD, ILOSXD, ILOD, ILXD, ILOXD}, {L1_GETX, L1_GETS}) {
- zz_recycleL1RequestQueue;
+ st_stallAndWaitL1RequestQueue;
}
transition({IFGX, IFGS, ISFGS, IFGXX, IFLXO, OFGX, ILOW, ILOXW, ILOSW, ILOSXW, SLSW, OLSW, ILSW, IW, ILXW, OW, SW, OXW, OLSXW, IFLS, IFLO, IFLOX, IFLOXX, IFLOSX,OLSXS, IGS, IGM, IGMLS, IGMO, MM, SS, OO, OI, MI, MII, OLSI, ILSI, SLSS, OLSS, OLSF, IGMIOFS, ILOSD, ILOSXD, ILOD, ILXD, ILOXD}, L2_Replacement) {
@@ -1674,6 +1676,7 @@
s_deallocateTBE;
da_sendDmaAckUnblock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOSX, Fwd_DMA, ILOSXD) {
@@ -1687,6 +1690,7 @@
s_deallocateTBE;
da_sendDmaAckUnblock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILO, Fwd_DMA, ILOD) {
@@ -1700,6 +1704,7 @@
s_deallocateTBE;
da_sendDmaAckUnblock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILX, Fwd_DMA, ILXD) {
@@ -1713,6 +1718,7 @@
s_deallocateTBE;
da_sendDmaAckUnblock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOX, Fwd_DMA, ILOXD) {
@@ -1726,6 +1732,7 @@
s_deallocateTBE;
da_sendDmaAckUnblock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition({ILOS, ILOSX, ILO, ILX, ILOX, ILXW}, Data) {
@@ -1740,6 +1747,7 @@
c_sendDataFromTBEToFwdGETS;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ISFGS, Data, ILOS) {
@@ -1747,6 +1755,7 @@
c_sendDataFromTBEToFwdGETS;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IFGS, Data_Exclusive, I) {
@@ -1755,6 +1764,7 @@
gg_clearLocalSharers;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -1771,6 +1781,7 @@
gg_clearLocalSharers;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition({ILOSX, ILOS}, Fwd_GETX, IFGXX) {
@@ -1801,6 +1812,7 @@
gg_clearLocalSharers;
s_deallocateTBE;
n_popTriggerQueue;
+ wa_wakeUpDependents;
}
@@ -1879,6 +1891,7 @@
s_deallocateTBE;
rr_deallocateL2CacheBlock;
n_popTriggerQueue;
+ wa_wakeUpDependents;
}
@@ -1927,6 +1940,7 @@
e_sendAck;
s_deallocateTBE;
n_popTriggerQueue;
+ wa_wakeUpDependents;
}
transition(S, Inv, I) {
@@ -1960,6 +1974,7 @@
g_recordLocalExclusive;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OLSX, L1_GETS, OLSXS) {
@@ -1972,6 +1987,7 @@
transition(OLSXS, Unblock, OLSX) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
// after this, can't get Fwd_GETX
@@ -2023,31 +2039,37 @@
transition(IFLOX, Unblock, ILOSX) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IFLS, Unblock, ILS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IFLOXX, Unblock, ILOSX) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IFLOSX, Unblock, ILOSX) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition({IFLOSX, IFLOXX}, Exclusive_Unblock, ILX) {
g_recordLocalExclusive;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IFLO, Unblock, ILOS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -2066,6 +2088,7 @@
g_recordLocalExclusive;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
// LOCAL REQUESTS THAT MUST ISSUE
@@ -2316,6 +2339,7 @@
f_sendUnblock;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IGS, Exclusive_Unblock, ILX) {
@@ -2323,6 +2347,7 @@
f_sendExclusiveUnblock;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IGMO, All_Acks) {
@@ -2335,6 +2360,7 @@
f_sendExclusiveUnblock;
s_deallocateTBE;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -2362,6 +2388,7 @@
transition(SLSS, Unblock, SLS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -2385,6 +2412,7 @@
transition(OLSS, Unblock, OLS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IGMO, Fwd_GETX, IGM) {
@@ -2418,6 +2446,7 @@
transition(MM, Exclusive_Unblock, ILX) {
g_recordLocalExclusive;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(M, L1_GETS, OO) {
@@ -2441,6 +2470,7 @@
transition(SS, Unblock, SLS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(O, L1_GETS, OO) {
@@ -2453,6 +2483,7 @@
transition(OO, Unblock, OLS) {
g_recordLocalSharer;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OO, Exclusive_Unblock, ILX) {
@@ -2460,6 +2491,7 @@
y_copyCacheStateToDir;
rr_deallocateL2CacheBlock;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -2495,11 +2527,13 @@
transition(ILSW, Unblock, ILS) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOW, Unblock, ILO) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOSX, L1_PUTS_only, ILOXW) {
@@ -2510,6 +2544,7 @@
transition(ILOXW, Unblock, ILOX) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
// hmmm...keep data or drop. Just drop for now
@@ -2526,11 +2561,13 @@
transition(ILOSW, Unblock, ILOS) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOSXW, Unblock, ILOSX) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(SLS, L1_PUTS, SLSW) {
@@ -2546,6 +2583,7 @@
transition(SW, {Unblock}, S) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OLS, L1_PUTS, OLSW) {
@@ -2581,16 +2619,19 @@
transition(OLSXW, {Unblock}, OLSX) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OW, {Unblock}, O) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OXW, {Unblock}, M) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILX, L1_PUTX, ILXW ) {
@@ -2604,6 +2645,7 @@
y_copyDirToCacheAndRemove;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
// clean writeback
@@ -2613,11 +2655,13 @@
y_copyDirToCacheAndRemove;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILXW, Unblock, ILX) {
// writeback canceled because L1 invalidated
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILSW, L1_WBCLEANDATA, SLS) {
@@ -2626,6 +2670,7 @@
u_writeDataToCache;
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(IW, L1_WBCLEANDATA, S) {
@@ -2634,7 +2679,7 @@
u_writeDataToCache;
gg_clearSharerFromL1Response;
n_popResponseQueue;
-
+ wa_wakeUpDependents;
}
// Owner can have dirty data
@@ -2644,6 +2689,7 @@
gg_clearOwnerFromL1Response;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOXW, L1_WBDIRTYDATA, M) {
@@ -2652,6 +2698,7 @@
gg_clearOwnerFromL1Response;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOXW, L1_WBCLEANDATA, M) {
@@ -2660,6 +2707,7 @@
gg_clearOwnerFromL1Response;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOSW, {L1_WBCLEANDATA, L1_WBDIRTYDATA}, OLS) {
@@ -2668,6 +2716,7 @@
gg_clearOwnerFromL1Response;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(ILOSXW, {L1_WBCLEANDATA, L1_WBDIRTYDATA}, OLSX) {
@@ -2676,17 +2725,20 @@
gg_clearOwnerFromL1Response;
u_writeDataToCache;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(SLSW, {Unblock}, SLS) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
transition(OLSW, {Unblock}, OLS) {
gg_clearSharerFromL1Response;
n_popResponseQueue;
+ wa_wakeUpDependents;
}
@@ -2787,11 +2839,13 @@
qq_sendDataFromTBEToMemory;
s_deallocateTBE;
m_popRequestQueue;
+ wa_wakeUpDependents;
}
transition(MII, Writeback_Nack, I) {
s_deallocateTBE;
m_popRequestQueue;
+ wa_wakeUpDependents;
}
transition(OI, Writeback_Nack) {
@@ -2803,17 +2857,20 @@
qq_sendDataFromTBEToMemory;
s_deallocateTBE;
m_popRequestQueue;
+ wa_wakeUpDependents;
}
transition(MII, Writeback_Ack, I) {
f_sendUnblock;
s_deallocateTBE;
m_popRequestQueue;
+ wa_wakeUpDependents;
}
transition(ILSI, Writeback_Ack, ILS) {
f_sendUnblock;
s_deallocateTBE;
m_popRequestQueue;
+ wa_wakeUpDependents;
}
}
