mem: Add snoop filters to L2 crossbars, and check size

This patch adds a snoop filter to the L2XBar. For now we refrain from
globally adding a snoop filter to the SystemXBar, since the latter is
also used in systems without caches. In scenarios without caches the
snoop filter will not see any writeback/clean evicts from the CPU
ports, despite the fact that they are snooping. To avoid inadvertent
use of the snoop filter in these cases we leave it out for now.

A size check is added to the snoop filter, merely to ensure it does
not grow beyond the total capacity of the caches above it. The size
has to be set manually, and a value of 8 MByte is choosen as suitably
high default.

diff --git a/src/mem/XBar.py b/src/mem/XBar.py
index a445b5e37ec45c7e8c2223d4b2baaae1d233af97..8614519b3ffdd2a8f1f62d2b6242677f64922b55 100644 (file)
--- a/src/mem/XBar.py
+++ b/src/mem/XBar.py
@@ -112,6 +112,9 @@ class SnoopFilter(SimObject):
 
     system = Param.System(Parent.any, "System that the crossbar belongs to.")
 
+    # Sanity check on max capacity to track, adjust if needed.
+    max_capacity = Param.MemorySize('8MB', "Maximum capacity of snoop filter")
+
 # We use a coherent crossbar to connect multiple masters to the L2
 # caches. Normally this crossbar would be part of the cache itself.
 class L2XBar(CoherentXBar):
@@ -125,6 +128,11 @@ class L2XBar(CoherentXBar):
     response_latency = 1
     snoop_response_latency = 1
 
+    # Use a snoop-filter by default, and set the latency to zero as
+    # the lookup is assumed to overlap with the frontend latency of
+    # the crossbar
+    snoop_filter = SnoopFilter(lookup_latency = 0)
+
 # One of the key coherent crossbar instances is the system
 # interconnect, tying together the CPU clusters, GPUs, and any I/O
 # coherent masters, and DRAM controllers.

diff --git a/src/mem/snoop_filter.cc b/src/mem/snoop_filter.cc
index f6e6ef1b4f8d8b086b86a4997ac03c46f1c6adc2..9b005cbc52151c1da0ad525897224c0fafd20dc2 100755 (executable)
--- a/src/mem/snoop_filter.cc
+++ b/src/mem/snoop_filter.cc
@@ -184,6 +184,10 @@ SnoopFilter::lookupSnoop(const Packet* cpkt)
     auto sf_it = cachedLocations.find(line_addr);
     bool is_hit = (sf_it != cachedLocations.end());
 
+    panic_if(!is_hit && (cachedLocations.size() >= maxEntryCount),
+             "snoop filter exceeded capacity of %d cache blocks\n",
+             maxEntryCount);
+
     // If the snoop filter has no entry and its an uncacheable
     // request, do not create a new snoop filter entry, simply return
     // a NULL portlist.

diff --git a/src/mem/snoop_filter.hh b/src/mem/snoop_filter.hh
index b1e33dc8f8260ef66260d43ea464aa13c811fd4b..be3018e6e8d5918984966ef2735e8b7ad3a89978 100755 (executable)
--- a/src/mem/snoop_filter.hh
+++ b/src/mem/snoop_filter.hh
@@ -90,7 +90,8 @@ class SnoopFilter : public SimObject {
 
     SnoopFilter (const SnoopFilterParams *p) :
         SimObject(p), reqLookupResult(cachedLocations.end()), retryItem{0, 0},
-        linesize(p->system->cacheLineSize()), lookupLatency(p->lookup_latency)
+        linesize(p->system->cacheLineSize()), lookupLatency(p->lookup_latency),
+        maxEntryCount(p->max_capacity / p->system->cacheLineSize())
     {
     }
 
@@ -254,6 +255,8 @@ class SnoopFilter : public SimObject {
     const unsigned linesize;
     /** Latency for doing a lookup in the filter */
     const Cycles lookupLatency;
+    /** Max capacity in terms of cache blocks tracked, for sanity checking */
+    const unsigned maxEntryCount;
 
     /** Statistics */
     Stats::Scalar totRequests;