Index: trunk/platforms/tsar_generic_leti/arch.py
===================================================================
--- trunk/platforms/tsar_generic_leti/arch.py	(revision 819)
+++ trunk/platforms/tsar_generic_leti/arch.py	(revision 820)
@@ -18,9 +18,9 @@
 #  - y_size         : number of clusters in a column
 #  - nb_procs       : number of processors per cluster
+#  - fbf_width      : frame_buffer width = frame_buffer heigth
 #
 #  The "hidden" parameters (defined below) are:
 #  - nb_ttys        : number of TTY channels
 #  - nb_nics        : number of NIC channels
-#  - fbf_width      : frame_buffer width = frame_buffer heigth
 #  - x_io           : cluster_io x coordinate
 #  - y_io           : cluster_io y coordinate
@@ -31,29 +31,38 @@
 #  - use_ramdisk    : use a ramdisk when True
 #  - peri_increment : address increment for replicated peripherals
-###############################################################################
+#
+# Regarding physical memory allocation, there is one allocator per cluster:
+# - We use only one big physical page (2 Mbytes) for the five boot vsegs,
+#   allocated in cluster[0,0], identity mapping.
+# - We use one big page per cluster for the kernel vsegs.
+#   The kernel_code, kernel_init and kernel_ptab can be replicated in all clusters.
+#   The kernel_data and kernel_uncdata shared vsegs are only mapped in cluster[0,0].
+# - We use 8 small physical pages (4 Kbytes) per cluster for the schedulers.
+# - We use one big page for each external peripheral in IO cluster,
+# - We use one small page per cluster for each internal peripheral.
+###################################################################################
 
 ########################
-def arch( x_size     = 2,
-          y_size     = 2,
-          nb_procs   = 2, 
-          fbf_width  = 128 ):
+def arch( x_size    = 2,
+          y_size    = 2,
+          nb_procs  = 2,
+          fbf_width = 128 ):
 
     ### define architecture constants
 
-    nb_ttys           = 1
-    nb_nics           = 2
-    x_io              = 0
-    y_io              = 0
-    x_width           = 4
-    y_width           = 4
-    p_width           = 4
-    paddr_width       = 40
-    irq_per_proc      = 4
-    use_ramdisk       = True
-    peri_increment    = 0x10000    # distributed peripherals vbase address increment
-    sched_increment   = 0x10000    # distributed schedulers vbase address increment 
-    ptab_increment    = 0x200000   # distributed page tables vbase address increment
-    reset_address     = 0x00000000
-    distributed_ptabs = False
+    nb_ttys         = 1
+    nb_nics         = 2
+    x_io            = 0
+    y_io            = 0
+    x_width         = 4
+    y_width         = 4
+    p_width         = 4
+    paddr_width     = 40
+    irq_per_proc    = 4
+    use_ramdisk     = True
+    peri_increment  = 0x10000    # distributed peripherals vbase address increment
+    sched_increment = 0x10000    # distributed schedulers vbase address increment
+    ptab_increment  = 0x200000   # distributed page tables vbase address increment
+    reset_address   = 0x00000000
 
     ### parameters checking
@@ -75,4 +84,5 @@
 
     ### define physical segments
+    ### These segments are replicated in all clusters
 
     ram_base = 0x0000000000
@@ -85,4 +95,7 @@
     mmc_size = 0x1000                      # 4 Kbytes
 
+    ### define physical segments for external peripherals
+    ## These segments are only defined in cluster_io
+
     offset_io = ((x_io << y_width) + y_io) << (paddr_width - x_width - y_width)
 
@@ -109,6 +122,9 @@
 
     ### define preloader & bootloader vsegs base addresses and sizes
-    ### we want to pack these 5 vsegs in the same big page
-
+    ### We want to pack these 5 vsegs in the same big page
+    ### => boot cost is one BPP in cluster[0][0]
+
+                               '__W_', vtype = 'PERI' , x = x , y = y , pseg =
+                               'XCU',
     preloader_vbase      = 0x00000000      # ident
     preloader_size       = 0x00010000      # 64 Kbytes
@@ -127,8 +143,12 @@
 
     ### define kernel vsegs base addresses and sizes
-    ### we want to pack code, init and data vsegs in one big page
-    ### we want to map the ptab vseg in one big page per cluster
-    ### we want to map the sched vseg in small pages
-    
+    ### code, init, ptab & sched vsegs are replicated in all clusters.
+    ### data & uncdata vsegs are only mapped in cluster[0][0].
+    ### - We pack code, init, data vsegs in the same BIG page.
+    ### - We use another BIG page for the ptab vseg.
+    ### - We use 2*nb_procs SMALL pages for the sched vseg.
+    ### - we use one SMALL page for uncdata
+    ### => kernel cost is 2 BPPs and (2*n + 1) SPPs per cluster.
+
     kernel_code_vbase    = 0x80000000
     kernel_code_size     = 0x00080000      # 512 Kbytes per cluster
@@ -138,5 +158,5 @@
 
     kernel_data_vbase    = 0x80100000
-    kernel_data_size     = 0x00100000      # 1 Mbytes  in cluster [0,0]
+    kernel_data_size     = 0x00100000      # 1 Mbytes in cluster[0][0]
 
     kernel_ptab_vbase    = 0xB0000000
@@ -146,6 +166,6 @@
     kernel_uncdata_size  = 0x00001000      # 4 Kbytes
 
-    kernel_sched_vbase   = 0xA0000000                # distributed in all clusters
-    kernel_sched_size    = 0x00002000 * nb_procs     # 8 kbytes per processor
+    kernel_sched_vbase   = 0xA0000000            # distributed in all clusters
+    kernel_sched_size    = 0x00002000 * nb_procs # 8 kbytes per processor
 
     ### create mapping
@@ -171,21 +191,15 @@
     ###  external peripherals (accessible in cluster[0,0] only for this mapping)
 
-    bdv = mapping.addPeriph( 'BDV', base = bdv_base, size = bdv_size,
-                             ptype = 'IOC', subtype = 'BDV' )
-
-    tty = mapping.addPeriph( 'TTY', base = tty_base, size = tty_size,
-                             ptype = 'TTY', channels = nb_ttys )
-
-    nic = mapping.addPeriph( 'NIC', base = nic_base, size = nic_size,
-                             ptype = 'NIC', channels = nb_nics )
-
-    cma = mapping.addPeriph( 'CMA', base = cma_base, size = cma_size,
-                             ptype = 'CMA', channels = 2*nb_nics )
-
-    fbf = mapping.addPeriph( 'FBF', base = fbf_base, size = fbf_size,
-                             ptype = 'FBF', arg = fbf_width )
-
-    pic = mapping.addPeriph( 'PIC', base = pic_base, size = pic_size,
-                             ptype = 'PIC', channels = 32 )
+    bdv = mapping.addPeriph( 'BDV', base = bdv_base, size = bdv_size, ptype = 'IOC', subtype = 'BDV' )
+
+    tty = mapping.addPeriph( 'TTY', base = tty_base, size = tty_size, ptype = 'TTY', channels = nb_ttys )
+
+    nic = mapping.addPeriph( 'NIC', base = nic_base, size = nic_size, ptype = 'NIC', channels = nb_nics )
+
+    cma = mapping.addPeriph( 'CMA', base = cma_base, size = cma_size, ptype = 'CMA', channels = 2*nb_nics )
+
+    fbf = mapping.addPeriph( 'FBF', base = fbf_base, size = fbf_size, ptype = 'FBF', arg = fbf_width )
+
+    pic = mapping.addPeriph( 'PIC', base = pic_base, size = pic_size, ptype = 'PIC', channels = 32 )
 
     mapping.addIrq( pic, index = 0 , isrtype = 'ISR_NIC_RX', channel = 0 )
@@ -207,14 +221,11 @@
             offset     = cluster_xy << (paddr_width - x_width - y_width)
 
-            ram = mapping.addRam( 'RAM', base = ram_base + offset,
-                                  size = ram_size )
-
-            mmc = mapping.addPeriph( 'MMC', base = mmc_base + offset,
-                                     size = mmc_size, ptype = 'MMC' )
-
-            xcu = mapping.addPeriph( 'XCU', base = xcu_base + offset,
-                                     size = xcu_size, ptype = 'XCU',
-                                     channels = nb_procs * irq_per_proc,
-                                     arg = 16 )
+            ram = mapping.addRam( 'RAM', base = ram_base + offset, size = ram_size )
+
+            mmc = mapping.addPeriph( 'MMC', base = mmc_base + offset, size = mmc_size,
+                                     ptype = 'MMC' )
+
+            xcu = mapping.addPeriph( 'XCU', base = xcu_base + offset, size = xcu_size,
+                                     ptype = 'XCU', channels = nb_procs * irq_per_proc, arg = 16 )
 
             # IRQs replicated in all clusters
@@ -230,103 +241,121 @@
                 mapping.addProc( x, y, p )
 
-    ### global vseg for preloader
+    ### global vsegs for preloader & boot_loader
+    ### we want to pack those 5 vsegs in the same big page
+    ### => same flags CXW_ / identity mapping / non local / big page
+
+    mapping.addGlobal( 'seg_preloader', preloader_vbase, preloader_size,
+                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
+                       identity = True, local = False, big = True )
+
+    mapping.addGlobal( 'seg_boot_mapping', boot_mapping_vbase, boot_mapping_size,
+                       'CXW_', vtype = 'BLOB'  , x = 0, y = 0, pseg = 'RAM',
+                       identity = True, local = False, big = True )
+
+    mapping.addGlobal( 'seg_boot_code', boot_code_vbase, boot_code_size,
+                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
+                       identity = True, local = False, big = True )
+
+    mapping.addGlobal( 'seg_boot_data', boot_data_vbase, boot_data_size,
+                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
+                       identity = True, local = False, big = True )
+
+    mapping.addGlobal( 'seg_boot_stack', boot_stack_vbase, boot_stack_size,
+                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
+                       identity = True, local = False, big = True )
+
+    ### global vsegs kernel_code, kernel_init : local / big page
+    ### replicated in all clusters with the same name (same vbase)
+    for x in xrange( x_size ):
+        for y in xrange( y_size ):
+            cluster_xy = (x << y_width) + y;
+
+            mapping.addGlobal( 'seg_kernel_code', kernel_code_vbase, kernel_code_size,
+                               'CXW_', vtype = 'ELF', x = x, y = y, pseg = 'RAM',
+                               binpath = 'build/kernel/kernel.elf',
+                               local = True, big = True )
+
+            mapping.addGlobal( 'seg_kernel_init', kernel_init_vbase, kernel_init_size,
+                               'CXW_', vtype = 'ELF', x = x, y = y, pseg = 'RAM',
+                               binpath = 'build/kernel/kernel.elf',
+                               local = True, big = True )
+
+    ### global vseg kernel_data: non local / big page
+    ### Only mapped in cluster[0][0]
+    mapping.addGlobal( 'seg_kernel_data', kernel_data_vbase, kernel_data_size,
+                       'C_W_', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM',
+                       binpath = 'build/kernel/kernel.elf', local = False )
+
+    ### global vseg kernel_uncdata: non local / small page
+    ### Only mapped in cluster[0][0]
+    mapping.addGlobal( 'seg_kernel_uncdata', kernel_uncdata_vbase, kernel_uncdata_size,
+                       '__W_', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM',
+                       binpath = 'build/kernel/kernel.elf', local = False )
+
+    for x in xrange( x_size ):
+        for y in xrange( y_size ):
+            cluster_xy = (x << y_width) + y;
+
+            ### Global vsegs kernel_ptab_x_y: non local / big pages
+            ### replicated in all clusters with name indexed by (x,y)
+            ### as vbase address is incremented by (cluster_xy * vseg_increment)
+            offset = cluster_xy * ptab_increment
+            mapping.addGlobal( 'seg_kernel_ptab_%d_%d' %(x,y), kernel_ptab_vbase + offset, kernel_ptab_size,
+                               'CXW_', vtype = 'PTAB', x = x, y = y, pseg = 'RAM',
+                               local = False, big = True )
+
+            ### global vsegs kernel_sched : non local / small pages
+            ### allocated in all clusters with name indexed by (x,y)
+            ### as vbase address is incremented by (cluster_xy * vseg_increment)
+            offset = cluster_xy * sched_increment
+            mapping.addGlobal( 'seg_kernel_sched_%d_%d' %(x,y), kernel_sched_vbase + offset , kernel_sched_size,
+                               'C_W_', vtype = 'SCHED', x = x, y = y, pseg = 'RAM',
+                               local = False, big = False )
 
     ### global vseg for ram disk
-
     if use_ramdisk:
         mapping.addGlobal( 'seg_rdk', rdk_base, rdk_size, '__W_',
                            vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
-                           identity = True )
-
-    ### global vsegs for external peripherals / identity mapping
-
+                           identity = True, local = False, big = True )
+
+    ### global vsegs for external peripherals: non local / big page
     mapping.addGlobal( 'seg_bdv', bdv_base, bdv_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'BDV',
-                       identity = True )
+                       local = False, big = True )
 
     mapping.addGlobal( 'seg_tty', tty_base, tty_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'TTY',
-                       identity = True )
+                       local = False, big = True )
 
     mapping.addGlobal( 'seg_nic', nic_base, nic_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'NIC',
-                       identity = True )
+                       local = False, big = True )
 
     mapping.addGlobal( 'seg_cma', cma_base, cma_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'CMA',
-                       identity = True )
+                       local = False, big = True )
 
     mapping.addGlobal( 'seg_fbf', fbf_base, fbf_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'FBF',
-                       identity = True )
+                       local = False, big = True )
 
     mapping.addGlobal( 'seg_pic', pic_base, pic_size, '__W_',
                        vtype = 'PERI', x = 0, y = 0, pseg = 'PIC',
-                       identity = True )
-
-    ### global vsegs for internal peripherals, and for schedulers
-    ### name is indexed by (x,y) / vbase address is incremented by (cluster_xy * peri_increment)
-
+                       local = False, big = True )
+
+    ### global vsegs for internal peripherals : non local / small pages
+    ### allocated in all clusters with name indexed by (x,y)
+    ### as vbase address is incremented by (cluster_xy * vseg_increment)
     for x in xrange( x_size ):
         for y in xrange( y_size ):
-            cluster_xy = (x << y_width) + y;
-            offset     = cluster_xy * peri_increment
+            offset = ((x << y_width) + y) * peri_increment
 
             mapping.addGlobal( 'seg_xcu_%d_%d' %(x,y), xcu_base + offset, xcu_size,
-                               '__W_', vtype = 'PERI' , x = x , y = y , pseg = 'XCU' )
+                               '__W_', vtype = 'PERI' , x = x , y = y , pseg = 'XCU',
+                               local = False, big = False )
 
             mapping.addGlobal( 'seg_mmc_%d_%d' %(x,y), mmc_base + offset, mmc_size,
-                               '__W_', vtype = 'PERI' , x = x , y = y , pseg = 'MMC' )
-
-            mapping.addGlobal( 'seg_sched_%d_%d' %(x,y), kernel_sched_vbase + offset, kernel_sched_size,
-                               'C_W_', vtype = 'SCHED', x = x , y = y , pseg = 'RAM' )
-
-    ### global vsegs for preloader and boot_loader 
-    ### identity mapping / non local / big page
-
-    mapping.addGlobal( 'seg_preloader', preloader_vbase, preloader_size, 
-                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
-                       identity = True, local = False, big = True )
-
-    mapping.addGlobal( 'seg_boot_mapping', boot_mapping_vbase, boot_mapping_size,
-                       'CXW_', vtype = 'BLOB'  , x = 0, y = 0, pseg = 'RAM',
-                       identity = True, local = False, big = True )
-
-    mapping.addGlobal( 'seg_boot_code', boot_code_vbase, boot_code_size,
-                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
-                       identity = True, local = False, big = True )
-
-    mapping.addGlobal( 'seg_boot_data', boot_data_vbase, boot_data_size,
-                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
-                       identity = True, local = False, big = True )
-
-    mapping.addGlobal( 'seg_boot_stack', boot_stack_vbase, boot_stack_size,
-                       'CXW_', vtype = 'BUFFER', x = 0, y = 0, pseg = 'RAM',
-                       identity = True, local = False, big = True )
-
-    ### global vsegs kernel_code, kernel_init
-    for x in xrange( x_size ):
-        for y in xrange( y_size ):
-            cluster_xy = (x << y_width) + y;
-
-            mapping.addGlobal( 'seg_kernel_code', kernel_code_vbase, kernel_code_size,
-                               'CXW_', vtype = 'ELF', x = x , y = y , pseg = 'RAM',
-                               binpath = 'build/kernel/kernel.elf', 
-                               local = True, big = True )
-
-            mapping.addGlobal( 'seg_kernel_init', kernel_init_vbase, kernel_init_size,
-                               'CXW_', vtype = 'ELF', x = x , y = y , pseg = 'RAM',
-                               binpath = 'build/kernel/kernel.elf',
-                               local = True, big = True )
-
-       
-
-    mapping.addGlobal( 'seg_kernel_data', kernel_data_vbase, kernel_data_size,
-                       'C_W_', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM',
-                       binpath = 'build/kernel/kernel.elf', local = False )
-
-    mapping.addGlobal( 'seg_kernel_uncdata', kernel_uncdata_vbase, kernel_uncdata_size,
-                       '__W_', vtype = 'ELF', x = 0, y = 0, pseg = 'RAM',
-                       binpath = 'build/kernel/kernel.elf', local = False )
+                               '__W_', vtype = 'PERI' , x = x , y = y , pseg = 'MMC',
+                               local = False, big = False )
 
     ### return mapping ###
