#!/usr/bin/env python from math import log, ceil from mapping import * ############################################################################### # file : arch.py (for the tsar_generic_leti architecture) # date : may 2014 # author : Alain Greiner ############################################################################### # This file contains a mapping generator for the "tsar_generic_leti" platform. # This includes both the hardware architecture (clusters, processors, # peripherals, physical space segmentation) and the mapping of all kernel # objects (global vsegs). # # The "constructor" parameters are: # - x_size : number of clusters in a row # - 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 # - x_io : cluster_io x coordinate # - y_io : cluster_io y coordinate # - x_width : number of bits for x coordinate # - y_width : number of bits for y coordinate # - paddr_width : number of bits for physical address # - irq_per_proc : number of input IRQs per processor # - 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 ): ### 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 ### parameters checking assert( nb_procs <= (1 << p_width) ) assert( (x_size == 1) or (x_size == 2) or (x_size == 4) or (x_size == 8) or (x_size == 16) ) assert( (y_size == 1) or (y_size == 2) or (y_size == 4) or (y_size == 8) or (y_size == 16) ) assert( nb_ttys == 1 ) assert( ((x_io == 0) and (y_io == 0)) or ((x_io == x_size-1) and (y_io == y_size-1)) ) platform_name = 'tsar_leti_%d_%d_%d' % ( x_size, y_size, nb_procs ) ### define physical segments ### These segments are replicated in all clusters ram_base = 0x0000000000 ram_size = 0x4000000 # 64 Mbytes xcu_base = 0x00F0000000 xcu_size = 0x1000 # 4 Kbytes mmc_base = 0x00E0000000 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) bdv_base = 0x00F2000000 + offset_io bdv_size = 0x1000 # 4kbytes tty_base = 0x00F4000000 + offset_io tty_size = 0x4000 # 16 Kbytes nic_base = 0x00F7000000 + offset_io nic_size = 0x80000 # 512 kbytes cma_base = 0x00F8000000 + offset_io cma_size = 0x1000 * 2 * nb_nics # 4 kbytes * 2 * nb_nics fbf_base = 0x00F3000000 + offset_io fbf_size = fbf_width * fbf_width # fbf_width * fbf_width bytes pic_base = 0x00F9000000 + offset_io pic_size = 0x1000 # 4 Kbytes rdk_base = 0x02000000 rdk_size = 0x02000000 # 32 Mbytes ### define preloader & bootloader vsegs base addresses and sizes ### 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 boot_mapping_vbase = 0x00010000 # ident boot_mapping_size = 0x00080000 # 512 Kbytes boot_code_vbase = 0x00090000 # ident boot_code_size = 0x00040000 # 256 Kbytes boot_data_vbase = 0x000D0000 # ident boot_data_size = 0x00080000 # 512 Kbytes boot_stack_vbase = 0x00150000 # ident boot_stack_size = 0x00050000 # 320 Kbytes ### define kernel vsegs base addresses and sizes ### 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 kernel_init_vbase = 0x80080000 kernel_init_size = 0x00080000 # 512 Kbytes per cluster kernel_data_vbase = 0x80100000 kernel_data_size = 0x00100000 # 1 Mbytes in cluster[0][0] kernel_ptab_vbase = 0xB0000000 kernel_ptab_size = 0x00200000 # 2 Mbytes per cluster kernel_uncdata_vbase = 0x90000000 kernel_uncdata_size = 0x00001000 # 4 Kbytes kernel_sched_vbase = 0xA0000000 # distributed in all clusters kernel_sched_size = 0x00002000 * nb_procs # 8 kbytes per processor ### create mapping mapping = Mapping( name = platform_name, x_size = x_size, y_size = y_size, nprocs = nb_procs, x_width = x_width, y_width = y_width, p_width = p_width, paddr_width = paddr_width, coherence = True, irq_per_proc = irq_per_proc, use_ramdisk = use_ramdisk, x_io = x_io, y_io = y_io, peri_increment = peri_increment, reset_address = reset_address, ram_base = ram_base, ram_size = ram_size ) ### 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 ) mapping.addIrq( pic, index = 0 , isrtype = 'ISR_NIC_RX', channel = 0 ) mapping.addIrq( pic, index = 1 , isrtype = 'ISR_NIC_RX', channel = 1 ) mapping.addIrq( pic, index = 2 , isrtype = 'ISR_NIC_TX', channel = 0 ) mapping.addIrq( pic, index = 3 , isrtype = 'ISR_NIC_TX', channel = 1 ) mapping.addIrq( pic, index = 4 , isrtype = 'ISR_CMA' , channel = 0 ) mapping.addIrq( pic, index = 5 , isrtype = 'ISR_CMA' , channel = 1 ) mapping.addIrq( pic, index = 6 , isrtype = 'ISR_CMA' , channel = 2 ) mapping.addIrq( pic, index = 7 , isrtype = 'ISR_CMA' , channel = 3 ) mapping.addIrq( pic, index = 8 , isrtype = 'ISR_BDV' , channel = 0 ) mapping.addIrq( pic, index = 16, isrtype = 'ISR_TTY_RX', channel = 0 ) ### hardware components replicated in all clusters for x in xrange( x_size ): for y in xrange( y_size ): cluster_xy = (x << y_width) + y; 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 ) # IRQs replicated in all clusters mapping.addIrq( xcu, index = 8, isrtype = 'ISR_MMC' ) # IRQ in IO cluster (0,0) if x == 0 and y == 0: mapping.addIrq( xcu, index = 9 , isrtype = 'ISR_BDV' ) mapping.addIrq( xcu, index = 10, isrtype = 'ISR_TTY_RX' ) # processors for p in xrange ( nb_procs ): mapping.addProc( x, y, p ) ### 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, 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', local = False, big = True ) mapping.addGlobal( 'seg_tty', tty_base, tty_size, '__W_', vtype = 'PERI', x = 0, y = 0, pseg = 'TTY', local = False, big = True ) mapping.addGlobal( 'seg_nic', nic_base, nic_size, '__W_', vtype = 'PERI', x = 0, y = 0, pseg = 'NIC', local = False, big = True ) mapping.addGlobal( 'seg_cma', cma_base, cma_size, '__W_', vtype = 'PERI', x = 0, y = 0, pseg = 'CMA', local = False, big = True ) mapping.addGlobal( 'seg_fbf', fbf_base, fbf_size, '__W_', vtype = 'PERI', x = 0, y = 0, pseg = 'FBF', local = False, big = True ) mapping.addGlobal( 'seg_pic', pic_base, pic_size, '__W_', vtype = 'PERI', x = 0, y = 0, pseg = 'PIC', 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 ): 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', 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', local = False, big = False ) ### return mapping ### return mapping ################################# platform test ####################################################### if __name__ == '__main__': mapping = arch( x_size = 2, y_size = 2, nb_procs = 2 ) # print mapping.netbsd_dts() print mapping.xml() # print mapping.giet_vsegs() # Local Variables: # tab-width: 4; # c-basic-offset: 4; # c-file-offsets:((innamespace . 0)(inline-open . 0)); # indent-tabs-mode: nil; # End: # # vim: filetype=python:expandtab:shiftwidth=4:tabstop=4:softtabstop=4