Index: /soft/giet_vm/giet_boot/boot.c =================================================================== --- /soft/giet_vm/giet_boot/boot.c (revision 818) +++ /soft/giet_vm/giet_boot/boot.c (revision 819) @@ -405,5 +405,5 @@ // The three flags (Local, Remote and Dirty) are set to 1 // to avoid hardware update for these flags, because GIET_VM - // does use these flags. + // does not use these flags. unsigned int flags = 0; if (vseg->mode & C_MODE_MASK) flags |= PTE_C; @@ -411,8 +411,11 @@ if (vseg->mode & W_MODE_MASK) flags |= PTE_W; if (vseg->mode & U_MODE_MASK) flags |= PTE_U; - if ( global ) flags |= PTE_G; flags |= PTE_L; flags |= PTE_R; flags |= PTE_D; + +#if GIET_USE_MMU_GLOBAL_FLAG + if ( global ) flags |= PTE_G; +#endif // compute VPN, PPN and number of pages (big or small) @@ -849,9 +852,9 @@ // idle_thread context, the HWI / PTI / WTI interrupt vectors, // and the XCU HWI / PTI / WTI masks. -// - In Step 2, it scan all threads in all vspaces to complete the threads contexts, +// - In Step 2, it scan all threads in all vspaces to complete the threads contexts, // initialisation as specified in the mapping_info data structure, -// and set the CP0_SCHED register. +// and set the CP0_SCHED register. //////////////////////////////////////////////////////////////////////////////////// -void boot_scheduler_init( unsigned int x, +void boot_scheduler_init( unsigned int x, unsigned int y ) { @@ -864,19 +867,19 @@ mapping_irq_t* irq = _get_irq_base(header); - unsigned int periph_id; + unsigned int periph_id; unsigned int irq_id; unsigned int vspace_id; unsigned int vseg_id; - unsigned int thread_id; - - unsigned int sched_vbase; // schedulers array vbase address + unsigned int thread_id; + + unsigned int sched_vbase; // schedulers array vbase address unsigned int sched_length; // schedulers array length static_scheduler_t* psched; // pointer on processor scheduler unsigned int cluster_id = (x * Y_SIZE) + y; - unsigned int cluster_xy = (x << Y_WIDTH) + y; + unsigned int cluster_xy = (x << Y_WIDTH) + y; unsigned int nprocs = cluster[cluster_id].procs; - unsigned int lpid; - + unsigned int lpid; + if ( nprocs > 8 ) { @@ -886,11 +889,11 @@ //////////////////////////////////////////////////////////////////////////////// - // Step 1 : - initialize the schedulers[] array of pointers, - // - initialize the "threads" and "current variables. + // Step 1 : - initialize the schedulers[] array of pointers, + // - initialize the "threads" and "current variables. // - initialise the idle_thread context. // - initialize the HWI, PTI and WTI interrupt vectors. // - initialize the XCU masks for HWI / WTI / PTI interrupts. // - // The general policy for interrupts routing is the following: + // The general policy for interrupts routing is the following: // - the local HWI are statically allocatedted to local processors. // - the nprocs first PTI are allocated for TICK (one per processor). @@ -955,5 +958,5 @@ unsigned int wti_mask[8] = {0,0,0,0,0,0,0,0}; - // scan local peripherals to get and check local XCU + // scan local peripherals to get and check local XCU mapping_periph_t* xcu = NULL; unsigned int min = cluster[cluster_id].periph_offset ; @@ -962,5 +965,5 @@ for ( periph_id = min ; periph_id < max ; periph_id++ ) { - if( periph[periph_id].type == PERIPH_TYPE_XCU ) + if( periph[periph_id].type == PERIPH_TYPE_XCU ) { xcu = &periph[periph_id]; @@ -996,8 +999,8 @@ } } - } + } if ( xcu == NULL ) - { + { _printf("\n[BOOT ERROR] missing XCU in cluster[%d,%d]\n", x , y ); _exit(); @@ -1005,5 +1008,5 @@ // HWI interrupt vector definition - // scan HWI connected to local XCU + // scan HWI connected to local XCU // for round-robin allocation to local processors lpid = 0; @@ -1029,5 +1032,5 @@ hwi_mask[lpid] = hwi_mask[lpid] | (1<vspaces; vspace_id++) - { - // We must set the PTPR depending on the vspace, because the start_vector + for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++) + { + // We must set the PTPR depending on the vspace, because the start_vector // and the stack address are defined in virtual space. _set_mmu_ptpr( (unsigned int)(_ptabs_paddr[vspace_id][x][y] >> 13) ); @@ -1094,10 +1097,21 @@ for (thread_id = vspace[vspace_id].thread_offset; thread_id < (vspace[vspace_id].thread_offset + vspace[vspace_id].threads); - thread_id++) + thread_id++) { // get the required thread placement coordinates [x,y,p] unsigned int req_x = cluster[thread[thread_id].clusterid].x; unsigned int req_y = cluster[thread[thread_id].clusterid].y; - unsigned int req_p = thread[thread_id].proclocid; + unsigned int req_p = thread[thread_id].proclocid; + + // skip this thread if it is allocated to another cluster + if ( ( req_x != x ) || ( req_y != y ) ) continue; + + if ( req_p >= NB_PROCS_MAX ) + { + _printf("\n[BOOT ERROR] Bad allocation of thread %s from vspace %s\n", + thread[thread_id].name, + vspace[vspace_id].name); + _exit(); + } // ctx_norun : two conditions to activate a thread @@ -1117,59 +1131,54 @@ // ctx_entry : Get the virtual address of the memory location containing - // the thread entry point : the start_vector is stored by GCC in the - // seg_data segment, and we must wait the application.elf loading to get + // the thread entry point : the start_vector is stored by GCC in the + // seg_data segment, and we must wait the application.elf loading to get // the entry point value... - vseg_id = vspace[vspace_id].start_vseg_id; + vseg_id = vspace[vspace_id].start_vseg_id; unsigned int ctx_entry = vseg[vseg_id].vbase + (thread[thread_id].startid)*4; - // ctx_sp : Get the vseg containing the stack + // ctx_sp : Get the vseg containing the stack // allocate 16 slots (64 bytes) for possible arguments. vseg_id = thread[thread_id].stack_vseg_id; unsigned int ctx_sp = vseg[vseg_id].vbase + vseg[vseg_id].length - 64; - // loop on the local processors - for ( lpid = 0 ; lpid < nprocs ; lpid++ ) - { - if ( (x == req_x) && (y == req_y) && (req_p == lpid) ) // fit - { - // pointer on selected scheduler - psched = _schedulers[x][y][lpid]; - - // ltid : compute local thread index in scheduler - unsigned int ltid = psched->threads; - - // update the threads field in scheduler: - psched->threads = ltid + 1; - - // ctx_trdid : compute pthread global identifier - unsigned int ctx_trdid = x << 24 | y<<16 | lpid<<8 | ltid; - - // initializes the thread context - psched->context[ltid].slot[CTX_CR_ID] = 0; - psched->context[ltid].slot[CTX_SR_ID] = GIET_SR_INIT_VALUE; - psched->context[ltid].slot[CTX_SP_ID] = ctx_sp; - psched->context[ltid].slot[CTX_EPC_ID] = ctx_entry; - psched->context[ltid].slot[CTX_ENTRY_ID] = ctx_entry; - psched->context[ltid].slot[CTX_PTPR_ID] = ctx_ptpr; - psched->context[ltid].slot[CTX_PTAB_ID] = ctx_ptab; - psched->context[ltid].slot[CTX_NPT2_ID] = ctx_npt2; - psched->context[ltid].slot[CTX_LTID_ID] = ltid; - psched->context[ltid].slot[CTX_TRDID_ID] = ctx_trdid; - psched->context[ltid].slot[CTX_VSID_ID] = vspace_id; - psched->context[ltid].slot[CTX_NORUN_ID] = ctx_norun; - psched->context[ltid].slot[CTX_SIGS_ID] = 0; - psched->context[ltid].slot[CTX_LOCKS_ID] = 0; - - psched->context[ltid].slot[CTX_TTY_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_CMA_FB_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_CMA_RX_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_CMA_TX_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_NIC_RX_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_NIC_TX_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_TIM_ID] = 0xFFFFFFFF; - psched->context[ltid].slot[CTX_HBA_ID] = 0xFFFFFFFF; - - // update thread ltid field in the mapping - thread[thread_id].ltid = ltid; + // pointer on selected scheduler + psched = _schedulers[x][y][req_p]; + + // ltid : compute local thread index in scheduler + unsigned int ltid = psched->threads; + + // update the threads field in scheduler: + psched->threads = ltid + 1; + + // ctx_trdid : compute pthread global identifier + unsigned int ctx_trdid = (x<<24) | (y<<16) | (req_p<<8) | ltid; + + // initializes the thread context + psched->context[ltid].slot[CTX_CR_ID] = 0; + psched->context[ltid].slot[CTX_SR_ID] = GIET_SR_INIT_VALUE; + psched->context[ltid].slot[CTX_SP_ID] = ctx_sp; + psched->context[ltid].slot[CTX_EPC_ID] = ctx_entry; + psched->context[ltid].slot[CTX_ENTRY_ID] = ctx_entry; + psched->context[ltid].slot[CTX_PTPR_ID] = ctx_ptpr; + psched->context[ltid].slot[CTX_PTAB_ID] = ctx_ptab; + psched->context[ltid].slot[CTX_NPT2_ID] = ctx_npt2; + psched->context[ltid].slot[CTX_LTID_ID] = ltid; + psched->context[ltid].slot[CTX_TRDID_ID] = ctx_trdid; + psched->context[ltid].slot[CTX_VSID_ID] = vspace_id; + psched->context[ltid].slot[CTX_NORUN_ID] = ctx_norun; + psched->context[ltid].slot[CTX_SIGS_ID] = 0; + psched->context[ltid].slot[CTX_LOCKS_ID] = 0; + + psched->context[ltid].slot[CTX_TTY_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_CMA_FB_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_CMA_RX_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_CMA_TX_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_NIC_RX_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_NIC_TX_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_TIM_ID] = 0xFFFFFFFF; + psched->context[ltid].slot[CTX_HBA_ID] = 0xFFFFFFFF; + + // update thread ltid field in the mapping + thread[thread_id].ltid = ltid; #if BOOT_DEBUG_SCHED @@ -1188,5 +1197,5 @@ thread[thread_id].name, vspace[vspace_id].name, - x, y, lpid, + x, y, req_p, psched->context[ltid].slot[CTX_LTID_ID], psched->context[ltid].slot[CTX_TRDID_ID], @@ -1201,6 +1210,4 @@ psched->context[ltid].slot[CTX_SIGS_ID] ); #endif - } // end if FIT - } // end for loop on local procs } // end loop on threads } // end loop on vspaces Index: /soft/giet_vm/giet_config.h =================================================================== --- /soft/giet_vm/giet_config.h (revision 818) +++ /soft/giet_vm/giet_config.h (revision 819) @@ -62,4 +62,5 @@ #define GIET_SR_INIT_VALUE 0x2000FF13 /* SR initial value (before eret) */ #define GIET_USE_HARD_FLOAT 0 /* hard float supported */ +#define GIET_USE_MMU_GLOBAL_FLAG 0 /* enable the use of the PTE_G flag */ #endif