source: trunk/kernel/arch/tsar/arch_init.c @ 223

/*
 * hal_devices_init.c - Devices intialization operations for the TSAR architecture
 *
 * Authors Ghassan Almaless (2008,2009,2010,2011,2012)
 *         Mohamed Lamine Karaoui (2015)
 *         Alain Greiner (2016)

 * Copyright (c) UPMC Sorbonne Universites
 *
 * This file is part of ALMOS-MKH.
 *
 * ALMOS-MKH.is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * ALMOS-MKH.is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH.; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <config.h>
#include <types.h>
#include <list.h>
#include <bits.h>
#include <cpu.h>
#include <system.h>
#include <cluster.h>
#include <chdev.h>
#include <driver.h>
#include <thread.h>
#include <process.h>
#include <kmem.h>
#include <ppm.h>
#include <pmm.h>
#include <page.h>
#include <kdmsg.h>
#include <devfs.h>
#include <drvdb.h>
#include <boot-info.h>
#include <soclib_xicu.h>
#include <soclib_iopic.h>
#define _ARCH_BIB_SIGNATURE_
#include <arch-bib.h>

#define die(args...) do {boot_dmsg(args); while(1);} while(0)

#define ICU_MASK     0xFFFFFFFF  // interrupt enabled for first 32 devices 
#define KATTR     (PMM_HUGE | PMM_READ | PMM_WRITE | PMM_EXECUTE | PMM_CACHED | PMM_GLOBAL | PMM_DIRTY | PMM_ACCESSED)
#define KDEV_ATTR (PMM_READ | PMM_WRITE | PMM_GLOBAL | PMM_DIRTY | PMM_ACCESSED)


/*********************************************************************************************
 * This function set the various masks for the XCU device in the local cluster. 
 * It should be executed by core[0] after the local cluster manager initialisation,
 * and after devices initialisation. The général policy is the following:
 * - PTI : it routes one PTI per core (with index = lid) for TICK.
 * - WTI : it routes one WTI per core (with index = lid) for wakup.
 * - HWI : it routes all local HWI to core[0].
 ********************************************************************************************/
void hal_xcu_set_mask()
{
    uint32_t       lid;
        cluster_t    * cluster;    // pointer on local cluster manager
        uint32_t       cores;      // number of cores in cluster
    device_t     * xcu;        // pointer on local XCU device descriptor        
    uint32_t       hwi_mask;   // HWI mask for XCU
    uint32_t       hwi_index;  // HWI index in XCU
    list_entry_t * iter;       // iterator to scan the device list
    device_t     * dev;        // pointer on local device descriptor
    dev_type_t     type;       // device type        

        cluster   = LOCAL_CLUSTER;
        cores     = cluster->cores_nr;
    xcu       = cluster->xcu;   

    // loop on all local devices to compute the hwi_mask 
    hwi_mask  = 0;
    hwi_index = 0;
    LIST_FOREACH( cluster->devlist , iter )
    {
        dev  = LIST_ELEMENT( iter , device_t , list );
        type = dev->type;

        // only replicated devices MMC and MWR generate HWI
        if( (type == DEV_TYPE_MMC) || (type == DEV_TYPE_MWR) )
        {
            hwi_mask = hwi_mask |= (1<<hwi_index);
            hwi_index++;
        }
    }

    // set the  three XCU masks
        for( lid = 0 ; lid < cores ; lid++ )
        {
        xcu_set_mask( xcu , (1<<lid) , XCU_PTI_TYPE , lid ); 
        xcu_set_mask( xcu , (1<<lid) , XCU_WTI_TYPE , lid ); 
        if( lid == 0 )
        {
            xcu_set_mask( xcu , hwi_mask , XCU_HWI_TYPE , lid ); 
        }
    }
}  // end hal_xcu_set_mask()


/*********************************************************************************************
 * This function returns a pointer on a cluster descriptor from the drvid. 
 ********************************************************************************************/
device_t * hal_dev_locate( list_entry_t * devlist_root,
                           uint32_t      drvid )
{
        struct list_entry *iter;
        struct device_s *dev;
  
        list_foreach(devlist_root, iter)
        {
                dev = list_element(iter, struct device_s, list);
                if(dev->op.drvid == drvid)
                        return dev;
        }
  
        return NULL;
}

/*********************************************************************************************
 * This function register all detected devices in given cluster.
 ********************************************************************************************/
void hal_dev_register( cluster_t       * cluster, 
                                   cluster_entry_s * entry )
{
        device_t     * xicu;
        list_entry_t * iter;
        device_t     * dev;
    cpu_t        * cpu_ptr;
        uint32_t       cpu;
        error_t        err;

        xicu = hal_dev_locate(&entry->devlist, SOCLIB_XICU_ID);
        dev  = hal_dev_locate(&entry->devlist, SOCLIB_DMA_ID);

        hal_entrys[entry->cxy].xicu = xicu;
        hal_entrys[entry->cxy].dma  = dev;
  
        if((cluster->id == entry->cxy) && (xicu == NULL))
                die("[ERROR]\tNo XICU is found for cluster %d\n", entry->cxy);
                
        //only local cluster get to set the irq
        if(cluster->id != entry->cxy)
                return;

        list_foreach((&entry->devlist), iter)
        {
                dev = list_element(iter, struct device_s, list);        

                if((dev != xicu) && (dev->irq != -1))
                {
                        err = xicu->op.icu.bind(xicu, dev);
      
                        if(err) 
                                die("[ERROR]\tFailed to bind device %s, irq %d, on xicu %s @%x [ err %d ]\n",
                                    dev->name, dev->irq, xicu->name, xicu, err);
                }

        }
                
        if(cluster->id != entry->cxy)
                return;

        hal_xicu_set_mask(cluster, xicu);

        for(cpu = 0; cpu < cluster->cpu_nr; cpu++)
        {
                cpu_ptr = &cluster->cpu_tbl[cpu];
                hal_cpu_set_irq_entry(cpu_ptr, 0, &xicu->action);
        }
}  // end hal_dev_register()

//////////////////////////////////////////////////
void hal_dev_iopic_register( cluster_t * cluster, 
                        struct cluster_entry_s *entry);

/** HAL-ARCH CPUs & Clusters counters */
static global_t hal_onln_cpus_nr;
static global_t hal_cpus_per_cluster;
static global_t hal_onln_clusters_nr;
static global_t hal_boot_cxy_;

/**/
struct hal_entry_s hal_entrys[CLUSTER_NR];
global_t __current_cxy;

/** Return Platform CPUs Online Number */
inline uint32_t hal_onln_cpu_nr(void)
{
        return hal_onln_cpus_nr.value;
}

/** Return Platform Clsuters Online Number */
inline uint32_t hal_onln_cluster_nr(void)
{
        return hal_onln_clusters_nr.value;
}

inline uint32_t hal_cpu_per_cluster(void)
{
        return hal_cpus_per_cluster.value;
}

inline cxy_t hal_boot_cxy(void)
{
        return hal_boot_cxy_.value;
}

extern uint32_t cpu_gid_tbl[CPU_PER_CLUSTER] CACHELINE;

inline void cpu_gid_tbl_init(struct boot_info_s *info)
{
        uint32_t i;
        uint32_t cpu_nr;

        cpu_nr = info->local_onln_cpu_nr;

        if(CPU_PER_CLUSTER < cpu_nr)
                while(1);//die("ERROR: This Kernel Is Compiled For only %d cpus per cluster\n", CPU_PER_CLUSTER);

        for(i=0; i < cpu_nr; i++)
                cpu_gid_tbl[i] = hal_cpu_gid(info->local_cluster_id, i); 
}

cxy_t Arch_cxy_To_Almos_cxy[CLUSTER_NR]; //address of info->hal_cxy_to_cxy array

static void fill_hal_cxy_to_almos_array(struct hal_bib_header_s *header)
{
        int i;
        cluster_info_t *cluster_tbl;

        cluster_tbl = (cluster_info_t*)((uint32_t)header + sizeof(header_info_t));
        assert(CLUSTER_NR >= header->onln_clstr_nr);

        for(i = 0; i < header->onln_clstr_nr; i++)
        {
                if(cluster_tbl[i].hal_cxy >= CLUSTER_NR)
                        continue;
                Arch_cxy_To_Almos_cxy[cluster_tbl[i].hal_cxy] = cluster_tbl[i].cxy;

        }
}

/*********************************************************************************************
 * This function makes the kernel virtual mapping of all hardware memory banks.
 * Initialize the boot table entries.
 ********************************************************************************************/
void hal_state_init( boot_info_t * info )
{
        struct process_s *process;
        struct thread_s *this;
        header_info_t *header;
        cluster_info_t *clusters;
        cluster_info_t *cluster_ptr;
        dev_info_t *dev_tbl;
        uint32_t vaddr_start;
        uint32_t vaddr_limit;
        uint32_t size;
        uint32_t cxy;
        uint32_t i;
        
        header = (header_info_t*) info->hal_info;
        

        clusters        = (cluster_info_t*) ((uint32_t)header + sizeof(header_info_t));
        cluster_ptr     = &clusters[info->local_cluster_id];
        dev_tbl         = (dev_info_t*)(cluster_ptr->offset + (uint32_t)header);
        cxy             = cluster_ptr->cxy;

        /* Global variables init */
        hal_onln_clusters_nr.value = header->onln_clstr_nr;
        hal_onln_cpus_nr.value     = header->onln_cpu_nr;
        hal_cpus_per_cluster.value = header->cpu_nr;
        hal_boot_cxy_.value         = info->boot_cluster_id;
        __current_cxy.value = cxy;
        fill_hal_cxy_to_almos_array(header);

        //used by remote_sb->hal_cxy_...
        for(i = 0; i < info->onln_clstr_nr; i++)
        {
                cluster_ptr = &clusters[i];
                cxy = cluster_ptr->cxy;
                hal_entrys[cxy].hal_cxy = cluster_ptr->hal_cxy;
        }

        cpu_gid_tbl_init(info);
        kboot_tty_init(info);

        size = dev_tbl[0].size;

        if(size == 0)
                die("[ERROR]\t%s: Unexpected memory size for cluster %u\n",             \
                                __FUNCTION__, cxy);

        this                         = CURRENT_THREAD;
        process                         = this->process;

        /* TODO: deal with offline clusters as well */
        for(i = 0; i < info->onln_clstr_nr; i++)
        {
                cluster_ptr = &clusters[i];
                cxy = cluster_ptr->cxy;
   
                if(cxy >= CLUSTER_NR)
                {
                        die("\n[ERROR]\t%s: This kernel version support up to %d clusters, found %d\n",         \
                                        __FUNCTION__, CLUSTER_NR, cxy);
                }

                dev_tbl = (dev_info_t*)(cluster_ptr->offset + (uint32_t)header);
                if(dev_tbl[0].id == SOCLIB_RAM_ID)
                {
                        vaddr_start     = dev_tbl[0].base;
                        size            = dev_tbl[0].size;
                }else
                {
                        vaddr_start = 0;
                        size = 0;
                }

                if((cluster_ptr->cpu_nr != 0) && (size == 0))
                        die("\n[ERROR]\t%s: This Kernel Version Do Not Support CPU-Only Clusters, cxy %d\n",    \
                                        __FUNCTION__, cxy);

                vaddr_limit          = vaddr_start + size;

                if(cxy == info->local_cluster_id)
                {
                        process->vmm.limit_addr         = vaddr_limit; 
                }


                boot_dmsg("[INFO]\tHardware initialization of cluster %u\t\t\t\t[ %d ]\n",          \
                                cxy, cpu_time_stamp());

                cluster_entry_init(cxy, vaddr_start, size);
        }
}

/*********************************************************************************************
 * This function makes the architecture specific initialisation.
 * It makes the local cluster initialisation, dynamically detects the cluster devices,
 * and associates the appropriate drivers.
 ********************************************************************************************/
void hal_init( boot_info_t * info )
{
        cluster_t      * cluster;
        header_info_t  * header;
        cluster_info_t * clusters;
        cluster_info_t * cluster_ptr;
        dev_info_t     * dev_tbl;
        cxy_t            iopic_cxy;     // should be the IO_cluster
        bool_t           iopic_found;
        error_t          error;
        uint32_t         rcxy;
        cxy_t            cxy;
        uint32_t         i;
  
/* Local variables init 
        header = (header_info_t *)info->hal_info;

        if( strncmp( header->signature , hal_bib_signature , 16 ) )
    {
                while(1);                       // TODO no TTY yet ??? [AG]
    }
        if( strncmp( header->arch , "TSAR" , 16))
    {
                while(1);           // TODO no TTY yet ??? [AG]
    }
*/
        iopic_found = false;
        iopic_cxy   = 0;
        clusters    = (cluster_info_t*)((uint32_t)header + sizeof(header_info_t));
        cluster_ptr = &clusters[info->local_cluster_id];
        dev_tbl     = (dev_info_t*)(cluster_ptr->offset + (uint32_t)header);
        cxy         = cluster_ptr->cxy;

        hal_state_init(info);

        /* init cluster: also init current_{cxy, cluster} macros */
        error = cluster_init(info, 
                           dev_tbl[0].base, 
                           dev_tbl[0].base + dev_tbl[0].size,
                           clusters_tbl[cxy].vmem_start);
        
        if(err)
                die("ERROR: Failed To Initialize Cluster %d, Err %d\n", cxy, err);

        cluster           = current_cluster;
        //TODO: move it to cluster.c
        //!!! THis value include the io_cluster !
        cluster->clstr_nr = info->onln_clstr_nr; // TODO: headr->x_max * header->y_max;
  
        for(i = 0; i < info->onln_clstr_nr; i++)
        {
                cluster_ptr = &clusters[i];

                dev_tbl     = (dev_info_t*)(cluster_ptr->offset + (uint32_t)header);

                rcxy = cluster_ptr->cxy;

                boot_dmsg("\n[INFO]\tDevices initialization for cluster %d\t\t\t\t[ %d ]\n",    \
                                rcxy, cpu_time_stamp());

                if(hal_dev_init(cluster, &clusters_tbl[rcxy], dev_tbl, cluster_ptr))
                {
                        iopic = true;
                        iopic_cxy = rcxy;
                }
        }

        if(iopic) //&& (cluster->io_clstr == cluster->id)) // Now it's dynamique
                hal_dev_iopic_register(cluster, &clusters_tbl[iopic_cxy]);

}  // end hal_init()


/*********************************************************************************************
 * This function initialises all devices descriptors associated to peripherals
 * contained in the local cluster, as specified by the boot_info structure.
 *********************************************************************************************
 * @ info    : pointer on the local boot-info structure.
 * @ returns 0 if success / returns ENOMEM or EINVAL if error
 ********************************************************************************************/
error_t hal_devices_init( boot_info_t * info )
{
        drvdb_entry_t * entry;
        list_entry_t  * devlist;        

    boot_device_t * dev_tbl;         // pointer on array of devices in boot_info
        uint32_t        dev_nr;          // actual number of devices in this cluster
    cluster_t     * cluster;         // pointer on local cluster manager
        cxy_t           local_cxy;       // local cluster identifier
    cxy-t           target_cxy;      // target cluster identifier
        device_t      * device;          // pointer on current device descriptor
        xptr_t          base;            // remote pointer on segment base
        uint32_t        size;            // segment size (bytes)
    dev_type_t      type;            // device type

        uint32_t        i;
        bool_t          pic_found;
        driver_t      * driver;
        error_t         error;
  
    // get local cluster pointer and identifier
    cluster = LOCAL_CLUSTER;
    cxy     = cluster->cxy;

    // get number of peripherals
        dev_nr      = info->devices_nr;
    dev_tbl     = info->dev;

    // loop on all peripherals in cluster
        for( i = 0 ; i < dev_nr ; i++ )
        {
        type = dev_tbl[i].type;

        // skip RAM and ROM
                if( (type == DEV_TYPE_RAM) || (type == DEV_TYPE_ROM) ) continue;

        // test PIC
                if( dev_tbl[i].type == DEV_TYPE_PIC )  pic_found = true;

        // replicated
        if( (type == DEV_TYPE_XCU ) || (type == DEV_TYPE_MMC) || (type == DEV_TYPE_MWR) )
        {
                    device = device_alloc( local);
                    if( device == NULL )
            {
                            printk(PANIC,"%s failed to allocate device descriptor in cluster %x\n",
                   __FUNCTION__ , cxy );
                return ENOMEM;
            }
        }
        else
        {

                driver          = drvdb_entry_get_driver(entry);

        // initialise device descriptor from boot_info
                spinlock_init( &device->lock );

        device->type    = dev_tbl[i].type;
                device->base    = dev_tbl[i].base; 
        device->size    = dev_tbl[i].size;
                device->isr         = dev_tbl[i].irq;

                error = driver->init( dev );

                if( error )
        {
                        die("[ERROR]\tFailed To Initialize Device %s [Cluster %d, Dev %d, Err %d]\n",       
                            drvdb_entry_get_name(entry), rcxy, i, err);
        }

                devfs_register( dev );
                list_add_last( devlist , &dev->list);

                boot_dmsg("[INFO]\tFound Device: %s\t\t\t\t\t\t[ %d ]\n[INFO]\t\tBase <0x%x> cxy %d Size <0x%x> Irq <%d>\t\t[ %d ]\n", \
                          drvdb_entry_get_name(entry), cpu_time_stamp(), (uint32_t)dev_base,           \
                          rcxy, dev_tbl[i].size, dev_tbl[i].irq, cpu_time_stamp());
        } // end loop on peripherals

        // register device descriptor in global list ???        
        hal_dev_register( cluster , centry );  // TODO ???

        return iopic_clstr;

}  // end hal_devices_init()



/*********************************************************************************************
 ********************************************************************************************/
void hal_xicu_set_wti(struct device_s *xicu, 
                        uint32_t cpu_inter, uint32_t wti_index, 
                        struct irq_action_s *action, 
                        struct device_s *dev)
{
        uint64_t mailbox;
        uint32_t irq_out;
        error_t err;

        mailbox = xicu->base_paddr + (wti_index << 2);
        irq_out = cpu_inter*OUTPUT_IRQ_PER_PROC;


        xicu->op.icu.set_mask(xicu,
                                (1 << wti_index),
                                XICU_MSK_WTI_ENABLE,
                                irq_out);

        boot_dmsg("[INFO]\t\t WTI %d <-> CPU %d (mailbox 0x%x%x)\t\t\t[ %d ]\n",        \
                        wti_index, cpu_inter, MSB(mailbox), LSB(mailbox),               \
                        cpu_time_stamp());

        err = xicu->op.icu.bind_wti(xicu, action, wti_index);

          
        if(err != 0) 
                die("[ERROR]\tFailed to bind %s via wti_mailbox 0x%x:%x , on xicu %s @%x [ err %d ]\n", \
                        dev ? dev->name : "IPI", MSB(mailbox),LSB(mailbox), xicu->name, xicu, err); 
          
        if(dev)
                dev->mailbox = mailbox;
        //The binding is (to be) done dynamically in the peripheral
        //iopic->op.iopic.bind_wti_and_irq(iopic, mailbox, dev->irq);
                        
}

//TODO: distibute iopic devices on multiple cluster ?
//For now all the devices of the IOPIC_Cluster (cluster 
//without cpus) and there IRQs are handled by the IO_Cluster.
void hal_dev_iopic_register(struct cluster_s *cluster, 
                        struct cluster_entry_s *entry)
{
        struct list_entry *iter;
        struct device_s *iopic;
        struct device_s *xicu;
        struct device_s *dev;
        uint32_t device_index;
        uint32_t wti_index;
        uint32_t cpu_inter;
        uint32_t cpu_nr;
        //error_t err;

        cpu_inter = 0;
        device_index = 0;
        cpu_nr = cluster->cpu_nr;
        wti_index = cpu_nr;//Leave one WTI per proc for IPI

        xicu = hal_entrys[cluster->id].xicu;
        iopic = hal_dev_locate(&entry->devlist, SOCLIB_IOPIC_ID);

        if(xicu == NULL)
                die("[ERROR]\tNo XICU Is Found for IOPIC cluster %d\n",                         \
                                entry->cxy);

        boot_dmsg("\n[INFO]\tIOPIC cluster configuration: cxy %d <0x%x,0x%x>\t[ %d ]\n",      \
                        entry->cxy, iopic->base, (iopic->base+iopic->size), cpu_time_stamp());

        list_foreach(&entry->devlist, iter)
        {
                dev = list_element(iter, struct device_s, list);

                if(wti_index > (IOPIC_WTI_PER_CLSTR - cpu_nr))
                        die("[ERROR]\tNo enough WTI in cluster %d\n", cluster->id);
                
                        
                if((dev != iopic) && (dev->irq != -1))
                {
                        boot_dmsg("[INFO]\tLinking device %s to XICU on cluster %d through IOPIC cluster\t[ %d ]\n",  \
                                        dev->name, xicu->cxy, cpu_time_stamp());

                        hal_xicu_set_wti(xicu, device_index%cpu_nr, wti_index, &dev->action, dev);

                        dev->iopic = iopic;

                        wti_index++;
                        device_index++;
                }

                devfs_register(dev);
        }
} // end hal_xcu_set_wti()