matos/arch/x86/paging.c

#include "allocArea.h"
#include "paging.h"
#include "errno.h"
#include "kernel.h"
#include "klibc.h"
#include "mem.h"
#include "mmuContext.h"
#include "stdarg.h"

// In a Vaddr, 10 first bit (MSB) are the index in the Page Directory. A Page Directory Entry
// point to a Page Table. The 10 next bits are then an index in this Page Table. A Page Table
// Entry then point to a physical address at which is added the remaining 12 bits. So they are
// 1024 entry in the PD, each of them pointing to a PT of 1024 entry. Each PTE pointing to 4K
// page. First address (up to page_desc from mem.c) are mapped such as Paddr == Vaddr. To make
// PD always accessible a (x86?) trick is used : The mirroring. A given entry N in the PD point
// to the PD (this is possible because PDE very looks like PTE in x86). So N << (10 + 12 = 4Mo)
// point to the Paddr of PD. Then, accessing N * 4Mo + I * 4Ko is accessing the PT of the Ieme
// entry in the PD (as MMU take the PD pointed by the PDE number N like a PT). More
// particularly, accessing N * 4Mo + N * 4ko is accessing the PD.
//
// PD is at Vaddr N * 4Mo and take 4ko. Each PT are allocated dynamically.
// Just make sure that N have not been used by identity mapping

#define PT_SHIFT 12
#define PTE_MASK 0x3ff // 10bits
#define PD_SHIFT 22
#define PD_MIRROR_PAGE_IDX 1023U

static unsigned long mappedPage = 0;

struct pde {
    uint32_t present : 1;
    uint32_t write : 1;         // 0 read - 1 RW
    uint32_t user : 1;          // 0 supervisor - 1 user
    uint32_t write_through : 1; // 0 write-back - 1 write_through
    uint32_t cache_disable : 1;
    uint32_t access : 1; // have been accessed
    uint32_t zero : 1;   // Not used
    uint32_t size : 1;   // 0 for 4Kb 1 for 4Mb
    uint32_t ignored : 1;
    uint32_t available : 3;
    uint32_t pt_addr : 20;
} __attribute__((packed));

struct pte {
    uint32_t present : 1;
    uint32_t write : 1;         // 0 read - 1 RW
    uint32_t user : 1;          // 0 supervisor - 1 user
    uint32_t write_through : 1; // 0 write-back - 1 write_through
    uint32_t cache_disable : 1;
    uint32_t access : 1; // have been accessed
    uint32_t dirty : 1;  // if set, indicates that page has been written to. This flag is
                         // not updated by the CPU, and once set will not unset itself.
    uint32_t zero : 1;   // if PAT is supported, shall indicate the memory type. Otherwise,
                         // it must be 0.
    uint32_t global : 1; // if set, prevents the TLB from updating the address in its cache
                         // if CR3 is reset. Note, that the page global enable bit in CR4
                         // must be set to enable this feature.
    uint32_t available : 3;
    uint32_t paddr : 20;
} __attribute__((packed));

struct pdbr {
    uint32_t zero1 : 3;          // reserved
    uint32_t write_through : 1;  // 0 write-back - 1 write-through
    uint32_t cache_disabled : 1; // 1=cache disabled
    uint32_t zero2 : 7;          // reserved
    uint32_t pd_paddr : 20;
} __attribute__((packed));

// invalidate the TLB entry for the page located at the given virtual address
static inline void __native_flush_tlb_single(unsigned long addr)
{
    asm volatile("invlpg (%0)" ::"r"(addr) : "memory");
}

int pagingSetup(paddr_t lowerKernelAddr, paddr_t upperKernelAddr)
{
    struct pdbr cr3;

    // x86 got 1024 of pde for 4Byte each: 4ko !
    struct pde *pd = (struct pde *)allocPhyPage(1);

    memset(pd, 0, PAGE_SIZE);
    memset(&cr3, 0x0, sizeof(struct pdbr));

    cr3.pd_paddr = ((paddr_t)pd) >> 12;

    // MMU not enabled for the moment. No need to use mirroring
    // Identity mapping up to upperKernelAddr
    for (paddr_t i = lowerKernelAddr; i < upperKernelAddr; i += PAGE_SIZE) {
        uint pdEntry = i >> (PD_SHIFT);
        uint ptEntry = (i >> PT_SHIFT) & PTE_MASK;
        struct pte *pt;
        if (pd[pdEntry].present) {
            pt = (struct pte *)(pd[pdEntry].pt_addr << PT_SHIFT);
            refPhyPage((paddr_t)pt);
        } else {
            pt = (struct pte *)allocPhyPage(1);

            memset(pt, 0, PAGE_SIZE);
            pd[pdEntry].present = 1;
            pd[pdEntry].write   = 1;
            pd[pdEntry].pt_addr = ((paddr_t)pt >> PT_SHIFT);
        }

        pt[ptEntry].present = 1;
        pt[ptEntry].write   = 1; // TODO set Kernel code as RO
        pt[ptEntry].paddr   = i >> PAGE_SHIFT;
    }

    // Setup mirroring
    pd[PAGING_MIRROR_VADDR >> PD_SHIFT].present = 1;
    pd[PAGING_MIRROR_VADDR >> PD_SHIFT].write   = 1;
    pd[PAGING_MIRROR_VADDR >> PD_SHIFT].pt_addr = ((paddr_t)pd >> PT_SHIFT);
    pd[PAGING_MIRROR_VADDR >> PD_SHIFT].user    = 0;

    // Loading of the PDBR in the MMU:
    asm volatile("movl %0,%%cr3\n\t"
                 "movl %%cr0,%%eax\n\t"
                 "orl $0x80010000, %%eax\n\t" /* bit 31 | bit 16 */
                 "movl %%eax,%%cr0\n\t"
                 "jmp 1f\n\t"
                 "1:\n\t"
                 "movl $2f, %%eax\n\t"
                 "jmp *%%eax\n\t"
                 "2:\n\t" ::"r"(cr3)
                 : "memory", "eax");
    return 0;
}

int pageMap(vaddr_t vaddr, paddr_t paddr, int flags)
{
    uint pdEntry = vaddr >> (PD_SHIFT);
    uint ptEntry = (vaddr >> PT_SHIFT) & PTE_MASK;

    if ((vaddr >= PAGING_MIRROR_VADDR) && (vaddr < PAGING_MIRROR_VADDR + PAGING_MIRROR_SIZE))
        return -EINVAL;

    // Thank to mirroring, we can access the PD
    struct pde *pd =
        (struct pde *)(PAGING_MIRROR_VADDR + PAGE_SIZE * (PAGING_MIRROR_VADDR >> PD_SHIFT));

    struct pte *pt = (struct pte *)((PAGING_MIRROR_VADDR) + (pdEntry * PAGE_SIZE));

    if (!pd[pdEntry].present) {
        paddr_t ptPhy = allocPhyPage(1);
        if (ptPhy == (vaddr_t)NULL)
            return -ENOMEM;

        pd[pdEntry].present = 1;
        pd[pdEntry].write   = 1;
        pd[pdEntry].pt_addr = (ptPhy >> PT_SHIFT);

        if (vaddr < PAGING_BASE_USER_ADDRESS) {
            pd[pdEntry].user = 0;
            mmuContextSyncKernelPDE(pdEntry, &pd[pdEntry], sizeof(struct pde));
        } else {
            assert(flags & PAGING_MEM_USER);
            pd[pdEntry].user = 1;
        }

        __native_flush_tlb_single((vaddr_t)pt);
        memset((void *)pt, 0, PAGE_SIZE);
    }
    {

        // Already mapped ? Remove old mapping
        if (pt[ptEntry].present) {
            unrefPhyPage(pt[ptEntry].paddr << PAGE_SHIFT);
        } // PTE not already used ? We will use it ! So increase the PT ref count
        else {
            refPhyPage(pd[pdEntry].pt_addr << PAGE_SHIFT);
        }
    }

    pt[ptEntry].user    = (flags & PAGING_MEM_USER) ? 1 : 0;
    pt[ptEntry].present = 1;
    pt[ptEntry].write   = (flags & PAGING_MEM_WRITE) ? 1 : 0;
    pt[ptEntry].paddr   = paddr >> PAGE_SHIFT;
    refPhyPage(paddr);

    __native_flush_tlb_single(vaddr);
    mappedPage++;
    return 0;
}

int pageUnmap(vaddr_t vaddr)
{
    uint pdEntry = vaddr >> (PD_SHIFT);
    uint ptEntry = (vaddr >> PT_SHIFT) & PTE_MASK;

    if ((vaddr >= PAGING_MIRROR_VADDR) && (vaddr < PAGING_MIRROR_VADDR + PAGING_MIRROR_SIZE))
        return -EINVAL;

    // Thank to mirroring, we can access the PD
    struct pde *pd =
        (struct pde *)(PAGING_MIRROR_VADDR + PAGE_SIZE * (PAGING_MIRROR_VADDR >> PD_SHIFT));

    struct pte *pt = (struct pte *)((PAGING_MIRROR_VADDR) + (pdEntry * PAGE_SIZE));
    if (!pd[pdEntry].present)
        return -EINVAL;
    if (!pt[ptEntry].present)
        return -EINVAL;

    unrefPhyPage(pt[ptEntry].paddr << PAGE_SHIFT);
    pt[ptEntry].present = 0;

    // PTE not used. Decrease refcount on it. Is PT not used anymore ?
    if (unrefPhyPage(pd[pdEntry].pt_addr << PT_SHIFT) == 0) {
        pd[pdEntry].present = 0;
        if (vaddr < PAGING_BASE_USER_ADDRESS) {
            mmuContextSyncKernelPDE(pdEntry, &pd[pdEntry], sizeof(struct pde));
        }
        __native_flush_tlb_single((vaddr_t)pt);
    }
    __native_flush_tlb_single(vaddr);
    mappedPage--;
    return 0;
}

paddr_t pagingGetPaddr(vaddr_t vaddr)
{
    /* Get the page directory entry and table entry index for this
       address */
    unsigned pdEntry    = vaddr >> PD_SHIFT;
    unsigned ptEntry    = vaddr >> PT_SHIFT & PTE_MASK;
    unsigned pageOffset = vaddr & PAGE_MASK;

    // Thank to mirroring, we can access the PD
    struct pde *pd =
        (struct pde *)(PAGING_MIRROR_VADDR + PAGE_SIZE * (PAGING_MIRROR_VADDR >> PD_SHIFT));

    struct pte *pt = (struct pte *)((PAGING_MIRROR_VADDR) + (pdEntry * PAGE_SIZE));

    /* No page mapped at this address ? */
    if (!pd[pdEntry].present)
        return (paddr_t)NULL;
    if (!pt[ptEntry].present)
        return (paddr_t)NULL;

    return (pt[ptEntry].paddr << PT_SHIFT) + pageOffset;
}

unsigned long getNbMappedPage(void)
{
    return mappedPage;
}

paddr_t pagingGetCurrentPDPaddr()
{
    struct pdbr pdbr;
    asm volatile("movl %%cr3, %0\n" : "=r"(pdbr));
    return (pdbr.pd_paddr << 12);
}

int pagingSetCurrentPDPaddr(paddr_t paddrPD)
{
    struct pdbr pdbr;

    assert(paddrPD != 0);
    assert(IS_ALIGNED(paddrPD, PAGE_SIZE));

    /* Setup the value of the PDBR */
    memset(&pdbr, 0x0, sizeof(struct pdbr)); /* Reset the PDBR */
    pdbr.pd_paddr = (paddrPD >> 12);

    /* Configure the MMU according to the PDBR */
    asm volatile("movl %0,%%cr3\n" ::"r"(pdbr));

    return 0;
}

// unmap page inside this MMU context
int pagingClearUserContext(vaddr_t vaddr_PD)
{
    struct pde *pd = (struct pde *)vaddr_PD;
    //Tmp pt to unref page they reference
    struct pte *pt = (struct pte *)areaAlloc(1, 0);

    if(pt == NULL)
        return -ENOMEM;

    for (int pdIdx = PAGING_BASE_USER_ADDRESS >> PD_SHIFT; pdIdx < 1024; pdIdx++) {

        if(!pd[pdIdx].present){
            memset(&pd[pdIdx], 0, sizeof(struct pde));

            continue;
        }
        paddr_t ptAddr = pd[pdIdx].pt_addr << PT_SHIFT;
        assert(!pageMap((vaddr_t)pt, ptAddr, PAGING_MEM_WRITE | PAGING_MEM_READ));
        for(int ptIdx = 0; ptIdx < 1024; ptIdx ++){
            if(!pt[ptIdx].present){
                memset(&pt[ptIdx], 0, sizeof(struct pte));

                continue;
            }
            unrefPhyPage(pt[ptIdx].paddr);
            memset(&pt[ptIdx], 0, sizeof(struct pte));
        }
        assert(!pageUnmap((vaddr_t)pt));
        memset(&pd[pdIdx], 0, sizeof(struct pde));

        unrefPhyPage(ptAddr);
    }

    areaFree((vaddr_t)pt);

    return 0;
}

int pagingCopyKernelSpace(vaddr_t destVaddrPD, paddr_t destPaddrPD, vaddr_t srcVaddrPD)
{
    struct pde *src_pd  = (struct pde *)srcVaddrPD;
    struct pde *dest_pd = (struct pde *)destVaddrPD;
    struct pde mirror_pde = {0};
    uint index_in_pd;

    /* Fill destination PD with zeros */
    memset((void *)destVaddrPD, 0x0, PAGE_SIZE);

    /* Synchronize it with the master Kernel MMU context. Stop just
       before the mirroring ! */
    for (index_in_pd = 0; index_in_pd < (PAGING_MIRROR_VADDR >> 22); /* 1 PDE = 1 PT
                                                = 1024 Pages
                                                = 4MB */
         index_in_pd++) {
        /* Copy the master's configuration */
        dest_pd[index_in_pd] = src_pd[index_in_pd];

        /* We DON'T mark the underlying PT and pages as referenced
       because all the PD are equivalent in the kernel space: as
       soon as a page is mapped in the kernel, it is mapped by X
       address spaces, and as soon as it is unmapped by 1 address
       space, it is unmapped in all the others. So that for X
       address spaces, the reference counter will be either 0 or X,
       and not something else: using the reference counter correctly
       won't be of any use and would consume some time in updating it. */
    }

    /* Setup the mirroring for the new address space */
    mirror_pde.present                 = TRUE;
    mirror_pde.write                   = 1;
    mirror_pde.user                    = 0; /* This is a KERNEL PDE */
    mirror_pde.pt_addr                 = (destPaddrPD >> 12);
    dest_pd[PAGING_MIRROR_VADDR >> 22] = mirror_pde;

    return 0;
}