/**
* Allocates one page.
*
* @param virtAddr The virtual address to which this page maybe mapped in
* the future.
*
* @returns Pointer to the allocated page, NULL on failure.
*/
static page_t *rtR0MemObjSolPageAlloc(caddr_t virtAddr)
{
u_offset_t offPage;
seg_t KernelSeg;
/*
* 16777215 terabytes of total memory for all VMs or
* restart 8000 1GB VMs 2147483 times until wraparound!
*/
mutex_enter(&g_OffsetMtx);
AssertCompileSize(u_offset_t, sizeof(uint64_t)); NOREF(RTASSERTVAR);
g_offPage = RT_ALIGN_64(g_offPage, PAGE_SIZE) + PAGE_SIZE;
offPage = g_offPage;
mutex_exit(&g_OffsetMtx);
KernelSeg.s_as = &kas;
page_t *pPage = page_create_va(&g_PageVnode, offPage, PAGE_SIZE, PG_WAIT | PG_NORELOC, &KernelSeg, virtAddr);
if (RT_LIKELY(pPage))
{
/*
* Lock this page into memory "long term" to prevent this page from being paged out
* when we drop the page lock temporarily (during free). Downgrade to a shared lock
* to prevent page relocation.
*/
page_pp_lock(pPage, 0 /* COW */, 1 /* Kernel */);
page_io_unlock(pPage);
page_downgrade(pPage);
Assert(PAGE_LOCKED_SE(pPage, SE_SHARED));
}
return pPage;
}
/*ARGSUSED*/
page_t *
segkmem_page_create(void *addr, size_t size, int vmflag, void *arg)
{
struct seg kseg;
int pgflags;
struct vnode *vp = arg;
if (vp == NULL)
vp = &kvp;
kseg.s_as = &kas;
pgflags = PG_EXCL;
if (segkmem_reloc == 0 || (vmflag & VM_NORELOC))
pgflags |= PG_NORELOC;
if ((vmflag & VM_NOSLEEP) == 0)
pgflags |= PG_WAIT;
if (vmflag & VM_PANIC)
pgflags |= PG_PANIC;
if (vmflag & VM_PUSHPAGE)
pgflags |= PG_PUSHPAGE;
if (vmflag & VM_NORMALPRI) {
ASSERT(vmflag & VM_NOSLEEP);
pgflags |= PG_NORMALPRI;
}
return (page_create_va(vp, (u_offset_t)(uintptr_t)addr, size,
pgflags, &kseg, addr));
}
/*ARGSUSED*/
static int
bootfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, enum seg_rw rw,
cred_t *cr)
{
bootfs_node_t *bnp = vp->v_data;
page_t *pp, *fpp;
pfn_t pfn;
for (;;) {
/* Easy case where the page exists */
pp = page_lookup(vp, off, rw == S_CREATE ? SE_EXCL : SE_SHARED);
if (pp != NULL) {
if (pl != NULL) {
pl[0] = pp;
pl[1] = NULL;
} else {
page_unlock(pp);
}
return (0);
}
pp = page_create_va(vp, off, PAGESIZE, PG_EXCL | PG_WAIT, seg,
addr);
/*
* If we didn't get the page, that means someone else beat us to
* creating this so we need to try again.
*/
if (pp != NULL)
break;
}
pfn = btop((bnp->bvn_addr + off) & PAGEMASK);
fpp = page_numtopp_nolock(pfn);
if (ppcopy(fpp, pp) == 0) {
pvn_read_done(pp, B_ERROR);
return (EIO);
}
if (pl != NULL) {
pvn_plist_init(pp, pl, plsz, off, PAGESIZE, rw);
} else {
pvn_io_done(pp);
}
return (0);
}
/*
* Find the largest contiguous block which contains `addr' for file offset
* `offset' in it while living within the file system block sizes (`vp_off'
* and `vp_len') and the address space limits for which no pages currently
* exist and which map to consecutive file offsets.
*/
page_t *
pvn_read_kluster(
struct vnode *vp,
u_offset_t off,
struct seg *seg,
caddr_t addr,
u_offset_t *offp, /* return values */
size_t *lenp, /* return values */
u_offset_t vp_off,
size_t vp_len,
int isra)
{
ssize_t deltaf, deltab;
page_t *pp;
page_t *plist = NULL;
spgcnt_t pagesavail;
u_offset_t vp_end;
ASSERT(off >= vp_off && off < vp_off + vp_len);
/*
* We only want to do klustering/read ahead if there
* is more than minfree pages currently available.
*/
pagesavail = freemem - minfree;
if (pagesavail <= 0)
if (isra)
return ((page_t *)NULL); /* ra case - give up */
else
pagesavail = 1; /* must return a page */
/* We calculate in pages instead of bytes due to 32-bit overflows */
if (pagesavail < (spgcnt_t)btopr(vp_len)) {
/*
* Don't have enough free memory for the
* max request, try sizing down vp request.
*/
deltab = (ssize_t)(off - vp_off);
vp_len -= deltab;
vp_off += deltab;
if (pagesavail < btopr(vp_len)) {
/*
* Still not enough memory, just settle for
* pagesavail which is at least 1.
*/
vp_len = ptob(pagesavail);
}
}
vp_end = vp_off + vp_len;
ASSERT(off >= vp_off && off < vp_end);
if (isra && SEGOP_KLUSTER(seg, addr, 0))
return ((page_t *)NULL); /* segment driver says no */
if ((plist = page_create_va(vp, off,
PAGESIZE, PG_EXCL | PG_WAIT, seg, addr)) == NULL)
return ((page_t *)NULL);
if (vp_len <= PAGESIZE || pvn_nofodklust) {
*offp = off;
*lenp = MIN(vp_len, PAGESIZE);
} else {
/*
* Scan back from front by incrementing "deltab" and
* comparing "off" with "vp_off + deltab" to avoid
* "signed" versus "unsigned" conversion problems.
*/
for (deltab = PAGESIZE; off >= vp_off + deltab;
deltab += PAGESIZE) {
/*
* Call back to the segment driver to verify that
* the klustering/read ahead operation makes sense.
*/
if (SEGOP_KLUSTER(seg, addr, -deltab))
break; /* page not eligible */
if ((pp = page_create_va(vp, off - deltab,
PAGESIZE, PG_EXCL, seg, addr - deltab))
== NULL)
break; /* already have the page */
/*
* Add page to front of page list.
*/
page_add(&plist, pp);
}
deltab -= PAGESIZE;
/* scan forward from front */
for (deltaf = PAGESIZE; off + deltaf < vp_end;
deltaf += PAGESIZE) {
/*
* Call back to the segment driver to verify that
* the klustering/read ahead operation makes sense.
*/
if (SEGOP_KLUSTER(seg, addr, deltaf))
break; /* page not file extension */
if ((pp = page_create_va(vp, off + deltaf,
PAGESIZE, PG_EXCL, seg, addr + deltaf))
== NULL)
break; /* already have page */
/*
* Add page to end of page list.
*/
page_add(&plist, pp);
plist = plist->p_next;
}
*offp = off = off - deltab;
*lenp = deltab + deltaf;
ASSERT(off >= vp_off);
/*
* If we ended up getting more than was actually
* requested, retract the returned length to only
* reflect what was requested. This might happen
* if we were allowed to kluster pages across a
* span of (say) 5 frags, and frag size is less
* than PAGESIZE. We need a whole number of
* pages to contain those frags, but the returned
* size should only allow the returned range to
* extend as far as the end of the frags.
*/
if ((vp_off + vp_len) < (off + *lenp)) {
ASSERT(vp_end > off);
*lenp = vp_end - off;
}
}
TRACE_3(TR_FAC_VM, TR_PVN_READ_KLUSTER,
"pvn_read_kluster:seg %p addr %x isra %x",
seg, addr, isra);
return (plist);
}
