/*
* Copy memory by briefly enabling incoherent cacheline-at-a-time mode.
*
* We set up our own source and destination PTEs that we fully control.
* This is the only way to guarantee that we don't race with another
* thread that is modifying the PTE; we can't afford to try the
* copy_{to,from}_user() technique of catching the interrupt, since
* we must run with interrupts disabled to avoid the risk of some
* other code seeing the incoherent data in our cache. (Recall that
* our cache is indexed by PA, so even if the other code doesn't use
* our kmap_atomic virtual addresses, they'll still hit in cache using
* the normal VAs that aren't supposed to hit in cache.)
*/
static void memcpy_multicache(void *dest, const void *source,
pte_t dst_pte, pte_t src_pte, int len)
{
int idx;
unsigned long flags, newsrc, newdst;
pmd_t *pmdp;
pte_t *ptep;
int type0, type1;
int cpu = get_cpu();
/*
* Disable interrupts so that we don't recurse into memcpy()
* in an interrupt handler, nor accidentally reference
* the PA of the source from an interrupt routine. Also
* notify the simulator that we're playing games so we don't
* generate spurious coherency warnings.
*/
local_irq_save(flags);
sim_allow_multiple_caching(1);
/* Set up the new dest mapping */
type0 = kmap_atomic_idx_push();
idx = FIX_KMAP_BEGIN + (KM_TYPE_NR * cpu) + type0;
newdst = __fix_to_virt(idx) + ((unsigned long)dest & (PAGE_SIZE-1));
pmdp = pmd_offset(pud_offset(pgd_offset_k(newdst), newdst), newdst);
ptep = pte_offset_kernel(pmdp, newdst);
if (pte_val(*ptep) != pte_val(dst_pte)) {
set_pte(ptep, dst_pte);
local_flush_tlb_page(NULL, newdst, PAGE_SIZE);
}
/* Set up the new source mapping */
type1 = kmap_atomic_idx_push();
idx += (type0 - type1);
src_pte = hv_pte_set_nc(src_pte);
src_pte = hv_pte_clear_writable(src_pte); /* be paranoid */
newsrc = __fix_to_virt(idx) + ((unsigned long)source & (PAGE_SIZE-1));
pmdp = pmd_offset(pud_offset(pgd_offset_k(newsrc), newsrc), newsrc);
ptep = pte_offset_kernel(pmdp, newsrc);
__set_pte(ptep, src_pte); /* set_pte() would be confused by this */
local_flush_tlb_page(NULL, newsrc, PAGE_SIZE);
/* Actually move the data. */
__memcpy_asm((void *)newdst, (const void *)newsrc, len);
/*
* Remap the source as locally-cached and not OLOC'ed so that
* we can inval without also invaling the remote cpu's cache.
* This also avoids known errata with inv'ing cacheable oloc data.
*/
src_pte = hv_pte_set_mode(src_pte, HV_PTE_MODE_CACHE_NO_L3);
src_pte = hv_pte_set_writable(src_pte); /* need write access for inv */
__set_pte(ptep, src_pte); /* set_pte() would be confused by this */
local_flush_tlb_page(NULL, newsrc, PAGE_SIZE);
/*
* Do the actual invalidation, covering the full L2 cache line
* at the end since __memcpy_asm() is somewhat aggressive.
*/
__inv_buffer((void *)newsrc, len);
/*
* We're done: notify the simulator that all is back to normal,
* and re-enable interrupts and pre-emption.
*/
kmap_atomic_idx_pop();
kmap_atomic_idx_pop();
sim_allow_multiple_caching(0);
local_irq_restore(flags);
put_cpu();
}
static void memcpy_multicache(void *dest, const void *source,
pte_t dst_pte, pte_t src_pte, int len)
{
int idx;
unsigned long flags, newsrc, newdst;
pmd_t *pmdp;
pte_t *ptep;
int type0, type1;
int cpu = get_cpu();
/*
*/
local_irq_save(flags);
sim_allow_multiple_caching(1);
/* */
type0 = kmap_atomic_idx_push();
idx = FIX_KMAP_BEGIN + (KM_TYPE_NR * cpu) + type0;
newdst = __fix_to_virt(idx) + ((unsigned long)dest & (PAGE_SIZE-1));
pmdp = pmd_offset(pud_offset(pgd_offset_k(newdst), newdst), newdst);
ptep = pte_offset_kernel(pmdp, newdst);
if (pte_val(*ptep) != pte_val(dst_pte)) {
set_pte(ptep, dst_pte);
local_flush_tlb_page(NULL, newdst, PAGE_SIZE);
}
/* */
type1 = kmap_atomic_idx_push();
idx += (type0 - type1);
src_pte = hv_pte_set_nc(src_pte);
src_pte = hv_pte_clear_writable(src_pte); /* */
newsrc = __fix_to_virt(idx) + ((unsigned long)source & (PAGE_SIZE-1));
pmdp = pmd_offset(pud_offset(pgd_offset_k(newsrc), newsrc), newsrc);
ptep = pte_offset_kernel(pmdp, newsrc);
__set_pte(ptep, src_pte); /* */
local_flush_tlb_page(NULL, newsrc, PAGE_SIZE);
/* */
__memcpy_asm((void *)newdst, (const void *)newsrc, len);
/*
*/
src_pte = hv_pte_set_mode(src_pte, HV_PTE_MODE_CACHE_NO_L3);
src_pte = hv_pte_set_writable(src_pte); /* */
__set_pte(ptep, src_pte); /* */
local_flush_tlb_page(NULL, newsrc, PAGE_SIZE);
/*
*/
__inv_buffer((void *)newsrc, len);
/*
*/
kmap_atomic_idx_pop();
kmap_atomic_idx_pop();
sim_allow_multiple_caching(0);
local_irq_restore(flags);
put_cpu();
}
