/*
* Identify large copies from remotely-cached memory, and copy them
* via memcpy_multicache() if they look good, otherwise fall back
* to the particular kind of copying passed as the memcpy_t function.
*/
static unsigned long fast_copy(void *dest, const void *source, int len,
memcpy_t func)
{
/*
* Check if it's big enough to bother with. We may end up doing a
* small copy via TLB manipulation if we're near a page boundary,
* but presumably we'll make it up when we hit the second page.
*/
while (len >= LARGE_COPY_CUTOFF) {
int copy_size, bytes_left_on_page;
pte_t *src_ptep, *dst_ptep;
pte_t src_pte, dst_pte;
struct page *src_page, *dst_page;
/* Is the source page oloc'ed to a remote cpu? */
retry_source:
src_ptep = virt_to_pte(current->mm, (unsigned long)source);
if (src_ptep == NULL)
break;
src_pte = *src_ptep;
if (!hv_pte_get_present(src_pte) ||
!hv_pte_get_readable(src_pte) ||
hv_pte_get_mode(src_pte) != HV_PTE_MODE_CACHE_TILE_L3)
break;
if (get_remote_cache_cpu(src_pte) == smp_processor_id())
break;
src_page = pfn_to_page(pte_pfn(src_pte));
get_page(src_page);
if (pte_val(src_pte) != pte_val(*src_ptep)) {
put_page(src_page);
goto retry_source;
}
if (pte_huge(src_pte)) {
/* Adjust the PTE to correspond to a small page */
int pfn = pte_pfn(src_pte);
pfn += (((unsigned long)source & (HPAGE_SIZE-1))
>> PAGE_SHIFT);
src_pte = pfn_pte(pfn, src_pte);
src_pte = pte_mksmall(src_pte);
}
/* Is the destination page writable? */
retry_dest:
dst_ptep = virt_to_pte(current->mm, (unsigned long)dest);
if (dst_ptep == NULL) {
put_page(src_page);
break;
}
dst_pte = *dst_ptep;
if (!hv_pte_get_present(dst_pte) ||
!hv_pte_get_writable(dst_pte)) {
put_page(src_page);
break;
}
dst_page = pfn_to_page(pte_pfn(dst_pte));
if (dst_page == src_page) {
/*
* Source and dest are on the same page; this
* potentially exposes us to incoherence if any
* part of src and dest overlap on a cache line.
* Just give up rather than trying to be precise.
*/
put_page(src_page);
break;
}
get_page(dst_page);
if (pte_val(dst_pte) != pte_val(*dst_ptep)) {
put_page(dst_page);
goto retry_dest;
}
if (pte_huge(dst_pte)) {
/* Adjust the PTE to correspond to a small page */
int pfn = pte_pfn(dst_pte);
pfn += (((unsigned long)dest & (HPAGE_SIZE-1))
>> PAGE_SHIFT);
dst_pte = pfn_pte(pfn, dst_pte);
dst_pte = pte_mksmall(dst_pte);
}
/* All looks good: create a cachable PTE and copy from it */
copy_size = len;
bytes_left_on_page =
PAGE_SIZE - (((int)source) & (PAGE_SIZE-1));
if (copy_size > bytes_left_on_page)
copy_size = bytes_left_on_page;
bytes_left_on_page =
PAGE_SIZE - (((int)dest) & (PAGE_SIZE-1));
if (copy_size > bytes_left_on_page)
copy_size = bytes_left_on_page;
memcpy_multicache(dest, source, dst_pte, src_pte, copy_size);
/* Release the pages */
put_page(dst_page);
put_page(src_page);
/* Continue on the next page */
dest += copy_size;
source += copy_size;
len -= copy_size;
}
return func(dest, source, len);
}
static unsigned long fast_copy(void *dest, const void *source, int len,
memcpy_t func)
{
/*
*/
while (len >= LARGE_COPY_CUTOFF) {
int copy_size, bytes_left_on_page;
pte_t *src_ptep, *dst_ptep;
pte_t src_pte, dst_pte;
struct page *src_page, *dst_page;
/* */
retry_source:
src_ptep = virt_to_pte(current->mm, (unsigned long)source);
if (src_ptep == NULL)
break;
src_pte = *src_ptep;
if (!hv_pte_get_present(src_pte) ||
!hv_pte_get_readable(src_pte) ||
hv_pte_get_mode(src_pte) != HV_PTE_MODE_CACHE_TILE_L3)
break;
if (get_remote_cache_cpu(src_pte) == smp_processor_id())
break;
src_page = pfn_to_page(hv_pte_get_pfn(src_pte));
get_page(src_page);
if (pte_val(src_pte) != pte_val(*src_ptep)) {
put_page(src_page);
goto retry_source;
}
if (pte_huge(src_pte)) {
/* */
int pfn = hv_pte_get_pfn(src_pte);
pfn += (((unsigned long)source & (HPAGE_SIZE-1))
>> PAGE_SHIFT);
src_pte = pfn_pte(pfn, src_pte);
src_pte = pte_mksmall(src_pte);
}
/* */
retry_dest:
dst_ptep = virt_to_pte(current->mm, (unsigned long)dest);
if (dst_ptep == NULL) {
put_page(src_page);
break;
}
dst_pte = *dst_ptep;
if (!hv_pte_get_present(dst_pte) ||
!hv_pte_get_writable(dst_pte)) {
put_page(src_page);
break;
}
dst_page = pfn_to_page(hv_pte_get_pfn(dst_pte));
if (dst_page == src_page) {
/*
*/
put_page(src_page);
break;
}
get_page(dst_page);
if (pte_val(dst_pte) != pte_val(*dst_ptep)) {
put_page(dst_page);
goto retry_dest;
}
if (pte_huge(dst_pte)) {
/* */
int pfn = hv_pte_get_pfn(dst_pte);
pfn += (((unsigned long)dest & (HPAGE_SIZE-1))
>> PAGE_SHIFT);
dst_pte = pfn_pte(pfn, dst_pte);
dst_pte = pte_mksmall(dst_pte);
}
/* */
copy_size = len;
bytes_left_on_page =
PAGE_SIZE - (((int)source) & (PAGE_SIZE-1));
if (copy_size > bytes_left_on_page)
copy_size = bytes_left_on_page;
bytes_left_on_page =
PAGE_SIZE - (((int)dest) & (PAGE_SIZE-1));
if (copy_size > bytes_left_on_page)
copy_size = bytes_left_on_page;
memcpy_multicache(dest, source, dst_pte, src_pte, copy_size);
/* */
put_page(dst_page);
put_page(src_page);
/* */
dest += copy_size;
source += copy_size;
len -= copy_size;
}
return func(dest, source, len);
}
