/*
* updates the status of erased blocks
*/
void ssd_update_free_block_status(int blk, int plane_num, ssd_element_metadata *metadata, ssd_t *s)
{
int bitpos;
// clear the bit corresponding to this block in the
// free blocks list for future use
bitpos = ssd_block_to_bitpos(s, blk);
ssd_clear_bit(metadata->free_blocks, bitpos);
metadata->block_usage[blk].state = SSD_BLOCK_CLEAN;
metadata->block_usage[blk].bsn = 0;
metadata->tot_free_blocks ++;
metadata->plane_meta[plane_num].free_blocks ++;
ssd_assert_free_blocks(s, metadata);
// there must be no valid pages in the erased block
ASSERT(metadata->block_usage[blk].num_valid == 0);
ssd_assert_valid_pages(plane_num, metadata, s);
}
/*
* migrate data from a cold block to "to_blk"
*/
int ssd_migrate_cold_data(int to_blk, double *mcost, int plane_num, int elem_num, ssd_t *s)
{
int i;
int from_blk = -1;
double oldest_erase_time = simtime;
double cost = 0;
int bitpos;
#if SSD_ASSERT_ALL
int f1;
int f2;
#endif
ssd_element_metadata *metadata = &(s->elements[elem_num].metadata);
// first select the coldest of all blocks.
// one way to select is to find the one that has the oldest
// erasure time.
if (plane_num == -1) {
for (i = 0; i < s->params.blocks_per_element; i ++) {
if (metadata->block_usage[i].num_valid > 0) {
if (metadata->block_usage[i].time_of_last_erasure < oldest_erase_time) {
oldest_erase_time = metadata->block_usage[i].time_of_last_erasure;
from_blk = i;
}
}
}
} else {
#if SSD_ASSERT_ALL
f1 = ssd_free_bits(plane_num, elem_num, metadata, s);
ASSERT(f1 == metadata->plane_meta[metadata->block_usage[to_blk].plane_num].free_blocks);
#endif
bitpos = plane_num * s->params.blocks_per_plane;
for (i = bitpos; i < bitpos + (int)s->params.blocks_per_plane; i ++) {
int block = ssd_bitpos_to_block(i, s);
ASSERT(metadata->block_usage[block].plane_num == plane_num);
if (metadata->block_usage[block].num_valid > 0) {
if (metadata->block_usage[block].time_of_last_erasure < oldest_erase_time) {
oldest_erase_time = metadata->block_usage[block].time_of_last_erasure;
from_blk = block;
}
}
}
}
ASSERT(from_blk != -1);
if (plane_num != -1) {
ASSERT(metadata->block_usage[from_blk].plane_num == metadata->block_usage[to_blk].plane_num);
}
// next, clean the block to which we'll transfer the
// cold data
cost += _ssd_clean_block_fully(to_blk, metadata->block_usage[to_blk].plane_num, elem_num, metadata, s);
#if SSD_ASSERT_ALL
if (plane_num != -1) {
f2 = ssd_free_bits(plane_num, elem_num, metadata, s);
ASSERT(f2 == metadata->plane_meta[metadata->block_usage[to_blk].plane_num].free_blocks);
}
#endif
// then, migrate the cold data to the worn out block.
// for which, we first read all the valid data
cost += metadata->block_usage[from_blk].num_valid * s->params.page_read_latency;
// include the write cost
cost += metadata->block_usage[from_blk].num_valid * s->params.page_write_latency;
// if the src and dest blocks are on different planes
// include the transfer cost also
cost += ssd_crossover_cost(s, metadata, from_blk, to_blk);
// the cost of erasing the cold block (represented by from_blk)
// will be added later ...
// finally, update the metadata
metadata->block_usage[to_blk].bsn = metadata->block_usage[from_blk].bsn;
metadata->block_usage[to_blk].num_valid = metadata->block_usage[from_blk].num_valid;
metadata->block_usage[from_blk].num_valid = 0;
for (i = 0; i < s->params.pages_per_block; i ++) {
int lpn = metadata->block_usage[from_blk].page[i];
if (lpn != -1) {
ASSERT(metadata->lba_table[lpn] == (from_blk * s->params.pages_per_block + i));
metadata->lba_table[lpn] = to_blk * s->params.pages_per_block + i;
}
metadata->block_usage[to_blk].page[i] = metadata->block_usage[from_blk].page[i];
}
metadata->block_usage[to_blk].state = metadata->block_usage[from_blk].state;
bitpos = ssd_block_to_bitpos(s, to_blk);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->tot_free_blocks --;
metadata->plane_meta[metadata->block_usage[to_blk].plane_num].free_blocks --;
#if SSD_ASSERT_ALL
if (plane_num != -1) {
f2 = ssd_free_bits(plane_num, elem_num, metadata, s);
ASSERT(f2 == metadata->plane_meta[metadata->block_usage[to_blk].plane_num].free_blocks);
}
#endif
ssd_assert_free_blocks(s, metadata);
ASSERT(metadata->block_usage[from_blk].num_valid == 0);
#if ASSERT_FREEBITS
if (plane_num != -1) {
f2 = ssd_free_bits(plane_num, elem_num, metadata, s);
ASSERT(f1 == f2);
}
#endif
*mcost = cost;
// stat
metadata->tot_migrations ++;
metadata->tot_pgs_migrated += metadata->block_usage[to_blk].num_valid;
metadata->mig_cost += cost;
return from_blk;
}
/*
* we clean a block only after it is fully used.
*/
int ssd_can_clean_block(ssd_t *s, ssd_element_metadata *metadata, int blk)
{
int bitpos = ssd_block_to_bitpos(s, blk);
return ((ssd_bit_on(metadata->free_blocks, bitpos)) && (metadata->block_usage[blk].state == SSD_BLOCK_SEALED));
}
/*
* vp
* description: this routine allocates and initializes the ssd element metadata
* structures. FIXME: if the systems is powered up, this init routine has to
* populate the structures by scanning the summary pages (to implement this,
* we can read from a disk checkpoint file). but, this is future work.
*/
void ssd_element_metadata_init(int elem_number, ssd_element_metadata *metadata, ssd_t *currdisk)
{
gang_metadata *g;
unsigned int ppage;
unsigned int i;
unsigned int bytes_to_alloc;
unsigned int tot_blocks = currdisk->params.blocks_per_element;
unsigned int tot_pages = tot_blocks * currdisk->params.pages_per_block;
unsigned int reserved_blocks, usable_blocks, export_size;
unsigned int reserved_blocks_per_plane, usable_blocks_per_plane;
unsigned int bitpos;
unsigned int active_block;
unsigned int elem_index;
unsigned int bsn = 1;
int plane_block_mapping = currdisk->params.plane_block_mapping;
//////////////////////////////////////////////////////////////////////////////
// active page starts at the 1st page on the reserved section
reserved_blocks_per_plane = (currdisk->params.reserve_blocks * currdisk->params.blocks_per_plane) / 100;
usable_blocks_per_plane = currdisk->params.blocks_per_plane - reserved_blocks_per_plane;
reserved_blocks = reserved_blocks_per_plane * currdisk->params.planes_per_pkg;
usable_blocks = usable_blocks_per_plane * currdisk->params.planes_per_pkg;
//////////////////////////////////////////////////////////////////////////////
// initialize the free blocks and free pages
metadata->tot_free_blocks = reserved_blocks;
//////////////////////////////////////////////////////////////////////////////
// assign the gang and init the element's free pages
metadata->gang_num = elem_number / currdisk->params.elements_per_gang;
currdisk->gang_meta[metadata->gang_num].elem_free_pages[elem_number] = \
metadata->tot_free_blocks * SSD_DATA_PAGES_PER_BLOCK(currdisk);
g = &currdisk->gang_meta[metadata->gang_num];
elem_index = elem_number % currdisk->params.elements_per_gang;
//////////////////////////////////////////////////////////////////////////////
// let's begin cleaning with the first plane
metadata->plane_to_clean = 0;
metadata->plane_to_write = 0;
metadata->block_alloc_pos = 0;
metadata->reqs_waiting = 0;
metadata->tot_migrations = 0;
metadata->tot_pgs_migrated = 0;
metadata->mig_cost = 0;
//////////////////////////////////////////////////////////////////////////////
// init the plane metadata
for (i = 0; i < (unsigned int)currdisk->params.planes_per_pkg; i ++) {
int blocks_to_skip;
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_CONCAT mapping");
blocks_to_skip = (i*currdisk->params.blocks_per_plane + usable_blocks_per_plane);
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_PAIRWISE_STRIPE mapping");
blocks_to_skip = (i/2)*(2*currdisk->params.blocks_per_plane) + (2*usable_blocks_per_plane) + i%2;
break;
case PLANE_BLOCKS_FULL_STRIPE:
blocks_to_skip = (currdisk->params.planes_per_pkg * usable_blocks_per_plane) + i;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
//metadata->plane_meta[i].active_page = blocks_to_skip*currdisk->params.pages_per_block;
metadata->plane_meta[i].free_blocks = reserved_blocks_per_plane;
metadata->plane_meta[i].valid_pages = 0;
metadata->plane_meta[i].clean_in_progress = 0;
metadata->plane_meta[i].clean_in_block = -1;
metadata->plane_meta[i].block_alloc_pos = i*currdisk->params.blocks_per_plane;
metadata->plane_meta[i].parunit_num = i / SSD_PLANES_PER_PARUNIT(currdisk);
metadata->plane_meta[i].num_cleans = 0;
metadata->plane_meta[i].dead_blocks = 0;
}
//////////////////////////////////////////////////////////////////////////////
// init the next plane to clean in a parunit
for (i = 0; i < (unsigned int) SSD_PARUNITS_PER_ELEM(currdisk); i ++) {
metadata->parunits[i].plane_to_clean = SSD_PLANES_PER_PARUNIT(currdisk)*i;
}
// since we reserve one page out of every block to store the summary info,
// the size exported by the flash disk is little less.
export_size = usable_blocks * SSD_DATA_PAGES_PER_BLOCK(currdisk);
currdisk->data_pages_per_elem = export_size;
//printf("res blks = %d, use blks = %d act page = %d exp size = %d\n",
// reserved_blocks, usable_blocks, metadata->active_page, export_size);
//////////////////////////////////////////////////////////////////////////////
// allocate the lba table
if ((metadata->lba_table = (int *)malloc(export_size * sizeof(int))) == NULL) {
fprintf(stderr, "Error: malloc to lba table in ssd_element_metadata_init failed\n");
#ifdef __x86_64__
fprintf(stderr, "Allocation size = %lu\n", export_size * sizeof(int));
#else
fprintf(stderr, "Allocation size = %u\n", export_size * sizeof(int));
#endif
exit(1);
}
//////////////////////////////////////////////////////////////////////////////
// allocate the free blocks bit map
// what if the no of blocks is not divisible by 8?
if ((tot_blocks % (sizeof(unsigned char) * 8)) != 0) {
fprintf(stderr, "This case is not yet handled\n");
exit(1);
}
bytes_to_alloc = tot_blocks / (sizeof(unsigned char) * 8);
if (!(metadata->free_blocks = (unsigned char *)malloc(bytes_to_alloc))) {
fprintf(stderr, "Error: malloc to free_blocks in ssd_element_metadata_init failed\n");
fprintf(stderr, "Allocation size = %d\n", bytes_to_alloc);
exit(1);
}
bzero(metadata->free_blocks, bytes_to_alloc);
//////////////////////////////////////////////////////////////////////////////
// allocate the block usage array and initialize it
if (!(metadata->block_usage = (block_metadata *)malloc(tot_blocks * sizeof(block_metadata)))) {
fprintf(stderr, "Error: malloc to block_usage in ssd_element_metadata_init failed\n");
#ifdef __x86_64__
fprintf(stderr, "Allocation size = %lu\n", tot_blocks * sizeof(block_metadata));
#else
fprintf(stderr, "Allocation size = %u\n", tot_blocks * sizeof(block_metadata));
#endif
exit(1);
}
bzero(metadata->block_usage, tot_blocks * sizeof(block_metadata));
for (i = 0; i < tot_blocks; i ++) {
int j;
metadata->block_usage[i].block_num = i;
metadata->block_usage[i].elem_num = elem_number;
metadata->block_usage[i].page = (ssd_page_metadata*)malloc(sizeof(ssd_page_metadata) * currdisk->params.pages_per_block);
// KJ: initialize all the page metadata
bzero(metadata->block_usage[i].page, currdisk->params.pages_per_block * sizeof(ssd_page_metadata));
metadata->block_usage[i].entry_in_refresh_queue= 0;
metadata->block_usage[i].total_pages_written = INITIAL_STRESSES*currdisk->params.pages_per_block;
metadata->block_usage[i].least_retention_page = &metadata->block_usage[i].page[0];
metadata->block_usage[i].logical_stresses = INITIAL_STRESSES*currdisk->params.pages_per_block;
metadata->block_usage[i].physical_stresses = INITIAL_STRESSES*currdisk->params.pages_per_block;
assert(metadata->block_usage[i].page);
for (j = 0; j < currdisk->params.pages_per_block; j ++) {
metadata->block_usage[i].page[j].lpn = -1;
metadata->block_usage[i].page[j].time_of_last_stress = 0;
metadata->block_usage[i].page[j].retention_period = 1e10;
metadata->block_usage[i].page[j].logical_stresses= INITIAL_STRESSES;
metadata->block_usage[i].page[j].physical_stresses = INITIAL_STRESSES;
metadata->block_usage[i].page[j].eqn_cycle = INITIAL_STRESSES;
metadata->block_usage[i].page[j].stress_increment = 1;
metadata->block_usage[i].page[j].recovery_period_total = 0;
}
// assign the plane number to each block
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
metadata->block_usage[i].plane_num = i / currdisk->params.blocks_per_plane;
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
metadata->block_usage[i].plane_num = (i/(2*currdisk->params.blocks_per_plane))*2 + i%2;
break;
case PLANE_BLOCKS_FULL_STRIPE:
metadata->block_usage[i].plane_num = i % currdisk->params.planes_per_pkg;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
// set the remaining life time and time of last erasure
metadata->block_usage[i].rem_lifetime = SSD_MAX_ERASURES;
metadata->block_usage[i].time_of_last_erasure = simtime;
// set the block state
metadata->block_usage[i].state = SSD_BLOCK_CLEAN;
// init the bsn to be zero
metadata->block_usage[i].bsn = 0;
}
//load the ssd state from the snapshot provided by trace_analyzer.
if(ta_snapshotfile)
ssd_init_stress_info_ta(currdisk,metadata,elem_number);
else if(ds_snapshotfile)
ssd_init_stress_info_ds(currdisk,metadata,elem_number);
//check for dead block status.
int dead_block_count = 0;
for (i=0;i<currdisk->params.blocks_per_element;i++) {
if (i < usable_blocks) //temporarily mark blocks as in use.
metadata->block_usage[i].state = SSD_BLOCK_INUSE;
if (ssd_block_dead(&metadata->block_usage[i],currdisk))
dead_block_count++;
else
metadata->block_usage[i].state = SSD_BLOCK_CLEAN;
}
fprintf(stderr,"Dead blocks in element #%d : %d\n",elem_number,dead_block_count);
if (dead_block_count >= reserved_blocks) { //if dead blocks more than overprovisioned blocks, then simulation cannot proceed further.
fprintf(stderr,"Total reserved blocks : %d\n",reserved_blocks);
fprintf(stderr,"Unable to proceed further as dead blocks more than reserved blocks\n");
fprintf(outputfile,"Total reserved blocks : %d\n",reserved_blocks);
fprintf(outputfile,"Unable to proceed further as dead blocks more than reserved blocks\n");
fclose(outputfile);
exit(1);
}
ssd_assert_free_blocks(currdisk,metadata); //////////////////////////////////////////////////////////////////////////////
// initially, we assume that every logical page is mapped
// onto a physical page. we start from the first phy page
// and continue to map, leaving the last page of every block
// to store the summary information.
ppage = 0;
i = 0;
while (i < export_size) {
int pgnum_in_gang;
int pp_index;
int plane_num;
unsigned int block = SSD_PAGE_TO_BLOCK(ppage, currdisk);
ASSERT(block < (unsigned int)currdisk->params.blocks_per_element);
if (metadata->block_usage[block].state == SSD_BLOCK_DEAD) {
//block dead. dont use this block for future.
bitpos = ssd_block_to_bitpos(currdisk, block);
ssd_set_bit(metadata->free_blocks, bitpos);
ppage+=currdisk->params.pages_per_block;
continue;
}
// if this is the last page in the block
if (ssd_last_page_in_block(ppage, currdisk)) {
// leave this physical page for summary page and
// seal the block
metadata->block_usage[block].state = SSD_BLOCK_SEALED;
//metadata->block_usage[block].page[currdisk->params.pages_per_block-1].logical_stresses++;
//metadata->block_usage[block].total_pages_written++;
// go to next block
ppage ++;
block = SSD_PAGE_TO_BLOCK(ppage, currdisk);
}
// if this block is in the reserved section, skip it
// and go to the next block.
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
{
assert(!"Unsupported plane block mapping PLANE_BLOCKS_CONCAT");
unsigned int block_index = block % currdisk->params.blocks_per_plane;
if ((block_index >= usable_blocks_per_plane) && (block_index < currdisk->params.blocks_per_plane)) {
// go to next block
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
}
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
{
assert(!"Unsupported plane block mapping PLANE_BLOCKS_PAIRWISE_STRIPE");
unsigned int block_index = block % (2*currdisk->params.blocks_per_plane);
if ((block_index >= 2*usable_blocks_per_plane) && (block_index < 2*currdisk->params.blocks_per_plane)) {
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
}
break;
case PLANE_BLOCKS_FULL_STRIPE:
if ((block >= usable_blocks+dead_block_count) && (block < (unsigned int)currdisk->params.blocks_per_element)) {
printf("Error: the control should not come here ...\n");
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
// when the control comes here, 'ppage' contains the next page
// that can be assigned to a logical page.
// find the index of the phy page within the block
pp_index = ppage % currdisk->params.pages_per_block;
// populate the lba table
metadata->lba_table[i] = ppage;
pgnum_in_gang = elem_index * export_size + i;
g->pg2elem[pgnum_in_gang].e = elem_number;
// mark this block as not free and its state as 'in use'.
// note that a block could be not free and its state be 'sealed'.
// it is enough if we set it once while working on the first phy page.
// also increment the block sequence number.
if (pp_index == 0) {
bitpos = ssd_block_to_bitpos(currdisk, block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[block].state = SSD_BLOCK_INUSE;
metadata->block_usage[block].bsn = bsn ++;
}
// increase the usage count per block
plane_num = metadata->block_usage[block].plane_num;
metadata->block_usage[block].page[pp_index].lpn = i;
metadata->block_usage[block].page[pp_index].time_of_last_stress = simtime;
metadata->block_usage[block].num_valid ++;
metadata->plane_meta[plane_num].valid_pages ++;
// go to the next physical page
ppage ++;
// go to the next logical page
i ++;
}
ssd_assert_free_blocks(currdisk,metadata); //////////////////////////////////////////////////////////////////////////////
// init the element's active page as well as the plane's active page.
int total_blocks = currdisk->params.blocks_per_plane * currdisk->params.planes_per_pkg;
for (i = 0; i < (unsigned int)currdisk->params.planes_per_pkg; i ++) {
int blocks_to_skip;
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_CONCAT mapping");
blocks_to_skip = (i*currdisk->params.blocks_per_plane + usable_blocks_per_plane);
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_PAIRWISE_STRIPE mapping");
blocks_to_skip = (i/2)*(2*currdisk->params.blocks_per_plane) + (2*usable_blocks_per_plane) + i%2;
break;
case PLANE_BLOCKS_FULL_STRIPE:
/*blocks_to_skip = (currdisk->params.planes_per_pkg * usable_blocks_per_plane) + i +
metadata->plane_meta[i].dead_blocks;*/
blocks_to_skip = i;
for (;blocks_to_skip<total_blocks;blocks_to_skip+=currdisk->params.planes_per_pkg)
if(metadata->block_usage[blocks_to_skip].state==SSD_BLOCK_CLEAN)
break;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
metadata->plane_meta[i].active_page = blocks_to_skip*currdisk->params.pages_per_block;
//metadata->plane_meta[i].free_blocks = reserved_blocks_per_plane-metadata->plane_meta[i].dead_blocks;
}
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_CONCAT mapping");
metadata->active_page = usable_blocks_per_plane * currdisk->params.pages_per_block;
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
assert(!"Refresh Operations not supported with PLANE_BLOCKS_PAIRWISE_STRIPE mapping");
metadata->active_page = (2 * usable_blocks_per_plane) * currdisk->params.pages_per_block;
break;
case PLANE_BLOCKS_FULL_STRIPE:
//metadata->active_page = currdisk->params.planes_per_pkg * (usable_blocks_per_plane+metadata->plane_meta[0].dead_blocks) * currdisk->params.pages_per_block;
metadata->active_page = metadata->plane_meta[0].active_page;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
//ASSERT(metadata->active_page == metadata->plane_meta[0].active_page);
active_block = metadata->active_page / currdisk->params.pages_per_block;
//////////////////////////////////////////////////////////////////////////////
// mark the block that corresponds to the active page
// as not free and 'in_use'.
ssd_assert_free_blocks(currdisk,metadata);
switch(currdisk->params.copy_back) {
case SSD_COPY_BACK_DISABLE:
assert(!"Refresh Operation not supported for copy back disabled parameter");
bitpos = ssd_block_to_bitpos(currdisk, active_block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[active_block].state = SSD_BLOCK_INUSE;
metadata->block_usage[active_block].bsn = bsn ++;
break;
case SSD_COPY_BACK_ENABLE:
for (i = 0; i < (unsigned int)currdisk->params.planes_per_pkg; i ++) {
int plane_active_block = SSD_PAGE_TO_BLOCK(metadata->plane_meta[i].active_page, currdisk);
bitpos = ssd_block_to_bitpos(currdisk, plane_active_block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[plane_active_block].state = SSD_BLOCK_INUSE;
metadata->block_usage[plane_active_block].bsn = bsn ++;
metadata->tot_free_blocks --;
metadata->plane_meta[i].free_blocks --;
}
ssd_assert_free_blocks(currdisk,metadata);
break;
default:
fprintf(stderr, "Error: invalid copy back policy %d\n",
currdisk->params.copy_back);
exit(1);
}
//////////////////////////////////////////////////////////////////////////////
// set the bsn for the ssd element
metadata->bsn = bsn;
//printf("set the bsn to %d\n", bsn);
}
/*
* vp
* description: this routine allocates and initializes the ssd element metadata
* structures. FIXME: if the systems is powered up, this init routine has to
* populate the structures by scanning the summary pages (to implement this,
* we can read from a disk checkpoint file). but, this is future work.
*/
void ssd_element_metadata_init(int elem_number, ssd_element_metadata *metadata, ssd_t *currdisk)
{
gang_metadata *g;
unsigned int ppage;
unsigned int i;
unsigned int bytes_to_alloc;
unsigned int tot_blocks = currdisk->params.blocks_per_element;
unsigned int tot_pages = tot_blocks * currdisk->params.pages_per_block;
unsigned int reserved_blocks, usable_blocks, export_size;
unsigned int reserved_blocks_per_plane, usable_blocks_per_plane;
unsigned int bitpos;
unsigned int active_block;
unsigned int elem_index;
unsigned int bsn = 1;
int plane_block_mapping = currdisk->params.plane_block_mapping;
//////////////////////////////////////////////////////////////////////////////
// active page starts at the 1st page on the reserved section
reserved_blocks_per_plane = (currdisk->params.reserve_blocks * currdisk->params.blocks_per_plane) / 100;
usable_blocks_per_plane = currdisk->params.blocks_per_plane - reserved_blocks_per_plane;
reserved_blocks = reserved_blocks_per_plane * currdisk->params.planes_per_pkg;
usable_blocks = usable_blocks_per_plane * currdisk->params.planes_per_pkg;
//////////////////////////////////////////////////////////////////////////////
// initialize the free blocks and free pages
metadata->tot_free_blocks = reserved_blocks;
//////////////////////////////////////////////////////////////////////////////
// assign the gang and init the element's free pages
metadata->gang_num = elem_number / currdisk->params.elements_per_gang;
currdisk->gang_meta[metadata->gang_num].elem_free_pages[elem_number] = \
metadata->tot_free_blocks * SSD_DATA_PAGES_PER_BLOCK(currdisk);
g = &currdisk->gang_meta[metadata->gang_num];
elem_index = elem_number % currdisk->params.elements_per_gang;
//////////////////////////////////////////////////////////////////////////////
// let's begin cleaning with the first plane
metadata->plane_to_clean = 0;
metadata->plane_to_write = 0;
metadata->block_alloc_pos = 0;
metadata->reqs_waiting = 0;
metadata->tot_migrations = 0;
metadata->tot_pgs_migrated = 0;
metadata->mig_cost = 0;
//////////////////////////////////////////////////////////////////////////////
// init the plane metadata
for (i = 0; i < (unsigned int)currdisk->params.planes_per_pkg; i ++) {
int blocks_to_skip;
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
blocks_to_skip = (i*currdisk->params.blocks_per_plane + usable_blocks_per_plane);
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
blocks_to_skip = (i/2)*(2*currdisk->params.blocks_per_plane) + (2*usable_blocks_per_plane) + i%2;
break;
case PLANE_BLOCKS_FULL_STRIPE:
blocks_to_skip = (currdisk->params.planes_per_pkg * usable_blocks_per_plane) + i;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
metadata->plane_meta[i].active_page = blocks_to_skip*currdisk->params.pages_per_block;
metadata->plane_meta[i].free_blocks = reserved_blocks_per_plane;
metadata->plane_meta[i].valid_pages = 0;
metadata->plane_meta[i].clean_in_progress = 0;
metadata->plane_meta[i].clean_in_block = -1;
metadata->plane_meta[i].block_alloc_pos = i*currdisk->params.blocks_per_plane;
metadata->plane_meta[i].parunit_num = i / SSD_PLANES_PER_PARUNIT(currdisk);
metadata->plane_meta[i].num_cleans = 0;
}
//////////////////////////////////////////////////////////////////////////////
// init the next plane to clean in a parunit
for (i = 0; i < (unsigned int) SSD_PARUNITS_PER_ELEM(currdisk); i ++) {
metadata->parunits[i].plane_to_clean = SSD_PLANES_PER_PARUNIT(currdisk)*i;
}
//////////////////////////////////////////////////////////////////////////////
// init the element's active page
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
metadata->active_page = usable_blocks_per_plane * currdisk->params.pages_per_block;
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
metadata->active_page = (2 * usable_blocks_per_plane) * currdisk->params.pages_per_block;
break;
case PLANE_BLOCKS_FULL_STRIPE:
metadata->active_page = (currdisk->params.planes_per_pkg * usable_blocks_per_plane) * currdisk->params.pages_per_block;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
ASSERT(metadata->active_page == metadata->plane_meta[0].active_page);
active_block = metadata->active_page / currdisk->params.pages_per_block;
// since we reserve one page out of every block to store the summary info,
// the size exported by the flash disk is little less.
export_size = usable_blocks * SSD_DATA_PAGES_PER_BLOCK(currdisk);
currdisk->data_pages_per_elem = export_size;
//printf("res blks = %d, use blks = %d act page = %d exp size = %d\n",
// reserved_blocks, usable_blocks, metadata->active_page, export_size);
//////////////////////////////////////////////////////////////////////////////
// allocate the lba table
if ((metadata->lba_table = (int *)malloc(export_size * sizeof(int))) == NULL) {
fprintf(stderr, "Error: malloc to lba table in ssd_element_metadata_init failed\n");
fprintf(stderr, "Allocation size = %d\n", export_size * sizeof(int));
exit(1);
}
//////////////////////////////////////////////////////////////////////////////
// allocate the free blocks bit map
// what if the no of blocks is not divisible by 8?
if ((tot_blocks % (sizeof(unsigned char) * 8)) != 0) {
fprintf(stderr, "This case is not yet handled\n");
exit(1);
}
bytes_to_alloc = tot_blocks / (sizeof(unsigned char) * 8);
if (!(metadata->free_blocks = (unsigned char *)malloc(bytes_to_alloc))) {
fprintf(stderr, "Error: malloc to free_blocks in ssd_element_metadata_init failed\n");
fprintf(stderr, "Allocation size = %d\n", bytes_to_alloc);
exit(1);
}
bzero(metadata->free_blocks, bytes_to_alloc);
//////////////////////////////////////////////////////////////////////////////
// allocate the block usage array and initialize it
if (!(metadata->block_usage = (block_metadata *)malloc(tot_blocks * sizeof(block_metadata)))) {
fprintf(stderr, "Error: malloc to block_usage in ssd_element_metadata_init failed\n");
fprintf(stderr, "Allocation size = %d\n", tot_blocks * sizeof(block_metadata));
exit(1);
}
bzero(metadata->block_usage, tot_blocks * sizeof(block_metadata));
for (i = 0; i < tot_blocks; i ++) {
int j;
metadata->block_usage[i].block_num = i;
metadata->block_usage[i].page = (int*)malloc(sizeof(int) * currdisk->params.pages_per_block);
for (j = 0; j < currdisk->params.pages_per_block; j ++) {
metadata->block_usage[i].page[j] = -1;
}
// assign the plane number to each block
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
metadata->block_usage[i].plane_num = i / currdisk->params.blocks_per_plane;
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
metadata->block_usage[i].plane_num = (i/(2*currdisk->params.blocks_per_plane))*2 + i%2;
break;
case PLANE_BLOCKS_FULL_STRIPE:
metadata->block_usage[i].plane_num = i % currdisk->params.planes_per_pkg;
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
// set the remaining life time and time of last erasure
metadata->block_usage[i].rem_lifetime = SSD_MAX_ERASURES;
metadata->block_usage[i].time_of_last_erasure = simtime;
// set the block state
metadata->block_usage[i].state = SSD_BLOCK_CLEAN;
// init the bsn to be zero
metadata->block_usage[i].bsn = 0;
}
//////////////////////////////////////////////////////////////////////////////
// initially, we assume that every logical page is mapped
// onto a physical page. we start from the first phy page
// and continue to map, leaving the last page of every block
// to store the summary information.
ppage = 0;
i = 0;
while (i < export_size) {
int pgnum_in_gang;
int pp_index;
int plane_num;
unsigned int block = SSD_PAGE_TO_BLOCK(ppage, currdisk);
ASSERT(block < (unsigned int)currdisk->params.blocks_per_element);
// if this is the last page in the block
if (ssd_last_page_in_block(ppage, currdisk)) {
// leave this physical page for summary page and
// seal the block
metadata->block_usage[block].state = SSD_BLOCK_SEALED;
// go to next block
ppage ++;
block = SSD_PAGE_TO_BLOCK(ppage, currdisk);
}
// if this block is in the reserved section, skip it
// and go to the next block.
switch(plane_block_mapping) {
case PLANE_BLOCKS_CONCAT:
{
unsigned int block_index = block % currdisk->params.blocks_per_plane;
if ((block_index >= usable_blocks_per_plane) && (block_index < currdisk->params.blocks_per_plane)) {
// go to next block
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
}
break;
case PLANE_BLOCKS_PAIRWISE_STRIPE:
{
unsigned int block_index = block % (2*currdisk->params.blocks_per_plane);
if ((block_index >= 2*usable_blocks_per_plane) && (block_index < 2*currdisk->params.blocks_per_plane)) {
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
}
break;
case PLANE_BLOCKS_FULL_STRIPE:
// ideally the control should not come here ...
if ((block >= usable_blocks) && (block < (unsigned int)currdisk->params.blocks_per_element)) {
printf("Error: the control should not come here ...\n");
ppage = ssd_first_page_in_next_block(ppage, currdisk);
continue;
}
break;
default:
fprintf(stderr, "Error: unknown plane_block_mapping %d\n", plane_block_mapping);
exit(1);
}
// when the control comes here, 'ppage' contains the next page
// that can be assigned to a logical page.
// find the index of the phy page within the block
pp_index = ppage % currdisk->params.pages_per_block;
// populate the lba table
metadata->lba_table[i] = ppage;
pgnum_in_gang = elem_index * export_size + i;
g->pg2elem[pgnum_in_gang].e = elem_number;
// mark this block as not free and its state as 'in use'.
// note that a block could be not free and its state be 'sealed'.
// it is enough if we set it once while working on the first phy page.
// also increment the block sequence number.
if (pp_index == 0) {
bitpos = ssd_block_to_bitpos(currdisk, block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[block].state = SSD_BLOCK_INUSE;
metadata->block_usage[block].bsn = bsn ++;
}
// increase the usage count per block
plane_num = metadata->block_usage[block].plane_num;
metadata->block_usage[block].page[pp_index] = i;
metadata->block_usage[block].num_valid ++;
metadata->plane_meta[plane_num].valid_pages ++;
// go to the next physical page
ppage ++;
// go to the next logical page
i ++;
}
//////////////////////////////////////////////////////////////////////////////
// mark the block that corresponds to the active page
// as not free and 'in_use'.
switch(currdisk->params.copy_back) {
case SSD_COPY_BACK_DISABLE:
bitpos = ssd_block_to_bitpos(currdisk, active_block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[active_block].state = SSD_BLOCK_INUSE;
metadata->block_usage[active_block].bsn = bsn ++;
break;
case SSD_COPY_BACK_ENABLE:
for (i = 0; i < (unsigned int)currdisk->params.planes_per_pkg; i ++) {
int plane_active_block = SSD_PAGE_TO_BLOCK(metadata->plane_meta[i].active_page, currdisk);
bitpos = ssd_block_to_bitpos(currdisk, plane_active_block);
ssd_set_bit(metadata->free_blocks, bitpos);
metadata->block_usage[plane_active_block].state = SSD_BLOCK_INUSE;
metadata->block_usage[plane_active_block].bsn = bsn ++;
metadata->tot_free_blocks --;
metadata->plane_meta[i].free_blocks --;
}
break;
default:
fprintf(stderr, "Error: invalid copy back policy %d\n",
currdisk->params.copy_back);
exit(1);
}
//////////////////////////////////////////////////////////////////////////////
// set the bsn for the ssd element
metadata->bsn = bsn;
//printf("set the bsn to %d\n", bsn);
}