1477 lines
45 KiB
C
1477 lines
45 KiB
C
|
// SPDX-License-Identifier: GPL-2.0
|
||
|
/*
|
||
|
* Copyright (c) 2019 MediaTek Inc.
|
||
|
*/
|
||
|
|
||
|
#include <linux/spinlock.h>
|
||
|
#include "m4u_priv.h"
|
||
|
#include "m4u_mva.h"
|
||
|
|
||
|
|
||
|
static short mvaGraph[MVA_MAX_BLOCK_NR + 1];
|
||
|
static void *mvaInfoGraph[MVA_MAX_BLOCK_NR + 1];
|
||
|
static DEFINE_SPINLOCK(gMvaGraph_lock);
|
||
|
|
||
|
void m4u_mvaGraph_init(void *priv_reserve)
|
||
|
{
|
||
|
int i;
|
||
|
unsigned int vpu_reset_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START);
|
||
|
unsigned int vpu_reset_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_END);
|
||
|
unsigned int vpu_fix_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_START);
|
||
|
unsigned int vpu_fix_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_END);
|
||
|
unsigned int ccu_fix_block_start = MVAGRAPH_INDEX(CCU_FIX_MVA_START);
|
||
|
unsigned int ccu_fix_block_end = MVAGRAPH_INDEX(CCU_FIX_MVA_END);
|
||
|
unsigned int ccu_nr;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
memset(mvaGraph, 0, sizeof(short) * (MVA_MAX_BLOCK_NR + 1));
|
||
|
memset(mvaInfoGraph, 0, sizeof(void *) * (MVA_MAX_BLOCK_NR + 1));
|
||
|
mvaGraph[0] = (short)(1 | MVA_BUSY_MASK);
|
||
|
mvaInfoGraph[0] = priv_reserve;
|
||
|
/*need to reserve two mva block region for vpu.
|
||
|
*[1,0x200],[0x360,0x500): shared with all ports.
|
||
|
*[0x500, 0x501): reserved for vpu reset vector.
|
||
|
*[0x501, 0x600): shared with all ports.
|
||
|
*[0x600, 0x7E0): reserved for vpu.
|
||
|
*[0x7E0, 0xFFF]: shared with all ports.
|
||
|
*/
|
||
|
/*[0x1,0x200],[0x360, 0x500]*/
|
||
|
mvaGraph[1] = GET_RANGE_SIZE(1, (ccu_fix_block_start - 1));
|
||
|
mvaGraph[ccu_fix_block_start - 1] =
|
||
|
GET_RANGE_SIZE(1, (ccu_fix_block_start - 1));
|
||
|
mvaInfoGraph[1] = priv_reserve;
|
||
|
mvaInfoGraph[ccu_fix_block_start - 1] = priv_reserve;
|
||
|
M4UINFO("[0x1, 0x200): start: 1 end: 0x%x mvaGraph: 0x%x\n",
|
||
|
(ccu_fix_block_start - 1), mvaGraph[1]);
|
||
|
|
||
|
/*CCU:[0x200,0x360]*/
|
||
|
ccu_nr = MVA_GRAPH_BLOCK_NR_ALIGNED(
|
||
|
CCU_FIX_MVA_END - CCU_FIX_MVA_START + 1);
|
||
|
for (i = 0; i < ccu_nr; i++)
|
||
|
MVA_SET_RESERVED(ccu_fix_block_start + i);
|
||
|
mvaGraph[ccu_fix_block_start] =
|
||
|
MVA_RESERVED_MASK | ccu_nr;
|
||
|
mvaGraph[ccu_fix_block_end] =
|
||
|
MVA_RESERVED_MASK | ccu_nr;
|
||
|
mvaInfoGraph[ccu_fix_block_start] = priv_reserve;
|
||
|
mvaInfoGraph[ccu_fix_block_end] = priv_reserve;
|
||
|
M4UINFO("%s,start:0x%x,end:0x%x,mvaGraph:0x%x\n",
|
||
|
__func__, ccu_fix_block_start, ccu_fix_block_end,
|
||
|
mvaGraph[ccu_fix_block_start]);
|
||
|
|
||
|
mvaGraph[ccu_fix_block_end + 1] =
|
||
|
GET_RANGE_SIZE((ccu_fix_block_end + 1),
|
||
|
(vpu_reset_block_start - 1));
|
||
|
mvaGraph[vpu_reset_block_start - 1] =
|
||
|
GET_RANGE_SIZE((ccu_fix_block_end + 1),
|
||
|
(vpu_reset_block_start - 1));
|
||
|
mvaInfoGraph[ccu_fix_block_end + 1] = priv_reserve;
|
||
|
mvaInfoGraph[vpu_reset_block_start - 1] = priv_reserve;
|
||
|
M4UINFO("[0x361, 0x500): start: 0x%x end: 0x%x mvaGraph: 0x%x\n",
|
||
|
(ccu_fix_block_end + 1),
|
||
|
(vpu_reset_block_start - 1),
|
||
|
mvaGraph[ccu_fix_block_end + 1]);
|
||
|
|
||
|
/*[0x500, 0x501)*/
|
||
|
/*set 1 to each bit14 of mvaGraph in vpu reserved region.
|
||
|
*the operation can prevent allocating mva in vpu region
|
||
|
*from m4u_do_mva_alloc.
|
||
|
*/
|
||
|
for (i = 0; i < VPU_RESET_VECTOR_BLOCK_NR; i++)
|
||
|
MVA_SET_RESERVED(vpu_reset_block_start + i);
|
||
|
mvaGraph[vpu_reset_block_start] =
|
||
|
MVA_RESERVED_MASK | VPU_RESET_VECTOR_BLOCK_NR;
|
||
|
mvaGraph[vpu_reset_block_end] =
|
||
|
MVA_RESERVED_MASK | VPU_RESET_VECTOR_BLOCK_NR;
|
||
|
mvaInfoGraph[vpu_reset_block_start] = priv_reserve;
|
||
|
mvaInfoGraph[vpu_reset_block_end] = priv_reserve;
|
||
|
M4UINFO("[0x500, 0x501): start: 0x%x end: 0x%x mvaGraph: 0x%x\n",
|
||
|
vpu_reset_block_start,
|
||
|
vpu_reset_block_end,
|
||
|
mvaGraph[vpu_reset_block_start]);
|
||
|
|
||
|
/*[0x501, 0x600)*/
|
||
|
mvaGraph[vpu_reset_block_end + 1] =
|
||
|
GET_RANGE_SIZE((vpu_reset_block_end + 1),
|
||
|
(vpu_fix_block_start - 1));
|
||
|
mvaGraph[vpu_fix_block_start - 1] =
|
||
|
GET_RANGE_SIZE((vpu_reset_block_end + 1),
|
||
|
vpu_fix_block_start - 1);
|
||
|
mvaInfoGraph[vpu_reset_block_end + 1] = priv_reserve;
|
||
|
mvaInfoGraph[vpu_fix_block_start - 1] = priv_reserve;
|
||
|
M4UINFO("[0x501, 0x600): start: 0x%x end: 0x%x mvaGraph: 0x%x\n",
|
||
|
(vpu_reset_block_end + 1),
|
||
|
(vpu_fix_block_start - 1),
|
||
|
mvaGraph[vpu_reset_block_end + 1]);
|
||
|
|
||
|
/*[0x600, 0x7E0)*/
|
||
|
/*set 1 to each bit14 of mvaGraph in vpu reserved region*/
|
||
|
for (i = 0; i < VPU_FIX_BLOCK_NR; i++)
|
||
|
MVA_SET_RESERVED(vpu_fix_block_start + i);
|
||
|
mvaGraph[vpu_fix_block_start] =
|
||
|
MVA_RESERVED_MASK | VPU_FIX_BLOCK_NR;
|
||
|
mvaGraph[vpu_fix_block_end] =
|
||
|
MVA_RESERVED_MASK | VPU_FIX_BLOCK_NR;
|
||
|
mvaInfoGraph[vpu_fix_block_start] = priv_reserve;
|
||
|
mvaInfoGraph[vpu_fix_block_end] = priv_reserve;
|
||
|
M4UINFO("[0x600, 0x7E0): start: 0x%x end: 0x%x mvaGraph: 0x%x\n",
|
||
|
vpu_fix_block_start,
|
||
|
vpu_fix_block_end,
|
||
|
mvaGraph[vpu_fix_block_start]);
|
||
|
|
||
|
/*[0x7E0, 0xFFF]*/
|
||
|
mvaGraph[vpu_fix_block_end + 1] =
|
||
|
GET_RANGE_SIZE((vpu_fix_block_end + 1), MVA_MAX_BLOCK_NR);
|
||
|
mvaGraph[MVA_MAX_BLOCK_NR] =
|
||
|
GET_RANGE_SIZE((vpu_fix_block_end + 1), MVA_MAX_BLOCK_NR);
|
||
|
mvaInfoGraph[vpu_fix_block_end + 1] = priv_reserve;
|
||
|
mvaInfoGraph[MVA_MAX_BLOCK_NR] = priv_reserve;
|
||
|
M4UINFO("[0x7E0, 0xFFF]: start: 0x%x end: 0x%x mvaGraph: 0x%x\n",
|
||
|
(vpu_fix_block_end + 1),
|
||
|
MVA_MAX_BLOCK_NR,
|
||
|
mvaGraph[vpu_fix_block_end + 1]);
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
}
|
||
|
|
||
|
void m4u_mvaGraph_dump_raw(void)
|
||
|
{
|
||
|
int i;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
M4ULOG_HIGH("[M4U_K] dump raw data of mvaGraph:============>\n");
|
||
|
for (i = 0; i < MVA_MAX_BLOCK_NR + 1; i++)
|
||
|
M4ULOG_HIGH("0x%4x: 0x%08x\n", i, mvaGraph[i]);
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
}
|
||
|
|
||
|
int is_in_vpu_region(unsigned int index, unsigned int nr)
|
||
|
{
|
||
|
unsigned int vpu_reset_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START);
|
||
|
unsigned int vpu_reset_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_END);
|
||
|
unsigned int vpu_fix_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_START);
|
||
|
unsigned int vpu_fix_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_END);
|
||
|
int ret = 0;
|
||
|
#if 0
|
||
|
M4UINFO("start = 0x%x, end = 0x%x nr = %x.\n",
|
||
|
index, GET_END_INDEX(index, nr), nr);
|
||
|
#endif
|
||
|
if ((index >= vpu_reset_block_start
|
||
|
&& GET_END_INDEX(index, nr) <= vpu_reset_block_end)
|
||
|
|| (index >= vpu_fix_block_start
|
||
|
&& GET_END_INDEX(index, nr) <= vpu_fix_block_end))
|
||
|
ret = 1;
|
||
|
#if 0
|
||
|
if (ret)
|
||
|
M4UINFO("input region[0x%x - 0x%x] is in the vpu region.\n",
|
||
|
index, GET_END_INDEX(index, nr));
|
||
|
#endif
|
||
|
return ret;
|
||
|
}
|
||
|
|
||
|
int is_intersected_with_vpu_region(unsigned int start, unsigned int nr)
|
||
|
{
|
||
|
unsigned int vpu_reset_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START);
|
||
|
unsigned int vpu_reset_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_END);
|
||
|
unsigned int vpu_fix_block_start =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_START);
|
||
|
unsigned int vpu_fix_block_end =
|
||
|
MVAGRAPH_INDEX(VPU_FIX_MVA_END);
|
||
|
int ret = 0;
|
||
|
|
||
|
/*4 cases:
|
||
|
* |--------------|------|---------|------------|----------|
|
||
|
* 0x0 0x500 0x501 0x600 0x800 0xFFF
|
||
|
*case 1: start | end
|
||
|
*case 2: start| end
|
||
|
*case 3: start | end
|
||
|
*case 4: start | end
|
||
|
*/
|
||
|
/*case 1: 1 <= start < 0x500 && 0x500<=end<=0xFFF*/
|
||
|
#if 0
|
||
|
M4UINFO("start = 0x%x, end = 0x%x nr = %x.\n",
|
||
|
start, GET_END_INDEX(start, nr), nr);
|
||
|
#endif
|
||
|
if ((start >= 1 && start < vpu_reset_block_start)
|
||
|
&& (GET_END_INDEX(start, nr) >= vpu_reset_block_start
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
|
||
|
/*case 2: 1 <= start < 0x501 && 0x501<=end<=0xFFF*/
|
||
|
if ((start >= 1 && start < (vpu_reset_block_end + 1))
|
||
|
&& (GET_END_INDEX(start, nr) >= (vpu_reset_block_end + 1)
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
/*case 3: 1 <= start < 0x600 && 0x600<=end<=0xFFF*/
|
||
|
if ((start >= 1 && start < vpu_fix_block_start)
|
||
|
&& (GET_END_INDEX(start, nr) >= vpu_fix_block_start
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
/*case 4: 1 <=start < 0x800 && 0x800<=end<=0xFFF*/
|
||
|
if ((start >= 1 && start < (vpu_fix_block_end + 1))
|
||
|
&& (GET_END_INDEX(start, nr) >= (vpu_fix_block_end + 1)
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
#if 0
|
||
|
if (ret)
|
||
|
M4UINFO("input region intersects to vpu region\n");
|
||
|
#endif
|
||
|
return ret;
|
||
|
|
||
|
}
|
||
|
|
||
|
int is_in_ccu_region(unsigned int index, unsigned int nr)
|
||
|
{
|
||
|
unsigned int ccu_fix_block_start = MVAGRAPH_INDEX(CCU_FIX_MVA_START);
|
||
|
unsigned int ccu_fix_block_end = MVAGRAPH_INDEX(CCU_FIX_MVA_END);
|
||
|
|
||
|
if (index >= ccu_fix_block_start &&
|
||
|
GET_END_INDEX(index, nr) <= ccu_fix_block_end)
|
||
|
return 1;
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
int is_intersected_with_ccu_region(unsigned int start, unsigned int nr)
|
||
|
{
|
||
|
unsigned int ccu_fix_block_start = MVAGRAPH_INDEX(CCU_FIX_MVA_START);
|
||
|
unsigned int ccu_fix_block_end = MVAGRAPH_INDEX(CCU_FIX_MVA_END);
|
||
|
int ret = 0;
|
||
|
|
||
|
M4ULOG_LOW("%s:start = 0x%x, end = 0x%x nr = %x.\n",
|
||
|
__func__, start, GET_END_INDEX(start, nr), nr);
|
||
|
|
||
|
/*case 1: 1 <= start < 0x200 && 0x300<=end<=0xFFF*/
|
||
|
if ((start >= 1 && start < ccu_fix_block_start)
|
||
|
&& (GET_END_INDEX(start, nr) >= ccu_fix_block_start
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
/*case 2: 1 <=start < 0x800 && 0x800<=end<=0xFFF*/
|
||
|
if ((start >= 1 && start < (ccu_fix_block_end + 1))
|
||
|
&& (GET_END_INDEX(start, nr) >= (ccu_fix_block_end + 1)
|
||
|
&& GET_END_INDEX(start, nr) <= MVA_MAX_BLOCK_NR))
|
||
|
ret = 1;
|
||
|
|
||
|
if (ret)
|
||
|
M4ULOG_LOW("input region intersects to ccu region\n");
|
||
|
|
||
|
return ret;
|
||
|
}
|
||
|
/*1: y; 0: n*/
|
||
|
int check_reserved_region_integrity(unsigned int start, unsigned int nr)
|
||
|
{
|
||
|
int i, integrity = 0;
|
||
|
|
||
|
for (i = 0; i < nr; i++) {
|
||
|
if (!MVA_IS_RESERVED(start + i))
|
||
|
break;
|
||
|
}
|
||
|
if (i == nr)
|
||
|
integrity = 1;
|
||
|
else {
|
||
|
M4UMSG("reserved blocks[0x%x-0x%x] corruptted at 0x%x\n",
|
||
|
start, GET_END_INDEX(start, nr), i);
|
||
|
}
|
||
|
return integrity;
|
||
|
}
|
||
|
|
||
|
|
||
|
/*need to print integrity of vpu region*/
|
||
|
void m4u_mvaGraph_dump(void)
|
||
|
{
|
||
|
unsigned int start = 0, end = 0, size = 0;
|
||
|
unsigned short index = 1, nr = 0;
|
||
|
int i, max_bit, is_busy, is_reserve, integrity = 0;
|
||
|
short frag[12] = { 0 };
|
||
|
unsigned short nr_free = 0, nr_alloc = 0;
|
||
|
|
||
|
M4ULOG_HIGH(
|
||
|
"[M4U_2.4] mva allocation info dump:====================>\n");
|
||
|
M4ULOG_HIGH(
|
||
|
"start end size blocknum busy reserve integrity\n");
|
||
|
|
||
|
for (index = 1; index < MVA_MAX_BLOCK_NR + 1; index += nr) {
|
||
|
start = index << MVA_BLOCK_SIZE_ORDER;
|
||
|
nr = MVA_GET_NR(index);
|
||
|
size = nr << MVA_BLOCK_SIZE_ORDER;
|
||
|
end = start + size - 1;
|
||
|
/* DO NOT call aee here directly to avoid recursive dump. */
|
||
|
if (nr == 0 || end <= start) {
|
||
|
M4ULOG_HIGH("%s err: nr=%d, start=0x08x\n",
|
||
|
__func__, nr, start);
|
||
|
break;
|
||
|
}
|
||
|
if (MVA_IS_BUSY(index)) {
|
||
|
is_busy = 1;
|
||
|
if (MVA_IS_RESERVED(index))
|
||
|
is_reserve = 1;
|
||
|
else
|
||
|
is_reserve = 0;
|
||
|
nr_alloc += nr;
|
||
|
} else { /* mva region is free */
|
||
|
/* mva region is free */
|
||
|
is_busy = 0;
|
||
|
if (MVA_IS_RESERVED(index))
|
||
|
is_reserve = 1;
|
||
|
else
|
||
|
is_reserve = 0;
|
||
|
nr_free += nr;
|
||
|
|
||
|
max_bit = 0;
|
||
|
for (i = 0; i < 12; i++) {
|
||
|
if (nr & (1 << i))
|
||
|
max_bit = i;
|
||
|
}
|
||
|
frag[max_bit]++;
|
||
|
}
|
||
|
/*verify if the reserve bit of each block in vpu region is set*/
|
||
|
if (is_in_ccu_region(index, nr)) {
|
||
|
for (i = 0; i < nr; i++) {
|
||
|
if (!MVA_IS_RESERVED(index + i))
|
||
|
break;
|
||
|
}
|
||
|
if (i == nr)
|
||
|
integrity = 1;
|
||
|
} else if (is_in_vpu_region(index, nr)) {
|
||
|
for (i = 0; i < nr; i++) {
|
||
|
if (!MVA_IS_RESERVED(index + i))
|
||
|
break;
|
||
|
}
|
||
|
if (i == nr)
|
||
|
integrity = 1;
|
||
|
}
|
||
|
|
||
|
M4ULOG_HIGH(
|
||
|
"0x%08x 0x%08x 0x%08x %4d %d %4d %8d\n",
|
||
|
start, end, size, nr, is_busy, is_reserve, integrity);
|
||
|
integrity = 0;
|
||
|
}
|
||
|
|
||
|
M4ULOG_HIGH("\n");
|
||
|
M4ULOG_HIGH(
|
||
|
"[M4U_2.4] mva alloc summary: (unit: blocks)========================>\n");
|
||
|
M4ULOG_HIGH("free: %d , alloc: %d, total: %d\n",
|
||
|
nr_free, nr_alloc, nr_free + nr_alloc);
|
||
|
M4ULOG_HIGH(
|
||
|
"[M4U_2.4] free region fragments in 2^x blocks unit:===============\n");
|
||
|
M4ULOG_HIGH(
|
||
|
" 0 1 2 3 4 5 6 7 8 9 10 11\n");
|
||
|
M4ULOG_HIGH(
|
||
|
"%4d %4d %4d %4d %4d %4d %4d %4d %4d %4d %4d %4d\n",
|
||
|
frag[0], frag[1], frag[2], frag[3],
|
||
|
frag[4], frag[5], frag[6],
|
||
|
frag[7], frag[8], frag[9], frag[10], frag[11]);
|
||
|
M4ULOG_HIGH(
|
||
|
"[M4U_2.4] mva alloc dump done=========================<\n");
|
||
|
}
|
||
|
|
||
|
void *mva_get_priv_ext(unsigned int mva)
|
||
|
{
|
||
|
void *priv = NULL;
|
||
|
unsigned int index;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
index = MVAGRAPH_INDEX(mva);
|
||
|
if (index == 0 || index > MVA_MAX_BLOCK_NR) {
|
||
|
M4UMSG("mvaGraph index is 0. mva=0x%x\n", mva);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
/* find prev head/tail of this region */
|
||
|
while (mvaGraph[index] == 0)
|
||
|
index--;
|
||
|
|
||
|
if (MVA_IS_BUSY(index))
|
||
|
priv = mvaInfoGraph[index];
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return priv;
|
||
|
}
|
||
|
|
||
|
int mva_foreach_priv(mva_buf_fn_t *fn, void *data)
|
||
|
{
|
||
|
unsigned short index = 1, nr = 0;
|
||
|
unsigned int mva;
|
||
|
void *priv;
|
||
|
int ret;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
for (index = 1; index < MVA_MAX_BLOCK_NR + 1; index += nr) {
|
||
|
mva = index << MVA_BLOCK_SIZE_ORDER;
|
||
|
nr = MVA_GET_NR(index);
|
||
|
if (MVA_IS_BUSY(index)) {
|
||
|
priv = mvaInfoGraph[index];
|
||
|
ret = fn(priv, mva, mva + nr * MVA_BLOCK_SIZE, data);
|
||
|
if (ret)
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
unsigned int get_first_valid_mva(void)
|
||
|
{
|
||
|
unsigned short index = 1, nr = 0;
|
||
|
unsigned int mva;
|
||
|
void *priv;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
for (index = 1; index < MVA_MAX_BLOCK_NR + 1; index += nr) {
|
||
|
mva = index << MVA_BLOCK_SIZE_ORDER;
|
||
|
nr = MVA_GET_NR(index);
|
||
|
if (MVA_IS_BUSY(index)) {
|
||
|
priv = mvaInfoGraph[index];
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return mva;
|
||
|
}
|
||
|
|
||
|
|
||
|
void *mva_get_priv(unsigned int mva)
|
||
|
{
|
||
|
void *priv = NULL;
|
||
|
unsigned int index;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
index = MVAGRAPH_INDEX(mva);
|
||
|
if (index == 0 || index > MVA_MAX_BLOCK_NR) {
|
||
|
M4UMSG("mvaGraph index is 0. mva=0x%x\n", mva);
|
||
|
return NULL;
|
||
|
}
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
if (MVA_IS_BUSY(index))
|
||
|
priv = mvaInfoGraph[index];
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return priv;
|
||
|
}
|
||
|
|
||
|
/*make sure @param priv is not NULL before call this function.
|
||
|
*return 0 means non-vpu region access.
|
||
|
*return -1 means input mva region
|
||
|
*allocated by non-vpu port is in vpu reserved region.
|
||
|
*or input mva region intersectst to vpu reserved region.
|
||
|
*return 1 means vpu port alloc mva in vpu reserved region.
|
||
|
*/
|
||
|
int m4u_check_mva_region(unsigned int startIdx, unsigned int nr, void *priv)
|
||
|
{
|
||
|
struct m4u_buf_info_t *pMvaInfo = (struct m4u_buf_info_t *)priv;
|
||
|
int is_in = 0, is_interseted = 0;
|
||
|
int is_vpu_port = 0;
|
||
|
#if defined(CONFIG_MACH_MT6775) || defined(CONFIG_MACH_MT6771)
|
||
|
is_vpu_port = (pMvaInfo->port == M4U_PORT_VPU0) ||
|
||
|
(pMvaInfo->port == M4U_PORT_VPU1);
|
||
|
#else
|
||
|
is_vpu_port = (pMvaInfo->port == M4U_PORT_VPU);
|
||
|
#endif
|
||
|
|
||
|
/*check if input mva region is in vpu region.
|
||
|
*if it's in vpu region, we check if it's non-vpu port
|
||
|
*/
|
||
|
#if 0
|
||
|
M4UINFO("%s: [0x%x - 0x%x]\n",
|
||
|
__func__, startIdx, GET_END_INDEX(startIdx, nr));
|
||
|
#endif
|
||
|
is_in = is_in_vpu_region(startIdx, nr);
|
||
|
if (is_in && is_vpu_port)
|
||
|
return 1;
|
||
|
else if (is_in && !is_vpu_port) {
|
||
|
M4UINFO(
|
||
|
"[0x%x - 0x%x] requested by port(%d) is in vpu reserved region!\n",
|
||
|
startIdx, GET_END_INDEX(startIdx, nr),
|
||
|
pMvaInfo->port);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
/*check if input mva region is intersected with vpu region*/
|
||
|
is_interseted = is_intersected_with_vpu_region(startIdx, nr);
|
||
|
/*return 0 means other port normal alloction.
|
||
|
*if it isn't in vpu region & insersected with vpu region.
|
||
|
*it's non-vpu port alloc non-vpu reserved region. then return 0.
|
||
|
*/
|
||
|
if (!is_in && !is_interseted)
|
||
|
return 0;
|
||
|
M4UINFO(
|
||
|
"[0x%x - 0x%x] requested by port(%d) intersects to vpu region!\n",
|
||
|
startIdx, GET_END_INDEX(startIdx, nr),
|
||
|
pMvaInfo->port);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
static int __check_ccu_mva_region(
|
||
|
unsigned int startIdx, unsigned int nr, void *priv)
|
||
|
{
|
||
|
struct m4u_buf_info_t *pMvaInfo = (struct m4u_buf_info_t *)priv;
|
||
|
int is_in = 0, is_interseted = 0;
|
||
|
int is_ccu_port = 0;
|
||
|
|
||
|
#if defined(CONFIG_MACH_MT6771)
|
||
|
is_ccu_port = (pMvaInfo->port == M4U_PORT_CCU0) ||
|
||
|
(pMvaInfo->port == M4U_PORT_CCU1) ||
|
||
|
(pMvaInfo->port == M4U_PORT_CAM_CCUI) ||
|
||
|
(pMvaInfo->port == M4U_PORT_CAM_CCUG)
|
||
|
|| (pMvaInfo->port == M4U_PORT_CAM_CCUO);
|
||
|
#else
|
||
|
is_ccu_port = (pMvaInfo->port == M4U_PORT_CAM_CCUI) ||
|
||
|
(pMvaInfo->port == M4U_PORT_CAM_CCUG)
|
||
|
|| (pMvaInfo->port == M4U_PORT_CAM_CCUO);
|
||
|
#endif
|
||
|
/*check if input mva region is in ccu region.
|
||
|
*if it's in ccu region, we check if it's non-ccu port
|
||
|
*/
|
||
|
M4ULOG_LOW("%s: [0x%x - 0x%x]\n",
|
||
|
__func__, startIdx, GET_END_INDEX(startIdx, nr));
|
||
|
|
||
|
is_in = is_in_ccu_region(startIdx, nr);
|
||
|
if (is_in && is_ccu_port)
|
||
|
return 1;
|
||
|
else if (is_in && !is_ccu_port) {
|
||
|
M4ULOG_MID(
|
||
|
"[0x%x - 0x%x] requested by port(%d) is in ccu reserved region!\n",
|
||
|
startIdx, GET_END_INDEX(startIdx, nr),
|
||
|
pMvaInfo->port);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
/*check if input mva region is intersected with vpu region*/
|
||
|
is_interseted = is_intersected_with_ccu_region(startIdx, nr);
|
||
|
/*return 0 means other port normal alloction.
|
||
|
*if it isn't in ccu region & insersected with ccu region.
|
||
|
*it's non-ccu port alloc non-ccu reserved region. then return 0.
|
||
|
*/
|
||
|
if (!is_in && !is_interseted)
|
||
|
return 0;
|
||
|
M4ULOG_LOW(
|
||
|
"[0x%x - 0x%x] requested by port(%d) intersects to ccu region!\n",
|
||
|
startIdx, GET_END_INDEX(startIdx, nr),
|
||
|
pMvaInfo->port);
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
/*m4u_do_mva_alloc should meet the following rules:
|
||
|
*(1) vpu port should not m4u_do_mva_alloc vpu fix region,
|
||
|
* but it can m4u_do_mva_alloc non-vpu fix region.
|
||
|
*(2) non-vpu port can m4u_do_mva_alloc non-vpu fix region.
|
||
|
*/
|
||
|
unsigned int m4u_do_mva_alloc(unsigned long va, unsigned int size, void *priv)
|
||
|
{
|
||
|
unsigned short s, end;
|
||
|
unsigned short new_start, new_end;
|
||
|
unsigned short nr = 0;
|
||
|
unsigned int mvaRegionStart;
|
||
|
unsigned long startRequire, endRequire, sizeRequire;
|
||
|
int region_status = 0;
|
||
|
const short fix_index0 = MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START);
|
||
|
const short fix_index1 = MVAGRAPH_INDEX(VPU_FIX_MVA_START);
|
||
|
short gap_start_idx = GET_END_INDEX(fix_index0,
|
||
|
VPU_RESET_VECTOR_BLOCK_NR) + 1;
|
||
|
short gap_end_idx = fix_index1 - 1;
|
||
|
short gap_nr = GET_RANGE_SIZE(gap_start_idx, gap_end_idx);
|
||
|
const short ccu_fix_index_start = MVAGRAPH_INDEX(CCU_FIX_MVA_START);
|
||
|
const short ccu_fix_index_end = MVAGRAPH_INDEX(CCU_FIX_MVA_END);
|
||
|
const short ccu_gap_nr = GET_RANGE_SIZE(ccu_fix_index_end, fix_index0);
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
if (size == 0)
|
||
|
return 0;
|
||
|
|
||
|
/* ----------------------------------------------------- */
|
||
|
/* calculate mva block number */
|
||
|
startRequire = va & (~M4U_PAGE_MASK);
|
||
|
endRequire = (va + size - 1) | M4U_PAGE_MASK;
|
||
|
sizeRequire = endRequire - startRequire + 1;
|
||
|
nr = (sizeRequire + MVA_BLOCK_ALIGN_MASK) >> MVA_BLOCK_SIZE_ORDER;
|
||
|
if (fix_index1 == fix_index0) {
|
||
|
gap_start_idx = MVAGRAPH_INDEX(VPU_FIX_MVA_START);
|
||
|
gap_end_idx = GET_END_INDEX(fix_index0,
|
||
|
VPU_RESET_VECTOR_BLOCK_NR) + 1;
|
||
|
gap_nr = GET_RANGE_SIZE(gap_start_idx, gap_end_idx);
|
||
|
}
|
||
|
/* ----------------------------------------------- */
|
||
|
/* find the proper mva graph on 3 stages:
|
||
|
* stage 1: find it in graph range [0x1-0x4FF ]
|
||
|
* every mva graph is neighbouring.
|
||
|
*cursor s navigate with plus the block nr in
|
||
|
* each graph from start to end.
|
||
|
*if there is one index whose graph's value is bigger
|
||
|
* than the number we need, that means we found the requeired region.
|
||
|
*/
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
/*find mva graph in range [0x1, 0x200]*/
|
||
|
for (s = 1; (s < ccu_fix_index_start) && (mvaGraph[s] < nr);
|
||
|
s += (mvaGraph[s] & MVA_BLOCK_NR_MASK))
|
||
|
;
|
||
|
/*if we didn't find the proper graph on stage 1,
|
||
|
*we will come to stage 2.
|
||
|
* in the case, all graph in [0x1-0x1FF ]
|
||
|
*is busy. jump ccu mva region.
|
||
|
*/
|
||
|
if (s == ccu_fix_index_start && nr <= ccu_gap_nr) {
|
||
|
s = ccu_fix_index_end + 1;
|
||
|
for (; (s < fix_index0) && (mvaGraph[s] < nr);
|
||
|
s += (mvaGraph[s] & MVA_BLOCK_NR_MASK))
|
||
|
;
|
||
|
}
|
||
|
|
||
|
if (s == fix_index0 && gap_nr >= 1) {
|
||
|
/* stage 2: jump vpu reserved region
|
||
|
*and find it in graph range [0x501-0x5FF ]
|
||
|
* MUST check if block number of gap region is enough to alloc.
|
||
|
* if not, we need to alloc from common region
|
||
|
* only vinson have stage2,for vpu reserved region num is 2.
|
||
|
* so for mt6775 and mt6771 we skip stage2.
|
||
|
*/
|
||
|
#if defined(CONFIG_MACH_MT6758)
|
||
|
if (nr <= gap_nr) {
|
||
|
M4ULOG_LOW(
|
||
|
"stage 2: stopped cursor(%d) on stage 1\n",
|
||
|
s);
|
||
|
s = gap_start_idx;
|
||
|
for (; (s <= gap_end_idx) && (mvaGraph[s] < nr);
|
||
|
s += (mvaGraph[s] & MVA_BLOCK_NR_MASK))
|
||
|
;
|
||
|
} else
|
||
|
goto stage3;
|
||
|
#else
|
||
|
goto stage3;
|
||
|
#endif
|
||
|
}
|
||
|
/*if we didn't find the proper graph on stage 2,
|
||
|
*we will come to stage 3.
|
||
|
* in the case, the requeired nr may be
|
||
|
*more than gap nr.
|
||
|
* Or, all graph in grap region is busy.
|
||
|
*/
|
||
|
if (s == fix_index1) {
|
||
|
/* stage 3: jump vpu reserved region and find it
|
||
|
*in graph range [VPU_FIX_MVA_END + 1-0xFFF ]
|
||
|
* allocate from common region directly.
|
||
|
*/
|
||
|
stage3:
|
||
|
M4ULOG_LOW("stage 3: stopped cursor(%d) on stage 2\n", s);
|
||
|
/*workaround for disp fb*/
|
||
|
#ifdef WORKAROUND_FOR_DISPLAY_FB
|
||
|
s = MVAGRAPH_INDEX(VPU_FIX_MVA_END + 1) +
|
||
|
(mvaGraph[MVAGRAPH_INDEX(VPU_FIX_MVA_END + 1)] &
|
||
|
MVA_BLOCK_NR_MASK);
|
||
|
#else
|
||
|
s = MVAGRAPH_INDEX(VPU_FIX_MVA_END + 1);
|
||
|
#endif
|
||
|
for (; (s < (MVA_MAX_BLOCK_NR + 1)) && (mvaGraph[s] < nr);
|
||
|
s += (mvaGraph[s] & MVA_BLOCK_NR_MASK))
|
||
|
;
|
||
|
}
|
||
|
|
||
|
/*double check if mva region we got is in vpu reserved region. */
|
||
|
region_status = m4u_check_mva_region(s, nr, priv) ||
|
||
|
__check_ccu_mva_region(s, nr, priv);
|
||
|
if (region_status) {
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
M4UMSG(
|
||
|
"mva_alloc error: fault cursor(0x%x) access vpu region\n",
|
||
|
s);
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
if (s > MVA_MAX_BLOCK_NR) {
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
M4UMSG(
|
||
|
"mva_alloc error: no available MVA region for %d blocks!\n",
|
||
|
nr);
|
||
|
#ifdef M4U_PROFILE
|
||
|
mmprofile_log_ex(M4U_MMP_Events[M4U_MMP_M4U_ERROR],
|
||
|
MMPROFILE_FLAG_PULSE, size, s);
|
||
|
#endif
|
||
|
return 0;
|
||
|
}
|
||
|
/* ----------------------------------------------- */
|
||
|
/* alloc a mva region */
|
||
|
end = s + mvaGraph[s] - 1;
|
||
|
|
||
|
if (unlikely(nr == mvaGraph[s])) {
|
||
|
MVA_SET_BUSY(s);
|
||
|
MVA_SET_BUSY(end);
|
||
|
mvaInfoGraph[s] = priv;
|
||
|
mvaInfoGraph[end] = priv;
|
||
|
} else {
|
||
|
new_end = s + nr - 1;
|
||
|
new_start = new_end + 1;
|
||
|
if (new_end > MVA_MAX_BLOCK_NR ||
|
||
|
new_start > MVA_MAX_BLOCK_NR) {
|
||
|
M4UMSG(
|
||
|
"mva_alloc error: mva region error! nr=%u, new_end=%u, s=%u, mvaGraph=0x%x\n",
|
||
|
nr, new_end, s, mvaGraph[s]);
|
||
|
return 0;
|
||
|
}
|
||
|
/* note: new_start may equals to end */
|
||
|
mvaGraph[new_start] = (mvaGraph[s] - nr);
|
||
|
mvaGraph[new_end] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[s] = mvaGraph[new_end];
|
||
|
mvaGraph[end] = mvaGraph[new_start];
|
||
|
|
||
|
mvaInfoGraph[s] = priv;
|
||
|
mvaInfoGraph[new_end] = priv;
|
||
|
}
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
mvaRegionStart = (unsigned int)s;
|
||
|
|
||
|
return (mvaRegionStart << MVA_BLOCK_SIZE_ORDER) + mva_pageOffset(va);
|
||
|
}
|
||
|
|
||
|
/*m4u_do_mva_alloc_fix is different with m4u_do_mva_alloc which find the free
|
||
|
* and suitable region from the first one to the last.
|
||
|
*Instead, m4u_do_mva_alloc_fix
|
||
|
* is only used to find the fix one started from
|
||
|
*formal parameter [mva] plus formal
|
||
|
* parameter [va]. It's similar to mmap.
|
||
|
*no matter how many [va] is, it will allocates
|
||
|
* the whole fix region. it should notice the following:
|
||
|
*(1) vpu fix region is protected by m4u_check_mva_region.
|
||
|
*(2) if vpu port alloc vpu fix region by this, it must make sure reserved bit
|
||
|
* mustn't be destroyed.
|
||
|
*(3)There is no protection to non-vpu fix region now. if other fix regions are
|
||
|
* taken by some ports, only m4u user can check this.
|
||
|
*(4)because formal parameter [mva] ==
|
||
|
*fix mva region start, va + size - 1 should be
|
||
|
*not more than fix mva region end.
|
||
|
*@Param va[in] virtual address used by cpu,
|
||
|
*we use it to get cpu page table offset.
|
||
|
*@Param mva[in] modified virtual start address used
|
||
|
*by m4u. user must provide the right one.
|
||
|
*@Param size[in] the size of requeired sub range.
|
||
|
*@Param priv[in] mva graph info.
|
||
|
*/
|
||
|
unsigned int m4u_do_mva_alloc_fix(unsigned long va,
|
||
|
unsigned int mva,
|
||
|
unsigned int size,
|
||
|
void *priv)
|
||
|
{
|
||
|
unsigned short nr = 0;
|
||
|
unsigned int startRequire, endRequire, sizeRequire;
|
||
|
unsigned short startIdx = mva >> MVA_BLOCK_SIZE_ORDER;
|
||
|
unsigned short endIdx;
|
||
|
unsigned short region_start, region_end;
|
||
|
int vpu_region_status = 0, is_in_vpu_region = 0;
|
||
|
int ccu_region_status, is_in_ccu_region = 0;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
if (size == 0) {
|
||
|
M4UMSG("%s: size = %d\n", __func__, size);
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
mva = mva | (va & M4U_PAGE_MASK);
|
||
|
/* ----------------------------------------------------- */
|
||
|
/* calculate mva block number */
|
||
|
startRequire = mva & (~MVA_BLOCK_ALIGN_MASK);
|
||
|
endRequire = (mva + size - 1) | MVA_BLOCK_ALIGN_MASK;
|
||
|
sizeRequire = endRequire - startRequire + 1;
|
||
|
nr = (sizeRequire + MVA_BLOCK_ALIGN_MASK) >> MVA_BLOCK_SIZE_ORDER;
|
||
|
|
||
|
ccu_region_status = __check_ccu_mva_region(startIdx, nr, priv);
|
||
|
vpu_region_status = m4u_check_mva_region(startIdx, nr, priv);
|
||
|
if (ccu_region_status == -1 && vpu_region_status == -1)
|
||
|
return 0;
|
||
|
else if (ccu_region_status == 1)
|
||
|
is_in_ccu_region = 1;
|
||
|
else if (vpu_region_status == 1)
|
||
|
is_in_vpu_region = 1;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
region_start = startIdx;
|
||
|
/* find prev head of this region. it may be the following relation:
|
||
|
* |-----------|-------------|----------------|--------|----------|
|
||
|
* 0x0 region_start startIdx endIdx region_end 0xFFF
|
||
|
*/
|
||
|
while (mvaGraph[region_start] == 0)
|
||
|
region_start--;
|
||
|
|
||
|
if (MVA_IS_BUSY(region_start)
|
||
|
|| (MVA_GET_NR(region_start) < nr + startIdx - region_start)) {
|
||
|
M4UMSG("mva is inuse index=0x%x, mvaGraph=0x%x\n",
|
||
|
region_start, mvaGraph[region_start]);
|
||
|
mva = 0;
|
||
|
goto out;
|
||
|
}
|
||
|
|
||
|
/* carveout startIdx~startIdx+nr-1 out of region_start */
|
||
|
endIdx = startIdx + nr - 1;
|
||
|
region_end = region_start + MVA_GET_NR(region_start) - 1;
|
||
|
|
||
|
if (startIdx == region_start && endIdx == region_end) {
|
||
|
/* case 1:
|
||
|
* |-----------|-----------------------------|-------------|
|
||
|
* 0x0 region_start region_end 0xFFF
|
||
|
* startIdx endIdx
|
||
|
* alloc(start) alloc(end)
|
||
|
*/
|
||
|
MVA_SET_BUSY(startIdx);
|
||
|
MVA_SET_BUSY(endIdx);
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
}
|
||
|
} else if (startIdx == region_start) {
|
||
|
/* case 2:
|
||
|
* |-----------|-------------------|----------|-------------|
|
||
|
* 0x0 region_start endIdx region_end 0xFFF
|
||
|
* startIdx
|
||
|
* alloc(start) alloc(end)
|
||
|
*/
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
mvaGraph[endIdx + 1] = region_end - endIdx;
|
||
|
mvaGraph[region_end] = mvaGraph[endIdx + 1];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
}
|
||
|
} else if (endIdx == region_end) {
|
||
|
/* case 3:
|
||
|
* |-----------|------------|---------------|-------------|
|
||
|
* 0x0 region_start startIdx region_end 0xFFF
|
||
|
* endIdx
|
||
|
* alloc(start) alloc(end)
|
||
|
*/
|
||
|
mvaGraph[region_start] = startIdx - region_start;
|
||
|
mvaGraph[startIdx - 1] = mvaGraph[region_start];
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
}
|
||
|
} else {
|
||
|
/* case 4:
|
||
|
* |-----------|-----|-------|--------|------|
|
||
|
* 0x0 region_start startIdx endIdx region_end 0xFFF
|
||
|
* alloc(start) alloc(end)
|
||
|
*/
|
||
|
mvaGraph[region_start] = startIdx - region_start;
|
||
|
mvaGraph[startIdx - 1] = mvaGraph[region_start];
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
mvaGraph[endIdx + 1] = region_end - endIdx;
|
||
|
mvaGraph[region_end] = mvaGraph[endIdx + 1];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
}
|
||
|
|
||
|
}
|
||
|
|
||
|
mvaInfoGraph[startIdx] = priv;
|
||
|
mvaInfoGraph[endIdx] = priv;
|
||
|
|
||
|
out:
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
return mva;
|
||
|
}
|
||
|
|
||
|
/*m4u_do_mva_alloc_start_from is used to allocate
|
||
|
*a mva region start from iova_start to iova_end
|
||
|
* for CCU & VPU. if [iova_start, iova_end] is busy,
|
||
|
*it will return alloc failed.
|
||
|
* for example: if [iova_start, iova_end] =
|
||
|
*[0x10000000, 0xffffffff] size = 0x100000,
|
||
|
* we will find a size = 0x100000 region started from
|
||
|
*0x10000000 to 0xfffffff, until there is
|
||
|
* a right one.
|
||
|
* NOTICE: vpu should not use the api.
|
||
|
*@Param va[in] virtual address used by cpu,
|
||
|
*we use it to get cpu page table offset.
|
||
|
*@Param mva[in] modified virtual start address used by
|
||
|
*m4u. user must provide the right one.
|
||
|
*@Param size[in] the size of requeired sub range.
|
||
|
*@Param priv[in] mva graph info.
|
||
|
*/
|
||
|
unsigned int m4u_do_mva_alloc_start_from(unsigned long va,
|
||
|
unsigned int mva,
|
||
|
unsigned int size,
|
||
|
void *priv)
|
||
|
{
|
||
|
unsigned short s = 0, end;
|
||
|
unsigned short new_start, new_end;
|
||
|
unsigned short nr = 0;
|
||
|
unsigned int mvaRegionStart;
|
||
|
unsigned long startRequire, endRequire, sizeRequire;
|
||
|
unsigned short startIdx, endIdx;
|
||
|
unsigned short region_start, region_end, next_region_start = 0;
|
||
|
int vpu_region_status = 0, is_in_vpu_region = 0;
|
||
|
int ccu_region_status, is_in_ccu_region = 0;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
if (size == 0 || priv == NULL) {
|
||
|
M4UMSG("%s: invalid size & port info\n", __func__);
|
||
|
return 0;
|
||
|
}
|
||
|
/*TODO:Need to check if sub range in fix
|
||
|
*range need to align to MVA_BLOCK_ALIGN_MASK
|
||
|
*/
|
||
|
/* find this region[startIdx ~ endIdx].
|
||
|
* |-----------|-|-----|----|--------|
|
||
|
* 0x0 iova_start_idx iova_end_idx 0xFFF
|
||
|
* startIdx endIdx
|
||
|
*/
|
||
|
startIdx = (mva + MVA_BLOCK_ALIGN_MASK) >> MVA_BLOCK_SIZE_ORDER;
|
||
|
|
||
|
/* ----------------------------------------------------- */
|
||
|
/* calculate mva block number */
|
||
|
startRequire = va & (~M4U_PAGE_MASK);
|
||
|
endRequire = (va + size - 1) | M4U_PAGE_MASK;
|
||
|
sizeRequire = endRequire - startRequire + 1;
|
||
|
nr = (sizeRequire + MVA_BLOCK_ALIGN_MASK) >> MVA_BLOCK_SIZE_ORDER;
|
||
|
endIdx = startIdx + nr - 1;
|
||
|
|
||
|
/*check [startIdx, endIdx] status*/
|
||
|
ccu_region_status = __check_ccu_mva_region(startIdx, nr, priv);
|
||
|
vpu_region_status = m4u_check_mva_region(startIdx, nr, priv);
|
||
|
if (ccu_region_status == -1 && vpu_region_status == -1)
|
||
|
return 0;
|
||
|
else if (ccu_region_status == 1)
|
||
|
is_in_ccu_region = 1;
|
||
|
else if (vpu_region_status == 1)
|
||
|
is_in_vpu_region = 1;
|
||
|
|
||
|
|
||
|
M4ULOG_LOW(
|
||
|
"%s: iova_start_idx:0x%x, startIdx=0x%x, endIdx = 0x%x, nr= 0x%x\n",
|
||
|
__func__, MVAGRAPH_INDEX(mva), startIdx, endIdx, nr);
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
/* use cursor region_start to find the region after the
|
||
|
* region including the "startIdx"th block.
|
||
|
* if we find the startIdx's neighbour and graph, we
|
||
|
*maybe need to split it.
|
||
|
* |-----|------|-|---|--------|------|----|-----|
|
||
|
* 0x0 iova_start_idx iova_end_idx 0xFFF
|
||
|
* startIdx endIdx(may be here)
|
||
|
* region start next region start next region end
|
||
|
*/
|
||
|
for (region_start = 1; region_start < (MVA_MAX_BLOCK_NR + 1);
|
||
|
region_start += MVA_GET_NR(region_start)) {
|
||
|
/*error check*/
|
||
|
if ((mvaGraph[region_start] & MVA_BLOCK_NR_MASK) == 0) {
|
||
|
m4u_mvaGraph_dump();
|
||
|
m4u_aee_print("%s: s=%d, 0x%x\n",
|
||
|
__func__, s, mvaGraph[region_start]);
|
||
|
}
|
||
|
if ((region_start + MVA_GET_NR(region_start)) > startIdx) {
|
||
|
next_region_start =
|
||
|
region_start + MVA_GET_NR(region_start);
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
if (region_start > MVA_MAX_BLOCK_NR) {
|
||
|
M4UMSG(
|
||
|
"%s:alloc mva fail,no available MVA for %d blocks\n",
|
||
|
__func__, nr);
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return 0;
|
||
|
}
|
||
|
region_end = region_start + MVA_GET_NR(region_start) - 1;
|
||
|
M4ULOG_LOW(
|
||
|
"%s: found region_start(0x%x) region_end(0x%x) next_region_start(0x%x)\n",
|
||
|
__func__, region_start, region_end, next_region_start);
|
||
|
|
||
|
/*if not found, it means error.*/
|
||
|
if (next_region_start == 0) {
|
||
|
M4UMSG("no enough mva to allocate.\n");
|
||
|
m4u_aee_print(
|
||
|
"%s: region_start: %d, region_end= %d, region block count= %d\n",
|
||
|
__func__, region_start,
|
||
|
region_end, MVA_GET_NR(region_start));
|
||
|
}
|
||
|
|
||
|
/*check the found region status.*/
|
||
|
if (MVA_IS_BUSY(region_start)) {
|
||
|
/*if [region_start, region_end] is busy, need to
|
||
|
*use another cursor s to continue.
|
||
|
* |-----|------|-----|---|-----|--|---|
|
||
|
* 0x0 iova_start_idx iova_end_idx 0xFFF
|
||
|
* startIdx endIdx(may be here)
|
||
|
* region start next region start next region end
|
||
|
* s
|
||
|
*/
|
||
|
s = region_start;
|
||
|
M4ULOG_LOW(
|
||
|
"found region is busy. need to traverse again from s=0x%x.\n",
|
||
|
s);
|
||
|
} else {
|
||
|
/*if [region_start, region_end] is free, need to check
|
||
|
*whether the nr of [region_start, region_end] is
|
||
|
*enough to alloc.
|
||
|
*if enough, alloc it. if not, need to continue with "s =
|
||
|
*next_region_start".
|
||
|
* |-----|---|-----|---|------|----|-----|
|
||
|
* 0x0 iova_start_idx iova_end_idx 0xFFF
|
||
|
* startIdx endIdx(may be here)
|
||
|
* region start next region start next region end
|
||
|
* s
|
||
|
*/
|
||
|
if ((region_end - startIdx + 1) < nr) {
|
||
|
s = next_region_start;
|
||
|
M4UINFO(
|
||
|
"the size of found region is not enough. need to traverse again from s=0x%x\n",
|
||
|
s);
|
||
|
} else
|
||
|
M4UINFO(
|
||
|
"found region is free. region_start=%d, s=%d\n",
|
||
|
region_start, s);
|
||
|
}
|
||
|
|
||
|
/* now [region_start, region_end] is not free or enough to allocate.
|
||
|
* so traverse mvaGraph to find the first right one.
|
||
|
* here we should also need to traverse mvaGraph.
|
||
|
*/
|
||
|
if (s != 0) {
|
||
|
/* find first match free region */
|
||
|
for (; s < (MVA_MAX_BLOCK_NR + 1);
|
||
|
s += (mvaGraph[s] & MVA_BLOCK_NR_MASK)) {
|
||
|
/*error check*/
|
||
|
if ((mvaGraph[s] & MVA_BLOCK_NR_MASK) == 0) {
|
||
|
m4u_aee_print("%s: s=%d, 0x%x\n",
|
||
|
__func__, s, mvaGraph[s]);
|
||
|
m4u_mvaGraph_dump();
|
||
|
}
|
||
|
|
||
|
if (MVA_GET_NR(s) > nr && !MVA_IS_BUSY(s)) {
|
||
|
/*check [s, s + MVA_GET_NR(s) -1] status*/
|
||
|
ccu_region_status = __check_ccu_mva_region(
|
||
|
s, MVA_GET_NR(s), priv);
|
||
|
vpu_region_status = m4u_check_mva_region(
|
||
|
s, MVA_GET_NR(s), priv);
|
||
|
if (ccu_region_status == -1 &&
|
||
|
vpu_region_status == -1)
|
||
|
continue;
|
||
|
else if (ccu_region_status == 1) {
|
||
|
is_in_ccu_region = 1;
|
||
|
break;
|
||
|
} else if (vpu_region_status == 1) {
|
||
|
is_in_vpu_region = 1;
|
||
|
break;
|
||
|
} else if (ccu_region_status == 0 &&
|
||
|
vpu_region_status == 0)
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (s > MVA_MAX_BLOCK_NR) {
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
M4UMSG(
|
||
|
"mva_alloc error: no available MVA region for %d blocks!\n",
|
||
|
nr);
|
||
|
#ifdef M4U_PROFILE
|
||
|
mmprofile_log_ex(M4U_MMP_Events[M4U_MMP_M4U_ERROR],
|
||
|
MMPROFILE_FLAG_PULSE, size, s);
|
||
|
#endif
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
if (s != 0)
|
||
|
M4ULOG_MID("after 2nd traverse, found region s = 0x%x\n", s);
|
||
|
|
||
|
/* ----------------------------------------------- */
|
||
|
/*s==0 means startIdx == mva_start >> 20*/
|
||
|
if (s == 0) {
|
||
|
/*[region_start, region_end] is free or enough to allocate.
|
||
|
*it's same to m4u_do_mva_alloc_fix
|
||
|
*/
|
||
|
if (startIdx == region_start && endIdx == region_end) {
|
||
|
MVA_SET_BUSY(startIdx);
|
||
|
MVA_SET_BUSY(endIdx);
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
}
|
||
|
|
||
|
} else if (startIdx == region_start) {
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
mvaGraph[endIdx + 1] = region_end - endIdx;
|
||
|
mvaGraph[region_end] = mvaGraph[endIdx + 1];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
}
|
||
|
} else if (endIdx == region_end) {
|
||
|
mvaGraph[region_start] = startIdx - region_start;
|
||
|
mvaGraph[startIdx - 1] = mvaGraph[region_start];
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
}
|
||
|
|
||
|
} else {
|
||
|
mvaGraph[region_start] = startIdx - region_start;
|
||
|
mvaGraph[startIdx - 1] = mvaGraph[region_start];
|
||
|
mvaGraph[startIdx] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[endIdx] = mvaGraph[startIdx];
|
||
|
mvaGraph[endIdx + 1] = region_end - endIdx;
|
||
|
mvaGraph[region_end] = mvaGraph[endIdx + 1];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(region_start);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
MVA_SET_RESERVED(region_end);
|
||
|
}
|
||
|
|
||
|
}
|
||
|
|
||
|
mvaInfoGraph[startIdx] = priv;
|
||
|
mvaInfoGraph[endIdx] = priv;
|
||
|
s = startIdx;
|
||
|
} else {
|
||
|
/* alloc a mva region */
|
||
|
end = s + MVA_GET_NR(s) - 1;
|
||
|
|
||
|
/*check [startIdx, endIdx] status*/
|
||
|
is_in_ccu_region = 0;
|
||
|
is_in_vpu_region = 0;
|
||
|
ccu_region_status = __check_ccu_mva_region(s, nr, priv);
|
||
|
if (ccu_region_status == 1)
|
||
|
is_in_ccu_region = 1;
|
||
|
vpu_region_status = m4u_check_mva_region(s,
|
||
|
MVA_GET_NR(s), priv);
|
||
|
if (vpu_region_status == 1)
|
||
|
is_in_vpu_region = 1;
|
||
|
|
||
|
if (unlikely(nr == MVA_GET_NR(s))) {
|
||
|
MVA_SET_BUSY(s);
|
||
|
MVA_SET_BUSY(end);
|
||
|
mvaInfoGraph[s] = priv;
|
||
|
mvaInfoGraph[end] = priv;
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(s);
|
||
|
MVA_SET_RESERVED(end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(s);
|
||
|
MVA_SET_RESERVED(end);
|
||
|
}
|
||
|
} else {
|
||
|
new_end = s + nr - 1;
|
||
|
new_start = new_end + 1;
|
||
|
/* note: new_start may equals to end */
|
||
|
mvaGraph[new_start] = (MVA_GET_NR(s) - nr);
|
||
|
mvaGraph[new_end] = nr | MVA_BUSY_MASK;
|
||
|
mvaGraph[s] = mvaGraph[new_end];
|
||
|
mvaGraph[end] = mvaGraph[new_start];
|
||
|
if (is_in_ccu_region) {
|
||
|
MVA_SET_RESERVED(new_start);
|
||
|
MVA_SET_RESERVED(new_end);
|
||
|
MVA_SET_RESERVED(s);
|
||
|
MVA_SET_RESERVED(end);
|
||
|
} else if (is_in_vpu_region) {
|
||
|
MVA_SET_RESERVED(new_start);
|
||
|
MVA_SET_RESERVED(new_end);
|
||
|
MVA_SET_RESERVED(s);
|
||
|
MVA_SET_RESERVED(end);
|
||
|
}
|
||
|
|
||
|
mvaInfoGraph[s] = priv;
|
||
|
mvaInfoGraph[new_end] = priv;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
mvaRegionStart = (unsigned int)s;
|
||
|
|
||
|
return (mvaRegionStart << MVA_BLOCK_SIZE_ORDER) + mva_pageOffset(va);
|
||
|
}
|
||
|
|
||
|
/*m4u_do_mva_free needs to care if non-vpu port wants to free vpu fix region.*/
|
||
|
#define RightWrong(x) ((x) ? "correct" : "error")
|
||
|
int m4u_do_mva_free(unsigned int mva, unsigned int size)
|
||
|
{
|
||
|
unsigned short startIdx;
|
||
|
unsigned short nr;
|
||
|
unsigned short endIdx;
|
||
|
unsigned int startRequire, endRequire, sizeRequire;
|
||
|
unsigned short nrRequire, nr_tmp = 0;
|
||
|
int vpu_region_status, is_in_vpu_region_flag = 0;
|
||
|
int ccu_region_status, is_in_ccu_region_flag = 0;
|
||
|
int ret = 0;
|
||
|
struct m4u_buf_info_t *p_mva_info;
|
||
|
int port;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
startIdx = mva >> MVA_BLOCK_SIZE_ORDER;
|
||
|
if (startIdx == 0 || startIdx > MVA_MAX_BLOCK_NR) {
|
||
|
M4UMSG("mvaGraph index is 0. mva=0x%x\n", mva);
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return -1;
|
||
|
}
|
||
|
nr = mvaGraph[startIdx] & MVA_BLOCK_NR_MASK;
|
||
|
endIdx = startIdx + nr - 1;
|
||
|
p_mva_info = (struct m4u_buf_info_t *)mvaInfoGraph[startIdx];
|
||
|
|
||
|
if (size == 0 || nr == 0 || p_mva_info == NULL) {
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
M4UMSG("%s error: the input size = %d nr = %d mva_info = %p.\n",
|
||
|
__func__, size, nr, p_mva_info);
|
||
|
if (p_mva_info)
|
||
|
M4UMSG("error port id = %d\n", p_mva_info->port);
|
||
|
m4u_mvaGraph_dump();
|
||
|
return -1;
|
||
|
}
|
||
|
port = p_mva_info->port;
|
||
|
|
||
|
/*check if reserved region meets free condition.*/
|
||
|
ccu_region_status = __check_ccu_mva_region(
|
||
|
startIdx, nr, (void *)p_mva_info);
|
||
|
vpu_region_status = m4u_check_mva_region(
|
||
|
startIdx, nr, (void *)p_mva_info);
|
||
|
if (ccu_region_status == -1 && vpu_region_status == -1) {
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
return -1;
|
||
|
} else if (ccu_region_status == 1)
|
||
|
is_in_ccu_region_flag = 1;
|
||
|
else if (vpu_region_status == 1)
|
||
|
is_in_vpu_region_flag = 1;
|
||
|
|
||
|
/* -------------------------------- */
|
||
|
/* check the input arguments */
|
||
|
/* right condition: startIdx is not NULL &&
|
||
|
* region is busy && right module && right size
|
||
|
*/
|
||
|
startRequire = mva & (unsigned int)(~M4U_PAGE_MASK);
|
||
|
endRequire = (mva + size - 1) | (unsigned int)M4U_PAGE_MASK;
|
||
|
sizeRequire = endRequire - startRequire + 1;
|
||
|
nrRequire = (sizeRequire + MVA_BLOCK_ALIGN_MASK)
|
||
|
>> MVA_BLOCK_SIZE_ORDER;
|
||
|
/* (sizeRequire>>MVA_BLOCK_SIZE_ORDER) +
|
||
|
* ((sizeRequire&MVA_BLOCK_ALIGN_MASK)!=0);
|
||
|
*/
|
||
|
if (!(startIdx != 0 /* startIdx is not NULL */
|
||
|
&& MVA_IS_BUSY(startIdx)
|
||
|
&& (nr == nrRequire))) {
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
M4UMSG("error to free mva========================>\n");
|
||
|
M4UMSG("BufSize=%d(unit:0x%xBytes) (expect %d) [%s]\n",
|
||
|
nrRequire, MVA_BLOCK_SIZE, nr,
|
||
|
RightWrong(nrRequire == nr));
|
||
|
M4UMSG("mva=0x%x, (IsBusy?)=%d (expect %d) [%s]\n",
|
||
|
mva, MVA_IS_BUSY(startIdx), 1,
|
||
|
RightWrong(MVA_IS_BUSY(startIdx)));
|
||
|
m4u_mvaGraph_dump();
|
||
|
/* m4u_mvaGraph_dump_raw(); */
|
||
|
return -1;
|
||
|
}
|
||
|
|
||
|
mvaInfoGraph[startIdx] = NULL;
|
||
|
mvaInfoGraph[endIdx] = NULL;
|
||
|
|
||
|
/*now do the mva deallocation. if mva to be freed is in vpu fix region,
|
||
|
*we should set reserved bit besides merging mva graph. if not in vpu
|
||
|
*fix region, we should make sure reserved bit
|
||
|
* is protected from destroying.
|
||
|
*/
|
||
|
if (is_in_ccu_region_flag || is_in_vpu_region_flag) {
|
||
|
/* merge with followed region.
|
||
|
* if endIdx + 1 is in vpu region, do the merge as before.
|
||
|
* or, since we have set the reserved bit when alloc,
|
||
|
* it's ok to do nothing.
|
||
|
*/
|
||
|
nr_tmp = MVA_GET_NR(endIdx + 1);
|
||
|
if (is_in_ccu_region(endIdx + 1, nr_tmp) &&
|
||
|
!MVA_IS_BUSY(endIdx + 1)) {
|
||
|
nr += nr_tmp;
|
||
|
mvaGraph[endIdx] = 0;
|
||
|
mvaGraph[endIdx + 1] = 0;
|
||
|
MVA_SET_RESERVED(endIdx);
|
||
|
MVA_SET_RESERVED(endIdx + 1);
|
||
|
}
|
||
|
/* merge with previous region
|
||
|
* same operation as merging with followed region
|
||
|
*/
|
||
|
nr_tmp = MVA_GET_NR(startIdx - 1);
|
||
|
if (is_in_ccu_region(
|
||
|
GET_START_INDEX((startIdx - 1), nr_tmp), nr_tmp)
|
||
|
&& (!MVA_IS_BUSY(startIdx - 1))) {
|
||
|
int pre_nr = nr_tmp;
|
||
|
|
||
|
mvaGraph[startIdx] = 0;
|
||
|
mvaGraph[startIdx - 1] = 0;
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(startIdx - 1);
|
||
|
startIdx -= pre_nr;
|
||
|
nr += pre_nr;
|
||
|
}
|
||
|
/* -------------------------------- */
|
||
|
/* set region flags */
|
||
|
mvaGraph[startIdx] = nr;
|
||
|
mvaGraph[startIdx + nr - 1] = nr;
|
||
|
MVA_SET_RESERVED(startIdx);
|
||
|
MVA_SET_RESERVED(startIdx + nr - 1);
|
||
|
} else {
|
||
|
/* merge with followed region.*/
|
||
|
nr_tmp = MVA_GET_NR(endIdx + 1);
|
||
|
if ((endIdx + 1 <= MVA_MAX_BLOCK_NR) &&
|
||
|
(!MVA_IS_BUSY(endIdx + 1))) {
|
||
|
/* check if the followed region is in vpu region.
|
||
|
* yes -> do nothing
|
||
|
* no -> do the merging
|
||
|
*/
|
||
|
if (!MVA_IS_RESERVED(endIdx + 1)) {
|
||
|
nr += nr_tmp;
|
||
|
mvaGraph[endIdx] = 0;
|
||
|
mvaGraph[endIdx + 1] = 0;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/* merge with previous region
|
||
|
* same operation as merging with followed region
|
||
|
*/
|
||
|
nr_tmp = MVA_GET_NR(startIdx - 1);
|
||
|
if ((startIdx - 1) > 0 && (!MVA_IS_BUSY(startIdx - 1))) {
|
||
|
if (!MVA_IS_RESERVED(startIdx - 1)) {
|
||
|
int pre_nr = nr_tmp;
|
||
|
|
||
|
mvaGraph[startIdx] = 0;
|
||
|
mvaGraph[startIdx - 1] = 0;
|
||
|
startIdx -= pre_nr;
|
||
|
nr += pre_nr;
|
||
|
}
|
||
|
}
|
||
|
/* -------------------------------- */
|
||
|
/* set region flags */
|
||
|
mvaGraph[startIdx] = nr;
|
||
|
mvaGraph[startIdx + nr - 1] = nr;
|
||
|
}
|
||
|
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
|
||
|
/*for debug*/
|
||
|
ret = check_reserved_region_integrity(
|
||
|
MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START),
|
||
|
VPU_RESET_VECTOR_BLOCK_NR);
|
||
|
if (!ret)
|
||
|
M4UMSG(
|
||
|
"VPU_RESET_VECTOR region is corruptted when port(%d) free mva(0x%x)\n",
|
||
|
port, mva);
|
||
|
|
||
|
ret = check_reserved_region_integrity(MVAGRAPH_INDEX(VPU_FIX_MVA_START),
|
||
|
VPU_FIX_BLOCK_NR);
|
||
|
if (!ret)
|
||
|
M4UMSG(
|
||
|
"VPU reserved data region is corruptted when port(%d) free mva(0x%x)\n",
|
||
|
port, mva);
|
||
|
|
||
|
return 0;
|
||
|
}
|
||
|
|
||
|
unsigned int get_last_free_graph_idx_in_stage1_region(void)
|
||
|
{
|
||
|
unsigned int index, nr;
|
||
|
unsigned long irq_flags;
|
||
|
|
||
|
index = MVAGRAPH_INDEX(VPU_RESET_VECTOR_FIX_MVA_START) - 1;
|
||
|
spin_lock_irqsave(&gMvaGraph_lock, irq_flags);
|
||
|
nr = MVA_GET_NR(index);
|
||
|
index = GET_START_INDEX(index, nr);
|
||
|
spin_unlock_irqrestore(&gMvaGraph_lock, irq_flags);
|
||
|
return index;
|
||
|
}
|
||
|
|
||
|
unsigned int get_first_free_idx(void)
|
||
|
{
|
||
|
int first_valid_mva, first_free_idx;
|
||
|
|
||
|
first_valid_mva = get_first_valid_mva();
|
||
|
first_free_idx = MVAGRAPH_INDEX(first_valid_mva) +
|
||
|
MVA_GET_NR(first_valid_mva);
|
||
|
return first_free_idx;
|
||
|
}
|