arm-trusted-firmware/services/std_svc/psci/psci_setup.c

369 lines
13 KiB
C
Raw Normal View History

2013-10-25 09:08:21 +01:00
/*
* Copyright (c) 2013-2014, ARM Limited and Contributors. All rights reserved.
2013-10-25 09:08:21 +01:00
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* Neither the name of ARM nor the names of its contributors may be used
* to endorse or promote products derived from this software without specific
* prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include <arch.h>
2013-10-25 09:08:21 +01:00
#include <arch_helpers.h>
#include <assert.h>
#include <bl_common.h>
#include <context.h>
#include <context_mgmt.h>
#include <platform.h>
#include <stddef.h>
#include "psci_private.h"
/*******************************************************************************
* Per cpu non-secure contexts used to program the architectural state prior
* return to the normal world.
* TODO: Use the memory allocator to set aside memory for the contexts instead
* of relying on platform defined constants. Using PSCI_NUM_AFFS will be an
* overkill.
******************************************************************************/
static cpu_context_t psci_ns_context[PLATFORM_CORE_COUNT];
2013-10-25 09:08:21 +01:00
/*******************************************************************************
* In a system, a certain number of affinity instances are present at an
* affinity level. The cumulative number of instances across all levels are
* stored in 'psci_aff_map'. The topology tree has been flattenned into this
* array. To retrieve nodes, information about the extents of each affinity
* level i.e. start index and end index needs to be present. 'psci_aff_limits'
* stores this information.
******************************************************************************/
static aff_limits_node_t psci_aff_limits[MPIDR_MAX_AFFLVL + 1];
2013-10-25 09:08:21 +01:00
/*******************************************************************************
* Routines for retrieving the node corresponding to an affinity level instance
* in the mpidr. The first one uses binary search to find the node corresponding
* to the mpidr (key) at a particular affinity level. The second routine decides
* extents of the binary search at each affinity level.
******************************************************************************/
static int psci_aff_map_get_idx(unsigned long key,
int min_idx,
int max_idx)
{
int mid;
/*
* Terminating condition: If the max and min indices have crossed paths
* during the binary search then the key has not been found.
*/
if (max_idx < min_idx)
return PSCI_E_INVALID_PARAMS;
/*
* Make sure we are within array limits.
*/
assert(min_idx >= 0 && max_idx < PSCI_NUM_AFFS);
2013-10-25 09:08:21 +01:00
/*
* Bisect the array around 'mid' and then recurse into the array chunk
* where the key is likely to be found. The mpidrs in each node in the
* 'psci_aff_map' for a given affinity level are stored in an ascending
* order which makes the binary search possible.
*/
mid = min_idx + ((max_idx - min_idx) >> 1); /* Divide by 2 */
2013-10-25 09:08:21 +01:00
if (psci_aff_map[mid].mpidr > key)
return psci_aff_map_get_idx(key, min_idx, mid - 1);
else if (psci_aff_map[mid].mpidr < key)
return psci_aff_map_get_idx(key, mid + 1, max_idx);
else
return mid;
}
aff_map_node_t *psci_get_aff_map_node(unsigned long mpidr, int aff_lvl)
2013-10-25 09:08:21 +01:00
{
int rc;
if (aff_lvl > get_max_afflvl())
return NULL;
2013-10-25 09:08:21 +01:00
/* Right shift the mpidr to the required affinity level */
mpidr = mpidr_mask_lower_afflvls(mpidr, aff_lvl);
rc = psci_aff_map_get_idx(mpidr,
psci_aff_limits[aff_lvl].min,
psci_aff_limits[aff_lvl].max);
if (rc >= 0)
return &psci_aff_map[rc];
else
return NULL;
}
/*******************************************************************************
* This function populates an array with nodes corresponding to a given range of
* affinity levels in an mpidr. It returns successfully only when the affinity
* levels are correct, the mpidr is valid i.e. no affinity level is absent from
* the topology tree & the affinity instance at level 0 is not absent.
******************************************************************************/
int psci_get_aff_map_nodes(unsigned long mpidr,
int start_afflvl,
int end_afflvl,
aff_map_node_t *mpidr_nodes[])
{
int rc = PSCI_E_INVALID_PARAMS, level;
aff_map_node_t *node;
rc = psci_check_afflvl_range(start_afflvl, end_afflvl);
if (rc != PSCI_E_SUCCESS)
return rc;
for (level = start_afflvl; level <= end_afflvl; level++) {
/*
* Grab the node for each affinity level. No affinity level
* can be missing as that would mean that the topology tree
* is corrupted.
*/
node = psci_get_aff_map_node(mpidr, level);
if (node == NULL) {
rc = PSCI_E_INVALID_PARAMS;
break;
}
/*
* Skip absent affinity levels unless it's afffinity level 0.
* An absent cpu means that the mpidr is invalid. Save the
* pointer to the node for the present affinity level
*/
if (!(node->state & PSCI_AFF_PRESENT)) {
if (level == MPIDR_AFFLVL0) {
rc = PSCI_E_INVALID_PARAMS;
break;
}
mpidr_nodes[level] = NULL;
} else
mpidr_nodes[level] = node;
}
return rc;
}
2013-10-25 09:08:21 +01:00
/*******************************************************************************
* Function which initializes the 'aff_map_node' corresponding to an affinity
* level instance. Each node has a unique mpidr, level and bakery lock. The data
* field is opaque and holds affinity level specific data e.g. for affinity
* level 0 it contains the index into arrays that hold the secure/non-secure
* state for a cpu that's been turned on/off
******************************************************************************/
static void psci_init_aff_map_node(unsigned long mpidr,
int level,
unsigned int idx)
{
unsigned char state;
uint32_t linear_id;
2013-10-25 09:08:21 +01:00
psci_aff_map[idx].mpidr = mpidr;
psci_aff_map[idx].level = level;
bakery_lock_init(&psci_aff_map[idx].lock);
/*
* If an affinity instance is present then mark it as OFF to begin with.
*/
state = plat_get_aff_state(level, mpidr);
psci_aff_map[idx].state = state;
if (level == MPIDR_AFFLVL0) {
psci: fix affinity level upgrade issue The psci implementation does not track target affinity level requests specified during cpu_suspend calls correctly as per the following example. 1. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0 2. Only the cpu0.cluster0 is powered down while cluster0 remains powered up 3. cpu1.cluster0 calls cpu_off to power itself down to highest possible affinity level 4. cluster0 will be powered off even though cpu0.cluster0 does not allow cluster shutdown This patch introduces reference counts at affinity levels > 0 to track the number of cpus which want an affinity instance at level X to remain powered up. This instance can be turned off only if its reference count is 0. Cpus still undergo the normal state transitions (ON, OFF, ON_PENDING, SUSPEND) but the higher levels can only be either ON or OFF depending upon their reference count. The above issue is thus fixed as follows: 1. cluster0's reference count is incremented by two when cpu0 and cpu1 are initially powered on. 2. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0. This does not affect the cluster0 reference count. 3. Only the cpu0.cluster0 is powered down while cluster0 remains powered up as it has a non-zero reference count. 4. cpu1.cluster0 call cpu_off to power itself down to highest possible affinity level. This decrements the cluster0 reference count. 5. cluster0 is still not powered off since its reference count will at least be 1 due to the restriction placed by cpu0. Change-Id: I433dfe82b946f5f6985b1602c2de87800504f7a9
2013-12-05 16:33:10 +00:00
/*
* Mark the cpu as OFF. Higher affinity level reference counts
* have already been memset to 0
*/
if (state & PSCI_AFF_PRESENT)
psci_set_state(&psci_aff_map[idx], PSCI_STATE_OFF);
/*
* Associate a non-secure context with this affinity
* instance through the context management library.
*/
linear_id = platform_get_core_pos(mpidr);
assert(linear_id < PLATFORM_CORE_COUNT);
/* Invalidate the suspend context for the node */
set_cpu_data_by_index(linear_id,
psci_svc_cpu_data.power_state,
PSCI_INVALID_DATA);
/*
* There is no state associated with the current execution
* context so ensure that any reads of the highest affinity
* level in a powered down state return PSCI_INVALID_DATA.
*/
set_cpu_data_by_index(linear_id,
psci_svc_cpu_data.max_phys_off_afflvl,
PSCI_INVALID_DATA);
cm_set_context_by_mpidr(mpidr,
(void *) &psci_ns_context[linear_id],
NON_SECURE);
2013-10-25 09:08:21 +01:00
}
return;
}
/*******************************************************************************
* Core routine used by the Breadth-First-Search algorithm to populate the
* affinity tree. Each level in the tree corresponds to an affinity level. This
* routine's aim is to traverse to the target affinity level and populate nodes
* in the 'psci_aff_map' for all the siblings at that level. It uses the current
* affinity level to keep track of how many levels from the root of the tree
* have been traversed. If the current affinity level != target affinity level,
* then the platform is asked to return the number of children that each
* affinity instance has at the current affinity level. Traversal is then done
* for each child at the next lower level i.e. current affinity level - 1.
*
* CAUTION: This routine assumes that affinity instance ids are allocated in a
* monotonically increasing manner at each affinity level in a mpidr starting
* from 0. If the platform breaks this assumption then this code will have to
* be reworked accordingly.
******************************************************************************/
static unsigned int psci_init_aff_map(unsigned long mpidr,
unsigned int affmap_idx,
int cur_afflvl,
int tgt_afflvl)
{
unsigned int ctr, aff_count;
assert(cur_afflvl >= tgt_afflvl);
/*
* Find the number of siblings at the current affinity level &
* assert if there are none 'cause then we have been invoked with
* an invalid mpidr.
*/
aff_count = plat_get_aff_count(cur_afflvl, mpidr);
assert(aff_count);
if (tgt_afflvl < cur_afflvl) {
for (ctr = 0; ctr < aff_count; ctr++) {
mpidr = mpidr_set_aff_inst(mpidr, ctr, cur_afflvl);
affmap_idx = psci_init_aff_map(mpidr,
affmap_idx,
cur_afflvl - 1,
tgt_afflvl);
}
} else {
for (ctr = 0; ctr < aff_count; ctr++, affmap_idx++) {
mpidr = mpidr_set_aff_inst(mpidr, ctr, cur_afflvl);
psci_init_aff_map_node(mpidr, cur_afflvl, affmap_idx);
}
/* affmap_idx is 1 greater than the max index of cur_afflvl */
psci_aff_limits[cur_afflvl].max = affmap_idx - 1;
}
return affmap_idx;
}
/*******************************************************************************
* This function initializes the topology tree by querying the platform. To do
* so, it's helper routines implement a Breadth-First-Search. At each affinity
* level the platform conveys the number of affinity instances that exist i.e.
* the affinity count. The algorithm populates the psci_aff_map recursively
* using this information. On a platform that implements two clusters of 4 cpus
* each, the populated aff_map_array would look like this:
*
* <- cpus cluster0 -><- cpus cluster1 ->
* ---------------------------------------------------
* | 0 | 1 | 0 | 1 | 2 | 3 | 0 | 1 | 2 | 3 |
* ---------------------------------------------------
* ^ ^
* cluster __| cpu __|
* limit limit
*
* The first 2 entries are of the cluster nodes. The next 4 entries are of cpus
* within cluster 0. The last 4 entries are of cpus within cluster 1.
* The 'psci_aff_limits' array contains the max & min index of each affinity
* level within the 'psci_aff_map' array. This allows restricting search of a
* node at an affinity level between the indices in the limits array.
******************************************************************************/
int32_t psci_setup(void)
2013-10-25 09:08:21 +01:00
{
unsigned long mpidr = read_mpidr();
int afflvl, affmap_idx, max_afflvl;
aff_map_node_t *node;
2013-10-25 09:08:21 +01:00
psci_plat_pm_ops = NULL;
/* Find out the maximum affinity level that the platform implements */
max_afflvl = get_max_afflvl();
assert(max_afflvl <= MPIDR_MAX_AFFLVL);
/*
* This call traverses the topology tree with help from the platform and
* populates the affinity map using a breadth-first-search recursively.
* We assume that the platform allocates affinity instance ids from 0
* onwards at each affinity level in the mpidr. FIRST_MPIDR = 0.0.0.0
*/
affmap_idx = 0;
for (afflvl = max_afflvl; afflvl >= MPIDR_AFFLVL0; afflvl--) {
affmap_idx = psci_init_aff_map(FIRST_MPIDR,
affmap_idx,
max_afflvl,
afflvl);
}
/*
* Set the bounds for the affinity counts of each level in the map. Also
* flush out the entire array so that it's visible to subsequent power
* management operations. The 'psci_aff_map' array is allocated in
* coherent memory so does not need flushing. The 'psci_aff_limits'
* array is allocated in normal memory. It will be accessed when the mmu
* is off e.g. after reset. Hence it needs to be flushed.
*/
for (afflvl = MPIDR_AFFLVL0; afflvl < max_afflvl; afflvl++) {
psci_aff_limits[afflvl].min =
psci_aff_limits[afflvl + 1].max + 1;
}
flush_dcache_range((unsigned long) psci_aff_limits,
sizeof(psci_aff_limits));
/*
* Mark the affinity instances in our mpidr as ON. No need to lock as
* this is the primary cpu.
*/
mpidr &= MPIDR_AFFINITY_MASK;
psci: fix affinity level upgrade issue The psci implementation does not track target affinity level requests specified during cpu_suspend calls correctly as per the following example. 1. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0 2. Only the cpu0.cluster0 is powered down while cluster0 remains powered up 3. cpu1.cluster0 calls cpu_off to power itself down to highest possible affinity level 4. cluster0 will be powered off even though cpu0.cluster0 does not allow cluster shutdown This patch introduces reference counts at affinity levels > 0 to track the number of cpus which want an affinity instance at level X to remain powered up. This instance can be turned off only if its reference count is 0. Cpus still undergo the normal state transitions (ON, OFF, ON_PENDING, SUSPEND) but the higher levels can only be either ON or OFF depending upon their reference count. The above issue is thus fixed as follows: 1. cluster0's reference count is incremented by two when cpu0 and cpu1 are initially powered on. 2. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0. This does not affect the cluster0 reference count. 3. Only the cpu0.cluster0 is powered down while cluster0 remains powered up as it has a non-zero reference count. 4. cpu1.cluster0 call cpu_off to power itself down to highest possible affinity level. This decrements the cluster0 reference count. 5. cluster0 is still not powered off since its reference count will at least be 1 due to the restriction placed by cpu0. Change-Id: I433dfe82b946f5f6985b1602c2de87800504f7a9
2013-12-05 16:33:10 +00:00
for (afflvl = MPIDR_AFFLVL0; afflvl <= max_afflvl; afflvl++) {
2013-10-25 09:08:21 +01:00
node = psci_get_aff_map_node(mpidr, afflvl);
assert(node);
/* Mark each present node as ON. */
psci: fix affinity level upgrade issue The psci implementation does not track target affinity level requests specified during cpu_suspend calls correctly as per the following example. 1. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0 2. Only the cpu0.cluster0 is powered down while cluster0 remains powered up 3. cpu1.cluster0 calls cpu_off to power itself down to highest possible affinity level 4. cluster0 will be powered off even though cpu0.cluster0 does not allow cluster shutdown This patch introduces reference counts at affinity levels > 0 to track the number of cpus which want an affinity instance at level X to remain powered up. This instance can be turned off only if its reference count is 0. Cpus still undergo the normal state transitions (ON, OFF, ON_PENDING, SUSPEND) but the higher levels can only be either ON or OFF depending upon their reference count. The above issue is thus fixed as follows: 1. cluster0's reference count is incremented by two when cpu0 and cpu1 are initially powered on. 2. cpu0.cluster0 calls cpu_suspend with the target affinity level as 0. This does not affect the cluster0 reference count. 3. Only the cpu0.cluster0 is powered down while cluster0 remains powered up as it has a non-zero reference count. 4. cpu1.cluster0 call cpu_off to power itself down to highest possible affinity level. This decrements the cluster0 reference count. 5. cluster0 is still not powered off since its reference count will at least be 1 due to the restriction placed by cpu0. Change-Id: I433dfe82b946f5f6985b1602c2de87800504f7a9
2013-12-05 16:33:10 +00:00
if (node->state & PSCI_AFF_PRESENT)
psci_set_state(node, PSCI_STATE_ON);
2013-10-25 09:08:21 +01:00
}
platform_setup_pm(&psci_plat_pm_ops);
2013-10-25 09:08:21 +01:00
assert(psci_plat_pm_ops);
return 0;
2013-10-25 09:08:21 +01:00
}