github/acrn-hypervisor
1
0
Fork 0
mirror of https://github.com/projectacrn/acrn-hypervisor.git synced 2025-08-10 20:43:48 +00:00
Code Issues Projects Releases Wiki Activity

hv: support CAT on hybrid platform

Browse Source 
On hybrid platform(e.g. ADL), there may be multiple instances of same level caches for different type of processors,
The current design only supports one global `rdt_info` for each RDT resource type.
In order to support hybrid platform, this patch introduce `rdt_ins` to represents the "instance".
Also, the number of `rdt_info` is dynamically generated by config-tool to match with physical board.

Tracked-On: projectacrn#6690
Signed-off-by: Tw <wei.tan@intel.com>
Acked-by: Eddie Dong <eddie.dong@Intel.com>
This commit is contained in:
Tw 2022-03-28 09:35:51 +08:00 committed by acrnsi-robot
parent 19da21c898
commit 3c384a489c
4 changed files with 89 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
hypervisor/arch/x86/guest/vcat.c
View File

@ -24,7 +24,10 @@
*/
bool is_l2_vcat_configured(const struct acrn_vm *vm)
{
return is_vcat_configured(vm) && (get_rdt_res_cap_info(RDT_RESOURCE_L2)->num_closids > 0U);
uint16_t pcpu = ffs64(vm->hw.cpu_affinity);
const struct rdt_ins *ins = get_rdt_res_ins(RDT_RESOURCE_L2, pcpu);
return is_vcat_configured(vm) && ins != NULL && (ins->num_closids > 0U);
}
/**
@ -32,7 +35,10 @@ bool is_l2_vcat_configured(const struct acrn_vm *vm)
*/
bool is_l3_vcat_configured(const struct acrn_vm *vm)
{
return is_vcat_configured(vm) && (get_rdt_res_cap_info(RDT_RESOURCE_L3)->num_closids > 0U);
uint16_t pcpu = ffs64(vm->hw.cpu_affinity);
const struct rdt_ins *ins = get_rdt_res_ins(RDT_RESOURCE_L3, pcpu);
return is_vcat_configured(vm) && ins != NULL && (ins->num_closids > 0U);
}
/**
@ -175,7 +181,7 @@ static bool is_l2_vcbm_msr(const struct acrn_vm *vm, uint32_t vmsr)
/* num_vcbm_msrs = num_vclosids */
uint16_t num_vcbm_msrs = vcat_get_num_vclosids(vm);
return ((get_rdt_res_cap_info(RDT_RESOURCE_L2)->num_closids > 0U)
return (is_l2_vcat_configured(vm)
&& (vmsr >= MSR_IA32_L2_MASK_BASE) && (vmsr < (MSR_IA32_L2_MASK_BASE + num_vcbm_msrs)));
}
@ -187,7 +193,7 @@ static bool is_l3_vcbm_msr(const struct acrn_vm *vm, uint32_t vmsr)
/* num_vcbm_msrs = num_vclosids */
uint16_t num_vcbm_msrs = vcat_get_num_vclosids(vm);
return ((get_rdt_res_cap_info(RDT_RESOURCE_L3)->num_closids > 0U)
return (is_l3_vcat_configured(vm)
&& (vmsr >= MSR_IA32_L3_MASK_BASE) && (vmsr < (MSR_IA32_L3_MASK_BASE + num_vcbm_msrs)));
}

119
hypervisor/arch/x86/rdt.c
View File

@ -23,98 +23,72 @@ const uint16_t hv_clos = 0U;
* each resource's clos max value to have consistent allocation.
*/
#ifdef CONFIG_RDT_ENABLED
static uint16_t common_num_closids = HV_SUPPORTED_MAX_CLOS;
static struct rdt_info res_cap_info[RDT_NUM_RESOURCES] = {
[RDT_RESOURCE_L3] = {
.res.cache = {
.bitmask = 0U,
.cbm_len = 0U,
.msr_qos_cfg = MSR_IA32_L3_QOS_CFG,
},
.num_closids = 0U,
.res_id = RDT_RESID_L3,
.msr_base = MSR_IA32_L3_MASK_BASE,
.platform_clos_array = platform_l3_clos_array,
},
[RDT_RESOURCE_L2] = {
.res.cache = {
.bitmask = 0U,
.cbm_len = 0U,
.msr_qos_cfg = MSR_IA32_L2_QOS_CFG,
},
.num_closids = 0U,
.res_id = RDT_RESID_L2,
.msr_base = MSR_IA32_L2_MASK_BASE,
.platform_clos_array = platform_l2_clos_array,
},
[RDT_RESOURCE_MBA] = {
.res.membw = {
.mba_max = 0U,
.delay_linear = true,
},
.num_closids = 0U,
.res_id = RDT_RESID_MBA,
.msr_base = MSR_IA32_MBA_MASK_BASE,
.platform_clos_array = platform_mba_clos_array,
},
};
/* TODO: once config-tool is ready to generate this information for us, we could remove these static definitions */
extern struct rdt_type res_cap_info[RDT_NUM_RESOURCES];
/*
* @pre res == RDT_RESOURCE_L3 || res == RDT_RESOURCE_L2 || res == RDT_RESOURCE_MBA
*/
const struct rdt_info *get_rdt_res_cap_info(int res)
const struct rdt_ins *get_rdt_res_ins(int res, uint16_t pcpu_id)
{
return &res_cap_info[res];
uint32_t i;
struct rdt_type *info = &res_cap_info[res];
struct rdt_ins *ins = NULL;
for (i = 0U; i < info->num_ins; i++) {
if (bitmap_test(pcpu_id, &info->ins_array[i].cpu_mask)) {
ins = &info->ins_array[i];
break;
}
}
return ins;
}
/*
* @pre res < RDT_NUM_RESOURCES
* @pre res_clos_info[i].mba_delay <= res_cap_info[res].res.membw.mba_max
* @pre length of res_clos_info[i].clos_mask <= cbm_len && all 1's in clos_mask is continuous
*/
static void setup_res_clos_msr(uint16_t pcpu_id, uint16_t res, struct platform_clos_info *res_clos_info)
static void setup_res_clos_msr(uint16_t pcpu_id, struct rdt_type *info, struct rdt_ins *ins)
{
uint16_t i, mask_array_size = common_num_closids;
uint16_t i;
uint32_t msr_index;
uint64_t val;
if (res != RDT_RESOURCE_MBA && res_cap_info[res].res.cache.is_cdp_enabled) {
mask_array_size = mask_array_size << 1U;
uint64_t val = 0;
uint32_t res = info->res_id;
union clos_config *cfg = ins->clos_config_array;
if (res != RDT_RESID_MBA && ins->res.cache.is_cdp_enabled) {
/* enable CDP before setting COS to simplify CAT mask remapping
* and prevent unintended behavior.
*/
msr_write(res_cap_info[res].res.cache.msr_qos_cfg, 0x1UL);
msr_write(info->msr_qos_cfg, 0x1UL);
}
for (i = 0U; i < mask_array_size; i++) {
for (i = 0U; i < ins->num_clos_config; i++) {
switch (res) {
case RDT_RESOURCE_L3:
case RDT_RESOURCE_L2:
val = (uint64_t)res_clos_info[i].value.clos_mask;
val = (uint64_t)cfg[i].clos_mask;
break;
case RDT_RESOURCE_MBA:
val = (uint64_t)res_clos_info[i].value.mba_delay;
val = (uint64_t)cfg[i].mba_delay;
break;
default:
ASSERT(res < RDT_NUM_RESOURCES, "Support only 3 RDT resources. res=%d is invalid", res);
}
msr_index = res_cap_info[res].msr_base + i;
msr_index = info->msr_base + i;
msr_write_pcpu(msr_index, val, pcpu_id);
}
}
void setup_clos(uint16_t pcpu_id)
{
uint16_t i;
uint16_t i, j;
struct rdt_type *info;
struct rdt_ins *ins;
for (i = 0U; i < RDT_NUM_RESOURCES; i++) {
/* If num_closids == 0, the resource is not supported
* so skip setting up resource MSR.
*/
if (res_cap_info[i].num_closids > 0U) {
setup_res_clos_msr(pcpu_id, i, res_cap_info[i].platform_clos_array);
info = &res_cap_info[i];
for (j = 0U; j < info->num_ins; j++) {
ins = &info->ins_array[j];
if (bitmap_test(pcpu_id, &ins->cpu_mask)) {
setup_res_clos_msr(pcpu_id, info, ins);
}
}
}
@ -132,13 +106,32 @@ uint64_t clos2pqr_msr(uint16_t clos)
return pqr_assoc;
}
static bool is_rdt_type_capable(struct rdt_type *info)
{
uint32_t i;
struct rdt_ins *ins;
bool ret = false;
if (info->num_ins > 0U) {
for (i = 0U; i < info->num_ins; i++) {
ins = &info->ins_array[i];
if (ins->num_closids > 0U) {
ret = true;
break;
}
}
}
return ret;
}
bool is_platform_rdt_capable(void)
{
bool ret = false;
if ((res_cap_info[RDT_RESOURCE_L3].num_closids > 0U) ||
(res_cap_info[RDT_RESOURCE_L2].num_closids > 0U) ||
(res_cap_info[RDT_RESOURCE_MBA].num_closids > 0U)) {
if (is_rdt_type_capable(&res_cap_info[RDT_RESOURCE_L3]) ||
is_rdt_type_capable(&res_cap_info[RDT_RESOURCE_L2]) ||
is_rdt_type_capable(&res_cap_info[RDT_RESOURCE_MBA])) {
ret = true;
}

16
hypervisor/include/arch/x86/asm/board.h
View File

@ -15,12 +15,10 @@
/* forward declarations */
struct acrn_vm;
struct platform_clos_info {
union {
uint16_t mba_delay;
uint32_t clos_mask;
}value;
uint32_t msr_index;
/* user configured mask and MSR info for each CLOS*/
union clos_config {
uint16_t mba_delay;
uint32_t clos_mask;
};
struct vmsix_on_msi_info {
@ -31,9 +29,9 @@ struct vmsix_on_msi_info {
extern struct dmar_info plat_dmar_info;
#ifdef CONFIG_RDT_ENABLED
extern struct platform_clos_info platform_l2_clos_array[MAX_CACHE_CLOS_NUM_ENTRIES];
extern struct platform_clos_info platform_l3_clos_array[MAX_CACHE_CLOS_NUM_ENTRIES];
extern struct platform_clos_info platform_mba_clos_array[MAX_MBA_CLOS_NUM_ENTRIES];
extern union clos_config platform_l2_clos_array[MAX_CACHE_CLOS_NUM_ENTRIES];
extern union clos_config platform_l3_clos_array[MAX_CACHE_CLOS_NUM_ENTRIES];
extern union clos_config platform_mba_clos_array[MAX_MBA_CLOS_NUM_ENTRIES];
#endif
extern const struct cpu_state_table board_cpu_state_tbl;

22
hypervisor/include/arch/x86/asm/rdt.h
View File

@ -22,15 +22,14 @@ enum {
extern const uint16_t hv_clos;
/* The intel Resource Director Tech(RDT) based Allocation Tech support */
struct rdt_info {
/* The instance of one RES_ID */
struct rdt_ins {
union {
struct {
uint32_t bitmask; /* A bitmask where each set bit indicates the corresponding cache way
may be used by other entities in the platform (e.g. GPU) */
uint16_t cbm_len; /* Length of Cache mask in bits */
bool is_cdp_enabled; /* True if support CDP */
uint32_t msr_qos_cfg; /* MSR addr to IA32_L3/L2_QOS_CFG */
} cache;
struct rdt_membw {
uint16_t mba_max; /* Max MBA delay throttling value supported */
@ -38,14 +37,25 @@ struct rdt_info {
} membw;
} res;
uint16_t num_closids; /* Number of CLOSIDs available, 0 indicates resource is not supported.*/
uint32_t res_id;
uint32_t num_clos_config; /* Number of element in clos_config_array */
union clos_config *clos_config_array;
uint64_t cpu_mask; /* the CPUs this RDT applies */
};
/* The intel Resource Director Tech(RDT) based Allocation Tech support */
struct rdt_type {
uint32_t res_id; /* RDT_RESID_L3/RDT_RESID_L2/RDT_RESID_MBA */
uint32_t msr_qos_cfg; /* MSR addr to IA32_L3/L2_QOS_CFG */
uint32_t msr_base; /* MSR base to program clos value */
struct platform_clos_info *platform_clos_array; /* user configured mask and MSR info for each CLOS*/
uint32_t num_ins; /* Number of element in ins_array */
struct rdt_ins *ins_array;
};
void setup_clos(uint16_t pcpu_id);
uint64_t clos2pqr_msr(uint16_t clos);
bool is_platform_rdt_capable(void);
const struct rdt_info *get_rdt_res_cap_info(int res);
const struct rdt_ins *get_rdt_res_ins(int res, uint16_t pcpu_id);
#endif /* RDT_H */