diff options
Diffstat (limited to 'src/arm/linux')
-rw-r--r-- | src/arm/linux/aarch32-isa.c | 9 | ||||
-rw-r--r-- | src/arm/linux/aarch64-isa.c | 18 | ||||
-rw-r--r-- | src/arm/linux/chipset.c | 94 | ||||
-rw-r--r-- | src/arm/linux/clusters.c | 20 | ||||
-rw-r--r-- | src/arm/linux/cpuinfo.c | 8 | ||||
-rw-r--r-- | src/arm/linux/init.c | 2 | ||||
-rw-r--r-- | src/arm/linux/midr.c | 6 |
7 files changed, 134 insertions, 23 deletions
diff --git a/src/arm/linux/aarch32-isa.c b/src/arm/linux/aarch32-isa.c index df68aa1..fb95ee9 100644 --- a/src/arm/linux/aarch32-isa.c +++ b/src/arm/linux/aarch32-isa.c @@ -64,6 +64,8 @@ void cpuinfo_arm_linux_decode_isa_from_proc_cpuinfo( * - Processors with Exynos M4 cores * - Processors with Exynos M5 cores * - Neoverse N1 cores + * - Neoverse V1 cores + * - Neoverse N2 cores */ if (chipset->series == cpuinfo_arm_chipset_series_samsung_exynos && chipset->model == 9810) { /* Only little cores of Exynos 9810 support FP16 & RDM */ @@ -73,9 +75,11 @@ void cpuinfo_arm_linux_decode_isa_from_proc_cpuinfo( case UINT32_C(0x4100D050): /* Cortex-A55 */ case UINT32_C(0x4100D060): /* Cortex-A65 */ case UINT32_C(0x4100D0B0): /* Cortex-A76 */ - case UINT32_C(0x4100D0C0): /* Neoverse N1 */ case UINT32_C(0x4100D0D0): /* Cortex-A77 */ case UINT32_C(0x4100D0E0): /* Cortex-A76AE */ + case UINT32_C(0x4100D460): /* Cortex-A510 */ + case UINT32_C(0x4100D470): /* Cortex-A710 */ + case UINT32_C(0x4100D480): /* Cortex-X2 */ case UINT32_C(0x4800D400): /* Cortex-A76 (HiSilicon) */ case UINT32_C(0x51008020): /* Kryo 385 Gold (Cortex-A75) */ case UINT32_C(0x51008030): /* Kryo 385 Silver (Cortex-A55) */ @@ -98,6 +102,9 @@ void cpuinfo_arm_linux_decode_isa_from_proc_cpuinfo( case UINT32_C(0x4100D0D0): /* Cortex-A77 */ case UINT32_C(0x4100D0E0): /* Cortex-A76AE */ case UINT32_C(0x4800D400): /* Cortex-A76 (HiSilicon) */ + case UINT32_C(0x4100D460): /* Cortex-A510 */ + case UINT32_C(0x4100D470): /* Cortex-A710 */ + case UINT32_C(0x4100D480): /* Cortex-X2 */ case UINT32_C(0x51008040): /* Kryo 485 Gold (Cortex-A76) */ case UINT32_C(0x51008050): /* Kryo 485 Silver (Cortex-A55) */ case UINT32_C(0x53000030): /* Exynos-M4 */ diff --git a/src/arm/linux/aarch64-isa.c b/src/arm/linux/aarch64-isa.c index 2000e1a..44a8f4d 100644 --- a/src/arm/linux/aarch64-isa.c +++ b/src/arm/linux/aarch64-isa.c @@ -41,6 +41,8 @@ void cpuinfo_arm64_linux_decode_isa_from_proc_cpuinfo( * - Processors with Exynos M4 cores * - Processors with Exynos M5 cores * - Neoverse N1 cores + * - Neoverse V1 cores + * - Neoverse N2 cores */ if (chipset->series == cpuinfo_arm_chipset_series_samsung_exynos && chipset->model == 9810) { /* Exynos 9810 reports that it supports FP16 compute, but in fact only little cores do */ @@ -54,6 +56,8 @@ void cpuinfo_arm64_linux_decode_isa_from_proc_cpuinfo( case UINT32_C(0x4100D0C0): /* Neoverse N1 */ case UINT32_C(0x4100D0D0): /* Cortex-A77 */ case UINT32_C(0x4100D0E0): /* Cortex-A76AE */ + case UINT32_C(0x4100D400): /* Neoverse V1 */ + case UINT32_C(0x4100D490): /* Neoverse N2 */ case UINT32_C(0x4800D400): /* Cortex-A76 (HiSilicon) */ case UINT32_C(0x51008020): /* Kryo 385 Gold (Cortex-A75) */ case UINT32_C(0x51008030): /* Kryo 385 Silver (Cortex-A55) */ @@ -78,6 +82,9 @@ void cpuinfo_arm64_linux_decode_isa_from_proc_cpuinfo( break; } } + if (features2 & CPUINFO_ARM_LINUX_FEATURE2_I8MM) { + isa->i8mm = true; + } /* * Many phones ship with an old kernel configuration that doesn't report UDOT/SDOT instructions. @@ -89,6 +96,8 @@ void cpuinfo_arm64_linux_decode_isa_from_proc_cpuinfo( case UINT32_C(0x4100D0C0): /* Neoverse N1 */ case UINT32_C(0x4100D0D0): /* Cortex-A77 */ case UINT32_C(0x4100D0E0): /* Cortex-A76AE */ + case UINT32_C(0x4100D400): /* Neoverse V1 */ + case UINT32_C(0x4100D490): /* Neoverse N2 */ case UINT32_C(0x4100D4A0): /* Neoverse E1 */ case UINT32_C(0x4800D400): /* Cortex-A76 (HiSilicon) */ case UINT32_C(0x51008040): /* Kryo 485 Gold (Cortex-A76) */ @@ -124,4 +133,13 @@ void cpuinfo_arm64_linux_decode_isa_from_proc_cpuinfo( if (features2 & CPUINFO_ARM_LINUX_FEATURE2_SVE2) { isa->sve2 = true; } + // SVEBF16 is set iff SVE and BF16 are both supported, but the SVEBF16 feature flag + // was added in Linux kernel before the BF16 feature flag, so we check for either. + if (features2 & (CPUINFO_ARM_LINUX_FEATURE2_BF16 | CPUINFO_ARM_LINUX_FEATURE2_SVEBF16)) { + isa->bf16 = true; + } + if (features & CPUINFO_ARM_LINUX_FEATURE_ASIMDFHM) { + isa->fhm = true; + } } + diff --git a/src/arm/linux/chipset.c b/src/arm/linux/chipset.c index e36283c..f2a002d 100644 --- a/src/arm/linux/chipset.c +++ b/src/arm/linux/chipset.c @@ -1,3 +1,4 @@ +#include <ctype.h> #include <stdbool.h> #include <stdint.h> #include <stdio.h> @@ -281,6 +282,82 @@ static bool match_sm( return true; } + +struct special_map_entry { + const char* platform; + uint16_t model; + uint8_t series; + char suffix; +}; + +static const struct special_map_entry qualcomm_hardware_map_entries[] = { + { + /* "Kona" -> Qualcomm Kona */ + .platform = "Kona", + .series = cpuinfo_arm_chipset_series_qualcomm_snapdragon, + .model = 865, + }, + { + /* "Bengal" -> Qualcomm Bengal */ + .platform = "Bengal", + .series = cpuinfo_arm_chipset_series_qualcomm_snapdragon, + .model = 662, + }, + { + /* "Bengalp" -> Qualcomm Bengalp */ + .platform = "Bengalp", + .series = cpuinfo_arm_chipset_series_qualcomm_snapdragon, + .model = 662, + }, + { + /* "Lito" -> Qualcomm Lito */ + .platform = "Lito", + .series = cpuinfo_arm_chipset_series_qualcomm_snapdragon, + .model = 765, + .suffix = 'G' + }, + { + /* "Lagoon" -> Qualcomm Lagoon */ + .platform = "Lagoon", + .series = cpuinfo_arm_chipset_series_qualcomm_snapdragon, + .model = 0, + }, +}; + + +int strcicmp(char const *a, char const *b) +{ + for (;; a++, b++) { + int d = tolower((unsigned char)*a) - tolower((unsigned char)*b); + if (d != 0 || !*a) + return d; + } +} + +static bool match_qualcomm_special( + const char* start, const char* end, + struct cpuinfo_arm_chipset chipset[restrict static 1]) +{ + for (size_t i = 0; i < CPUINFO_COUNT_OF(qualcomm_hardware_map_entries); i++) { + int length = end - start; + if (strcicmp(qualcomm_hardware_map_entries[i].platform, start) == 0 && + qualcomm_hardware_map_entries[i].platform[length] == 0) + { + *chipset = (struct cpuinfo_arm_chipset) { + .vendor = chipset_series_vendor[qualcomm_hardware_map_entries[i].series], + .series = (enum cpuinfo_arm_chipset_series) qualcomm_hardware_map_entries[i].series, + .model = qualcomm_hardware_map_entries[i].model, + .suffix = { + [0] = qualcomm_hardware_map_entries[i].suffix, + }, + }; + return true; + } + } + return false; + +} + /** * Tries to match /Samsung Exynos\d{4}$/ signature (case-insensitive) for Samsung Exynos chipsets. * If match successful, extracts model information into \p chipset argument. @@ -1351,7 +1428,7 @@ static bool match_and_parse_sunxi( return false; } - /* Compare sunXi platform id and number of cores to tabluted values to decode chipset name */ + /* Compare sunXi platform id and number of cores to tabulated values to decode chipset name */ uint32_t model = 0; char suffix = 0; for (size_t i = 0; i < CPUINFO_COUNT_OF(sunxi_map_entries); i++) { @@ -1752,13 +1829,6 @@ static bool is_tegra(const char* start, const char* end) { return (length == 5 || start[5] == '3'); } -struct special_map_entry { - const char* platform; - uint16_t model; - uint8_t series; - char suffix; -}; - static const struct special_map_entry special_hardware_map_entries[] = { #if CPUINFO_ARCH_ARM { @@ -2317,6 +2387,14 @@ struct cpuinfo_arm_chipset cpuinfo_arm_linux_decode_chipset_from_proc_cpuinfo_ha (int) hardware_length, hardware); return chipset; } + + if (match_qualcomm_special(pos, hardware_end, &chipset)) { + cpuinfo_log_debug( + "matched Qualcomm signature in /proc/cpuinfo Hardware string \"%.*s\"", + (int) hardware_length, hardware); + return chipset; + } + } word_start = false; break; diff --git a/src/arm/linux/clusters.c b/src/arm/linux/clusters.c index c7a4045..430773d 100644 --- a/src/arm/linux/clusters.c +++ b/src/arm/linux/clusters.c @@ -48,7 +48,7 @@ static inline bool bitmask_all(uint32_t bitfield, uint32_t mask) { * @param usable_processors - number of processors in the @p processors array with CPUINFO_LINUX_FLAG_VALID flags. * @param max_processors - number of elements in the @p processors array. * @param[in,out] processors - processor descriptors with pre-parsed POSSIBLE and PRESENT flags, minimum/maximum - * frequency, MIDR infromation, and core cluster (package siblings list) information. + * frequency, MIDR information, and core cluster (package siblings list) information. * * @retval true if the heuristic successfully assigned all processors into clusters of cores. * @retval false if known details about processors contradict the heuristic configuration of core clusters. @@ -292,9 +292,9 @@ bool cpuinfo_arm_linux_detect_core_clusters_by_heuristic( * - Processors assigned to these clusters stay assigned to the same clusters * - No new processors are added to these clusters * - Processors without pre-assigned cluster are clustered in one sequential scan: - * - If known details (min/max frequency, MIDR components) of a processor are compatible with a preceeding - * processor, without pre-assigned cluster, the processor is assigned to the cluster of the preceeding processor. - * - If known details (min/max frequency, MIDR components) of a processor are not compatible with a preceeding + * - If known details (min/max frequency, MIDR components) of a processor are compatible with a preceding + * processor, without pre-assigned cluster, the processor is assigned to the cluster of the preceding processor. + * - If known details (min/max frequency, MIDR components) of a processor are not compatible with a preceding * processor, the processor is assigned to a newly created cluster. * * The function must be called after parsing OS-provided information on core clusters, and usually is called only @@ -309,7 +309,7 @@ bool cpuinfo_arm_linux_detect_core_clusters_by_heuristic( * * @param max_processors - number of elements in the @p processors array. * @param[in,out] processors - processor descriptors with pre-parsed POSSIBLE and PRESENT flags, minimum/maximum - * frequency, MIDR infromation, and core cluster (package siblings list) information. + * frequency, MIDR information, and core cluster (package siblings list) information. * * @retval true if the heuristic successfully assigned all processors into clusters of cores. * @retval false if known details about processors contradict the heuristic configuration of core clusters. @@ -331,7 +331,7 @@ void cpuinfo_arm_linux_detect_core_clusters_by_sequential_scan( if (cluster_flags & CPUINFO_LINUX_FLAG_MIN_FREQUENCY) { if (cluster_min_frequency != processors[i].min_frequency) { cpuinfo_log_info( - "minimum frequency of processor %"PRIu32" (%"PRIu32" KHz) is different than of preceeding cluster (%"PRIu32" KHz); " + "minimum frequency of processor %"PRIu32" (%"PRIu32" KHz) is different than of preceding cluster (%"PRIu32" KHz); " "processor %"PRIu32" starts to a new cluster", i, processors[i].min_frequency, cluster_min_frequency, i); goto new_cluster; @@ -346,7 +346,7 @@ void cpuinfo_arm_linux_detect_core_clusters_by_sequential_scan( if (cluster_flags & CPUINFO_LINUX_FLAG_MAX_FREQUENCY) { if (cluster_max_frequency != processors[i].max_frequency) { cpuinfo_log_debug( - "maximum frequency of processor %"PRIu32" (%"PRIu32" KHz) is different than of preceeding cluster (%"PRIu32" KHz); " + "maximum frequency of processor %"PRIu32" (%"PRIu32" KHz) is different than of preceding cluster (%"PRIu32" KHz); " "processor %"PRIu32" starts a new cluster", i, processors[i].max_frequency, cluster_max_frequency, i); goto new_cluster; @@ -361,7 +361,7 @@ void cpuinfo_arm_linux_detect_core_clusters_by_sequential_scan( if (cluster_flags & CPUINFO_ARM_LINUX_VALID_IMPLEMENTER) { if ((cluster_midr & CPUINFO_ARM_MIDR_IMPLEMENTER_MASK) != (processors[i].midr & CPUINFO_ARM_MIDR_IMPLEMENTER_MASK)) { cpuinfo_log_debug( - "CPU Implementer of processor %"PRIu32" (0x%02"PRIx32") is different than of preceeding cluster (0x%02"PRIx32"); " + "CPU Implementer of processor %"PRIu32" (0x%02"PRIx32") is different than of preceding cluster (0x%02"PRIx32"); " "processor %"PRIu32" starts to a new cluster", i, midr_get_implementer(processors[i].midr), midr_get_implementer(cluster_midr), i); goto new_cluster; @@ -417,11 +417,11 @@ void cpuinfo_arm_linux_detect_core_clusters_by_sequential_scan( } } - /* All checks passed, attach processor to the preceeding cluster */ + /* All checks passed, attach processor to the preceding cluster */ cluster_processors++; processors[i].package_leader_id = cluster_start; processors[i].flags |= CPUINFO_LINUX_FLAG_PACKAGE_CLUSTER; - cpuinfo_log_debug("assigned processor %"PRIu32" to preceeding cluster of processor %"PRIu32, i, cluster_start); + cpuinfo_log_debug("assigned processor %"PRIu32" to preceding cluster of processor %"PRIu32, i, cluster_start); continue; new_cluster: diff --git a/src/arm/linux/cpuinfo.c b/src/arm/linux/cpuinfo.c index 90e1631..817da12 100644 --- a/src/arm/linux/cpuinfo.c +++ b/src/arm/linux/cpuinfo.c @@ -177,6 +177,10 @@ static void parse_features( #if CPUINFO_ARCH_ARM64 processor->features |= CPUINFO_ARM_LINUX_FEATURE_FCMA; #endif + } else if (memcmp(feature_start, "i8mm", feature_length) == 0) { + #if CPUINFO_ARCH_ARM64 + processor->features2 |= CPUINFO_ARM_LINUX_FEATURE2_I8MM; + #endif #if CPUINFO_ARCH_ARM } else if (memcmp(feature_start, "half", feature_length) == 0) { processor->features |= CPUINFO_ARM_LINUX_FEATURE_HALF; @@ -283,6 +287,10 @@ static void parse_features( #if CPUINFO_ARCH_ARM64 processor->features |= CPUINFO_ARM_LINUX_FEATURE_ASIMDRDM; #endif + } else if (memcmp(feature_start, "asimdfhm", feature_length) == 0) { + #if CPUINFO_ARCH_ARM64 + processor->features |= CPUINFO_ARM_LINUX_FEATURE_ASIMDFHM; + #endif #if CPUINFO_ARCH_ARM } else if (memcmp(feature_start, "fastmult", feature_length) == 0) { processor->features |= CPUINFO_ARM_LINUX_FEATURE_FASTMULT; diff --git a/src/arm/linux/init.c b/src/arm/linux/init.c index 23d8439..d3da5a9 100644 --- a/src/arm/linux/init.c +++ b/src/arm/linux/init.c @@ -510,7 +510,7 @@ void cpuinfo_arm_linux_init(void) { uint32_t l2_count = 0, l3_count = 0, big_l3_size = 0, cluster_id = UINT32_MAX; /* Indication whether L3 (if it exists) is shared between all cores */ bool shared_l3 = true; - /* Populate cache infromation structures in l1i, l1d */ + /* Populate cache information structures in l1i, l1d */ for (uint32_t i = 0; i < valid_processors; i++) { if (arm_linux_processors[i].package_leader_id == arm_linux_processors[i].system_processor_id) { cluster_id += 1; diff --git a/src/arm/linux/midr.c b/src/arm/linux/midr.c index 2c3116b..0d8f03f 100644 --- a/src/arm/linux/midr.c +++ b/src/arm/linux/midr.c @@ -675,10 +675,10 @@ static bool cpuinfo_arm_linux_detect_cluster_midr_by_big_little_heuristic( /* * Initializes MIDR for leaders of core clusters in a single sequential scan: - * - Clusters preceeding the first reported MIDR value are assumed to have default MIDR value. + * - Clusters preceding the first reported MIDR value are assumed to have default MIDR value. * - Clusters following any reported MIDR value to have that MIDR value. * - * @param default_midr - MIDR value that will be assigned to cluster leaders preceeding any reported MIDR value. + * @param default_midr - MIDR value that will be assigned to cluster leaders preceding any reported MIDR value. * @param processors_count - number of logical processor descriptions in the @p processors array. * @param[in,out] processors - array of logical processor descriptions with pre-parsed MIDR, maximum frequency, * and decoded core cluster (package_leader_id) information. @@ -833,7 +833,7 @@ uint32_t cpuinfo_arm_linux_detect_cluster_midr( * 2. For systems with 2 clusters and MIDR known for one cluster, assume big.LITTLE configuration, * and estimate MIDR for the other cluster under assumption that MIDR for the big cluster is known. * 3. Initialize MIDRs for core clusters in a single sequential scan: - * - Clusters preceeding the first reported MIDR value are assumed to have the last reported MIDR value. + * - Clusters preceding the first reported MIDR value are assumed to have the last reported MIDR value. * - Clusters following any reported MIDR value to have that MIDR value. */ |