From f99f5f3ea7efd54ba0529c4f2d7c72712918a522 Mon Sep 17 00:00:00 2001 From: Sarah Walker Date: Wed, 22 Nov 2023 16:34:31 +0000 Subject: drm/imagination: Add GPU ID parsing and firmware loading Read the GPU ID register at probe time and select the correct features/quirks/enhancements. Use the GPU ID to form the firmware file name and load the firmware. The features/quirks/enhancements arrays are currently hardcoded in the driver for the supported GPUs. We are looking at moving this information to the firmware image. Changes since v8: - Corrected license identifiers Changes since v7: - Fix kerneldoc for pvr_device_info_set_enhancements() Changes since v5: - Add BRN 71242 to device info Changes since v4: - Retrieve device information from firmware header - Pull forward firmware header parsing from FW infrastructure patch - Use devm_add_action_or_reset to release firmware Changes since v3: - Use drm_dev_{enter,exit} Co-developed-by: Frank Binns Signed-off-by: Frank Binns Co-developed-by: Matt Coster Signed-off-by: Matt Coster Co-developed-by: Donald Robson Signed-off-by: Donald Robson Signed-off-by: Sarah Walker Link: https://lore.kernel.org/r/1ff76f7a5b45c742279c78910f8491b8a5e7f6e6.1700668843.git.donald.robson@imgtec.com Signed-off-by: Maxime Ripard --- drivers/gpu/drm/imagination/pvr_drv.c | 521 +++++++++++++++++++++++++++++++++- 1 file changed, 520 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/imagination/pvr_drv.c') diff --git a/drivers/gpu/drm/imagination/pvr_drv.c b/drivers/gpu/drm/imagination/pvr_drv.c index 597188b31f96..1e4408f33e94 100644 --- a/drivers/gpu/drm/imagination/pvr_drv.c +++ b/drivers/gpu/drm/imagination/pvr_drv.c @@ -3,6 +3,9 @@ #include "pvr_device.h" #include "pvr_drv.h" +#include "pvr_rogue_defs.h" +#include "pvr_rogue_fwif_client.h" +#include "pvr_rogue_fwif_shared.h" #include @@ -87,6 +90,382 @@ pvr_ioctl_get_bo_mmap_offset(struct drm_device *drm_dev, void *raw_args, return -ENOTTY; } +static __always_inline u64 +pvr_fw_version_packed(u32 major, u32 minor) +{ + return ((u64)major << 32) | minor; +} + +static u32 +rogue_get_common_store_partition_space_size(struct pvr_device *pvr_dev) +{ + u32 max_partitions = 0; + u32 tile_size_x = 0; + u32 tile_size_y = 0; + + PVR_FEATURE_VALUE(pvr_dev, tile_size_x, &tile_size_x); + PVR_FEATURE_VALUE(pvr_dev, tile_size_y, &tile_size_y); + PVR_FEATURE_VALUE(pvr_dev, max_partitions, &max_partitions); + + if (tile_size_x == 16 && tile_size_y == 16) { + u32 usc_min_output_registers_per_pix = 0; + + PVR_FEATURE_VALUE(pvr_dev, usc_min_output_registers_per_pix, + &usc_min_output_registers_per_pix); + + return tile_size_x * tile_size_y * max_partitions * + usc_min_output_registers_per_pix; + } + + return max_partitions * 1024; +} + +static u32 +rogue_get_common_store_alloc_region_size(struct pvr_device *pvr_dev) +{ + u32 common_store_size_in_dwords = 512 * 4 * 4; + u32 alloc_region_size; + + PVR_FEATURE_VALUE(pvr_dev, common_store_size_in_dwords, &common_store_size_in_dwords); + + alloc_region_size = common_store_size_in_dwords - (256U * 4U) - + rogue_get_common_store_partition_space_size(pvr_dev); + + if (PVR_HAS_QUIRK(pvr_dev, 44079)) { + u32 common_store_split_point = (768U * 4U * 4U); + + return min(common_store_split_point - (256U * 4U), alloc_region_size); + } + + return alloc_region_size; +} + +static inline u32 +rogue_get_num_phantoms(struct pvr_device *pvr_dev) +{ + u32 num_clusters = 1; + + PVR_FEATURE_VALUE(pvr_dev, num_clusters, &num_clusters); + + return ROGUE_REQ_NUM_PHANTOMS(num_clusters); +} + +static inline u32 +rogue_get_max_coeffs(struct pvr_device *pvr_dev) +{ + u32 max_coeff_additional_portion = ROGUE_MAX_VERTEX_SHARED_REGISTERS; + u32 pending_allocation_shared_regs = 2U * 1024U; + u32 pending_allocation_coeff_regs = 0U; + u32 num_phantoms = rogue_get_num_phantoms(pvr_dev); + u32 tiles_in_flight = 0; + u32 max_coeff_pixel_portion; + + PVR_FEATURE_VALUE(pvr_dev, isp_max_tiles_in_flight, &tiles_in_flight); + max_coeff_pixel_portion = DIV_ROUND_UP(tiles_in_flight, num_phantoms); + max_coeff_pixel_portion *= ROGUE_MAX_PIXEL_SHARED_REGISTERS; + + /* + * Compute tasks on cores with BRN48492 and without compute overlap may lock + * up without two additional lines of coeffs. + */ + if (PVR_HAS_QUIRK(pvr_dev, 48492) && !PVR_HAS_FEATURE(pvr_dev, compute_overlap)) + pending_allocation_coeff_regs = 2U * 1024U; + + if (PVR_HAS_ENHANCEMENT(pvr_dev, 38748)) + pending_allocation_shared_regs = 0; + + if (PVR_HAS_ENHANCEMENT(pvr_dev, 38020)) + max_coeff_additional_portion += ROGUE_MAX_COMPUTE_SHARED_REGISTERS; + + return rogue_get_common_store_alloc_region_size(pvr_dev) + pending_allocation_coeff_regs - + (max_coeff_pixel_portion + max_coeff_additional_portion + + pending_allocation_shared_regs); +} + +static inline u32 +rogue_get_cdm_max_local_mem_size_regs(struct pvr_device *pvr_dev) +{ + u32 available_coeffs_in_dwords = rogue_get_max_coeffs(pvr_dev); + + if (PVR_HAS_QUIRK(pvr_dev, 48492) && PVR_HAS_FEATURE(pvr_dev, roguexe) && + !PVR_HAS_FEATURE(pvr_dev, compute_overlap)) { + /* Driver must not use the 2 reserved lines. */ + available_coeffs_in_dwords -= ROGUE_CSRM_LINE_SIZE_IN_DWORDS * 2; + } + + /* + * The maximum amount of local memory available to a kernel is the minimum + * of the total number of coefficient registers available and the max common + * store allocation size which can be made by the CDM. + * + * If any coeff lines are reserved for tessellation or pixel then we need to + * subtract those too. + */ + return min(available_coeffs_in_dwords, (u32)ROGUE_MAX_PER_KERNEL_LOCAL_MEM_SIZE_REGS); +} + +/** + * pvr_dev_query_gpu_info_get() + * @pvr_dev: Device pointer. + * @args: [IN] Device query arguments containing a pointer to a userspace + * struct drm_pvr_dev_query_gpu_info. + * + * If the query object pointer is NULL, the size field is updated with the + * expected size of the query object. + * + * Returns: + * * 0 on success, or if size is requested using a NULL pointer, or + * * -%E2BIG if the indicated length of the allocation is less than is + * required to contain the copied data, or + * * -%EFAULT if local memory could not be copied to userspace. + */ +static int +pvr_dev_query_gpu_info_get(struct pvr_device *pvr_dev, + struct drm_pvr_ioctl_dev_query_args *args) +{ + struct drm_pvr_dev_query_gpu_info gpu_info = {0}; + int err; + + if (!args->pointer) { + args->size = sizeof(struct drm_pvr_dev_query_gpu_info); + return 0; + } + + gpu_info.gpu_id = + pvr_gpu_id_to_packed_bvnc(&pvr_dev->gpu_id); + gpu_info.num_phantoms = rogue_get_num_phantoms(pvr_dev); + + err = PVR_UOBJ_SET(args->pointer, args->size, gpu_info); + if (err < 0) + return err; + + if (args->size > sizeof(gpu_info)) + args->size = sizeof(gpu_info); + return 0; +} + +/** + * pvr_dev_query_runtime_info_get() + * @pvr_dev: Device pointer. + * @args: [IN] Device query arguments containing a pointer to a userspace + * struct drm_pvr_dev_query_runtime_info. + * + * If the query object pointer is NULL, the size field is updated with the + * expected size of the query object. + * + * Returns: + * * 0 on success, or if size is requested using a NULL pointer, or + * * -%E2BIG if the indicated length of the allocation is less than is + * required to contain the copied data, or + * * -%EFAULT if local memory could not be copied to userspace. + */ +static int +pvr_dev_query_runtime_info_get(struct pvr_device *pvr_dev, + struct drm_pvr_ioctl_dev_query_args *args) +{ + struct drm_pvr_dev_query_runtime_info runtime_info = {0}; + int err; + + if (!args->pointer) { + args->size = sizeof(struct drm_pvr_dev_query_runtime_info); + return 0; + } + + runtime_info.free_list_min_pages = 0; /* FIXME */ + runtime_info.free_list_max_pages = + ROGUE_PM_MAX_FREELIST_SIZE / ROGUE_PM_PAGE_SIZE; + runtime_info.common_store_alloc_region_size = + rogue_get_common_store_alloc_region_size(pvr_dev); + runtime_info.common_store_partition_space_size = + rogue_get_common_store_partition_space_size(pvr_dev); + runtime_info.max_coeffs = rogue_get_max_coeffs(pvr_dev); + runtime_info.cdm_max_local_mem_size_regs = + rogue_get_cdm_max_local_mem_size_regs(pvr_dev); + + err = PVR_UOBJ_SET(args->pointer, args->size, runtime_info); + if (err < 0) + return err; + + if (args->size > sizeof(runtime_info)) + args->size = sizeof(runtime_info); + return 0; +} + +/** + * pvr_dev_query_quirks_get() - Unpack array of quirks at the address given + * in a struct drm_pvr_dev_query_quirks, or gets the amount of space required + * for it. + * @pvr_dev: Device pointer. + * @args: [IN] Device query arguments containing a pointer to a userspace + * struct drm_pvr_dev_query_query_quirks. + * + * If the query object pointer is NULL, the size field is updated with the + * expected size of the query object. + * If the userspace pointer in the query object is NULL, or the count is + * short, no data is copied. + * The count field will be updated to that copied, or if either pointer is + * NULL, that which would have been copied. + * The size field in the query object will be updated to the size copied. + * + * Returns: + * * 0 on success, or if size/count is requested using a NULL pointer, or + * * -%EINVAL if args contained non-zero reserved fields, or + * * -%E2BIG if the indicated length of the allocation is less than is + * required to contain the copied data, or + * * -%EFAULT if local memory could not be copied to userspace. + */ +static int +pvr_dev_query_quirks_get(struct pvr_device *pvr_dev, + struct drm_pvr_ioctl_dev_query_args *args) +{ + /* + * @FIXME - hardcoding of numbers here is intended as an + * intermediate step so the UAPI can be fixed, but requires a + * a refactor in the future to store them in a more appropriate + * location + */ + static const u32 umd_quirks_musthave[] = { + 47217, + 49927, + 62269, + }; + static const u32 umd_quirks[] = { + 48545, + 51764, + }; + struct drm_pvr_dev_query_quirks query; + u32 out[ARRAY_SIZE(umd_quirks_musthave) + ARRAY_SIZE(umd_quirks)]; + size_t out_musthave_count = 0; + size_t out_count = 0; + int err; + + if (!args->pointer) { + args->size = sizeof(struct drm_pvr_dev_query_quirks); + return 0; + } + + err = PVR_UOBJ_GET(query, args->size, args->pointer); + + if (err < 0) + return err; + if (query._padding_c) + return -EINVAL; + + for (int i = 0; i < ARRAY_SIZE(umd_quirks_musthave); i++) { + if (pvr_device_has_uapi_quirk(pvr_dev, umd_quirks_musthave[i])) { + out[out_count++] = umd_quirks_musthave[i]; + out_musthave_count++; + } + } + + for (int i = 0; i < ARRAY_SIZE(umd_quirks); i++) { + if (pvr_device_has_uapi_quirk(pvr_dev, umd_quirks[i])) + out[out_count++] = umd_quirks[i]; + } + + if (!query.quirks) + goto copy_out; + if (query.count < out_count) + return -E2BIG; + + if (copy_to_user(u64_to_user_ptr(query.quirks), out, + out_count * sizeof(u32))) { + return -EFAULT; + } + + query.musthave_count = out_musthave_count; + +copy_out: + query.count = out_count; + err = PVR_UOBJ_SET(args->pointer, args->size, query); + if (err < 0) + return err; + + args->size = sizeof(query); + return 0; +} + +/** + * pvr_dev_query_enhancements_get() - Unpack array of enhancements at the + * address given in a struct drm_pvr_dev_query_enhancements, or gets the amount + * of space required for it. + * @pvr_dev: Device pointer. + * @args: [IN] Device query arguments containing a pointer to a userspace + * struct drm_pvr_dev_query_enhancements. + * + * If the query object pointer is NULL, the size field is updated with the + * expected size of the query object. + * If the userspace pointer in the query object is NULL, or the count is + * short, no data is copied. + * The count field will be updated to that copied, or if either pointer is + * NULL, that which would have been copied. + * The size field in the query object will be updated to the size copied. + * + * Returns: + * * 0 on success, or if size/count is requested using a NULL pointer, or + * * -%EINVAL if args contained non-zero reserved fields, or + * * -%E2BIG if the indicated length of the allocation is less than is + * required to contain the copied data, or + * * -%EFAULT if local memory could not be copied to userspace. + */ +static int +pvr_dev_query_enhancements_get(struct pvr_device *pvr_dev, + struct drm_pvr_ioctl_dev_query_args *args) +{ + /* + * @FIXME - hardcoding of numbers here is intended as an + * intermediate step so the UAPI can be fixed, but requires a + * a refactor in the future to store them in a more appropriate + * location + */ + const u32 umd_enhancements[] = { + 35421, + 42064, + }; + struct drm_pvr_dev_query_enhancements query; + u32 out[ARRAY_SIZE(umd_enhancements)]; + size_t out_idx = 0; + int err; + + if (!args->pointer) { + args->size = sizeof(struct drm_pvr_dev_query_enhancements); + return 0; + } + + err = PVR_UOBJ_GET(query, args->size, args->pointer); + + if (err < 0) + return err; + if (query._padding_a) + return -EINVAL; + if (query._padding_c) + return -EINVAL; + + for (int i = 0; i < ARRAY_SIZE(umd_enhancements); i++) { + if (pvr_device_has_uapi_enhancement(pvr_dev, umd_enhancements[i])) + out[out_idx++] = umd_enhancements[i]; + } + + if (!query.enhancements) + goto copy_out; + if (query.count < out_idx) + return -E2BIG; + + if (copy_to_user(u64_to_user_ptr(query.enhancements), out, + out_idx * sizeof(u32))) { + return -EFAULT; + } + +copy_out: + query.count = out_idx; + err = PVR_UOBJ_SET(args->pointer, args->size, query); + if (err < 0) + return err; + + args->size = sizeof(query); + return 0; +} + /** * pvr_ioctl_dev_query() - IOCTL to copy information about a device * @drm_dev: [IN] DRM device. @@ -111,7 +490,41 @@ static int pvr_ioctl_dev_query(struct drm_device *drm_dev, void *raw_args, struct drm_file *file) { - return -ENOTTY; + struct pvr_device *pvr_dev = to_pvr_device(drm_dev); + struct drm_pvr_ioctl_dev_query_args *args = raw_args; + int idx; + int ret = -EINVAL; + + if (!drm_dev_enter(drm_dev, &idx)) + return -EIO; + + switch ((enum drm_pvr_dev_query)args->type) { + case DRM_PVR_DEV_QUERY_GPU_INFO_GET: + ret = pvr_dev_query_gpu_info_get(pvr_dev, args); + break; + + case DRM_PVR_DEV_QUERY_RUNTIME_INFO_GET: + ret = pvr_dev_query_runtime_info_get(pvr_dev, args); + break; + + case DRM_PVR_DEV_QUERY_QUIRKS_GET: + ret = pvr_dev_query_quirks_get(pvr_dev, args); + break; + + case DRM_PVR_DEV_QUERY_ENHANCEMENTS_GET: + ret = pvr_dev_query_enhancements_get(pvr_dev, args); + break; + + case DRM_PVR_DEV_QUERY_HEAP_INFO_GET: + return -EINVAL; + + case DRM_PVR_DEV_QUERY_STATIC_DATA_AREAS_GET: + return -EINVAL; + } + + drm_dev_exit(idx); + + return ret; } /** @@ -349,6 +762,112 @@ pvr_ioctl_submit_jobs(struct drm_device *drm_dev, void *raw_args, return -ENOTTY; } +int +pvr_get_uobj(u64 usr_ptr, u32 usr_stride, u32 min_stride, u32 obj_size, void *out) +{ + if (usr_stride < min_stride) + return -EINVAL; + + return copy_struct_from_user(out, obj_size, u64_to_user_ptr(usr_ptr), usr_stride); +} + +int +pvr_set_uobj(u64 usr_ptr, u32 usr_stride, u32 min_stride, u32 obj_size, const void *in) +{ + if (usr_stride < min_stride) + return -EINVAL; + + if (copy_to_user(u64_to_user_ptr(usr_ptr), in, min_t(u32, usr_stride, obj_size))) + return -EFAULT; + + if (usr_stride > obj_size && + clear_user(u64_to_user_ptr(usr_ptr + obj_size), usr_stride - obj_size)) { + return -EFAULT; + } + + return 0; +} + +int +pvr_get_uobj_array(const struct drm_pvr_obj_array *in, u32 min_stride, u32 obj_size, void **out) +{ + int ret = 0; + void *out_alloc; + + if (in->stride < min_stride) + return -EINVAL; + + if (!in->count) + return 0; + + out_alloc = kvmalloc_array(in->count, obj_size, GFP_KERNEL); + if (!out_alloc) + return -ENOMEM; + + if (obj_size == in->stride) { + if (copy_from_user(out_alloc, u64_to_user_ptr(in->array), + (unsigned long)obj_size * in->count)) + ret = -EFAULT; + } else { + void __user *in_ptr = u64_to_user_ptr(in->array); + void *out_ptr = out_alloc; + + for (u32 i = 0; i < in->count; i++) { + ret = copy_struct_from_user(out_ptr, obj_size, in_ptr, in->stride); + if (ret) + break; + + out_ptr += obj_size; + in_ptr += in->stride; + } + } + + if (ret) { + kvfree(out_alloc); + return ret; + } + + *out = out_alloc; + return 0; +} + +int +pvr_set_uobj_array(const struct drm_pvr_obj_array *out, u32 min_stride, u32 obj_size, + const void *in) +{ + if (out->stride < min_stride) + return -EINVAL; + + if (!out->count) + return 0; + + if (obj_size == out->stride) { + if (copy_to_user(u64_to_user_ptr(out->array), in, + (unsigned long)obj_size * out->count)) + return -EFAULT; + } else { + u32 cpy_elem_size = min_t(u32, out->stride, obj_size); + void __user *out_ptr = u64_to_user_ptr(out->array); + const void *in_ptr = in; + + for (u32 i = 0; i < out->count; i++) { + if (copy_to_user(out_ptr, in_ptr, cpy_elem_size)) + return -EFAULT; + + out_ptr += obj_size; + in_ptr += out->stride; + } + + if (out->stride > obj_size && + clear_user(u64_to_user_ptr(out->array + obj_size), + out->stride - obj_size)) { + return -EFAULT; + } + } + + return 0; +} + #define DRM_PVR_IOCTL(_name, _func, _flags) \ DRM_IOCTL_DEF_DRV(PVR_##_name, pvr_ioctl_##_func, _flags) -- cgit v1.2.3