From d5a5fb3563121a4904be9185914c563070ffd021 Mon Sep 17 00:00:00 2001 From: Sunny He Date: Mon, 31 Jul 2017 16:10:14 -0700 Subject: gpu: nvgpu: remove kind map inheritance Currently, the kind map initialization inherits from earlier chips' kind map definitions and simply adds on newly supported features. This is dangerous as changes in older architectures may affect newer architectures in unpredictable ways. This patch removes inheritance between subsequent architectures' kind map initializations. Jira NVGPU-74 Change-Id: Ie5c33f613db7ba109564b4cb2312e47ee3332afc Signed-off-by: Sunny He Reviewed-on: https://git-master.nvidia.com/r/1537745 Reviewed-by: Automatic_Commit_Validation_User Reviewed-by: svccoveritychecker GVS: Gerrit_Virtual_Submit Reviewed-by: Terje Bergstrom --- drivers/gpu/nvgpu/gk20a/kind_gk20a.c | 319 +++++++++++++------------- drivers/gpu/nvgpu/gk20a/kind_gk20a.h | 18 +- drivers/gpu/nvgpu/gm20b/fb_gm20b.c | 383 ++++++++++++++++++++++++++++++-- drivers/gpu/nvgpu/gp10b/fb_gp10b.c | 418 +++++++++++++++++++++++++++++++++-- 4 files changed, 927 insertions(+), 211 deletions(-) diff --git a/drivers/gpu/nvgpu/gk20a/kind_gk20a.c b/drivers/gpu/nvgpu/gk20a/kind_gk20a.c index 57cf028b..1d7a0ebf 100644 --- a/drivers/gpu/nvgpu/gk20a/kind_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/kind_gk20a.c @@ -23,22 +23,6 @@ /* TBD: generate these from kind_macros.h */ -/* TBD: not sure on the work creation for gk20a, doubtful */ -static inline bool gk20a_kind_work_creation_sked(u8 k) -{ - return false; -} -static inline bool gk20a_kind_work_creation_host(u8 k) -{ - return false; -} - -static inline bool gk20a_kind_work_creation(u8 k) -{ - return gk20a_kind_work_creation_sked(k) || - gk20a_kind_work_creation_host(k); -} - /* * Use noinline_for_stack for these huge functions, otherwise GCC 4.7 will * blow up when building a coverage-enabled kernel. (error: the frame size of @@ -48,177 +32,174 @@ static inline bool gk20a_kind_work_creation(u8 k) /* note: taken from the !2cs_compression case */ static noinline_for_stack bool gk20a_kind_supported(u8 k) { - return gk20a_kind_work_creation(k) || - (k == gmmu_pte_kind_invalid_v()) || - (k == gmmu_pte_kind_pitch_v()) || - (k >= gmmu_pte_kind_z16_v() && - k <= gmmu_pte_kind_z16_ms8_2c_v()) || - (k >= gmmu_pte_kind_z16_2z_v() && - k <= gmmu_pte_kind_z16_ms8_2z_v()) || - (k == gmmu_pte_kind_s8z24_v()) || - (k >= gmmu_pte_kind_s8z24_2cz_v() && - k <= gmmu_pte_kind_s8z24_ms8_2cz_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()) || - (k == gmmu_pte_kind_z24s8_v()) || - (k >= gmmu_pte_kind_z24s8_2cz_v() && - k <= gmmu_pte_kind_z24s8_ms8_2cz_v()) || - (k == gmmu_pte_kind_zf32_v()) || - (k >= gmmu_pte_kind_zf32_2cz_v() && - k <= gmmu_pte_kind_zf32_ms8_2cz_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v() && - k <= gmmu_pte_kind_zf32_x24s8_v()) || - (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && - k <= gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()) || - (k == gmmu_pte_kind_generic_16bx2_v()) || - (k == gmmu_pte_kind_c32_2c_v()) || - (k == gmmu_pte_kind_c32_2cra_v()) || - (k == gmmu_pte_kind_c32_ms2_2c_v()) || - (k == gmmu_pte_kind_c32_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms4_2c_v() && - k <= gmmu_pte_kind_c32_ms4_2cbr_v()) || - (k >= gmmu_pte_kind_c32_ms4_2cra_v() && - k <= gmmu_pte_kind_c64_2c_v()) || - (k == gmmu_pte_kind_c64_2cra_v()) || - (k == gmmu_pte_kind_c64_ms2_2c_v()) || - (k == gmmu_pte_kind_c64_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms4_2c_v() && - k <= gmmu_pte_kind_c64_ms4_2cbr_v()) || - (k >= gmmu_pte_kind_c64_ms4_2cra_v() && - k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()) || - (k == gmmu_pte_kind_pitch_no_swizzle_v()); - } + return gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_invalid_v()) + || (k == gmmu_pte_kind_pitch_v()) + || (k >= gmmu_pte_kind_z16_v() && + k <= gmmu_pte_kind_z16_ms8_2c_v()) + || (k >= gmmu_pte_kind_z16_2z_v() && + k <= gmmu_pte_kind_z16_ms8_2z_v()) + || (k == gmmu_pte_kind_s8z24_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms8_2cz_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()) + || (k == gmmu_pte_kind_z24s8_v()) + || (k >= gmmu_pte_kind_z24s8_2cz_v() && + k <= gmmu_pte_kind_z24s8_ms8_2cz_v()) + || (k == gmmu_pte_kind_zf32_v()) + || (k >= gmmu_pte_kind_zf32_2cz_v() && + k <= gmmu_pte_kind_zf32_ms8_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k == gmmu_pte_kind_c32_2c_v()) + || (k == gmmu_pte_kind_c32_2cra_v()) + || (k == gmmu_pte_kind_c32_ms2_2c_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c32_ms4_2cra_v() && + k <= gmmu_pte_kind_c64_2c_v()) + || (k == gmmu_pte_kind_c64_2cra_v()) + || (k == gmmu_pte_kind_c64_ms2_2c_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c64_ms4_2cra_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()) + || (k == gmmu_pte_kind_pitch_no_swizzle_v()); +} static noinline_for_stack bool gk20a_kind_z(u8 k) { return (k >= gmmu_pte_kind_z16_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) - /* || - (k >= gmmu_pte_kind_zv32_x24s8_2cszv_v() && - k <= gmmu_pte_kind_xf32_x24s8_ms16_2cs_v())*/; + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()); } static noinline_for_stack bool gk20a_kind_c(u8 k) { - return gk20a_kind_work_creation(k) || - (k == gmmu_pte_kind_pitch_v()) || - (k == gmmu_pte_kind_generic_16bx2_v()) || - (k >= gmmu_pte_kind_c32_2c_v() && - k <= gmmu_pte_kind_c32_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c32_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms4_2c_v() && - k <= gmmu_pte_kind_c64_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c64_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms4_2c_v() && - k <= gmmu_pte_kind_pitch_no_swizzle_v()); + return gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_pitch_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k >= gmmu_pte_kind_c32_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_pitch_no_swizzle_v()); } static noinline_for_stack bool gk20a_kind_compressible(u8 k) { return (k >= gmmu_pte_kind_z16_2c_v() && - k <= gmmu_pte_kind_z16_ms16_4cz_v()) || - (k >= gmmu_pte_kind_s8z24_1z_v() && - k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) || - (k >= gmmu_pte_kind_z24s8_1z_v() && - k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) || - (k >= gmmu_pte_kind_zf32_1z_v() && - k <= gmmu_pte_kind_zf32_ms16_2cz_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) || - (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && - k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) || - (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && - k <= gmmu_pte_kind_c32_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c32_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms4_2c_v() && - k <= gmmu_pte_kind_c64_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c64_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms4_2c_v() && - k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_1z_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_1z_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_1z_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } static noinline_for_stack bool gk20a_kind_zbc(u8 k) { return (k >= gmmu_pte_kind_z16_2c_v() && - k <= gmmu_pte_kind_z16_ms16_2c_v()) || - (k >= gmmu_pte_kind_z16_4cz_v() && - k <= gmmu_pte_kind_z16_ms16_4cz_v()) || - (k >= gmmu_pte_kind_s8z24_2cz_v() && - k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_2cs_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()) || - (k >= gmmu_pte_kind_v8z24_ms4_vc12_4cszv_v() && - k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) || - (k >= gmmu_pte_kind_z24s8_2cs_v() && - k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_2cs_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()) || - (k >= gmmu_pte_kind_z24v8_ms4_vc12_4cszv_v() && - k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) || - (k >= gmmu_pte_kind_zf32_2cs_v() && - k <= gmmu_pte_kind_zf32_ms16_2cz_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1czv_v() && - k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) || - (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1czv_v() && - k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) || - (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && - k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) || - (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && - k <= gmmu_pte_kind_c32_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms2_2c_v() && - k <= gmmu_pte_kind_c32_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c32_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms4_2c_v() && - k <= gmmu_pte_kind_c32_ms4_2cra_v()) || - (k >= gmmu_pte_kind_c32_ms8_ms16_2c_v() && - k <= gmmu_pte_kind_c64_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms2_2c_v() && - k <= gmmu_pte_kind_c64_ms2_2cbr_v()) || - (k == gmmu_pte_kind_c64_ms2_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms4_2c_v() && - k <= gmmu_pte_kind_c64_ms4_2cra_v()) || - (k >= gmmu_pte_kind_c64_ms8_ms16_2c_v() && - k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); + k <= gmmu_pte_kind_z16_ms16_2c_v()) + || (k >= gmmu_pte_kind_z16_4cz_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_2cs_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_2cs_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms2_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c64_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms2_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } u8 gk20a_uc_kind_map[256]; diff --git a/drivers/gpu/nvgpu/gk20a/kind_gk20a.h b/drivers/gpu/nvgpu/gk20a/kind_gk20a.h index 28d5802c..dcc33f7d 100644 --- a/drivers/gpu/nvgpu/gk20a/kind_gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/kind_gk20a.h @@ -3,7 +3,7 @@ * * GK20A memory kind management * - * Copyright (c) 2011-2014, NVIDIA CORPORATION. All rights reserved. + * Copyright (c) 2011-2017, NVIDIA CORPORATION. All rights reserved. * * This program is free software; you can redistribute it and/or modify it * under the terms and conditions of the GNU General Public License, @@ -36,6 +36,22 @@ extern u16 gk20a_kind_attr[]; #define GK20A_KIND_ATTR_C BIT(3) #define GK20A_KIND_ATTR_ZBC BIT(4) +/* TBD: not sure on the work creation for gk20a, doubtful */ +static inline bool gk20a_kind_work_creation_sked(u8 k) +{ + return false; +} +static inline bool gk20a_kind_work_creation_host(u8 k) +{ + return false; +} + +static inline bool gk20a_kind_work_creation(u8 k) +{ + return gk20a_kind_work_creation_sked(k) || + gk20a_kind_work_creation_host(k); +} + static inline bool gk20a_kind_is_supported(u8 k) { return !!(gk20a_kind_attr[k] & GK20A_KIND_ATTR_SUPPORTED); diff --git a/drivers/gpu/nvgpu/gm20b/fb_gm20b.c b/drivers/gpu/nvgpu/gm20b/fb_gm20b.c index 8b811cbe..47b4313b 100644 --- a/drivers/gpu/nvgpu/gm20b/fb_gm20b.c +++ b/drivers/gpu/nvgpu/gm20b/fb_gm20b.c @@ -33,53 +33,402 @@ static void fb_gm20b_init_fs_state(struct gk20a *g) g->ltc_count); } -void gm20b_init_uncompressed_kind_map(void) +noinline_for_stack void gm20b_init_uncompressed_kind_map(void) { - gk20a_init_uncompressed_kind_map(); + int i; + for (i = 0; i < 256; i++) + gk20a_uc_kind_map[i] = gmmu_pte_kind_invalid_v(); + + /* From gm20b */ gk20a_uc_kind_map[gmmu_pte_kind_s8_v()] = gk20a_uc_kind_map[gmmu_pte_kind_s8_2s_v()] = gmmu_pte_kind_s8_v(); + + /* From gk20a */ + gk20a_uc_kind_map[gmmu_pte_kind_z16_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms8_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms2_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms4_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms8_2z_v()] = + gmmu_pte_kind_z16_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms8_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms8_2cs_v()] = + gmmu_pte_kind_s8z24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2zv_v()] = + gmmu_pte_kind_v8z24_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2zv_v()] = + gmmu_pte_kind_v8z24_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2zv_v()] = + gmmu_pte_kind_v8z24_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()] = + gmmu_pte_kind_v8z24_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms8_2cz_v()] = + gmmu_pte_kind_z24s8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms8_2cz_v()] = + gmmu_pte_kind_zf32_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms2_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms4_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms8_2cs_v()] = + gmmu_pte_kind_zf32_x24s8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_c32_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms2_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms8_ms16_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms2_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms8_ms16_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms2_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms4_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms8_ms16_2cr_v()] = + gmmu_pte_kind_generic_16bx2_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2zv_v()] = + gmmu_pte_kind_z24v8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2zv_v()] = + gmmu_pte_kind_z24v8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2zv_v()] = + gmmu_pte_kind_z24v8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2zv_v()] = + gmmu_pte_kind_z24v8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8c24_v()] = + gmmu_pte_kind_x8c24_v(); } -static bool gm20b_kind_supported(u8 k) +static noinline_for_stack bool gm20b_kind_supported(u8 k) { - return k == gmmu_pte_kind_smsked_message_v() + return /* From gm20b */ + (k == gmmu_pte_kind_smsked_message_v()) || (k >= gmmu_pte_kind_s8_v() && - k <= gmmu_pte_kind_s8_2s_v()); + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_invalid_v()) + || (k == gmmu_pte_kind_pitch_v()) + || (k >= gmmu_pte_kind_z16_v() && + k <= gmmu_pte_kind_z16_ms8_2c_v()) + || (k >= gmmu_pte_kind_z16_2z_v() && + k <= gmmu_pte_kind_z16_ms8_2z_v()) + || (k == gmmu_pte_kind_s8z24_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms8_2cz_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()) + || (k == gmmu_pte_kind_z24s8_v()) + || (k >= gmmu_pte_kind_z24s8_2cz_v() && + k <= gmmu_pte_kind_z24s8_ms8_2cz_v()) + || (k == gmmu_pte_kind_zf32_v()) + || (k >= gmmu_pte_kind_zf32_2cz_v() && + k <= gmmu_pte_kind_zf32_ms8_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k == gmmu_pte_kind_c32_2c_v()) + || (k == gmmu_pte_kind_c32_2cra_v()) + || (k == gmmu_pte_kind_c32_ms2_2c_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c32_ms4_2cra_v() && + k <= gmmu_pte_kind_c64_2c_v()) + || (k == gmmu_pte_kind_c64_2cra_v()) + || (k == gmmu_pte_kind_c64_ms2_2c_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c64_ms4_2cra_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()) + || (k == gmmu_pte_kind_pitch_no_swizzle_v()); +} + +static noinline_for_stack bool gm20b_kind_z(u8 k) +{ + return /* From gm20b */ + (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()); } -static bool gm20b_kind_z(u8 k) +static noinline_for_stack bool gm20b_kind_c(u8 k) { - return (k >= gmmu_pte_kind_s8_v() && - k <= gmmu_pte_kind_s8_2s_v()); + return gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_pitch_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k >= gmmu_pte_kind_c32_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_pitch_no_swizzle_v()); } -static bool gm20b_kind_compressible(u8 k) +static noinline_for_stack bool gm20b_kind_compressible(u8 k) { - return (k >= gmmu_pte_kind_s8_v() && - k <= gmmu_pte_kind_s8_2s_v()); + return /* From gm20b */ + (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_2c_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_1z_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_1z_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_1z_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } -static bool gm20b_kind_zbc(u8 k) +static noinline_for_stack bool gm20b_kind_zbc(u8 k) { - return (k >= gmmu_pte_kind_s8_v() && - k <= gmmu_pte_kind_s8_2s_v()); + return /* From gm20b */ + (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_2c_v() && + k <= gmmu_pte_kind_z16_ms16_2c_v()) + || (k >= gmmu_pte_kind_z16_4cz_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_2cs_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_2cs_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms2_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c64_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms2_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } void gm20b_init_kind_attr(void) { u16 k; - gk20a_init_kind_attr(); - - for (k = 0; k < 256; k++) { + for (k = 0; k < NV_KIND_ATTR_SIZE; k++) { + gk20a_kind_attr[k] = 0; if (gm20b_kind_supported((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_SUPPORTED; if (gm20b_kind_compressible((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_COMPRESSIBLE; if (gm20b_kind_z((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_Z; + if (gm20b_kind_c((u8)k)) + gk20a_kind_attr[k] |= GK20A_KIND_ATTR_C; if (gm20b_kind_zbc((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_ZBC; } diff --git a/drivers/gpu/nvgpu/gp10b/fb_gp10b.c b/drivers/gpu/nvgpu/gp10b/fb_gp10b.c index 06cd94f2..8ba9ff8e 100644 --- a/drivers/gpu/nvgpu/gp10b/fb_gp10b.c +++ b/drivers/gpu/nvgpu/gp10b/fb_gp10b.c @@ -20,8 +20,14 @@ #include -static void gp10b_init_uncompressed_kind_map(void) +static noinline_for_stack void gp10b_init_uncompressed_kind_map(void) { + int i; + + for (i = 0; i < 256; i++) + gk20a_uc_kind_map[i] = gmmu_pte_kind_invalid_v(); + + /* From gp10b */ gk20a_uc_kind_map[gmmu_pte_kind_z16_2cz_v()] = gk20a_uc_kind_map[gmmu_pte_kind_z16_ms2_2cz_v()] = gk20a_uc_kind_map[gmmu_pte_kind_z16_ms4_2cz_v()] = @@ -32,55 +38,419 @@ static void gp10b_init_uncompressed_kind_map(void) gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_4cbra_v()] = gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_4cbra_v()] = gmmu_pte_kind_generic_16bx2_v(); + + /* From gm20b */ + gk20a_uc_kind_map[gmmu_pte_kind_s8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8_2s_v()] = + gmmu_pte_kind_s8_v(); + + /* From gk20a */ + gk20a_uc_kind_map[gmmu_pte_kind_z16_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms8_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms2_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms4_2z_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z16_ms8_2z_v()] = + gmmu_pte_kind_z16_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms8_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_s8z24_ms8_2cs_v()] = + gmmu_pte_kind_s8z24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc4_2zv_v()] = + gmmu_pte_kind_v8z24_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc8_2zv_v()] = + gmmu_pte_kind_v8z24_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms4_vc12_2zv_v()] = + gmmu_pte_kind_v8z24_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()] = + gmmu_pte_kind_v8z24_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24s8_ms8_2cz_v()] = + gmmu_pte_kind_z24s8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms2_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms4_2cz_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_ms8_2cz_v()] = + gmmu_pte_kind_zf32_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()] = + gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()] = + gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms2_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms4_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms2_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_zf32_x24s8_ms8_2cs_v()] = + gmmu_pte_kind_zf32_x24s8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_c32_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms2_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms4_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c32_ms8_ms16_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms2_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cbr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cba_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms4_2bra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c64_ms8_ms16_2cra_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms2_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms2_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms4_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms4_2cr_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms8_ms16_2c_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_c128_ms8_ms16_2cr_v()] = + gmmu_pte_kind_generic_16bx2_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc4_2zv_v()] = + gmmu_pte_kind_z24v8_ms4_vc4_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms4_vc12_2zv_v()] = + gmmu_pte_kind_z24v8_ms4_vc12_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc8_2zv_v()] = + gmmu_pte_kind_z24v8_ms8_vc8_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()] = + gk20a_uc_kind_map[gmmu_pte_kind_z24v8_ms8_vc24_2zv_v()] = + gmmu_pte_kind_z24v8_ms8_vc24_v(); + + gk20a_uc_kind_map[gmmu_pte_kind_x8c24_v()] = + gmmu_pte_kind_x8c24_v(); } -static bool gp10b_kind_supported(u8 k) +static noinline_for_stack bool gp10b_kind_supported(u8 k) { - return (k >= gmmu_pte_kind_z16_2cz_v() && - k <= gmmu_pte_kind_z16_ms8_2cz_v()) + return /* From gp10b */ + (k >= gmmu_pte_kind_z16_2cz_v() && + k <= gmmu_pte_kind_z16_ms8_2cz_v()) || k == gmmu_pte_kind_z16_ms16_2cz_v() || k == gmmu_pte_kind_c32_ms4_4cbra_v() - || k == gmmu_pte_kind_c64_ms4_4cbra_v(); + || k == gmmu_pte_kind_c64_ms4_4cbra_v() + /* From gm20b */ + || (k == gmmu_pte_kind_smsked_message_v()) + || (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_invalid_v()) + || (k == gmmu_pte_kind_pitch_v()) + || (k >= gmmu_pte_kind_z16_v() && + k <= gmmu_pte_kind_z16_ms8_2c_v()) + || (k >= gmmu_pte_kind_z16_2z_v() && + k <= gmmu_pte_kind_z16_ms8_2z_v()) + || (k == gmmu_pte_kind_s8z24_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms8_2cz_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2zv_v()) + || (k == gmmu_pte_kind_z24s8_v()) + || (k >= gmmu_pte_kind_z24s8_2cz_v() && + k <= gmmu_pte_kind_z24s8_ms8_2cz_v()) + || (k == gmmu_pte_kind_zf32_v()) + || (k >= gmmu_pte_kind_zf32_2cz_v() && + k <= gmmu_pte_kind_zf32_ms8_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms8_2cszv_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k == gmmu_pte_kind_c32_2c_v()) + || (k == gmmu_pte_kind_c32_2cra_v()) + || (k == gmmu_pte_kind_c32_ms2_2c_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c32_ms4_2cra_v() && + k <= gmmu_pte_kind_c64_2c_v()) + || (k == gmmu_pte_kind_c64_2cra_v()) + || (k == gmmu_pte_kind_c64_ms2_2c_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cbr_v()) + || (k >= gmmu_pte_kind_c64_ms4_2cra_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()) + || (k == gmmu_pte_kind_pitch_no_swizzle_v()); +} + +static noinline_for_stack bool gp10b_kind_z(u8 k) +{ + return /* From gp10b */ + (k >= gmmu_pte_kind_z16_2cz_v() && + k <= gmmu_pte_kind_z16_ms8_2cz_v()) + || (k == gmmu_pte_kind_z16_ms16_2cz_v()) + /* From gm20b */ + || (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()); } -static bool gp10b_kind_z(u8 k) +static noinline_for_stack bool gp10b_kind_c(u8 k) { - return (k >= gmmu_pte_kind_z16_2cz_v() && - k <= gmmu_pte_kind_z16_ms8_2cz_v()) || - k == gmmu_pte_kind_z16_ms16_2cz_v(); + return gk20a_kind_work_creation(k) + || (k == gmmu_pte_kind_pitch_v()) + || (k == gmmu_pte_kind_generic_16bx2_v()) + || (k >= gmmu_pte_kind_c32_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_pitch_no_swizzle_v()); } -static bool gp10b_kind_compressible(u8 k) +static noinline_for_stack bool gp10b_kind_compressible(u8 k) { - return (k >= gmmu_pte_kind_z16_2cz_v() && - k <= gmmu_pte_kind_z16_ms8_2cz_v()) || - k == gmmu_pte_kind_z16_ms16_2cz_v() || - (k >= gmmu_pte_kind_z16_4cz_v() && - k <= gmmu_pte_kind_z16_ms16_4cz_v()) || - k == gmmu_pte_kind_c32_ms4_4cbra_v() || - k == gmmu_pte_kind_c64_ms4_4cbra_v(); + return /* From gp10b */ + (k >= gmmu_pte_kind_z16_2cz_v() && + k <= gmmu_pte_kind_z16_ms8_2cz_v()) + || (k == gmmu_pte_kind_z16_ms16_2cz_v()) + || (k >= gmmu_pte_kind_z16_4cz_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k == gmmu_pte_kind_c32_ms4_4cbra_v()) + || (k == gmmu_pte_kind_c64_ms4_4cbra_v()) + /* From gm20b */ + || (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_2c_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_1z_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_1z_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_1z_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1zv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } -static bool gp10b_kind_zbc(u8 k) +static noinline_for_stack bool gp10b_kind_zbc(u8 k) { - return (k >= gmmu_pte_kind_z16_2cz_v() && - k <= gmmu_pte_kind_z16_ms8_2cz_v()) || - k == gmmu_pte_kind_z16_ms16_2cz_v() || - k == gmmu_pte_kind_c32_ms4_4cbra_v() || - k == gmmu_pte_kind_c64_ms4_4cbra_v(); + return /* From gp10b */ + (k >= gmmu_pte_kind_z16_2cz_v() && + k <= gmmu_pte_kind_z16_ms8_2cz_v()) + || (k == gmmu_pte_kind_z16_ms16_2cz_v()) + || (k == gmmu_pte_kind_c32_ms4_4cbra_v()) + || (k == gmmu_pte_kind_c64_ms4_4cbra_v()) + /* From gm20b */ + || (k >= gmmu_pte_kind_s8_v() && + k <= gmmu_pte_kind_s8_2s_v()) + /* From gk20a */ + || (k >= gmmu_pte_kind_z16_2c_v() && + k <= gmmu_pte_kind_z16_ms16_2c_v()) + || (k >= gmmu_pte_kind_z16_4cz_v() && + k <= gmmu_pte_kind_z16_ms16_4cz_v()) + || (k >= gmmu_pte_kind_s8z24_2cz_v() && + k <= gmmu_pte_kind_s8z24_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_v8z24_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_v8z24_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_z24s8_2cs_v() && + k <= gmmu_pte_kind_z24s8_ms16_4cszv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2cs_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2cs_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_2czv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_2czv_v()) + || (k >= gmmu_pte_kind_z24v8_ms4_vc12_4cszv_v() && + k <= gmmu_pte_kind_z24v8_ms8_vc24_4cszv_v()) + || (k >= gmmu_pte_kind_zf32_2cs_v() && + k <= gmmu_pte_kind_zf32_ms16_2cz_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_x8z24_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_x8z24_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1cs_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x16v8s8_ms4_vc12_1czv_v() && + k <= gmmu_pte_kind_zf32_x16v8s8_ms8_vc24_2cszv_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_1cs_v() && + k <= gmmu_pte_kind_zf32_x24s8_ms16_1cs_v()) + || (k >= gmmu_pte_kind_zf32_x24s8_2cszv_v() && + k <= gmmu_pte_kind_c32_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms2_2c_v() && + k <= gmmu_pte_kind_c32_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c32_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms4_2c_v() && + k <= gmmu_pte_kind_c32_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c32_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c64_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms2_2c_v() && + k <= gmmu_pte_kind_c64_ms2_2cbr_v()) + || (k == gmmu_pte_kind_c64_ms2_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms4_2c_v() && + k <= gmmu_pte_kind_c64_ms4_2cra_v()) + || (k >= gmmu_pte_kind_c64_ms8_ms16_2c_v() && + k <= gmmu_pte_kind_c128_ms8_ms16_2cr_v()); } static void gp10b_init_kind_attr(void) { u16 k; - for (k = 0; k < 256; k++) { + for (k = 0; k < NV_KIND_ATTR_SIZE; k++) { if (gp10b_kind_supported((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_SUPPORTED; if (gp10b_kind_compressible((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_COMPRESSIBLE; if (gp10b_kind_z((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_Z; + if (gp10b_kind_c((u8)k)) + gk20a_kind_attr[k] |= GK20A_KIND_ATTR_C; if (gp10b_kind_zbc((u8)k)) gk20a_kind_attr[k] |= GK20A_KIND_ATTR_ZBC; } -- cgit v1.2.2