nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-24 10:34:43 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: tu104: implement l2 sector promotion

Browse Source 
Introduce new HAL gops_ltc.set_l2_sector_promotion to configure L2
sector promotion policy. The follow three promotion settings are support:
- NVGPU_GPU_IOCTL_TSG_L2_SECTOR_PROMOTE_FLAG_NONE
- NVGPU_GPU_IOCTL_TSG_L2_SECTOR_PROMOTE_FLAG_64B
- NVGPU_GPU_IOCTL_TSG_L2_SECTOR_PROMOTE_FLAG_128B

Add ioctl "NVGPU_TSG_IOCTL_SET_L2_SECTOR_PROMOTION" to the gpu tsg node
to support l2 sector promotion. On chips which do not support sector
promotion, the ioctl returns 0.

Bug 200656177

Change-Id: Iad835a5c954d3b10da436cfafb388aaaa04f44c7
Signed-off-by: Antony Clince Alex <aalex@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2460553
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-cert <svc-mobile-cert@nvidia.com>
Reviewed-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
GVS: Gerrit_Virtual_Submit
This commit is contained in:
Antony Clince Alex
2020-12-16 12:09:03 +05:30
committed by mobile promotions
parent 5517e14e57
commit 95bfa039f5
8 changed files with 246 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
drivers/gpu/nvgpu/hal/init/hal_tu104.c
View File

@@ -330,6 +330,7 @@ static const struct gops_ltc tu104_ops_ltc = {
.split_lts_broadcast_addr = gm20b_ltc_split_lts_broadcast_addr,
.split_ltc_broadcast_addr = gm20b_ltc_split_ltc_broadcast_addr,
.pri_is_lts_tstg_addr = tu104_ltc_pri_is_lts_tstg_addr,
.set_l2_sector_promotion = tu104_set_l2_sector_promotion,
#endif /* CONFIG_NVGPU_DEBUGGER */
};

105
drivers/gpu/nvgpu/hal/ltc/ltc_tu104.c
View File

@@ -26,8 +26,10 @@
#include <nvgpu/timers.h>
#include <nvgpu/gk20a.h>
#include <nvgpu/trace.h>
#include "ltc_tu104.h"
#include <nvgpu/regops.h>
#include "hal/gr/gr/gr_gk20a.h"
#include "ltc_tu104.h"
#include "ltc_gv11b.h"
#include <nvgpu/hw/tu104/hw_ltc_tu104.h>
@@ -69,4 +71,105 @@ u32 tu104_ltc_pri_is_lts_tstg_addr(struct gk20a *g, u32 addr)
return (lts_addr >= LTS_TSTG_BASE && lts_addr <= LTS_TSTG_EXTENT) ?
true : false;
}
int tu104_set_l2_sector_promotion(struct gk20a *g, struct nvgpu_tsg *tsg,
u32 policy)
{
int err = 0;
struct nvgpu_dbg_reg_op cfg_ops[2] = {
{
.op = REGOP(READ_32),
.type = REGOP(TYPE_GR_CTX),
.offset = ltc_ltcs_ltss_tstg_cfg2_r()
},
{
.op = REGOP(READ_32),
.type = REGOP(TYPE_GR_CTX),
.offset = ltc_ltcs_ltss_tstg_cfg3_r()
},
};
u32 flags = NVGPU_REG_OP_FLAG_MODE_ALL_OR_NONE;
u32 num_ops = 2U;
u32 cfg2_vidmem = 0U, cfg3_sysmem = 0U;
/*
* Read current value for ltc_ltcs_ltss_tstg_cfg(2,3)_r
*/
err = gr_gk20a_exec_ctx_ops(tsg, cfg_ops, num_ops, 0, num_ops, &flags);
if (err != 0) {
nvgpu_err(g, "failed to read ltcs_ltss_tstg_cfg(2,3)_r");
goto fail;
}
cfg2_vidmem = cfg_ops[0].value_lo;
cfg3_sysmem = cfg_ops[1].value_lo;
#define APPLY_SECTOR_PROMOTION_POLICY(cfg, unit, policy) \
do { \
switch (policy) { \
case NVGPU_L2_SECTOR_PROMOTE_FLAG_NONE: \
cfg = set_field(cfg, \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_m(), \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_f( \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_none_v() \
)); \
break; \
case NVGPU_L2_SECTOR_PROMOTE_FLAG_64B: \
cfg = set_field(cfg, \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_m(), \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_f( \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_64b_v() \
)); \
break; \
case NVGPU_L2_SECTOR_PROMOTE_FLAG_128B: \
cfg = set_field(cfg, \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_m(), \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_f( \
ltc_ltcs_ltss_tstg_##cfg##_##unit##_promote_128b_v() \
)); \
break; \
} \
} while (0)
/*
* Update T1_PROMOTE and L1_PROMOTE fields of cfg2_vidmem and
* cfg3_sysmem.
*/
APPLY_SECTOR_PROMOTION_POLICY(cfg2_vidmem, t1, policy);
APPLY_SECTOR_PROMOTION_POLICY(cfg2_vidmem, l1, policy);
APPLY_SECTOR_PROMOTION_POLICY(cfg3_sysmem, t1, policy);
APPLY_SECTOR_PROMOTION_POLICY(cfg3_sysmem, l1, policy);
#undef APPLY_SECTOR_PROMOTION_POLICY
cfg_ops[0].op = REGOP(WRITE_32);
cfg_ops[0].value_lo = cfg2_vidmem;
cfg_ops[1].op = REGOP(WRITE_32);
cfg_ops[1].value_lo = cfg3_sysmem;
err = gr_gk20a_exec_ctx_ops(tsg, cfg_ops, num_ops, num_ops, 0, &flags);
if (err != 0) {
nvgpu_err(g, "failed to update ltcs_ltss_tstg_cfg(2,3)_r");
goto fail;
}
/* Readback and verify the write */
cfg_ops[0].op = REGOP(READ_32);
cfg_ops[0].value_lo = 0U;
cfg_ops[1].op = REGOP(READ_32);
cfg_ops[1].value_lo = 0U;
err = gr_gk20a_exec_ctx_ops(tsg, cfg_ops, num_ops, 0, num_ops, &flags);
if (err != 0) {
nvgpu_err(g, "failed to read ltcs_ltss_tstg_cfg(2,3)_r");
goto fail;
}
if (cfg2_vidmem != cfg_ops[0].value_lo || cfg3_sysmem != cfg_ops[1].value_lo) {
nvgpu_err(g, "mismatch: cfg2: wrote(0x%x) read(0x%x)",
cfg_ops[0].value_lo, cfg2_vidmem);
nvgpu_err(g, " cfg3: wrote(0x%x) read(0x%x)",
cfg_ops[1].value_lo, cfg3_sysmem);
err = -EINVAL;
}
fail:
return err;
}
#endif

2
drivers/gpu/nvgpu/hal/ltc/ltc_tu104.h
View File

@@ -41,6 +41,8 @@ struct gk20a;
void ltc_tu104_init_fs_state(struct gk20a *g);
#ifdef CONFIG_NVGPU_DEBUGGER
u32 tu104_ltc_pri_is_lts_tstg_addr(struct gk20a *g, u32 addr);
int tu104_set_l2_sector_promotion(struct gk20a *g, struct nvgpu_tsg *tsg,
u32 policy);
#endif
#endif /* LTC_TU104_H */