[Intel-xe] [PATCH v7 02/10] drm/xe: Log and count the GT hardware errors.
Aravind Iddamsetty
aravind.iddamsetty at linux.intel.com
Thu Oct 19 08:24:25 UTC 2023
On 18/10/23 09:30, Himal Prasad Ghimiray wrote:
> For the errors reported by GT unit, read the GT error register.
> Log and count these errors and clear the error register.
>
> Bspec: 53088, 53089, 53090
>
> v6
> - define the BIT and use it.
> - Limit the GT error reporting to DG2 and PVC only.
> - Rename function to xe_gt_hw_error_log_status_reg from
> xe_gt_hw_error_status_reg_handler. (Aravind)
>
> v7
> - ci fixes
>
> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
> Cc: Aravind Iddamsetty <aravind.iddamsetty at intel.com>
> Cc: Matthew Brost <matthew.brost at intel.com>
> Cc: Matt Roper <matthew.d.roper at intel.com>
> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> Cc: Jani Nikula <jani.nikula at intel.com>
> Signed-off-by: Himal Prasad Ghimiray <himal.prasad.ghimiray at intel.com>
> ---
> drivers/gpu/drm/xe/regs/xe_gt_error_regs.h | 13 +++
> drivers/gpu/drm/xe/regs/xe_tile_error_regs.h | 1 +
> drivers/gpu/drm/xe/xe_device.c | 4 +
> drivers/gpu/drm/xe/xe_device_types.h | 1 +
> drivers/gpu/drm/xe/xe_gt.c | 1 +
> drivers/gpu/drm/xe/xe_gt_types.h | 7 ++
> drivers/gpu/drm/xe/xe_hw_error.c | 97 +++++++++++++++++++-
> drivers/gpu/drm/xe/xe_hw_error.h | 24 +++++
> 8 files changed, 147 insertions(+), 1 deletion(-)
> create mode 100644 drivers/gpu/drm/xe/regs/xe_gt_error_regs.h
>
> diff --git a/drivers/gpu/drm/xe/regs/xe_gt_error_regs.h b/drivers/gpu/drm/xe/regs/xe_gt_error_regs.h
> new file mode 100644
> index 000000000000..6180704a6149
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/regs/xe_gt_error_regs.h
> @@ -0,0 +1,13 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright © 2023 Intel Corporation
> + */
> +#ifndef XE_GT_ERROR_REGS_H_
> +#define XE_GT_ERROR_REGS_H_
> +
> +#define _ERR_STAT_GT_COR 0x100160
> +#define _ERR_STAT_GT_NONFATAL 0x100164
> +#define ERR_STAT_GT_REG(x) XE_REG(_PICK_EVEN((x), \
> + _ERR_STAT_GT_COR, \
> + _ERR_STAT_GT_NONFATAL))
> +#endif
> diff --git a/drivers/gpu/drm/xe/regs/xe_tile_error_regs.h b/drivers/gpu/drm/xe/regs/xe_tile_error_regs.h
> index db78d6687213..2224f7d328e5 100644
> --- a/drivers/gpu/drm/xe/regs/xe_tile_error_regs.h
> +++ b/drivers/gpu/drm/xe/regs/xe_tile_error_regs.h
> @@ -12,4 +12,5 @@
> #define DEV_ERR_STAT_REG(x) XE_REG(_PICK_EVEN((x), \
> _DEV_ERR_STAT_CORRECTABLE, \
> _DEV_ERR_STAT_NONFATAL))
> +#define XE_GT_ERROR 0
as it is a register field an indent is needed here
> #endif
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index 79685348cc69..2b8b9a0713b1 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -391,10 +391,14 @@ static void xe_device_remove_display(struct xe_device *xe)
> static void xe_hw_error_fini(struct xe_device *xe)
> {
> struct xe_tile *tile;
> + struct xe_gt *gt;
> int i;
>
> for_each_tile(tile, xe, i)
> xa_destroy(&tile->errors.hw_error);
if we restrict the xarray initialization to only primary gt
then we can do xa_destroy(&tile->primary_gt->errors.hw_error)
and avoid the below.
> +
> + for_each_gt(gt, xe, i)
> + xa_destroy(>->errors.hw_error);
> }
>
> void xe_device_remove(struct xe_device *xe)
> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> index c4464dcef56f..dbc04a1f6dc1 100644
> --- a/drivers/gpu/drm/xe/xe_device_types.h
> +++ b/drivers/gpu/drm/xe/xe_device_types.h
> @@ -414,6 +414,7 @@ struct xe_device {
> /** @hw_err_regs: list of hw error regs*/
> struct hardware_errors_regs {
> const struct err_name_index_pair *dev_err_stat[HARDWARE_ERROR_MAX];
> + const struct err_name_index_pair *err_stat_gt[HARDWARE_ERROR_MAX];
> } hw_err_regs;
>
> /* private: */
> diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> index 74e1f47bd401..112fc159fd4f 100644
> --- a/drivers/gpu/drm/xe/xe_gt.c
> +++ b/drivers/gpu/drm/xe/xe_gt.c
> @@ -282,6 +282,7 @@ int xe_gt_init_early(struct xe_gt *gt)
> {
> int err;
>
> + xa_init(>->errors.hw_error);
do this only for primary_gt
> xe_force_wake_init_gt(gt, gt_to_fw(gt));
>
> err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
> diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
> index d4310be3e1e7..ac26e8e8de59 100644
> --- a/drivers/gpu/drm/xe/xe_gt_types.h
> +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> @@ -9,10 +9,12 @@
> #include "xe_force_wake_types.h"
> #include "xe_gt_idle_sysfs_types.h"
> #include "xe_hw_engine_types.h"
> +#include "xe_hw_error.h"
> #include "xe_hw_fence_types.h"
> #include "xe_reg_sr_types.h"
> #include "xe_sa_types.h"
> #include "xe_uc_types.h"
> +#include "regs/xe_gt_error_regs.h"
order it.
>
> struct xe_exec_queue_ops;
> struct xe_migrate;
> @@ -347,6 +349,11 @@ struct xe_gt {
> /** @oob: bitmap with active OOB workaroudns */
> unsigned long *oob;
> } wa_active;
> +
> + /** @errors: hardware errors reported for the gt */
append with "count of"
> + struct gt_hw_errors {
> + struct xarray hw_error;
> + } errors;
> };
>
> #endif
> diff --git a/drivers/gpu/drm/xe/xe_hw_error.c b/drivers/gpu/drm/xe/xe_hw_error.c
> index ac25072db6c0..941f71609abd 100644
> --- a/drivers/gpu/drm/xe/xe_hw_error.c
> +++ b/drivers/gpu/drm/xe/xe_hw_error.c
> @@ -100,9 +100,40 @@ static const struct err_name_index_pair pvc_err_stat_correctable_reg[] = {
> [9 ... 31] = {"Undefined", XE_HW_ERR_TILE_CORR_UNKNOWN},
> };
>
> +static const struct err_name_index_pair dg2_stat_gt_fatal_reg[] = {
> + [0] = {"Undefined", XE_HW_ERR_GT_FATAL_UNKNOWN},
> + [1] = {"Array BIST", XE_HW_ERR_GT_FATAL_ARR_BIST},
> + [2] = {"Undefined", XE_HW_ERR_GT_FATAL_UNKNOWN},
> + [3] = {"FPU", XE_HW_ERR_GT_FATAL_FPU},
> + [4] = {"L3 Double", XE_HW_ERR_GT_FATAL_L3_DOUB},
> + [5] = {"L3 ECC Checker", XE_HW_ERR_GT_FATAL_L3_ECC_CHK},
> + [6] = {"GUC SRAM", XE_HW_ERR_GT_FATAL_GUC},
> + [7] = {"Undefined", XE_HW_ERR_GT_FATAL_UNKNOWN},
> + [8] = {"IDI PARITY", XE_HW_ERR_GT_FATAL_IDI_PAR},
> + [9] = {"SQIDI", XE_HW_ERR_GT_FATAL_SQIDI},
> + [10 ... 11] = {"Undefined", XE_HW_ERR_GT_FATAL_UNKNOWN},
> + [12] = {"SAMPLER", XE_HW_ERR_GT_FATAL_SAMPLER},
> + [13] = {"SLM", XE_HW_ERR_GT_FATAL_SLM},
> + [14] = {"EU IC", XE_HW_ERR_GT_FATAL_EU_IC},
> + [15] = {"EU GRF", XE_HW_ERR_GT_FATAL_EU_GRF},
> + [16 ... 31] = {"Undefined", XE_HW_ERR_GT_FATAL_UNKNOWN},
> +};
> +
> +static const struct err_name_index_pair dg2_stat_gt_correctable_reg[] = {
> + [0] = {"L3 SINGLE", XE_HW_ERR_GT_CORR_L3_SNG},
> + [1] = {"SINGLE BIT GUC SRAM", XE_HW_ERR_GT_CORR_GUC},
> + [2 ... 11] = {"Undefined", XE_HW_ERR_GT_CORR_UNKNOWN},
> + [12] = {"SINGLE BIT SAMPLER", XE_HW_ERR_GT_CORR_SAMPLER},
> + [13] = {"SINGLE BIT SLM", XE_HW_ERR_GT_CORR_SLM},
> + [14] = {"SINGLE BIT EU IC", XE_HW_ERR_GT_CORR_EU_IC},
> + [15] = {"SINGLE BIT EU GRF", XE_HW_ERR_GT_CORR_EU_GRF},
> + [16 ... 31] = {"Undefined", XE_HW_ERR_GT_CORR_UNKNOWN},
> +};
> +
> void xe_assign_hw_err_regs(struct xe_device *xe)
> {
> const struct err_name_index_pair **dev_err_stat = xe->hw_err_regs.dev_err_stat;
> + const struct err_name_index_pair **err_stat_gt = xe->hw_err_regs.err_stat_gt;
>
> /* Error reporting is supported only for DG2 and
> * PVC currently. Error reporting support for other
> @@ -112,6 +143,8 @@ void xe_assign_hw_err_regs(struct xe_device *xe)
> dev_err_stat[HARDWARE_ERROR_CORRECTABLE] = dg2_err_stat_correctable_reg;
> dev_err_stat[HARDWARE_ERROR_NONFATAL] = dg2_err_stat_nonfatal_reg;
> dev_err_stat[HARDWARE_ERROR_FATAL] = dg2_err_stat_fatal_reg;
> + err_stat_gt[HARDWARE_ERROR_CORRECTABLE] = dg2_stat_gt_correctable_reg;
> + err_stat_gt[HARDWARE_ERROR_FATAL] = dg2_stat_gt_fatal_reg;
> }
>
> if (xe->info.platform == XE_PVC) {
> @@ -119,6 +152,7 @@ void xe_assign_hw_err_regs(struct xe_device *xe)
> dev_err_stat[HARDWARE_ERROR_NONFATAL] = pvc_err_stat_nonfatal_reg;
> dev_err_stat[HARDWARE_ERROR_FATAL] = pvc_err_stat_fatal_reg;
> }
> +
> }
>
> static bool xe_ras_enabled(struct xe_device *xe)
> @@ -145,6 +179,66 @@ xe_update_hw_error_cnt(struct drm_device *drm, struct xarray *hw_error, unsigned
> xa_unlock_irqrestore(hw_error, flags);
> }
>
> +static void
> +xe_gt_hw_error_log_status_reg(struct xe_gt *gt, const enum hardware_error hw_err)
> +{
> + const char *hw_err_str = hardware_error_type_to_str(hw_err);
> + const struct err_name_index_pair *errstat;
> + struct hardware_errors_regs *err_regs;
> + unsigned long errsrc;
> + const char *name;
> + u32 indx;
> + u32 errbit;
> +
> + err_regs = >_to_xe(gt)->hw_err_regs;
> + errsrc = xe_mmio_read32(gt, ERR_STAT_GT_REG(hw_err));
> + if (!errsrc) {
> + drm_err_ratelimited(>_to_xe(gt)->drm, HW_ERR
> + "GT%d reported ERR_STAT_GT_REG_%s blank!\n",
> + gt->info.id, hw_err_str);
> + return;
> + }
> +
> + drm_dbg(>_to_xe(gt)->drm, HW_ERR "GT%d ERR_STAT_GT_REG_%s=0x%08lx\n",
> + gt->info.id, hw_err_str, errsrc);
> +
> + if (hw_err == HARDWARE_ERROR_NONFATAL) {
> + /* The GT Non Fatal Error Status Register has only reserved bits
> + * Nothing to service.
> + */
> + drm_err_ratelimited(>_to_xe(gt)->drm, HW_ERR "GT%d reported %s error\n",
> + gt->info.id, hw_err_str);
> + goto clear_reg;
> + }
> +
> + errstat = err_regs->err_stat_gt[hw_err];
> + for_each_set_bit(errbit, &errsrc, XE_RAS_REG_SIZE) {
> + name = errstat[errbit].name;
> + indx = errstat[errbit].index;
> +
> + if (hw_err == HARDWARE_ERROR_FATAL)
> + drm_err_ratelimited(>_to_xe(gt)->drm, HW_ERR
> + "GT%d reported %s %s error, bit[%d] is set\n",
> + gt->info.id, name, hw_err_str, errbit);
for gt errors use helpers in xe_gt_printk.h or if we want HARDWARE_ERROR to be at the beginning
define a new helper.
> + else
> + drm_warn(>_to_xe(gt)->drm, HW_ERR
> + "GT%d reported %s %s error, bit[%d] is set\n",
> + gt->info.id, name, hw_err_str, errbit);
> +
> + xe_update_hw_error_cnt(>_to_xe(gt)->drm, >->errors.hw_error, indx);
> + }
> +clear_reg: xe_mmio_write32(gt, ERR_STAT_GT_REG(hw_err), errsrc);
new line missing after label.
> +}
> +
> +static void
> +xe_gt_hw_error_handler(struct xe_gt *gt, const enum hardware_error hw_err)
> +{
> + lockdep_assert_held(>_to_xe(gt)->irq.lock);
> +
> + if (gt_to_xe(gt)->info.platform == XE_DG2)
> + xe_gt_hw_error_log_status_reg(gt, hw_err);
> +}
> +
> static void
> xe_hw_error_source_handler(struct xe_tile *tile, const enum hardware_error hw_err)
> {
> @@ -199,8 +293,9 @@ xe_hw_error_source_handler(struct xe_tile *tile, const enum hardware_error hw_er
> if (indx != XE_HW_ERR_TILE_UNSPEC)
> xe_update_hw_error_cnt(&tile_to_xe(tile)->drm,
> &tile->errors.hw_error, indx);
> + if (errbit == XE_GT_ERROR)
> + xe_gt_hw_error_handler(tile->primary_gt, hw_err);
> }
> -
> xe_mmio_write32(gt, DEV_ERR_STAT_REG(hw_err), errsrc);
> unlock:
> spin_unlock_irqrestore(&tile_to_xe(tile)->irq.lock, flags);
> diff --git a/drivers/gpu/drm/xe/xe_hw_error.h b/drivers/gpu/drm/xe/xe_hw_error.h
> index e74dd6fc6faf..df69ddd8d015 100644
> --- a/drivers/gpu/drm/xe/xe_hw_error.h
> +++ b/drivers/gpu/drm/xe/xe_hw_error.h
> @@ -38,6 +38,30 @@ enum xe_tile_hw_errors {
> XE_HW_ERROR_TILE_MAX,
> };
>
> +/* Count of GT Correctable and FATAL HW ERRORS */
> +enum xe_gt_hw_errors {
> + XE_HW_ERR_GT_CORR_L3_SNG,
> + XE_HW_ERR_GT_CORR_GUC,
> + XE_HW_ERR_GT_CORR_SAMPLER,
> + XE_HW_ERR_GT_CORR_SLM,
> + XE_HW_ERR_GT_CORR_EU_IC,
> + XE_HW_ERR_GT_CORR_EU_GRF,
> + XE_HW_ERR_GT_CORR_UNKNOWN,
> + XE_HW_ERR_GT_FATAL_ARR_BIST,
> + XE_HW_ERR_GT_FATAL_FPU,
> + XE_HW_ERR_GT_FATAL_L3_DOUB,
> + XE_HW_ERR_GT_FATAL_L3_ECC_CHK,
> + XE_HW_ERR_GT_FATAL_GUC,
> + XE_HW_ERR_GT_FATAL_IDI_PAR,
> + XE_HW_ERR_GT_FATAL_SQIDI,
> + XE_HW_ERR_GT_FATAL_SAMPLER,
> + XE_HW_ERR_GT_FATAL_SLM,
> + XE_HW_ERR_GT_FATAL_EU_IC,
> + XE_HW_ERR_GT_FATAL_EU_GRF,
> + XE_HW_ERR_GT_FATAL_UNKNOWN,
> + XE_HW_ERR_GT_MAX,
> +};
> +
> struct err_name_index_pair {
> const char *name;
> const u32 index;
Thanks,
Aravind.
More information about the Intel-xe
mailing list