Minki/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2024-11-14 16:12:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

drm/i915/uc: use different ggtt pin offsets for uc loads

Browse Source 
Our current FW loading process is the same for all FWs:

- Pin FW to GGTT at the start of the ggtt->uc_fw node
- Load the FW
- Unpin

This worked because we didn't have a case where 2 FWs would be loaded on
the same GGTT at the same time. On MTL, however, this can happen if both
GTs are reset at the same time, so we can't pin everything in the same
spot and we need to use separate offset. For simplicity, instead of
calculating the exact required size, we reserve a 2MB slot for each fw.

v2: fail fetch if FW is > 2MBs, improve comments (John)
v3: more comment improvements (John)

Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <john.c.harrison@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221108020600.3575467-3-daniele.ceraolospurio@intel.com
This commit is contained in:
Daniele Ceraolo Spurio 2022-11-07 18:05:56 -08:00
parent 9deca79836
commit 016241168d
2 changed files with 43 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
drivers/gpu/drm/i915/gt/uc/intel_uc_fw.c
View File

@ -575,6 +575,17 @@ int intel_uc_fw_fetch(struct intel_uc_fw *uc_fw)
err = firmware_request_nowarn(&fw, uc_fw->file_selected.path, dev);
memcpy(&file_ideal, &uc_fw->file_wanted, sizeof(file_ideal));
if (!err && fw->size > INTEL_UC_RSVD_GGTT_PER_FW) {
drm_err(&i915->drm,
"%s firmware %s: size (%zuKB) exceeds max supported size (%uKB)\n",
intel_uc_fw_type_repr(uc_fw->type), uc_fw->file_selected.path,
fw->size / SZ_1K, INTEL_UC_RSVD_GGTT_PER_FW / SZ_1K);
/* try to find another blob to load */
release_firmware(fw);
err = -ENOENT;
}
/* Any error is terminal if overriding. Don't bother searching for older versions */
if (err && intel_uc_fw_is_overridden(uc_fw))
goto fail;
@ -677,14 +688,30 @@ fail:
static u32 uc_fw_ggtt_offset(struct intel_uc_fw *uc_fw)
{
struct i915_ggtt *ggtt = __uc_fw_to_gt(uc_fw)->ggtt;
struct intel_gt *gt = __uc_fw_to_gt(uc_fw);
struct i915_ggtt *ggtt = gt->ggtt;
struct drm_mm_node *node = &ggtt->uc_fw;
u32 offset = uc_fw->type * INTEL_UC_RSVD_GGTT_PER_FW;
/*
* The media GT shares the GGTT with the root GT, which means that
* we need to use different offsets for the binaries on the media GT.
* To keep the math simple, we use 8MB for the root tile and 8MB for
* the media one. This will need to be updated if we ever have more
* than 1 media GT.
*/
BUILD_BUG_ON(INTEL_UC_FW_NUM_TYPES * INTEL_UC_RSVD_GGTT_PER_FW > SZ_8M);
GEM_BUG_ON(gt->type == GT_MEDIA && gt->info.id > 1);
if (gt->type == GT_MEDIA)
offset += SZ_8M;
GEM_BUG_ON(!drm_mm_node_allocated(node));
GEM_BUG_ON(upper_32_bits(node->start));
GEM_BUG_ON(upper_32_bits(node->start + node->size - 1));
GEM_BUG_ON(offset + uc_fw->obj->base.size > node->size);
GEM_BUG_ON(uc_fw->obj->base.size > INTEL_UC_RSVD_GGTT_PER_FW);
return lower_32_bits(node->start);
return lower_32_bits(node->start + offset);
}
static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw)
@ -699,7 +726,6 @@ static void uc_fw_bind_ggtt(struct intel_uc_fw *uc_fw)
dummy->bi.pages = obj->mm.pages;
GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
GEM_BUG_ON(dummy->node_size > ggtt->uc_fw.size);
/* uc_fw->obj cache domains were not controlled across suspend */
if (i915_gem_object_has_struct_page(obj))

14
drivers/gpu/drm/i915/gt/uc/intel_uc_fw.h
View File

@ -6,6 +6,7 @@
#ifndef _INTEL_UC_FW_H_
#define _INTEL_UC_FW_H_
#include <linux/sizes.h>
#include <linux/types.h>
#include "intel_uc_fw_abi.h"
#include "intel_device_info.h"
@ -114,6 +115,19 @@ struct intel_uc_fw {
(uc)->fw.file_selected.minor_ver, \
(uc)->fw.file_selected.patch_ver))
/*
* When we load the uC binaries, we pin them in a reserved section at the top of
* the GGTT, which is ~18 MBs. On multi-GT systems where the GTs share the GGTT,
* we also need to make sure that each binary is pinned to a unique location
* during load, because the different GT can go through the FW load at the same
* time (see uc_fw_ggtt_offset() for details).
* Given that the available space is much greater than what is required by the
* binaries, to keep things simple instead of dynamically partitioning the
* reserved section to make space for all the blobs we can just reserve a static
* chunk for each binary.
*/
#define INTEL_UC_RSVD_GGTT_PER_FW SZ_2M
#ifdef CONFIG_DRM_I915_DEBUG_GUC
void intel_uc_fw_change_status(struct intel_uc_fw *uc_fw,
enum intel_uc_fw_status status);