From: Matthew Brost <matthew.brost@intel.com>
To: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
Cc: <intel-xe@lists.freedesktop.org>,
<dri-devel@lists.freedesktop.org>,
<himal.prasad.ghimiray@intel.com>, <apopple@nvidia.com>,
<airlied@gmail.com>, <simona.vetter@ffwll.ch>,
<felix.kuehling@amd.com>, <dakr@kernel.org>
Subject: Re: [PATCH v4 29/33] drm/xe: Basic SVM BO eviction
Date: Tue, 11 Feb 2025 11:21:39 -0800 [thread overview]
Message-ID: <Z6ujQ0G6Wj4QQezs@lstrano-desk.jf.intel.com> (raw)
In-Reply-To: <ec4bcb3b178a3de6e605216e588e2eba9cc3217a.camel@linux.intel.com>
On Fri, Feb 07, 2025 at 03:45:51PM +0100, Thomas Hellström wrote:
> On Wed, 2025-01-29 at 11:52 -0800, Matthew Brost wrote:
> > Wire xe_bo_move to GPU SVM migration via new helper xe_svm_bo_evict.
> >
> > v2:
> > - Use xe_svm_bo_evict
> > - Drop bo->range
> > v3:
> > - Kernel doc (Thomas)
> > v4:
> > - Add missing xe_bo.c code
> >
> > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
>
> I think in the long run, we'd want to do the svm eviction / unbind in
> move_notify(), since that's where we're supposed to unbind other
> subsystems. And then just purge the bo using a NULL placement, but
> since this is equivalent let's postpone that to a more general
> xe_bo_move() cleanup. It's getting pretty hard to follow.
>
Agree xe_bo_move() could use some cleanup.
Matt
> Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>
>
> > ---
> > drivers/gpu/drm/xe/xe_bo.c | 19 +++++++++++++++++++
> > drivers/gpu/drm/xe/xe_svm.c | 15 ++++++++++++++-
> > drivers/gpu/drm/xe/xe_svm.h | 3 +++
> > 3 files changed, 36 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> > index 20c96709e267..657687ee70d0 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -255,6 +255,8 @@ int xe_bo_placement_for_flags(struct xe_device
> > *xe, struct xe_bo *bo,
> > static void xe_evict_flags(struct ttm_buffer_object *tbo,
> > struct ttm_placement *placement)
> > {
> > + struct xe_bo *bo;
> > +
> > if (!xe_bo_is_xe_bo(tbo)) {
> > /* Don't handle scatter gather BOs */
> > if (tbo->type == ttm_bo_type_sg) {
> > @@ -266,6 +268,12 @@ static void xe_evict_flags(struct
> > ttm_buffer_object *tbo,
> > return;
> > }
> >
> > + bo = ttm_to_xe_bo(tbo);
> > + if (bo->flags & XE_BO_FLAG_CPU_ADDR_MIRROR) {
> > + *placement = sys_placement;
> > + return;
> > + }
> > +
> > /*
> > * For xe, sg bos that are evicted to system just triggers a
> > * rebind of the sg list upon subsequent validation to
> > XE_PL_TT.
> > @@ -710,6 +718,17 @@ static int xe_bo_move(struct ttm_buffer_object
> > *ttm_bo, bool evict,
> > goto out;
> > }
> >
> > + if (!move_lacks_source && (bo->flags &
> > XE_BO_FLAG_CPU_ADDR_MIRROR) &&
> > + new_mem->mem_type == XE_PL_SYSTEM) {
> > + ret = xe_svm_bo_evict(bo);
> > + if (!ret) {
> > + drm_dbg(&xe->drm, "Evict system allocator BO
> > success\n");
> > + ttm_bo_move_null(ttm_bo, new_mem);
> > + }
> > +
> > + goto out;
> > + }
> > +
> > if (old_mem_type == XE_PL_SYSTEM && new_mem->mem_type ==
> > XE_PL_TT && !handle_system_ccs) {
> > ttm_bo_move_null(ttm_bo, new_mem);
> > goto out;
> > diff --git a/drivers/gpu/drm/xe/xe_svm.c
> > b/drivers/gpu/drm/xe/xe_svm.c
> > index fc030855d078..dafc5061eb42 100644
> > --- a/drivers/gpu/drm/xe/xe_svm.c
> > +++ b/drivers/gpu/drm/xe/xe_svm.c
> > @@ -768,6 +768,20 @@ bool xe_svm_has_mapping(struct xe_vm *vm, u64
> > start, u64 end)
> > return drm_gpusvm_has_mapping(&vm->svm.gpusvm, start, end);
> > }
> >
> > +/**
> > + * xe_svm_bo_evict() - SVM evict BO to system memory
> > + * @bo: BO to evict
> > + *
> > + * SVM evict BO to system memory. GPU SVM layer ensures all device
> > pages
> > + * are evicted before returning.
> > + *
> > + * Return: 0 on success standard error code otherwise
> > + */
> > +int xe_svm_bo_evict(struct xe_bo *bo)
> > +{
> > + return drm_gpusvm_evict_to_ram(&bo->devmem_allocation);
> > +}
> > +
> > #if IS_ENABLED(CONFIG_DRM_XE_DEVMEM_MIRROR)
> > static struct drm_pagemap_dma_addr
> > xe_drm_pagemap_map_dma(struct drm_pagemap *dpagemap,
> > @@ -795,7 +809,6 @@ static const struct drm_pagemap_ops
> > xe_drm_pagemap_ops = {
> > .map_dma = xe_drm_pagemap_map_dma,
> > };
> >
> > ->>>>>>> 133db8ade5f0 (drm/xe: Add drm_pagemap ops to SVM)
> > /**
> > * xe_devm_add: Remap and provide memmap backing for device memory
> > * @tile: tile that the memory region belongs to
> > diff --git a/drivers/gpu/drm/xe/xe_svm.h
> > b/drivers/gpu/drm/xe/xe_svm.h
> > index 4c2576162c39..77dec5aae0ee 100644
> > --- a/drivers/gpu/drm/xe/xe_svm.h
> > +++ b/drivers/gpu/drm/xe/xe_svm.h
> > @@ -11,6 +11,7 @@
> >
> > #define XE_INTERCONNECT_VRAM DRM_INTERCONNECT_DRIVER
> >
> > +struct xe_bo;
> > struct xe_mem_region;
> > struct xe_tile;
> > struct xe_vm;
> > @@ -56,6 +57,8 @@ int xe_svm_handle_pagefault(struct xe_vm *vm,
> > struct xe_vma *vma,
> >
> > bool xe_svm_has_mapping(struct xe_vm *vm, u64 start, u64 end);
> >
> > +int xe_svm_bo_evict(struct xe_bo *bo);
> > +
> > static inline bool xe_svm_range_pages_valid(struct xe_svm_range
> > *range)
> > {
> > return drm_gpusvm_range_pages_valid(range->base.gpusvm,
> > &range->base);
>
next prev parent reply other threads:[~2025-02-11 19:20 UTC|newest]
Thread overview: 103+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-01-29 19:51 [PATCH v4 00/33] Introduce GPU SVM and Xe SVM implementation Matthew Brost
2025-01-29 19:51 ` [PATCH v4 01/33] drm/xe: Retry BO allocation Matthew Brost
2025-01-29 19:51 ` [PATCH v4 02/33] mm/migrate: Add migrate_device_pfns Matthew Brost
2025-01-31 5:24 ` Alistair Popple
2025-01-31 7:47 ` Gwan-gyeong Mun
2025-02-04 22:17 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 03/33] mm/migrate: Trylock device page in do_swap_page Matthew Brost
2025-01-29 19:51 ` [PATCH v4 04/33] drm/pagemap: Add DRM pagemap Matthew Brost
2025-02-07 8:34 ` Thomas Hellström
2025-02-10 18:41 ` Matthew Brost
2025-02-11 16:03 ` Thomas Hellström
2025-02-11 18:17 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 05/33] drm/xe/bo: Introduce xe_bo_put_async Matthew Brost
2025-01-30 8:49 ` Thomas Hellström
2025-01-30 16:26 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 06/33] drm/gpusvm: Add support for GPU Shared Virtual Memory Matthew Brost
2025-01-30 9:13 ` Thomas Hellström
2025-01-30 11:17 ` Matthew Auld
2025-01-30 13:13 ` Gwan-gyeong Mun
2025-01-30 16:42 ` Matthew Brost
2025-02-07 9:06 ` Thomas Hellström
2025-02-10 17:31 ` Matthew Brost
2025-02-11 15:17 ` Thomas Hellström
2025-02-11 18:05 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 07/33] drm/xe: Select DRM_GPUSVM Kconfig Matthew Brost
2025-02-07 3:18 ` Ghimiray, Himal Prasad
2025-02-07 9:30 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 08/33] drm/xe/uapi: Add DRM_XE_VM_BIND_FLAG_CPU_ADDR_MIRROR flag Matthew Brost
2025-02-07 9:37 ` Thomas Hellström
2025-02-07 12:11 ` Ghimiray, Himal Prasad
2025-02-07 13:47 ` Upadhyay, Tejas
2025-02-10 19:08 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 09/33] drm/xe: Add SVM init / close / fini to faulting VMs Matthew Brost
2025-02-07 3:24 ` Ghimiray, Himal Prasad
2025-02-07 9:43 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 10/33] drm/xe: Add dma_addr res cursor Matthew Brost
2025-02-10 19:11 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 11/33] drm/xe: Nuke VM's mapping upon close Matthew Brost
2025-01-30 10:50 ` Matthew Auld
2025-01-30 16:28 ` Matthew Brost
2025-02-07 10:15 ` Thomas Hellström
2025-02-10 19:16 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 12/33] drm/xe: Add SVM range invalidation and page fault handler Matthew Brost
2025-02-07 10:32 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 13/33] drm/gpuvm: Add DRM_GPUVA_OP_DRIVER Matthew Brost
2025-02-07 10:36 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 14/33] drm/xe: Add (re)bind to SVM page fault handler Matthew Brost
2025-01-29 19:51 ` [PATCH v4 15/33] drm/xe: Add SVM garbage collector Matthew Brost
2025-02-07 12:42 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 16/33] drm/xe: Add unbind to " Matthew Brost
2025-02-07 12:55 ` Thomas Hellström
2025-02-10 21:17 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 17/33] drm/xe: Do not allow CPU address mirror VMA unbind if the GPU has bindings Matthew Brost
2025-02-07 13:01 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 18/33] drm/xe: Enable CPU address mirror uAPI Matthew Brost
2025-02-07 13:02 ` Thomas Hellström
2025-01-29 19:51 ` [PATCH v4 19/33] drm/xe/uapi: Add DRM_XE_QUERY_CONFIG_FLAG_HAS_CPU_ADDR_MIRROR Matthew Brost
2025-02-07 11:35 ` Ghimiray, Himal Prasad
2025-02-07 11:35 ` Ghimiray, Himal Prasad
2025-02-07 13:04 ` Thomas Hellström
2025-02-07 13:43 ` Upadhyay, Tejas
2025-02-10 19:15 ` Matthew Brost
2025-01-29 19:51 ` [PATCH v4 20/33] drm/xe: Add migrate layer functions for SVM support Matthew Brost
2025-02-07 13:07 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 21/33] drm/xe: Add SVM device memory mirroring Matthew Brost
2025-02-07 13:29 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 22/33] drm/xe: Add drm_gpusvm_devmem to xe_bo Matthew Brost
2025-01-29 19:52 ` [PATCH v4 23/33] drm/xe: Add drm_pagemap ops to SVM Matthew Brost
2025-01-30 10:54 ` Matthew Auld
2025-01-30 13:24 ` Gwan-gyeong Mun
2025-01-30 16:24 ` Matthew Brost
2025-01-29 19:52 ` [PATCH v4 24/33] drm/xe: Add GPUSVM device memory copy vfunc functions Matthew Brost
2025-02-07 13:32 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 25/33] drm/xe: Add Xe SVM populate_devmem_pfn GPU SVM vfunc Matthew Brost
2025-01-29 19:52 ` [PATCH v4 26/33] drm/xe: Add Xe SVM devmem_release " Matthew Brost
2025-01-29 19:52 ` [PATCH v4 27/33] drm/xe: Add BO flags required for SVM Matthew Brost
2025-02-07 13:54 ` Thomas Hellström
2025-02-11 19:19 ` Matthew Brost
2025-02-11 19:36 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 28/33] drm/xe: Add SVM VRAM migration Matthew Brost
2025-01-30 14:22 ` Matthew Auld
2025-01-30 16:32 ` Matthew Brost
2025-01-30 16:41 ` Thomas Hellström
2025-01-30 16:56 ` Matthew Auld
2025-01-30 17:31 ` Matthew Brost
2025-01-30 18:51 ` Thomas Hellström
2025-01-31 17:30 ` Matthew Brost
2025-02-07 13:57 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 29/33] drm/xe: Basic SVM BO eviction Matthew Brost
2025-02-07 14:45 ` Thomas Hellström
2025-02-11 19:21 ` Matthew Brost [this message]
2025-01-29 19:52 ` [PATCH v4 30/33] drm/xe: Add SVM debug Matthew Brost
2025-02-07 14:46 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 31/33] drm/xe: Add modparam for SVM notifier size Matthew Brost
2025-02-07 14:48 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 32/33] drm/xe: Add always_migrate_to_vram modparam Matthew Brost
2025-02-07 14:50 ` Thomas Hellström
2025-01-29 19:52 ` [PATCH v4 33/33] drm/doc: gpusvm: Add GPU SVM documentation Matthew Brost
2025-02-07 14:54 ` Thomas Hellström
2025-01-29 21:04 ` ✓ CI.Patch_applied: success for Introduce GPU SVM and Xe SVM implementation (rev4) Patchwork
2025-01-29 21:05 ` ✗ CI.checkpatch: warning " Patchwork
2025-01-29 21:06 ` ✗ CI.KUnit: failure " Patchwork
2025-01-30 13:52 ` [PATCH v4 00/33] Introduce GPU SVM and Xe SVM implementation Gwan-gyeong Mun
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=Z6ujQ0G6Wj4QQezs@lstrano-desk.jf.intel.com \
--to=matthew.brost@intel.com \
--cc=airlied@gmail.com \
--cc=apopple@nvidia.com \
--cc=dakr@kernel.org \
--cc=dri-devel@lists.freedesktop.org \
--cc=felix.kuehling@amd.com \
--cc=himal.prasad.ghimiray@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=simona.vetter@ffwll.ch \
--cc=thomas.hellstrom@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox