Re: [PATCH drm-next v2 00/16] [RFC] DRM GPUVA Manager & Nouveau VM_BIND UAPI

From: Danilo Krummrich
Date: Fri Mar 10 2023 - 11:50:12 EST


Hi Boris,

On 3/9/23 10:48, Boris Brezillon wrote:
On Thu, 9 Mar 2023 10:12:43 +0100
Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx> wrote:

Hi Danilo,

On Fri, 17 Feb 2023 14:44:06 +0100
Danilo Krummrich <dakr@xxxxxxxxxx> wrote:

Changes in V2:
==============
Nouveau:
- Reworked the Nouveau VM_BIND UAPI to avoid memory allocations in fence
signalling critical sections. Updates to the VA space are split up in three
separate stages, where only the 2. stage executes in a fence signalling
critical section:

1. update the VA space, allocate new structures and page tables

Sorry for the silly question, but I didn't find where the page tables
pre-allocation happens. Mind pointing it to me? It's also unclear when
this step happens. Is this at bind-job submission time, when the job is
not necessarily ready to run, potentially waiting for other deps to be
signaled. Or is it done when all deps are met, as an extra step before
jumping to step 2. If that's the former, then I don't see how the VA
space update can happen, since the bind-job might depend on other
bind-jobs modifying the same portion of the VA space (unbind ops might
lead to intermediate page table levels disappearing while we were
waiting for deps). If it's the latter, I wonder why this is not
considered as an allocation in the fence signaling path (for the
bind-job out-fence to be signaled, you need these allocations to
succeed, unless failing to allocate page-tables is considered like a HW
misbehavior and the fence is signaled with an error in that case).

Ok, so I just noticed you only have one bind queue per drm_file
(cli->sched_entity), and jobs are executed in-order on a given queue,
so I guess that allows you to modify the VA space at submit time
without risking any modifications to the VA space coming from other
bind-queues targeting the same VM. And, if I'm correct, synchronous
bind/unbind ops take the same path, so no risk for those to modify the
VA space either (just wonder if it's a good thing to have to sync
bind/unbind operations waiting on async ones, but that's a different
topic).

Yes, that's all correct.

The page table allocation happens through nouveau_uvmm_vmm_get() which either allocates the corresponding page tables or increases the reference count, in case they already exist, accordingly.
The call goes all the way through nvif into the nvkm layer (not the easiest to follow the call chain) and ends up in nvkm_vmm_ptes_get().

There are multiple reasons for updating the VA space at submit time in Nouveau.

1) Subsequent EXEC ioctl() calls would need to wait for the bind jobs they depend on within the ioctl() rather than in the scheduler queue, because at the point of time where the ioctl() happens the VA space wouldn't be up-to-date.

2) Let's assume a new mapping is requested and within it's range other mappings already exist. Let's also assume that those existing mappings aren't contiguous, such that there are gaps between them. In such a case I need to allocate page tables only for the gaps between the existing mappings, or alternatively, allocate them for the whole range of the new mapping, but free / decrease the reference count of the page tables for the ranges of the previously existing mappings afterwards.
In the first case I need to know the gaps to allocate page tables for when submitting the job, which means the VA space must be up-to-date. In the latter one I must save the ranges of the previously existing mappings somewhere in order to clean them up, hence I need to allocate memory to store this information. Since I can't allocate this memory in the jobs run() callback (fence signalling critical section) I need to do it when submitting the job already and hence the VA space must be up-to-date again.
However, this is due to how page table management currently works in Nouveau and we might change that in the future.

Synchronous binds/unbinds taking the same path through the scheduler is a downside of this approach.

- Danilo



Note that I'm not familiar at all with Nouveau or TTM, and it might
be something that's solved by another component, or I'm just
misunderstanding how the whole thing is supposed to work. This being
said, I'd really like to implement a VM_BIND-like uAPI in pancsf using
the gpuva_manager infra you're proposing here, so please bare with me
:-).

2. (un-)map the requested memory bindings
3. free structures and page tables

- Separated generic job scheduler code from specific job implementations.
- Separated the EXEC and VM_BIND implementation of the UAPI.
- Reworked the locking parts of the nvkm/vmm RAW interface, such that
(un-)map operations can be executed in fence signalling critical sections.

Regards,

Boris