Re: [PATCH v3 05/15] habanalabs: add command buffer module

From: Oded Gabbay
Date: Fri Feb 08 2019 - 14:53:55 EST


On Fri, Feb 8, 2019 at 2:06 PM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Feb 04, 2019 at 10:32:44PM +0200, Oded Gabbay wrote:
> > +int hl_cb_ioctl(struct hl_fpriv *hpriv, void *data)
> > +{
> > + union hl_cb_args *args = data;
> > + struct hl_device *hdev = hpriv->hdev;
> > + u64 handle;
> > + int rc;
> > +
> > + switch (args->in.op) {
> > + case HL_CB_OP_CREATE:
> > + rc = hl_cb_create(hdev, &hpriv->cb_mgr, args->in.cb_size,
> > + &handle, hpriv->ctx->asid);
>
> so cb_size comes from userspace, ok, you check for the value to be too
> small, but not too big. That means someone can try to allocate too much
> memory, possibly crashing things, not good :(
Yes, correct, but even if I limit a single allocation to, let's say,
1MB, what's stopping a userspace process from allocating multiple CBs
and draining the system memory ? I'm counting on the oom module to
kill that process if it mis-behaves.
And, btw, I assumed there is hard limit of 4MB on a single
dma_alloc_coherent. i.e. I was never able to allocate more then 4MB
through that API. So I never thought I need to check for max size
because of that hard limit.
Am I missing something here ?

>
>
> > + memset(args, 0, sizeof(*args));
> > + args->out.cb_handle = handle;
> > + break;
> > + case HL_CB_OP_DESTROY:
> > + rc = hl_cb_destroy(hdev, &hpriv->cb_mgr,
> > + args->in.cb_handle);
> > + memset(args, 0, sizeof(*args));
>
> Why zero this if it's not copied back to userspace?
Fixed
>
>
>
> > + break;
> > + default:
> > + rc = -EINVAL;
>
> -ENOTTY is normally the "invalid ioctl value", right?
Fixed
>
> > + break;
> > + }
> > +
> > + return rc;
> > +}
>
> thanks,
>
> greg k-h