On Tue, May 31, 2022 at 10:17 AM Huacai Chen <chenhuacai@xxxxxxxxxx> wrote:
> On Tue, May 31, 2022 at 4:09 PM Arnd Bergmann <arnd@xxxxxxxxxx> wrote:
> >
> > On Tue, May 31, 2022 at 9:50 AM Huacai Chen <chenhuacai@xxxxxxxxxx> wrote:
> > > On Mon, May 30, 2022 at 11:56 PM Arnd Bergmann <arnd@xxxxxxxxxx> wrote:
> > > > On Mon, May 30, 2022 at 5:00 PM WANG Xuerui <kernel@xxxxxxxxxx> wrote:
> > > > > Now I see
> > > > > the loongarch-next HEAD is already rebased on top of what I believe to
> > > > > be the current main branch, however I vaguely remember that it's not
> > > > > good to base one's patches on top of "some random commit", so I wonder
> > > > > whether the current branch state is appropriate for a PR?
> > > >
> > > > You are correct, a pull request should always be based on an -rc, orat least
> > > > have the minimum set of dependencies. The branch was previously
> > > > based on top of the spinlock implementation, which is still the best
> > > > place to start here.
> > > I have a difficult problem to select the base. Take swiotlb_init() as
> > > an example: If I select 5.18-rc1, I should use swiotlb_init(1); if I
> > > select Linus' latest tree, I should use swiotlb_init(true,
> > > SWIOTLB_VERBOSE). However, if I select 5.18-rc1, linux-next will have
> > > a build error because the code there expect swiotlb_init(true,
> > > SWIOTLB_VERBOSE).
> >
> > Ok, I see. This is the kind of thing we normally prevent by having everything
> > in linux-next for a few weeks before the merge window. How many issues
> > like this are you aware of? If it's just the swiotlb, you could try merging
> > the swiotlb branch that is in mainline now on top of the spinlock branch,
> > and still get a minimum set of dependencies. If there are many more,
> > then basing on top of the current mainline is probably less intrusive after
> > all.
> I have 3 issues:
> 1, swiotlb_init(1) --> swiotlb_init(true, SWIOTLB_VERBOSE);
> 2, the prototype of handle_kernel_image() should be changed from 5
> parameters to 6 parameters;
> 3, the return value type of huge_ptep_get_and_clear() should be
> changed from void to pte_t (and the function implementation should be
> also changed).
Ok, I see. Let's stay with the base on top of a mainline snapshot then.
Arnd
