Re: [PATCH v7 2/3] PCI: rockchip: Simplify reset control handling by using reset_control_bulk*() function

[Manivannan Sadhasivam]

On Sat, Oct 12, 2024 at 03:34:25PM +0530, Anand Moon wrote:
> Hi Manivannan,
> 
> On Sat, 12 Oct 2024 at 13:30, Manivannan Sadhasivam
> <manivannan.sadhasivam@xxxxxxxxxx> wrote:
> >
> > On Sat, Oct 12, 2024 at 12:55:32PM +0530, Anand Moon wrote:
> > > Hi Manivannan,
> > >
> > > Thanks for your review comments.
> > >
> > > On Sat, 12 Oct 2024 at 11:48, Manivannan Sadhasivam
> > > <manivannan.sadhasivam@xxxxxxxxxx> wrote:
> > > >
> > > > On Sat, Oct 12, 2024 at 10:36:04AM +0530, Anand Moon wrote:
> > > > > Refactor the reset control handling in the Rockchip PCIe driver,
> > > > > introducing a more robust and efficient method for assert and
> > > > > deassert reset controller using reset_control_bulk*() API. Using the
> > > > > reset_control_bulk APIs, the reset handling for the core clocks reset
> > > > > unit becomes much simpler.
> > > > >
> > > >
> > > > Same comments as previous patch.
> > > >
> > > I will explain more about this.
> > > > > Spilt the reset controller in two groups as pre the RK3399 TRM.
> > > >
> > > > *per
> > > >
> > > > Also please state the TRM name and section for reference.
> > > >
> > > Yes
> > > > > After power up, the software driver should de-assert the reset of PCIe PHY,
> > > > > then wait the PLL locked by polling the status, if PLL
> > > > > has locked, then can de-assert the reset simultaneously
> > > > > driver need to De-assert the reset pins simultionaly.
> > > > >
> > > > >   PIPE_RESET_N/MGMT_STICKY_RESET_N/MGMT_RESET_N/RESET_N.
> > > > >
> > > > > - replace devm_reset_control_get_exclusive() with
> > > > >       devm_reset_control_bulk_get_exclusive().
> > > > > - replace reset_control_assert with
> > > > >       reset_control_bulk_assert().
> > > > > - replace reset_control_deassert with
> > > > >       reset_control_bulk_deassert().
> > > > >
> > > > > Signed-off-by: Anand Moon <linux.amoon@xxxxxxxxx>
> > > > > ---
> > > > > v7: replace devm_reset_control_bulk_get_optional_exclusive()
> > > > >         with devm_reset_control_bulk_get_exclusive()
> > > > >     update the functional changes.
> > > > > V6: Add reason for the split of the RESET pins.
> > > > > v5: Fix the De-assert reset core as per the TRM
> > > > >     De-assert the PIPE_RESET_N/MGMT_STICKY_RESET_N/MGMT_RESET_N/RESET_N
> > > > >     simultaneously.
> > > > > v4: use dev_err_probe in error path.
> > > > > v3: Fix typo in commit message, dropped reported by.
> > > > > v2: Fix compilation error reported by Intel test robot
> > > > >     fixed checkpatch warning.
> > > > > ---
> > > > >  drivers/pci/controller/pcie-rockchip.c | 151 +++++--------------------
> > > > >  drivers/pci/controller/pcie-rockchip.h |  26 +++--
> > > > >  2 files changed, 49 insertions(+), 128 deletions(-)
> > > > >
> > > > > diff --git a/drivers/pci/controller/pcie-rockchip.c b/drivers/pci/controller/pcie-rockchip.c
> > > > > index 2777ef0cb599..9a118e2b8cbd 100644
> > > > > --- a/drivers/pci/controller/pcie-rockchip.c
> > > > > +++ b/drivers/pci/controller/pcie-rockchip.c
> >
> > [...]
> >
> > > > > @@ -256,31 +181,15 @@ int rockchip_pcie_init_port(struct rockchip_pcie *rockchip)
> > > > >        * Please don't reorder the deassert sequence of the following
> > > > >        * four reset pins.
> > > >
> > > > I don't think my earlier comment on this addressed. Why are you changing the
> > > > reset order? Why can't you have the resets in below (older) order?
> > > >
> > > > static const char * const rockchip_pci_core_rsts[] = {
> > > >         mgmt-sticky",
> > > >         "core",
> > > >         "mgmt",
> > > >         "pipe",
> > > > };
> > >  I will add a comment on this above.
> 
> I get your point, I missed your point.
> [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/pcie-rockchip.c?h=v6.12-rc2#n275
> 
> Actually I had these changes, but it got missed out in rebase.
> 
> >
> > Sorry, I don't get your response. My suggestion was to keep the resets sorted as
> > the original order (also indicated by my above snippet).
> 
> I will go through all the suggestions and modify them accordingly.
> 
> As per the RK3399 TRM
> [2] https://rockchip.fr/Rockchip%20RK3399%20TRM%20V1.3%20Part2.pdf
> 
> 17.5.2.2 Reset Application
> 17.5.2.2.2 System Reset (describe all the core reset feature)
> (name as per the dts mapping)
> RESET_N: - core
> MGMT_RESET_N:- mgmt
> MGMT_STICKY_RESET_N:- mgmt-sticky
> PIPE_RESET_N: - pipe
> AXI_RESET_N - aclk
> APB_RESET_N: pclk
> PM_RESET_N: - pm
> PCIE_PHY_RESET_N: - phy reset (used in the phy driver).
> 
> This is the reason for the split of the clk and core reset.
> 
> Further down in
> 17.5.8 PCIe Operation
> 17.5.8.1 PCIe Initialization Sequence
> 17.5.8.1.1 PCIe as Root Complex
> 
> 6. De-assert the PIPE_RESET_N/MGMT_STICKY_RESET_N/MGMT_RESET_N/RESET_N
> simultaneously
> 
> Should I follow this order ? or the above order.
> static const char * const rockchip_pci_core_rsts[] = {
>         "pipe",
>         "mgmt-sticky",
>          "mgmt",
>          "core",
> };

Ok, thanks for clarifying. I was worried about the comment in the driver that
warns against changing the order. But TRM rececommends a different order :/

But since no one ever reported any issue, let's go with the existing order. If
you want to follow TRM, then I'd like to get an Ack from Rockchip Engineers who
knows the hardware better.

- Mani

-- 
மணிவண்ணன் சதாசிவம்