Re: [PATCH v4 2/2] PCI: iproc: Add outbound configuration for 32-bit I/O region

From: Lorenzo Pieralisi
Date: Tue Apr 02 2019 - 06:26:50 EST

On Tue, Apr 02, 2019 at 03:20:21PM +0530, Srinath Mannam wrote:
> Hi Ray,
>
> Thanks for detailed explanation.
> Please see some more details below.
>
> On Tue, Apr 2, 2019 at 3:33 AM Ray Jui <ray.jui@xxxxxxxxxxxx> wrote:
> >
> > Hi Lorenzo/Srinath,
> >
> > I look at the commit message again and indeed it looks quite confusing.
> >
> > I'll add my comment inline in the code section. I hope that will help to
> > make it more clear.
> >
> > On 4/1/2019 9:44 AM, Lorenzo Pieralisi wrote:
> > > On Mon, Apr 01, 2019 at 11:04:48AM +0530, Srinath Mannam wrote:
> > >> Hi Lorenzo,
> > >>
> > >> Please see my reply below,
> > >>
> > >> On Fri, Mar 29, 2019 at 11:06 PM Lorenzo Pieralisi
> > >> <lorenzo.pieralisi@xxxxxxx> wrote:
> > >>>
> > >>> On Fri, Mar 01, 2019 at 10:22:16AM +0530, Srinath Mannam wrote:
> > >>>> In the present driver outbound window configuration is done to map above
> > >>>> 32-bit address I/O regions with corresponding PCI memory range given in
> > >>>> ranges DT property.
> > >>>>
> > >>>> This patch add outbound window configuration to map below 32-bit I/O range
> > >>>> with corresponding PCI memory, which helps to access I/O region in ARM
> > >>>> 32-bit and one to one mapping of I/O region to PCI memory.
> > >>>>
> > >>>> Ex:
> > >>>> 1. ranges DT property given for current driver is,
> > >>>> ranges = <0x83000000 0x0 0x40000000 0x4 0x00000000 0 0x40000000>;
> > >>>> I/O region address is 0x400000000
> > >>>> 2. ranges DT property can be given after this patch,
> > >>>> ranges = <0x83000000 0x0 0x42000000 0x0 0x42000000 0 0x2000000>;
> > >>>> I/O region address is 0x42000000
> > >>>
> > >>> I was applying this patch but I don't understand the commit log and
> > >>> how it matches the code, please explain it to me and I will reword
> > >>> it.
> > >> Iproc PCIe host controller supports outbound address translation
> > >> feature to translate AXI address to PCI bus address.
> > >> IO address ranges (AXI and PCI) given through ranges DT property have
> > >> to program to controller outbound window registers.
> > >> Present driver has the support for only 64bit AXI address. so that
> > >> ranges DT property has given as 64bit AXI address and 32 bit
> > >> PCI bus address.
> > >> But with this patch 32-bit AXI address also could be programmed to
> > >> Iproc host controller outbound window registers. so that ranges
> > >> DT property can have 32bit AXI address which can map to 32-bit PCI bus address.
> > >
> > > The code change seems to add a check for the window size, I see no
> > > notion of 64 vs 32 bit addressing there so I am pretty sure there is
> > > something you are not telling me that is implicit in the IProc outbound
> > > window configuration, for instance why is the lowest index window
> > > considered for 32-bit.
> > >
> > > AFAICS you are adding code to allow a window whose size is < than
> > > the lowest index in the ob_map array. How this relates to 64 vs
> > > 32 bit addresses is not clear but it should be.
> > >
> > > When I read your commit log it is impossible to understand how it
> > > correlates to the code you are changing - I still have not figured it
> > > out myself.
> > >
> > > Please explain in detail to me how this works, forget DT changes I
> > > want to understand how HW works.
> > >
> > > Lorenzo
> > >
> > >> Example given in commit log is describing ranges DT property changes
> > >> with and without this patch.
> > >> In the case, without this patch AXI address is more than 32bit
> > >> "0x400000000". and with this patch AXI address is 32-bit "0x42000000".
> > >> PCI bus address is 32 bit address in both the cases "0x40000000" and 0x42000000.
> > >>
> > >> Regards,
> > >> Srinath.
> > >>> Thanks,
> > >>> Lorenzo
> > >>>
> > >>>> Signed-off-by: Srinath Mannam <srinath.mannam@xxxxxxxxxxxx>
> > >>>> Signed-off-by: Abhishek Shah <abhishek.shah@xxxxxxxxxxxx>
> > >>>> Signed-off-by: Ray Jui <ray.jui@xxxxxxxxxxxx>
> > >>>> ---
> > >>>> drivers/pci/controller/pcie-iproc.c | 21 +++++++++++++++++++--
> > >>>> 1 file changed, 19 insertions(+), 2 deletions(-)
> > >>>>
> > >>>> diff --git a/drivers/pci/controller/pcie-iproc.c b/drivers/pci/controller/pcie-iproc.c
> > >>>> index b882255..080f142 100644
> > >>>> --- a/drivers/pci/controller/pcie-iproc.c
> > >>>> +++ b/drivers/pci/controller/pcie-iproc.c
> > >>>> @@ -955,8 +955,25 @@ static int iproc_pcie_setup_ob(struct iproc_pcie *pcie, u64 axi_addr,
> > >>>> resource_size_t window_size =
> > >>>> ob_map->window_sizes[size_idx] * SZ_1M;
> > >>>>
> > >>>> - if (size < window_size)
> > >>>> - continue;
> > >>>> + /*
> > >>>> + * Keep iterating until we reach the last window and
> > >>>> + * with the minimal window size at index zero. In this
> > >>>> + * case, we take a compromise by mapping it using the
> > >>>> + * minimum window size that can be supported
> > >>>> + */
> >
> > I think the code comment above is much more clear than the commit
> > message. It looks like the commit message was to describe a particular
> > use case that prompts the code change. However, if I remember correctly,
> > during our internal review, I already made some modification to the code
> > to make the change much more generic than dealing with a special use
> > case. This patch contains the generic change I made.
> >
> > Basically, 'size' is the intended outbound mapping size (from DT or ACPI
> > or whatever, it does not really matter). 'window_size' is a specific
> > mapping window size our PCIe controller can support. Our PCIe controller
> > supports a fixed set of outbound mapping window sizes. I don't think
> > this is much different from some other PCIe controllers.
> >
> > It looks like, there are cases where one cannot find an exact match
> > between 'size' and 'window_size'. In this case, and when we know we have
> > already exhausted all possible mapping windows, i.e., 'size_idx' == 0
> > AND 'window_idx' == 0, we take a compromise by programming the outbound
> > mapping by using a window size that's actually larger than the 'size'.
> >
> > Srinath, please correct me if I'm wrong. But I carefully reviewed the
> > code again and I believe this is essentially what it is.
> >
> 128MB is the minimum size of window available to configure outbound
> memory. But size of 32bit AXI address region
> of our controller is 32MB. Also this 32bit AXI address region has to
> configure in OARR0. so that 32bit AXI address region
> is configured in 128MB window of OARR0.
> this is the reason in the below to allow ob window configure only when
> size_idx is 0 which means window size 128MB
> and window_idx is 0 means OARR0 else continue using (size_idx > 0 ||
> window_idx > 0) condition.

Ok - I start to understand. What does it mean in HW terms that your
32bit AXI address region size is 32MB ? Please explain to me in details.

IIUC you are using an OARR0 of 128MB in size to map a 32MB address
region, that's what I understand this patch does (and the lowest index
corresponds to the smallest possible size - it is far from clear by
looking at the patch).

Please clarify.

Thanks,
Lorenzo

> Regards,
> Srinath.
> > If so, then the commit message is quite misleading and can be changed to
> > above descriptions.
> >
> > >>>> + if (size < window_size) {
> > >>>> + if (size_idx > 0 || window_idx > 0)
> > >>>> + continue;
> > >>>> +
> > >>>> + /*
> > >>>> + * For the corner case of reaching the minimal
> > >>>> + * window size that can be supported on the
> > >>>> + * last window
> > >>>> + */
> > >>>> + axi_addr = ALIGN_DOWN(axi_addr, window_size);
> > >>>> + pci_addr = ALIGN_DOWN(pci_addr, window_size);
> > >>>> + size = window_size;
> > >>>> + }
> > >>>>
> > >>>> if (!IS_ALIGNED(axi_addr, window_size) ||
> > >>>> !IS_ALIGNED(pci_addr, window_size)) {
> > >>>> --
> > >>>> 2.7.4
> > >>>>