[RESEND PATCH v3 5/7] PCI: dwc: all: Modify dbi accessors to access data of 4/2/1 bytes

Niklas Cassel niklas.cassel at axis.com
Fri Mar 10 04:56:17 PST 2017

On 03/10/2017 01:04 PM, Kishon Vijay Abraham I wrote:
> Hi,
> On Thursday 09 March 2017 08:18 PM, Niklas Cassel wrote:
>> On 03/09/2017 07:39 AM, Kishon Vijay Abraham I wrote:
>>> Previously dbi accessors can be used to access data of size 4
>>> bytes. But there might be situations (like accessing
>>> MSI_MESSAGE_CONTROL in order to set/get the number of required
>>> MSI interrupts in EP mode) where dbi accessors must
>>> be used to access data of size 2. This is in preparation for
>>> adding endpoint mode support to designware driver.
>> Hello Kishon
>> I don't really like the idea of adding an extra argument to every existing read/write.
>> Will not a read/write of length != 32 be quite uncommon compared to
>> a read/write of length == 32?
>> How about adding some defines to pcie-designware.h:
>> #define dw_pcie_writel_dbi(pci, base, reg, val) dw_pcie_write_dbi(pci, base, reg, 0x4, val)
>> #define dw_pcie_readl_dbi(pci, base, reg) dw_pcie_read_dbi(pci, base, reg, 0x4)
>> That way we don't have to change every existing read/write.
>> Is there a reason why we can't just do:
>> vial = dw_pcie_readl_dbi(pci, base, MSI_MESSAGE_CONTROL);
> MSI_MESSAGE_CONTROL is 0x52 (MSI capability offset + 2). I'm not sure if we can
> do a readl that crosses the alignment boundary in all platforms. The other
> option is to readl from "MSI capability offset + 0" and extract the last 16
> bits. I felt this is more clear since we are interested only in the
>> <shifting+masking the bits we need to get/set>
>> dw_pcie_writel_dbi(pci, base, MSI_MESSAGE_CONTROL, val);
>> Or are we going to be doing read/writes of length != 32 so often that
>> you think that it's cleaner to have this abstraction?
> it's used mainly for accessing configuration space header fields. Even the pci
> core uses *pci_read_config_word* for accessing such fields.

I see. Adding an extra size argument is a good thing then,
since it's consistent with the pci generic code.

However, I still think that having defines for writel/readl is a
good thing :)

> Thanks
> Kishon

More information about the linux-arm-kernel mailing list