Re: [PATCH V6 1/5] LIB: Indirect ISA/LPC port IO introduced

[John Garry]

On 30/01/2017 17:12, Alexander Graf wrote:
> On 01/24/2017 08:05 AM, zhichang.yuan wrote:
> > Low-pin-count interface is integrated into some SoCs. The accesses to
> > those
> > peripherals under LPC make use of I/O ports rather than the memory
> > mapped I/O.
> >
> > To drive these devices, this patch introduces a method named indirect-IO.
> > In this method the in/out() accessor in include/asm-generic/io.h will be
> > redefined. When upper layer drivers call in/out() with those known
> > legacy port
> > addresses to access the peripherals, the I/O operations will be routed
> > to the
> > right hooks which are registered specific to the host device, such as
> > LPC.
> > Then the hardware relevant manupulations are finished by the
> > corresponding
> > host.
> >
> > According to the comments on V5, this patch adds a common indirect-IO
> > driver
> > which support this I/O indirection to the generic directory.
> >
> > In the later pathches, some host-relevant drivers are implemented to
> > support
> > the specific I/O hooks and register them.
> > Based on these, the upper layer drivers which depend on in/out() can
> > work well
> > without any extra work or any changes.
> >
> > Signed-off-by: zhichang.yuan <yuanzhichang@xxxxxxxxxxxxx>
> > Signed-off-by: Gabriele Paoloni <gabriele.paoloni@xxxxxxxxxx>
> > Signed-off-by: John Garry <john.garry@xxxxxxxxxx>
>
> I like the extio idea. That allows us to handle all PIO requests on
> platforms that don't have native PIO support via different routes
> depending on the region they're in. Unfortunately we now we have 2
> frameworks for handling sparse PIO regions: One in extio, one in PCI.
>
> Why don't we just merge the two? Most of the code that has #ifdef
> PCI_IOBASE throughout the code base sounds like an ideal candidate to
> get migrated to extio instead. Then we only have a single framework to
> worry about ...

To be clear, are you suggesting we merge the functionality from 
pci_register_io_range(), pci_pio_to_address(), pci_address_to_pio() into 
extio, so extio manages all PIO? And having a single type of node to 
register PIO ranges, by amalgamating struct extio_node and io_range (as 
Bjorn mentioned)?

It would make sense. We would be somewhat decoupling PIO from PCI.

I think that other architectures, like PPC, and other code would need to 
be fixed up to handle this.

We need to consider all the other challenges/obstacles to this.

> > ---
> >   include/asm-generic/io.h |  50 ++++++++++++++++
> >   include/linux/extio.h    |  85 +++++++++++++++++++++++++++
> >   include/linux/io.h       |   1 +
> >   lib/Kconfig              |   8 +++
> >   lib/Makefile             |   2 +
> >   lib/extio.c              | 147
> > +++++++++++++++++++++++++++++++++++++++++++++++ xc>>   create mode 100644 include/linux/extio.h
> >   create mode 100644 lib/extio.c

<snip>

> > + * Copyright (C) 2016 Hisilicon Limited, All Rights Reserved.
> > + * Author: Zhichang Yuan <yuanzhichang@xxxxxxxxxxxxx>
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + *
> > + * This program is distributed in the hope that it will be useful,
> > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> > + * GNU General Public License for more details.
> > + *
> > + * You should have received a copy of the GNU General Public License
> > + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
> > + */
> > +
> > +#include <linux/io.h>
> > +#include <linux/spinlock.h>
> > +
> > +static LIST_HEAD(extio_dev_list);
> > +static DEFINE_RWLOCK(extio_list_lock);
>
> Why not just make the list an RCU list? Then you don't need read locks.
> We also wouldn't create potential lock contention between devices that
> could easily have parallel PIO operations (say a PCI device and an LPC
> device).

OK

> > +
> > +void register_extio(struct extio_node *node)
> > +{
> > +    write_lock(&extio_list_lock);
> > +    list_add_tail(&node->list, &extio_dev_list);
> > +    write_unlock(&extio_list_lock);
> > +}
> > +
> > +static struct extio_node *find_extio_token(unsigned long addr)
> > +{
> > +    struct extio_node *extio_entry;
> > +
> > +    read_lock(&extio_list_lock);
> > +    list_for_each_entry(extio_entry, &extio_dev_list, list) {
> > +        if ((addr < extio_entry->io_start + extio_entry->range_size) &&
> > +            (addr >= extio_entry->io_start))
> > +            break;
> > +    }
> > +    read_unlock(&extio_list_lock);
> > +    return (&extio_entry->list == &extio_dev_list) ? NULL : extio_entry;
> > +}
> > +
> > +struct extio_node *extio_find_node(struct fwnode_handle *node)
> > +{
> > +    struct extio_node *entry;
> > +
> > +    read_lock(&extio_list_lock);
> > +    list_for_each_entry(entry, &extio_dev_list, list) {
> > +        if (entry->fwnode == node)
> > +            break;
> > +    }
> > +    read_unlock(&extio_list_lock);
> > +
> > +    return (&entry->list == &extio_dev_list) ? NULL : entry;
> > +}
> > +
> > +unsigned long extio_translate(struct fwnode_handle *node,
> > +        unsigned long bus_addr)
> > +{
> > +    struct extio_node *entry;
> > +    unsigned long port_id = -1;
> > +
> > +    read_lock(&extio_list_lock);
> > +    list_for_each_entry(entry, &extio_dev_list, list) {
> > +        if (entry->fwnode == node &&
> > +            bus_addr >= entry->bus_start &&
> > +            bus_addr - entry->bus_start < entry->range_size)
> > +            port_id = entry->io_start + bus_addr -
> > +                    entry->bus_start;
> > +    }
> > +    read_unlock(&extio_list_lock);
> > +
> > +    return port_id;
> > +}
> > +
> > +#ifdef PCI_IOBASE
> > +
> > +#define BUILD_EXTIO(bw, type)                        \
> > +type extio_in##bw(unsigned long addr)                    \
> > +{                                    \
> > +    struct extio_node *extio_entry = find_extio_token(addr);    \
> > +                                    \
> > +    if (!extio_entry)                        \
> > +        return read##bw(PCI_IOBASE + addr);            \
> > +    return extio_entry->ops->pfin ?                    \
> > +            extio_entry->ops->pfin(extio_entry->devpara,    \
> > +            addr, sizeof(type)) : -1;            \
> > +}                                    \
> > +                                    \
> > +void extio_out##bw(type value, unsigned long addr)            \
> > +{                                    \
> > +    struct extio_node *extio_entry = find_extio_token(addr);    \
> > +                                    \
> > +    if (!extio_entry)                        \
> > +        write##bw(value, PCI_IOBASE + addr);            \
>
> All of the fallback code would also disappear as a nice side effect of
> making pci pio handling a user of extio :).

Is your idea that PCI IO space will also register accessors, which would 
be the same read{b,w,l}/write{b,w,l}?

>

It would be nice to have a quicker way to so the lookup from address to 
node, as we loop all nodes in find_extio_token() every single time.

> Alex
Thanks,
John

> .