Re: [RFC PATCH 05/20] dev_dax_iomap: Add dax_operations for use by fs-dax on devdax

From: John Groves
Date: Mon Feb 26 2024 - 11:09:20 EST


On 24/02/26 12:32PM, Jonathan Cameron wrote:
> On Fri, 23 Feb 2024 11:41:49 -0600
> John Groves <John@xxxxxxxxxx> wrote:
>
> > Notes about this commit:
> >
> > * These methods are based somewhat loosely on pmem_dax_ops from
> > drivers/nvdimm/pmem.c
> >
> > * dev_dax_direct_access() is returns the hpa, pfn and kva. The kva was
> > newly stored as dev_dax->virt_addr by dev_dax_probe().
> >
> > * The hpa/pfn are used for mmap (dax_iomap_fault()), and the kva is used
> > for read/write (dax_iomap_rw())
> >
> > * dev_dax_recovery_write() and dev_dax_zero_page_range() have not been
> > tested yet. I'm looking for suggestions as to how to test those.
> >
> > Signed-off-by: John Groves <john@xxxxxxxxxx>
> > ---
> > drivers/dax/bus.c | 107 ++++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 107 insertions(+)
> >
> > diff --git a/drivers/dax/bus.c b/drivers/dax/bus.c
> > index 664e8c1b9930..06fcda810674 100644
> > --- a/drivers/dax/bus.c
> > +++ b/drivers/dax/bus.c
> > @@ -10,6 +10,12 @@
> > #include "dax-private.h"
> > #include "bus.h"
> >
> > +#if IS_ENABLED(CONFIG_DEV_DAX_IOMAP)
> > +#include <linux/backing-dev.h>
> > +#include <linux/pfn_t.h>
> > +#include <linux/range.h>
> > +#endif
> > +
>
> Is it worth avoiding includes based on config? Probably not.

Just trying to demonstrate that I can be tedious :D
I'll drop the #if unless somebody disagrees.

>
> > static DEFINE_MUTEX(dax_bus_lock);
> >
> > #define DAX_NAME_LEN 30
> > @@ -1349,6 +1355,101 @@ __weak phys_addr_t dax_pgoff_to_phys(struct dev_dax *dev_dax, pgoff_t pgoff,
> > }
> > EXPORT_SYMBOL_GPL(dax_pgoff_to_phys);
> >
> > +#if IS_ENABLED(CONFIG_DEV_DAX_IOMAP)
> > +
>
> > +
> > +static long __dev_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
> > + long nr_pages, enum dax_access_mode mode, void **kaddr,
> > + pfn_t *pfn)
> > +{
> > + struct dev_dax *dev_dax = dax_get_private(dax_dev);
> > + size_t dax_size = dev_dax_size(dev_dax);
> > + size_t size = nr_pages << PAGE_SHIFT;
> > + size_t offset = pgoff << PAGE_SHIFT;
> > + phys_addr_t phys;
> > + u64 virt_addr = dev_dax->virt_addr + offset;
> > + pfn_t local_pfn;
> > + u64 flags = PFN_DEV|PFN_MAP;
> > +
> > + WARN_ON(!dev_dax->virt_addr); /* virt_addr must be saved for direct_access */
> Fair enough, but from local code point of view, does it make sense to check this
> if !kaddr as we won't use this.

Hmm. This gets called with kaddr=NULL for mmap faults, and with non-NULL
kaddr for read/write (which need the virt_addr to do a memcpy variant).
If dev_dax->virt-addr is NULL, mmap will work but read/write will hork.

I lean toward keeping the warning. With these updates, it's broken if
read/write are broken.

> > +
> > + phys = dax_pgoff_to_phys(dev_dax, pgoff, nr_pages << PAGE_SHIFT);
> > +
> > + if (kaddr)
> > + *kaddr = (void *)virt_addr;
>
> Back to earlier comment on virt_addr as a void *. Definitely looking like
> that would be more accurate and simpler! Also not much point in computing
> virt_addr unless kaddr is good.

Yes, done (the void *).

the computation is copied directly from drivers/nvdimm/__pmem_direct_access() -
which does not warn if virt_addr is null. Actually I suppose this code should
just trust that dev_dax_probe sets virt_addr, and not warn?

So I'm now contradicting myself above...

>
> > +
> > + local_pfn = phys_to_pfn_t(phys, flags); /* are flags correct? */
> If you aren't going to do anything with it for !pfn, move it under the if (pfn).
>
> > + if (pfn)
> > + *pfn = local_pfn;
> > +
> > + /* This the valid size at the specified address */
> > + return PHYS_PFN(min_t(size_t, size, dax_size - offset));
> > +}
> > +
>
> > +
> > +static const struct dax_operations dev_dax_ops = {
> > + .direct_access = dev_dax_direct_access,
> > + .zero_page_range = dev_dax_zero_page_range,
> > + .recovery_write = dev_dax_recovery_write,
> > +};
> > +
> > +#endif /* IS_ENABLED(CONFIG_DEV_DAX_IOMAP) */
> > +
> > struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
> > {
> > struct dax_region *dax_region = data->dax_region;
> > @@ -1404,11 +1505,17 @@ struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
> > }
> > }
> >
>
> If we were to make this
>
> if (IS_ENABLED(CONFIG_DEV_DAX_IOMAP))
>
> etc can we avoid the ifdef stuff above and let dead code removal deal with it?
> Might need a few stubs - I haven't tried.

Better, thanks. No stubs needed.

>
> > +#if IS_ENABLED(CONFIG_DEV_DAX_IOMAP)
> > + /* holder_ops currently populated separately in a slightly hacky way */
> > + dax_dev = alloc_dax(dev_dax, &dev_dax_ops);
> > +#else
> > /*
> > * No dax_operations since there is no access to this device outside of
> > * mmap of the resulting character device.
> > */
> > dax_dev = alloc_dax(dev_dax, NULL);
> > +#endif

Thanks!
John