Re: [PATCH 33/46] resource: Introduce alloc_free_mem_region()

[Dan Williams <dan.j.williams@intel.com>]

Jonathan Cameron wrote:
> On Thu, 23 Jun 2022 21:19:37 -0700
> Dan Williams <dan.j.williams@intel.com> wrote:
> 
> > The core of devm_request_free_mem_region() is a helper that searches for
> > free space in iomem_resource and performs __request_region_locked() on
> > the result of that search. The policy choices of the implementation
> > conform to what CONFIG_DEVICE_PRIVATE users want which is memory that is
> > immediately marked busy, and a preference to search for the first-fit
> > free range in descending order from the top of the physical address
> > space.
> > 
> > CXL has a need for a similar allocator, but with the following tweaks:
> > 
> > 1/ Search for free space in ascending order
> > 
> > 2/ Search for free space relative to a given CXL window
> > 
> > 3/ 'insert' rather than 'request' the new resource given downstream
> >    drivers from the CXL Region driver (like the pmem or dax drivers) are
> >    responsible for request_mem_region() when they activate the memory
> >    range.
> > 
> > Rework __request_free_mem_region() into get_free_mem_region() which
> > takes a set of GFR_* (Get Free Region) flags to control the allocation
> > policy (ascending vs descending), and "busy" policy (insert_resource()
> > vs request_region()).
> > 
> > Suggested-by: Jason Gunthorpe <jgg@nvidia.com>
> > Link: https://lore.kernel.org/linux-cxl/20220420143406.GY2120790@nvidia.com/
> > Cc: Matthew Wilcox <willy@infradead.org>
> > Cc: Christoph Hellwig <hch@infradead.org>
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> 
> A few things inline,
> 
> Thanks,
> 
> Jonathan
> 
> > ---
> >  include/linux/ioport.h |   2 +
> >  kernel/resource.c      | 174 ++++++++++++++++++++++++++++++++---------
> >  mm/Kconfig             |   5 ++
> >  3 files changed, 146 insertions(+), 35 deletions(-)
> > 
> > diff --git a/include/linux/ioport.h b/include/linux/ioport.h
> > index ec5f71f7135b..ed03518347aa 100644
> > --- a/include/linux/ioport.h
> > +++ b/include/linux/ioport.h
> > @@ -329,6 +329,8 @@ struct resource *devm_request_free_mem_region(struct device *dev,
> >  		struct resource *base, unsigned long size);
> >  struct resource *request_free_mem_region(struct resource *base,
> >  		unsigned long size, const char *name);
> > +struct resource *alloc_free_mem_region(struct resource *base,
> > +		unsigned long size, unsigned long align, const char *name);
> >  
> >  static inline void irqresource_disabled(struct resource *res, u32 irq)
> >  {
> > diff --git a/kernel/resource.c b/kernel/resource.c
> > index 53a534db350e..9fc990274106 100644
> > --- a/kernel/resource.c
> > +++ b/kernel/resource.c
> 
> 
> > +static bool gfr_continue(struct resource *base, resource_size_t addr,
> > +			 resource_size_t size, unsigned long flags)
> > +{
> > +	if (flags & GFR_DESCENDING)
> > +		return addr > size && addr >= base->start;
> > +	return addr > addr - size &&
> 
> Is this checking for wrap around?  If so maybe a comment to call that out?

Yes, and ok.

> 
> > +	       addr <= min_t(resource_size_t, base->end,
> > +			     (1ULL << MAX_PHYSMEM_BITS) - 1);
> > +}
> > +
> > +static resource_size_t gfr_next(resource_size_t addr, resource_size_t size,
> > +				unsigned long flags)
> > +{
> > +	if (flags & GFR_DESCENDING)
> > +		return addr - size;
> > +	return addr + size;
> > +}
> > +
> > +static void remove_free_mem_region(void *_res)
> >  {
> > -	resource_size_t end, addr;
> > +	struct resource *res = _res;
> > +
> > +	if (res->parent)
> > +		remove_resource(res);
> > +	free_resource(res);
> > +}
> > +
> > +static struct resource *
> > +get_free_mem_region(struct device *dev, struct resource *base,
> > +		    resource_size_t size, const unsigned long align,
> > +		    const char *name, const unsigned long desc,
> > +		    const unsigned long flags)
> > +{
> > +	resource_size_t addr;
> >  	struct resource *res;
> >  	struct region_devres *dr = NULL;
> >  
> > -	size = ALIGN(size, 1UL << PA_SECTION_SHIFT);
> > -	end = min_t(unsigned long, base->end, (1UL << MAX_PHYSMEM_BITS) - 1);
> > -	addr = end - size + 1UL;
> > +	size = ALIGN(size, align);
> >  
> >  	res = alloc_resource(GFP_KERNEL);
> >  	if (!res)
> >  		return ERR_PTR(-ENOMEM);
> >  
> > -	if (dev) {
> > +	if (dev && (flags & GFR_REQUEST_REGION)) {
> >  		dr = devres_alloc(devm_region_release,
> >  				sizeof(struct region_devres), GFP_KERNEL);
> >  		if (!dr) {
> >  			free_resource(res);
> >  			return ERR_PTR(-ENOMEM);
> >  		}
> > +	} else if (dev) {
> > +		if (devm_add_action_or_reset(dev, remove_free_mem_region, res))
> > +			return ERR_PTR(-ENOMEM);
> 
> slightly nicer to return whatever value you got back from devm_add_action_or_reset()

Yes, but it is known to only return -ENOMEM on failure and saves adding
a local @rc variable.

> 
> >  	}
> >  
> >  	write_lock(&resource_lock);
> > -	for (; addr > size && addr >= base->start; addr -= size) {
> > -		if (__region_intersects(addr, size, 0, IORES_DESC_NONE) !=
> > -				REGION_DISJOINT)
> > +	for (addr = gfr_start(base, size, align, flags);
> > +	     gfr_continue(base, addr, size, flags);
> > +	     addr = gfr_next(addr, size, flags)) {
> > +		if (__region_intersects(base, addr, size, 0, IORES_DESC_NONE) !=
> > +		    REGION_DISJOINT)
> >  			continue;
> >  
> > -		if (__request_region_locked(res, &iomem_resource, addr, size,
> > -						name, 0))
> > -			break;
> > +		if (flags & GFR_REQUEST_REGION) {
> > +			if (__request_region_locked(res, &iomem_resource, addr,
> > +						    size, name, 0))
> > +				break;
> >  
> > -		if (dev) {
> > -			dr->parent = &iomem_resource;
> > -			dr->start = addr;
> > -			dr->n = size;
> > -			devres_add(dev, dr);
> > -		}
> > +			if (dev) {
> > +				dr->parent = &iomem_resource;
> > +				dr->start = addr;
> > +				dr->n = size;
> > +				devres_add(dev, dr);
> > +			}
> >  
> > -		res->desc = IORES_DESC_DEVICE_PRIVATE_MEMORY;
> > -		write_unlock(&resource_lock);
> > +			res->desc = desc;
> > +			write_unlock(&resource_lock);
> > +
> > +
> > +			/*
> > +			 * A driver is claiming this region so revoke any
> > +			 * mappings.
> > +			 */
> > +			revoke_iomem(res);
> > +		} else {
> > +			res->start = addr;
> > +			res->end = addr + size - 1;
> > +			res->name = name;
> > +			res->desc = desc;
> > +			res->flags = IORESOURCE_MEM;
> > +
> > +			/*
> > +			 * Only succeed if the resource hosts an exclusive
> > +			 * range after the insert
> > +			 */
> > +			if (__insert_resource(base, res) || res->child)
> > +				break;
> > +
> > +			write_unlock(&resource_lock);
> > +		}
> >  
> > -		/*
> > -		 * A driver is claiming this region so revoke any mappings.
> > -		 */
> > -		revoke_iomem(res);
> >  		return res;
> >  	}
> >  	write_unlock(&resource_lock);
> >  
> > -	free_resource(res);
> > -	if (dr)
> > +	if (flags & GFR_REQUEST_REGION) {
> > +		free_resource(res);
> >  		devres_free(dr);
> 
> The original if (dr) was unnecessary as devres_free() checks.
> 
> Looking just at this patch it looks like you aren't covering the
> corner case of dev == NULL and GFR_REQUEST_REGION.
> 
> Perhaps worth a tiny comment in patch description? (doesn't seem worth
> pulling this change out as a precursor given it's so small).
> Of add the extra if (dr) back in to 'document' that no change...

Added to the changelog:

As part of the consolidation of the legacy GFR_REQUEST_REGION case with
the new default of just inserting a new resource into the free space
some minor cleanups like not checking for NULL before calling
devres_free() (which does its own check) is included.

> 
> 
> > +	} else if (dev)
> > +		devm_release_action(dev, remove_free_mem_region, res);
> >  
> >  	return ERR_PTR(-ERANGE);
> >  }
> > @@ -1854,18 +1928,48 @@ static struct resource *__request_free_mem_region(struct device *dev,
> >  struct resource *devm_request_free_mem_region(struct device *dev,
> >  		struct resource *base, unsigned long size)
> >  {
> > -	return __request_free_mem_region(dev, base, size, dev_name(dev));
> > +	unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
> > +
> > +	return get_free_mem_region(dev, base, size, GFR_DEFAULT_ALIGN,
> > +				   dev_name(dev),
> > +				   IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
> >  }
> >  EXPORT_SYMBOL_GPL(devm_request_free_mem_region);
> >  
> >  struct resource *request_free_mem_region(struct resource *base,
> >  		unsigned long size, const char *name)
> >  {
> > -	return __request_free_mem_region(NULL, base, size, name);
> > +	unsigned long flags = GFR_DESCENDING | GFR_REQUEST_REGION;
> > +
> > +	return get_free_mem_region(NULL, base, size, GFR_DEFAULT_ALIGN, name,
> > +				   IORES_DESC_DEVICE_PRIVATE_MEMORY, flags);
> >  }
> >  EXPORT_SYMBOL_GPL(request_free_mem_region);
> >  
> > -#endif /* CONFIG_DEVICE_PRIVATE */
> > +/**
> > + * alloc_free_mem_region - find a free region relative to @base
> > + * @base: resource that will parent the new resource
> > + * @size: size in bytes of memory to allocate from @base
> > + * @align: alignment requirements for the allocation
> > + * @name: resource name
> > + *
> > + * Buses like CXL, that can dynamically instantiate new memory regions,
> > + * need a method to allocate physical address space for those regions.
> > + * Allocate and insert a new resource to cover a free, unclaimed by a
> > + * descendant of @base, range in the span of @base.
> > + */
> > +struct resource *alloc_free_mem_region(struct resource *base,
> Given the extra align parameter, does it make sense to give this a naming
> that highlights that vs the other two interfaces above?
> 
> alloc_free_mem_region_aligned()

The other variants are also aligned, they just aren't variably aligned,
they are implicitly aligned to GFR_DEFAULT_ALIGN. So I think calling
this one _aligned() betrays what is happening in the other cases.

> > +				       unsigned long size, unsigned long align,
> > +				       const char *name)
> > +{
> > +	/* GFR_ASCENDING | GFR_INSERT_RESOURCE */
> 
> Given those flags don't exist and some fool like me might grep for them
> perhaps better to describe it in text
> 
> 	/* Default of ascending direction and insert resource */

Ok.