Re: [PATCH 14/46] cxl/hdm: Enumerate allocated DPA

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

Jonathan Cameron wrote:
> On Thu, 23 Jun 2022 19:46:44 -0700
> Dan Williams <dan.j.williams@intel.com> wrote:
> 
> > In preparation for provisioining CXL regions, add accounting for the DPA
> > space consumed by existing regions / decoders. Recall, a CXL region is a
> > memory range comrpised from one or more endpoint devices contributing a
> > mapping of their DPA into HPA space through a decoder.
> > 
> > Record the DPA ranges covered by committed decoders at initial probe of
> > endpoint ports relative to a per-device resource tree of the DPA type
> > (pmem or volaltile-ram).
> > 
> > The cxl_dpa_rwsem semaphore is introduced to globally synchronize DPA
> > state across all endpoints and their decoders at once. The vast majority
> > of DPA operations are reads as region creation is expected to be as rare
> > as disk partitioning and volume creation. The device_lock() for this
> > synchronization is specifically avoided for concern of entangling with
> > sysfs attribute removal.
> > 
> > Co-developed-by: Ben Widawsky <bwidawsk@kernel.org>
> > Signed-off-by: Ben Widawsky <bwidawsk@kernel.org>
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >  drivers/cxl/core/hdm.c |  148 ++++++++++++++++++++++++++++++++++++++++++++----
> >  drivers/cxl/cxl.h      |    2 +
> >  drivers/cxl/cxlmem.h   |   13 ++++
> >  3 files changed, 152 insertions(+), 11 deletions(-)
> > 
> > diff --git a/drivers/cxl/core/hdm.c b/drivers/cxl/core/hdm.c
> > index c940a4911fee..daae6e533146 100644
> > --- a/drivers/cxl/core/hdm.c
> > +++ b/drivers/cxl/core/hdm.c
> > @@ -7,6 +7,8 @@
> >  #include "cxlmem.h"
> >  #include "core.h"
> >  
> > +static DECLARE_RWSEM(cxl_dpa_rwsem);
> 
> I've not checked many files, but pci.c has equivalent static defines after
> the DOC: entry so for consistency move this below that?

ok.

> 
> 
> > +
> >  /**
> >   * DOC: cxl core hdm
> >   *
> > @@ -128,10 +130,108 @@ struct cxl_hdm *devm_cxl_setup_hdm(struct cxl_port *port)
> >  }
> >  EXPORT_SYMBOL_NS_GPL(devm_cxl_setup_hdm, CXL);
> >  
> > +/*
> > + * Must be called in a context that synchronizes against this decoder's
> > + * port ->remove() callback (like an endpoint decoder sysfs attribute)
> > + */
> > +static void cxl_dpa_release(void *cxled);
> > +static void __cxl_dpa_release(struct cxl_endpoint_decoder *cxled, bool remove_action)
> > +{
> > +	struct cxl_port *port = cxled_to_port(cxled);
> > +	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
> > +	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > +	struct resource *res = cxled->dpa_res;
> > +
> > +	lockdep_assert_held_write(&cxl_dpa_rwsem);
> > +
> > +	if (remove_action)
> > +		devm_remove_action(&port->dev, cxl_dpa_release, cxled);
> 
> This code organization is more surprising than I'd like. Why not move this to
> a wrapper that is like devm_kfree() and similar which do the free now and
> remove from the devm list?

True. I see how this got here incrementally, but this end state can
definitely now be fixed up to be more devm idiomatic.

> 
> static void __cxl_dpa_release(struct cxl_endpoint_decoder *cxled)
> {
> 	struct cxl_port *port = cxled_to_port(cxled);
> 	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
> 	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> 	struct resource *res = cxled->dpa_res;
> 
> 	if (cxled->skip)
> 		__release_region(&cxlds->dpa_res, res->start - cxled->skip,
> 				 cxled->skip);
> 	cxled->skip = 0;
> 	__release_region(&cxlds->dpa_res, res->start, resource_size(res));
> 	cxled->dpa_res = NULL;
> }
> 
> /* possibly add some underscores to this name to indicate it's special
>    in when you can safely call it */
> static void devm_cxl_dpa_release(struct cxl_endpoint_decoder *cxled)
> {
> 	struct cxl_port *port = cxled_to_port(cxled);
> 	lockdep_assert_held_write(&cxl_dpa_rwsem);
> 	devm_remove_action(&port->dev, cxl_dpa_release, cxled);
> 	__cxl_dpa_release(cxled);
> }
> 
> static void cxl_dpa_release(void *cxled)
> {
> 	down_write(&cxl_dpa_rwsem);
> 	__cxl_dpa_release(cxled, false);
> 	up_write(&cxl_dpa_rwsem);
> }
> 
> > +
> > +	if (cxled->skip)
> > +		__release_region(&cxlds->dpa_res, res->start - cxled->skip,
> > +				 cxled->skip);
> > +	cxled->skip = 0;
> > +	__release_region(&cxlds->dpa_res, res->start, resource_size(res));
> > +	cxled->dpa_res = NULL;
> > +}
> > +
> > +static void cxl_dpa_release(void *cxled)
> > +{
> > +	down_write(&cxl_dpa_rwsem);
> > +	__cxl_dpa_release(cxled, false);
> > +	up_write(&cxl_dpa_rwsem);
> > +}
> > +
> > +static int __cxl_dpa_reserve(struct cxl_endpoint_decoder *cxled,
> > +			     resource_size_t base, resource_size_t len,
> > +			     resource_size_t skip)
> > +{
> > +	struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
> > +	struct cxl_port *port = cxled_to_port(cxled);
> > +	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > +	struct device *dev = &port->dev;
> > +	struct resource *res;
> > +
> > +	lockdep_assert_held_write(&cxl_dpa_rwsem);
> > +
> > +	if (!len)
> > +		return 0;
> > +
> > +	if (cxled->dpa_res) {
> > +		dev_dbg(dev, "decoder%d.%d: existing allocation %pr assigned\n",
> > +			port->id, cxled->cxld.id, cxled->dpa_res);
> > +		return -EBUSY;
> > +	}
> > +
> > +	if (skip) {
> > +		res = __request_region(&cxlds->dpa_res, base - skip, skip,
> > +				       dev_name(dev), 0);
> 
> 
> Interface that uses a backwards definition of skip as what to skip before
> the base parameter is a little odd can we rename base parameter to something
> like 'current_top' then have base = current_top + skip?  current_top naming
> not great though...

How about just name it "skipped" instead of "skip"? As the parameter is
how many bytes were skipped to allow a new allocation to start at base.