Re: [PATCH v2 06/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Robert Richter <rrichter@amd.com>]

On 20.10.22 22:17:07, Dan Williams wrote:
> Robert Richter wrote:
> > A downstream port must be connected to a component register block.
> > For restricted hosts the base address is determined from the RCRB. The
> > RCRB is provided by the host's CEDT CHBS entry. Rework CEDT parser to
> > get the RCRB and add code to extract the component register block from
> > it.
> > 
> > RCRB's BAR[0..1] point to the component block containing CXL subsystem
> > component registers. MEMBAR extraction follows the PCI base spec here,
> > esp. 64 bit extraction and memory range alignment (6.0, 7.5.1.2.1).
> > 
> > Note: Right now the component register block is used for HDM decoder
> > capability only which is optional for RCDs. If unsupported by the RCD,
> > the HDM init will fail. It is future work to bypass it in this case.
> > 
> > Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> > Signed-off-by: Robert Richter <rrichter@amd.com>
> > ---
> >  drivers/cxl/acpi.c | 79 ++++++++++++++++++++++++++++++++++++++++------
> >  1 file changed, 69 insertions(+), 10 deletions(-)
> > 
> > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > index fb9f72813067..a92d5d7b7a92 100644
> > --- a/drivers/cxl/acpi.c
> > +++ b/drivers/cxl/acpi.c
> > @@ -9,6 +9,8 @@
> >  #include "cxlpci.h"
> >  #include "cxl.h"
> >  
> > +#define CXL_RCRB_SIZE	SZ_8K
> > +
> >  static unsigned long cfmws_to_decoder_flags(int restrictions)
> >  {
> >  	unsigned long flags = CXL_DECODER_F_ENABLE;
> > @@ -229,27 +231,82 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> >  struct cxl_chbs_context {
> >  	struct device *dev;
> >  	unsigned long long uid;
> > -	resource_size_t chbcr;
> > +	struct acpi_cedt_chbs chbs;
> >  };
> >  
> > -static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
> > -			 const unsigned long end)
> > +static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
> > +			const unsigned long end)
> >  {
> >  	struct cxl_chbs_context *ctx = arg;
> >  	struct acpi_cedt_chbs *chbs;
> >  
> > -	if (ctx->chbcr)
> > +	if (ctx->chbs.base)
> >  		return 0;
> >  
> >  	chbs = (struct acpi_cedt_chbs *) header;
> >  
> >  	if (ctx->uid != chbs->uid)
> >  		return 0;
> > -	ctx->chbcr = chbs->base;
> > +	ctx->chbs = *chbs;
> >  
> >  	return 0;
> >  }
> >  
> > +static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
> > +{
> 
> The core logic of this looks good, but this wants to be shared with the
> upstream port component register discovery.
> 
> Full disclosure I am reconciling these patches with an attempt that Dave
> Jiang made at this topic. Since your series hit the list first I will
> let it take the lead, but then fill it in with comments and learnings
> from Dave's effort.
> 
> So in this case Dave moved this into the drivers/cxl/core/regs.c with a
> function signature like:
> 
> enum cxl_rcrb {
>        CXL_RCRB_DOWNSTREAM,
>        CXL_RCRB_UPSTREAM,
> };
> 
> resource_size_t cxl_rcrb_to_component(struct device *dev,
>                                       resource_size_t rcrb_base, int len,
>                                       enum cxl_rcrb which);
> 
> ...where @which alternates when called by cxl_acpi for the downstream
> case, or cxl_mem for the upstream case.

Ok, I see where to go here. Could you point me to Dave's postings you
are referring to? I checked linux-cxl and could not find anything
related to RCRB or that changes regs.c.

> 
> 
> > +	struct acpi_cedt_chbs *chbs = &ctx->chbs;
> > +	resource_size_t component_reg_phys, rcrb;
> > +	u32 bar0, bar1;
> > +	void *addr;
> > +
> > +	if (!chbs->base)
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
> > +		return chbs->base;
> > +
> > +	/* Extract RCRB */
> > +
> > +	if (chbs->length != CXL_RCRB_SIZE)
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	rcrb = chbs->base;
> > +
> > +	dev_dbg(ctx->dev, "RCRB found for UID %lld: 0x%08llx\n",
> > +		ctx->uid, (u64)rcrb);
> > +
> > +	/*
> > +	 * RCRB's BAR[0..1] point to component block containing CXL
> > +	 * subsystem component registers. MEMBAR extraction follows
> > +	 * the PCI Base spec here, esp. 64 bit extraction and memory
> > +	 * ranges alignment (6.0, 7.5.1.2.1).
> > +	 */
> > +	addr = ioremap(rcrb, PCI_BASE_ADDRESS_0 + SZ_8);
> 
> No failure check? This also only needs to map 4K at a time.

Right, will add that.

> 
> > +	bar0 = readl(addr + PCI_BASE_ADDRESS_0);
> > +	bar1 = readl(addr + PCI_BASE_ADDRESS_1);
> > +	iounmap(addr);
> > +
> > +	/* sanity check */
> > +	if (bar0 & (PCI_BASE_ADDRESS_MEM_TYPE_1M | PCI_BASE_ADDRESS_SPACE_IO))
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	component_reg_phys = bar0 & PCI_BASE_ADDRESS_MEM_MASK;
> > +	if (bar0 & PCI_BASE_ADDRESS_MEM_TYPE_64)
> > +		component_reg_phys |= ((u64)bar1) << 32;
> > +
> > +	if (!component_reg_phys)
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	/*
> > +	 * Must be 8k aligned (size of combined CXL 1.1 Downstream and
> > +	 * Upstream Port RCRBs).
> > +	 */
> > +	if (component_reg_phys & (CXL_RCRB_SIZE - 1))
> > +		return CXL_RESOURCE_NONE;
> 
> This is open-coding the IS_ALIGNED() macro. More importantly, why is it
> using RCRB size for the component register block alignment? The
> component lock is 64K, and at least for CXL 2.0 devices it is 64K
> aligned (8.1.9.1 Register Block Offset Low), so I am not sure what this
> check is for?

True, this is a mistake and needs to be corrected. It is the component
reg range which is 64k. Will also use IS_ALIGNED().

> 
> ---
> 
> Given that there are actual CXL RCH platforms in the wild I want this
> topic branch to be the first thing queued for v6.2. To help us
> coordinate I pushed:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/log/?h=rch
> 
> ...with the patches from this set accepted so far. You can use that as
> the baseline for the next spin.

Yes, thanks for that branch and applying the first part.

-Robert