On Wed, 29 May 2024 14:38:58 -0700 Dave Jiang <dave.jiang@xxxxxxxxx> wrote: > The current bandwidth calculation aggregates all the targets. This simple > method does not take into account where multiple targets sharing under > a switch where the aggregated bandwidth can be less or greater than the > upstream link of the switch. > > To accurately account for the shared upstream uplink cases, a new update > function is introduced by walking from the leaves to the root of the > hierachy and adjust the bandwidth in the process. This process is done when > all the targets for a region is present but before the final values are are present > send to the HMAT handling code cached access_coordinate targets. > > The original perf calculation path was kept to calculate the configurations > without switches and also keep the latency data as that does not require > special calculation. The shared upstream link calculation is done as a > second pass when all the endpoints have arrived. > > Suggested-by: Jonathan Cameron <jonathan.cameron@xxxxxxxxxx> > Link: https://lore.kernel.org/linux-cxl/20240501152503.00002e60@xxxxxxxxxx/ > Signed-off-by: Dave Jiang <dave.jiang@xxxxxxxxx> > > --- > v3: > - Redo calculation based on Jonathan's suggestion. (Jonathan) > --- > drivers/cxl/core/cdat.c | 380 ++++++++++++++++++++++++++++++++++++-- Hi Dave, The one case I couldn't quite figure out is that of multiple RP on a given RC. Imagine bw as follows Host | 3 from GP/HMAT _____|_____ RP RP 2| |2 __|__ ___|__ |1 |1 1| |1 EP EP EP EP I'm not sure following through the code that we get 3 (reported aggregate bw to the host bridge) rather than 4 (min elsewhere) I might be missing why this case works... Various minor comments inline. Jonathan > drivers/cxl/core/core.h | 4 +- > drivers/cxl/core/pci.c | 21 +++ > drivers/cxl/core/port.c | 26 +++ > drivers/cxl/core/region.c | 4 + > drivers/cxl/cxl.h | 1 + > 6 files changed, 418 insertions(+), 18 deletions(-) > > diff --git a/drivers/cxl/core/cdat.c b/drivers/cxl/core/cdat.c > index fea214340d4b..e77b416b0fe0 100644 > --- a/drivers/cxl/core/cdat.c > +++ b/drivers/cxl/core/cdat.c > @@ -548,32 +548,378 @@ void cxl_coordinates_combine(struct access_coordinate *out, > > MODULE_IMPORT_NS(CXL); > > -void cxl_region_perf_data_calculate(struct cxl_region *cxlr, > - struct cxl_endpoint_decoder *cxled) > +static void cxl_bandwidth_add(struct access_coordinate *coord, > + struct access_coordinate *c1, > + struct access_coordinate *c2) > { > - struct cxl_memdev *cxlmd = cxled_to_memdev(cxled); > - struct cxl_dev_state *cxlds = cxlmd->cxlds; > - struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds); > - struct range dpa = { > - .start = cxled->dpa_res->start, > - .end = cxled->dpa_res->end, > - }; > - struct cxl_dpa_perf *perf; > + for (int i = 0; i < ACCESS_COORDINATE_MAX; i++) { > + coord[i].read_bandwidth = c1[i].read_bandwidth + > + c2[i].read_bandwidth; > + coord[i].write_bandwidth = c1[i].write_bandwidth + > + c2[i].write_bandwidth; > + } > +} > > - switch (cxlr->mode) { > +struct cxl_perf_ctx { > + struct access_coordinate coord[ACCESS_COORDINATE_MAX]; > + struct cxl_port *port; > +}; ... > +static bool dpa_perf_contains(struct cxl_dpa_perf *perf, > + struct resource *dpa_res) > +{ > + struct range dpa = { > + .start = dpa_res->start, > + .end = dpa_res->end, over indented. > + }; > > if (!range_contains(&perf->dpa_range, &dpa)) > + return false; > + > + return true; > +} > + > +static int cxl_endpoint_gather_coordinates(struct cxl_region *cxlr, > + struct cxl_endpoint_decoder *cxled, > + struct xarray *usp_xa) > +{ > + struct cxl_port *endpoint = to_cxl_port(cxled->cxld.dev.parent); > + struct access_coordinate pci_coord[ACCESS_COORDINATE_MAX]; > + struct access_coordinate sw_coord[ACCESS_COORDINATE_MAX]; > + struct access_coordinate ep_coord[ACCESS_COORDINATE_MAX]; > + struct cxl_memdev *cxlmd = cxled_to_memdev(cxled); > + struct cxl_dev_state *cxlds = cxlmd->cxlds; > + struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds); > + struct pci_dev *pdev = to_pci_dev(cxlds->dev); > + struct cxl_perf_ctx *c __free(kfree) = NULL; Linus isn't fond of the fact free ordering is odd in these cases. Doesn't matter here, but it might if we refactor in future. You can reduce the scope anyway to make that a moot point. > + struct cxl_perf_ctx *perf_ctx; > + struct cxl_port *parent_port; > + struct cxl_dpa_perf *perf; > + unsigned long index; > + void *ptr; > + int rc; > + > + if (cxlds->rcd) > + return -ENODEV; > + > + parent_port = to_cxl_port(endpoint->dev.parent); > + /* No CXL switch upstream then skip out. */ > + if (is_cxl_root(parent_port)) > + return -ENODEV; > + > + perf = cxl_memdev_get_dpa_perf(mds, cxlr->mode); > + if (IS_ERR(perf)) > + return PTR_ERR(perf); > + > + if (!dpa_perf_contains(perf, cxled->dpa_res)) > + return -EINVAL; > + > + /* > + * The index for the xarray is the upstream port device of the upstream > + * CXL switch. > + */ > + index = (unsigned long)(unsigned long *)parent_port->uport_dev; Random side note for another day. That seems rather inefficient as xarrays love densely clustered indexes and this will be aligned so bottom bunch of bits are 0. Ah well - improving that would make for messy code or some nice macros. Why do you need the (unsigned long *) cast? > + perf_ctx = xa_load(usp_xa, index); > + if (!perf_ctx) { > + c = kzalloc(sizeof(*perf_ctx), GFP_KERNEL); scope of c is just in here I think, so drag the definition and destructor down here? > + if (!c) > + return -ENOMEM; > + ptr = xa_store(usp_xa, index, c, GFP_KERNEL); > + if (xa_is_err(ptr)) > + return xa_err(ptr); > + perf_ctx = no_free_ptr(c); > + } > + > + /* Upstream link bandwidth */ Too may upstreams in here. I'd be more specific. /* Direct upstream link from EP bandwidth */ > + rc = cxl_pci_get_bandwidth(pdev, pci_coord); > + if (rc < 0) > + return rc; > + > + /* Min of upstream link bandwidth and Endpoint CDAT bandwidth from DSLBIS */ > + cxl_coordinates_combine(ep_coord, pci_coord, perf->cdat_coord); > + > + /* > + * Retrieve the switch SSLBIS for switch downstream port associated with > + * the endpoint bandwidth. > + */ > + rc = cxl_port_get_switch_dport_bandwidth(endpoint, sw_coord); > + if (rc) > + return rc; > + > + /* Min of the earlier coordinates with the switch SSLBIS bandwidth */ > + cxl_coordinates_combine(ep_coord, ep_coord, sw_coord); > + > + /* > + * Aggregate the computed bandwidth with the current aggregated bandwidth > + * of the endpoints with the same upstream switch port. "switch upstream port" Arguable the DSP could be the upstream switch port (from the EP) > + */ > + cxl_bandwidth_add(perf_ctx->coord, perf_ctx->coord, ep_coord); > + perf_ctx->port = parent_port; > + > + return 0; > +} > + > +static struct xarray *cxl_switch_iterate_coordinates(struct xarray *input_xa) > +{ > + struct xarray *res_xa __free(kfree) = kzalloc(sizeof(*res_xa), GFP_KERNEL); > + struct access_coordinate coords[ACCESS_COORDINATE_MAX]; > + struct cxl_perf_ctx *n __free(kfree) = NULL; Same comment as below. Reduce the scope of this and you'll have the destructor and constructor nicely paired. > + struct cxl_perf_ctx *ctx, *us_ctx; > + unsigned long index, key; > + bool is_root = false; > + void *ptr; > + int rc; > + > + if (!res_xa) > + return ERR_PTR(-ENOMEM); > + xa_init(res_xa); > + > + xa_for_each(input_xa, index, ctx) { > + struct device *dev = (struct device *)(unsigned long *)index; Why cast via an (unsigned long *) ? > + struct cxl_port *parent_port, *port; > + struct cxl_dport *dport; > + struct pci_dev *pdev; > + > + port = ctx->port; > + parent_port = to_cxl_port(port->dev.parent); > + dport = port->parent_dport; > + /* > + * If parent port is CXL root port, then take the min of the parent > + * dport bandwidth, which comes from the generic target via ACPI > + * tables (SRAT & HMAT), and the bandwidth in the xarray. Then > + * update the bandwidth in the xarray. > + */ > + if (is_cxl_root(parent_port)) { > + ctx->port = parent_port; > + cxl_coordinates_combine(ctx->coord, ctx->coord, > + dport->coord); I'm a bit lost in all the levels of iteration so may have missed it. Do we assume that GP BW (which is the root bridge) is shared across multiple root ports on that host bridge if they are both part of the interleave set? > + is_root = true; > + continue; > + } > + > + /* Below is the calculation for another switch upstream */ > + > + /* > + * If the device isn't an upstream PCIe port, there's something > + * wrong with the topology. > + */ > + if (!dev_is_pci(dev)) > + return ERR_PTR(-EINVAL); > + > + /* Retrieve the upstream link bandwidth */ > + pdev = to_pci_dev(dev); > + rc = cxl_pci_get_bandwidth(pdev, coords); > + if (rc) > + return ERR_PTR(-ENXIO); > + > + /* > + * Take the min of downstream bandwidth and the upstream link > + * bandwidth. > + */ > + cxl_coordinates_combine(coords, coords, ctx->coord); > + /* > + * Take the min of the calculated bandwdith and the upstream > + * switch SSLBIS bandwidth. > + */ > + cxl_coordinates_combine(coords, coords, dport->coord); > + > + /* > + * Create an xarray entry with the key of the upstream > + * port of the upstream switch. > + */ > + key = (unsigned long)(unsigned long *)parent_port->uport_dev; > + us_ctx = xa_load(res_xa, key); > + if (!us_ctx) { as above, define n here. > + n = kzalloc(sizeof(*n), GFP_KERNEL); > + if (!n) > + return ERR_PTR(-ENOMEM); > + ptr = xa_store(res_xa, key, n, GFP_KERNEL); > + if (xa_is_err(ptr)) > + return ERR_PTR(xa_err(ptr)); > + us_ctx = no_free_ptr(n); > + } > + > + /* > + * Aggregate the bandwidth based on the upstream switch. > + */ > + cxl_bandwidth_add(us_ctx->coord, us_ctx->coord, coords); > + us_ctx->port = parent_port; > + } > + > + if (is_root) > + return NULL; > + > + return_ptr(res_xa); > +} > + > +static void cxl_region_update_access_coordinate(struct cxl_region *cxlr, > + struct xarray *input_xa) > +{ > + struct access_coordinate coord[ACCESS_COORDINATE_MAX]; > + struct cxl_perf_ctx *ctx; > + unsigned long index; > + > + memset(coord, 0, sizeof(coord)); > + xa_for_each(input_xa, index, ctx) > + cxl_bandwidth_add(coord, coord, ctx->coord); > + > + for (int i = 0; i < ACCESS_COORDINATE_MAX; i++) { > + cxlr->coord[i].read_bandwidth = coord[i].read_bandwidth; > + cxlr->coord[i].write_bandwidth = coord[i].write_bandwidth; > + } > +} > + > +static void free_perf_xa(struct xarray *xa) > +{ > + struct cxl_perf_ctx *ctx; > + unsigned long index; > + > + if (!xa) > + return; > + > + xa_for_each(xa, index, ctx) > + kfree(ctx); > + xa_destroy(xa); > + kfree(xa); > +} > + > +/* > + * cxl_region_shared_upstream_perf_update - Recalculate the access coordinates > + * @cxl_region: the cxl region to recalculate > + * > + * For certain region construction with endpoints behind CXL switches, > + * there is the possibility of the total bandwdith for all the endpoints > + * behind a switch between less or more than the switch upstream link. The switch being less or more > + * algorithm assumes the configuration is a symmetric topology that puts topology as that maximizes performance. > + * max performance into consideration. For a region topology without switch > + * involvement, the code exits without doing computation. > + * I suspect this restriction might bite us one day, but meh, it's complex enough for now and clearly stated what the limitation is! > + * An example hierachy: > + * > + * CFMWS 0 > + * | > + * _________|_________ > + * | | > + * GP0/HB0/ACPI0017-0 GP1/HB1/ACPI0017-1 > + * | | | | > + * RP0 RP1 RP2 RP3 > + * | | | | > + * SW 0 SW 1 SW 2 SW 3 > + * | | | | | | | | > + * EP0 EP1 EP2 EP3 EP4 EP5 EP6 EP7 > + * > + * Computation for the example hierachy: > + * > + * Min (GP0 to CPU BW, > + * Min(SW 0 Upstream Link to RP0 BW, > + * Min(SW0SSLBIS for SW0DSP0 (EP0), EP0 DSLBIS, EP0 Upstream Link) + > + * Min(SW0SSLBIS for SW0DSP1 (EP1), EP1 DSLBIS, EP1 Upstream link)) + > + * Min(SW 1 Upstream Link to RP1 BW, > + * Min(SW1SSLBIS for SW1DSP0 (EP2), EP2 DSLBIS, EP2 Upstream Link) + > + * Min(SW1SSLBIS for SW1DSP1 (EP3), EP3 DSLBIS, EP3 Upstream link))) + > + * Min (GP1 to CPU BW, > + * Min(SW 2 Upstream Link to RP2 BW, > + * Min(SW2SSLBIS for SW2DSP0 (EP4), EP4 DSLBIS, EP4 Upstream Link) + > + * Min(SW2SSLBIS for SW2DSP1 (EP5), EP5 DSLBIS, EP5 Upstream link)) + > + * Min(SW 3 Upstream Link to RP3 BW, > + * Min(SW3SSLBIS for SW3DSP0 (EP6), EP6 DSLBIS, EP6 Upstream Link) + > + * Min(SW3SSLBIS for SW3DSP1 (EP7), EP7 DSLBIS, EP7 Upstream link)))) > + */ > +void cxl_region_shared_upstream_perf_update(struct cxl_region *cxlr) > +{ > + struct xarray *usp_xa, *iter_xa, *working_xa; > + int rc; > + > + lockdep_assert_held(&cxl_dpa_rwsem); > + > + usp_xa = kzalloc(sizeof(*usp_xa), GFP_KERNEL); > + if (!usp_xa) > + return; > + > + xa_init(usp_xa); > + > + /* > + * Collect aggregated endpoint bandwidth and store the bandwidth in > + * an xarray indexed by the upstream port of the switch. The bandwidth > + * is aggregated per switch. Each endpoint consists of the minumum of > + * bandwidth from DSLBIS from the endpoint CDAT, the endpoint upstream > + * link bandwidth, and the bandwidth from the SSLBIS of the switch > + * CDAT for the downstream port that's associated with the endpoint. CDAT for the switch upstream port to the downstream port that's ... Should have p2p downstream to downstream numbers in there too I think and we don't yet care about those. > + */ > + for (int i = 0; i < cxlr->params.nr_targets; i++) { > + struct cxl_endpoint_decoder *cxled = cxlr->params.targets[i]; > + > + rc = cxl_endpoint_gather_coordinates(cxlr, cxled, usp_xa); > + if (rc) { > + free_perf_xa(usp_xa); > + return; > + } > + } > + > + /* > + * Iterate through the components in the xarray and aggregate any > + * component that share the same upstream switch. The iteration Nitpick but worth making it clear this is the VCS of a switch or avoid it by referring to sharing the upstream link from the switch. > + * takes consideration of multi-level switch hierachy. > + * > + * When cxl_switch_iterate_coordinates() detect the next level > + * upstream is a root port, it updates the bandwidth in the > + * xarray by taking the min of the provided bandwidth and > + * the bandwidth from the generic port. The function > + * returns NULL to indicate that the input xarray has been > + * updated. > + */ > + iter_xa = usp_xa; > + usp_xa = NULL; > + do { > + working_xa = cxl_switch_iterate_coordinates(iter_xa); > + if (IS_ERR(working_xa)) { Maybe worth a traditional end of function cleanup and goto as there is always one of these to free. You could use __free() magic but that becomes less obvious when iterating so up to you. (there are examples in tree such as in property.h that do iteration with __free() magic but I'd not take them as good precedence given the author (me) :) ) err: free_perf_xa(iter_xa); > + free_perf_xa(iter_xa); > + return; > + } > + if (working_xa) { > + free_perf_xa(iter_xa); > + iter_xa = working_xa; > + } > + } while (working_xa); > + > + /* > + * Aggregate the bandwidths in the xarray and update the region > + * bandwidths with the newly calculated bandwidths. > + */ > + cxl_region_update_access_coordinate(cxlr, iter_xa); > + free_perf_xa(iter_xa); > +} > + > +void cxl_region_perf_data_calculate(struct cxl_region *cxlr, > + struct cxl_endpoint_decoder *cxled) > +{ > + struct cxl_memdev *cxlmd = cxled_to_memdev(cxled); > + struct cxl_dev_state *cxlds = cxlmd->cxlds; > + struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlds); Trivial and obviously this is in original code but you could take opportunity for struct cxl_memdev_state *mds = to_cxl_memdev_state(cxlmd->cxlds); as cxlds isn't accessed in here otherwise. Nor is cxlmd, but folding that in as well feels too far to go. > + struct cxl_dpa_perf *perf; > + > + perf = cxl_memdev_get_dpa_perf(mds, cxlr->mode); > + if (IS_ERR(perf)) > + return; > + > + lockdep_assert_held(&cxl_dpa_rwsem); > + > + if (!dpa_perf_contains(perf, cxled->dpa_res)) > return; > > for (int i = 0; i < ACCESS_COORDINATE_MAX; i++) { > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c > index 887ed6e358fb..2c7509a9a943 100644 > --- a/drivers/cxl/core/port.c > +++ b/drivers/cxl/core/port.c > @@ -2259,6 +2259,32 @@ int cxl_endpoint_get_perf_coordinates(struct cxl_port *port, > } > EXPORT_SYMBOL_NS_GPL(cxl_endpoint_get_perf_coordinates, CXL); > > +int cxl_port_get_switch_dport_bandwidth(struct cxl_port *port, > + struct access_coordinate *c) > +{ > + struct cxl_dport *dport; > + struct cxl_port *iter; The iter naming seems odd as no iterating going on.. It's only there I think to check we aren't on a root port? So struct cxl_dport *dport = port->parent_dport; /* Check this port is connected to a switch DSP and not an RP */ if (parent_port_is_cxl_root(to_cxl_port(port->dev.parent)) return -ENODEV; if (!coord... as before > + > + iter = port; > + dport = port->parent_dport; > + iter = to_cxl_port(iter->dev.parent); > + > + if (parent_port_is_cxl_root(iter)) > + return -ENODEV; > + > + if (!coordinates_valid(dport->coord)) { > + dev_warn(dport->dport_dev, "XXX %s invalid coords\n", __func__); XXX? I guess a marker you were using to find it during debuging that wants to go away now. > + return -EINVAL; > + } > + > + for (int i = 0; i < ACCESS_COORDINATE_MAX; i++) { > + c[i].read_bandwidth = dport->coord[i].read_bandwidth; > + c[i].write_bandwidth = dport->coord[i].write_bandwidth; > + } > + > + return 0; > +} > +
