CXL XOR Interleave Arithmetic
Subsystems (3)
[email protected]
maintainerv4 cover CXL XOR Interleave Arithmetic
From: Alison Schofield <[email protected]>
Changes in v4:
- Use GENMASK_ULL to fix i386 arch build (0-day)
- Use do_div to fix ARM arch build (0-day)
- Update comments in ACPICA patch to reflect new state of the ACPICA patch - pending again in github.
Changes in v3:
- Fix the 3, 6, 12 way interleave (again).
- Do not look for a CXIMS when not needed for x1 & x3 interleaves
- New cxl_test patch: Add cxl_test module support for this feature
- In a separate ndctl patch, cxl test: cxl_xor_region is added
Changes in v2:
- Use ilog2() of the decoded interleave ways to determine number of xormaps, instead of using encoded ways directly. This fixes 3, 6, and 12 way interleaves. (Dan)
Add support for the new 'XOR' Interleave Arithmetic as defined in the CXL 3.0 Specification:
https://www.computeexpresslink.org/download-the-specification
The prereq patch is: tools/testing/cxl: Add an x1 host-bridge with 4 devices
Alison Schofield (3):
For ACPICA: Add the CXIMS structure definition to the CEDT table cxl/acpi: Support CXL XOR Interleave Math (CXIMS) tools/testing/cxl: Add XOR math support
drivers/cxl/cxl.h | 2 + include/acpi/actbl1.h | 14 ++- drivers/cxl/acpi.c | 133 +++++++++++++++++++++++++++-- tools/testing/cxl/test/cxl.c | 161 ++++++++++++++++++++++++++++++++++- 4 files changed, 300 insertions(+), 10 deletions(-)
base-commit: 1cd8a2537eb07751d405ab7e2223f20338a90506
prerequisite-patch-id: 7475f0a9ed0870e85bb4ba2b3eb25618325614bf
Signature · 1 line
2.31.1
[email protected]
maintainerPATCH v4 1/3 For ACPICA: Add the CXIMS structure definition to the CEDT table
From: Alison Schofield <[email protected]>
A linux-ized ACPI patch is included here for reference. The actual ACPICA pull request is pending: https://github.com/acpica/acpica/pull/795
The CXL XOR Interleave Math Structure (CXIMS) is added to the CXL Early Discovery Table (CEDT). This new structure is defined in the CXL 3.0 specification Section 9.17.1.4
https://www.computeexpresslink.org/spec-landing
--- include/acpi/actbl1.h | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
include/acpi/actbl1.h +13 -1
@@ -329,7 +329,8 @@ struct acpi_cedt_header {enum acpi_cedt_type {ACPI_CEDT_TYPE_CHBS =0 ,ACPI_CEDT_TYPE_CFMWS =1 , -ACPI_CEDT_TYPE_RESERVED =2 , +ACPI_CEDT_TYPE_CXIMS =2 , +ACPI_CEDT_TYPE_RESERVED =3 , };/* Values for version field above */
@@ -380,6 +381,7 @@ struct acpi_cedt_cfmws_target_element {/* Values for Interleave Arithmetic field above */ #define ACPI_CEDT_CFMWS_ARITHMETIC_MODULO (0) +#define ACPI_CEDT_CFMWS_ARITHMETIC_XOR (1)/* Values for Restrictions field above */
@@ -389,6 +391,16 @@ struct acpi_cedt_cfmws_target_element {#define ACPI_CEDT_CFMWS_RESTRICT_PMEM (1<<3)#define ACPI_CEDT_CFMWS_RESTRICT_FIXED (1<<4) +/* 2: CXL XOR Interleave Math Structure */ + +struct acpi_cedt_cxims { +struct acpi_cedt_header header; +u16 reserved1; +u8 hbig; +u8 nr_xormaps; +u64 xormap_list[]; +}; +/******************************************************************************* * * CPEP - Corrected Platform Error Polling table (ACPI 4.0)
Signature · 1 line
2.31.1
[email protected]
maintainerPATCH v4 2/3 cxl/acpi: Support CXL XOR Interleave Math (CXIMS)
In reply to [email protected]From: Alison Schofield <[email protected]>
When the CFMWS is using XOR math, parse the corresponding CXIMS structure and store the xormaps in the root decoder structure. Use the xormaps in a new lookup, cxl_hb_xor(), to find a targets entry in the host bridge interleave target list.
Defined in CXL Specfication 3.0 Section: 9.17.1
--- drivers/cxl/cxl.h | 2 + drivers/cxl/acpi.c | 133 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 130 insertions(+), 5 deletions(-)
drivers/cxl/cxl.h +2 -0
@@ -330,12 +330,14 @@ struct cxl_switch_decoder {* @res: host / parent resource for region allocations * @region_id: region id for next region provisioning event * @calc_hb: which host bridge covers the n'th position by granularity +* @platform_data: platform specific configuration data * @cxlsd: base cxl switch decoder */ struct cxl_root_decoder {struct resource *res;atomic_t region_id;struct cxl_dport *(*calc_hb)(struct cxl_root_decoder *cxlrd,int pos); +void *platform_data;struct cxl_switch_decoder cxlsd; };
drivers/cxl/acpi.c +128 -5
161 lines not on this page · raw message
Signature · 1 line
2.31.1
[email protected]
maintainerPATCH v4 3/3 tools/testing/cxl: Add XOR math support
In reply to [email protected]From: Alison Schofield <[email protected]>
Expand the cxl_test topology to include CFMWS's that use XOR math for interleave arithmetic, as defined in the CXL Specification 3.0, and to include an additional host bridge for testing a 4-way host bridge interleave.
With this expanded topology, cxl_test is useful for testing:
x1 - no interleave, target index is always 0
x2 - 2 way interleave uses one xormap
x3 - 3 way interleave uses no xormaps and relies on a modulo calc. x4 - 4 way interleave uses two xormaps
Interleave ways not offered here (x6, 8, 12, 16) follow patterns tested in the set above.
Avoid memory bloat in cxl_test by defining a set of CFMWS entries that use XOR interleave arithmetic and can be allocated using a new module parameter 'interleave_arithmetic'. The default remains as Modulo (0).
modprobe cxl_test interleave_arithmetic=1
--- tools/testing/cxl/test/cxl.c | 161 ++++++++++++++++++++++++++++++++++- 1 file changed, 157 insertions(+), 4 deletions(-)
tools/testing/cxl/test/cxl.c +157 -4
@@ -11,7 +11,9 @@#include <cxlmem.h> #include "mock.h" -#define NR_CXL_HOST_BRIDGES 2 +static int interleave_arithmetic; + +#define NR_CXL_HOST_BRIDGES 3#define NR_CXL_DIRECT_HOST_BRIDGES 1#define NR_CXL_ROOT_PORTS 2#define NR_CXL_DIRECT_ROOT_PORTS 4
@@ -45,6 +47,9 @@ host_bridge[NR_CXL_HOST_BRIDGES + NR_CXL_DIRECT_HOST_BRIDGES] = { [2 ] = { .handle = &host_bridge[2 ], }, + [3 ] = { + .handle = &host_bridge[3 ], + }, };static bool is_mock_dev (struct device *dev)
@@ -101,6 +106,26 @@ static struct {struct acpi_cedt_cfmws cfmws;u32 target[1 ]; } cfmws5; +struct { +struct acpi_cedt_cfmws cfmws; +u32 target[1 ]; + } cfmws6; +struct { +struct acpi_cedt_cfmws cfmws; +u32 target[2 ]; + } cfmws7; +struct { +struct acpi_cedt_cfmws cfmws; +u32 target[3 ]; + } cfmws8; +struct { +struct acpi_cedt_cfmws cfmws; +u32 target[4 ]; + } cfmws9; +struct { +struct acpi_cedt_cxims cxims; +u64 xormap_list[2 ]; + } cxims0; }__packed mock_cedt = { .cedt = { .header = {
@@ -133,6 +158,14 @@ static struct { .uid = 2, .cxl_version = ACPI_CEDT_CHBS_VERSION_CXL20, }, + .chbs[3] = { + .header = { + .type = ACPI_CEDT_TYPE_CHBS, + .length = sizeof(mock_cedt.chbs[0]), + }, + .uid = 3, + .cxl_version = ACPI_CEDT_CHBS_VERSION_CXL20, + }, .cfmws0 = { .cfmws = { .header = {
@@ -223,15 +256,113 @@ static struct { }, .target = { 2 }, }, + +/* CFMWS #6 through #9 use XOR Math (interleave_arithmetic == 1) */ + .cfmws6 = { + .cfmws = { + .header = { + .type = ACPI_CEDT_TYPE_CFMWS, + .length = sizeof(mock_cedt.cfmws6), + }, + .interleave_arithmetic = 1, +/* No interleaving */ + .interleave_ways = 0, + .granularity = 0, + .restrictions =ACPI_CEDT_CFMWS_RESTRICT_TYPE3 | + ACPI_CEDT_CFMWS_RESTRICT_PMEM, + .qtg_id = 0, + .window_size = SZ_256M * 4UL, + }, + .target = { 0 }, + }, + .cfmws7 = { + .cfmws = { + .header = { + .type = ACPI_CEDT_TYPE_CFMWS, + .length = sizeof(mock_cedt.cfmws7), + }, + .interleave_arithmetic = 1, +/* 2-way interleave */ + .interleave_ways = 1, + .granularity = 0, + .restrictions =ACPI_CEDT_CFMWS_RESTRICT_TYPE3 | + ACPI_CEDT_CFMWS_RESTRICT_PMEM, + .qtg_id = 0, + .window_size = SZ_256M * 8UL, + }, + .target = { 0, 1, }, + }, + .cfmws8 = { + .cfmws = { + .header = { + .type =ACPI_CEDT_TYPE_CFMWS , + .length = sizeof(mock_cedt.cfmws8), + }, + .interleave_arithmetic =1 , +/* 3-way interleave */ + .interleave_ways = 8, + .granularity = 0, + .restrictions = ACPI_CEDT_CFMWS_RESTRICT_TYPE3 | +ACPI_CEDT_CFMWS_RESTRICT_PMEM , + .qtg_id = 0, + .window_size = SZ_256M * 12UL, + }, + .target = { 0, 1, 2 }, + }, + .cfmws9 = { + .cfmws = { + .header = { + .type =ACPI_CEDT_TYPE_CFMWS , + .length = sizeof(mock_cedt.cfmws9), + }, + .interleave_arithmetic =1 , +/* 4-way interleave */ + .interleave_ways = 2, + .granularity = 0, + .restrictions = ACPI_CEDT_CFMWS_RESTRICT_TYPE3 | +ACPI_CEDT_CFMWS_RESTRICT_PMEM , + .qtg_id = 0, + .window_size = SZ_256M * 16UL, + }, + .target = { 0, 1, 2, 3 }, + }, + .cxims0 = { + .cxims = { + .header = { + .type =ACPI_CEDT_TYPE_CXIMS , + .length = sizeof(mock_cedt.cxims0), + }, + .hbig =0 , + .nr_xormaps = 2, + }, + .xormap_list = { 0x404100, 0x808200 }, + }, + }; -struct acpi_cedt_cfmws *mock_cfmws[6 ] = { +struct acpi_cedt_cfmws *mock_cfmws[10 ] = { [0 ] = &mock_cedt.cfmws0.cfmws, [1 ] = &mock_cedt.cfmws1.cfmws, [2 ] = &mock_cedt.cfmws2.cfmws, [3 ] = &mock_cedt.cfmws3.cfmws, [4 ] = &mock_cedt.cfmws4.cfmws, [5 ] = &mock_cedt.cfmws5.cfmws, +/* Modulo Math above, XOR Math below */ + [6 ] = &mock_cedt.cfmws6.cfmws, + [7 ] = &mock_cedt.cfmws7.cfmws, + [8 ] = &mock_cedt.cfmws8.cfmws, + [9 ] = &mock_cedt.cfmws9.cfmws, +}; + +static int cfmws_start; +static int cfmws_end; +#define CFMWS_MOD_ARRAY_START 0 +#define CFMWS_MOD_ARRAY_END 5 +#define CFMWS_XOR_ARRAY_START 6 +#define CFMWS_XOR_ARRAY_END 9 + +struct acpi_cedt_cxims *mock_cxims[1 ] = { + [0 ] = &mock_cedt.cxims0.cxims, }; struct cxl_mock_res {
@@ -303,7 +434,7 @@ static int populate_cedt(void) chbs->length = size; } -for (i =0 ; i <ARRAY_SIZE (mock_cfmws); i++) { +for (i = cfmws_start; i <= cfmws_end; i++) {struct acpi_cedt_cfmws *window = mock_cfmws[i]; res =alloc_mock_res (window->window_size);
@@ -346,12 +477,19 @@ static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id, }if (id ==ACPI_CEDT_TYPE_CFMWS ) -for (i =0 ; i <ARRAY_SIZE (mock_cfmws); i++) { +for (i = cfmws_start; i <= cfmws_end; i++) { h = (union acpi_subtable_headers *) mock_cfmws[i]; end = (unsigned long ) h + mock_cfmws[i]->header.length;handler_arg (h, arg, end); } +if (id ==ACPI_CEDT_TYPE_CXIMS ) +for (i =0 ; i <ARRAY_SIZE (mock_cxims); i++) { + h = (union acpi_subtable_headers *)mock_cxims[i]; + end = (unsigned long )h + mock_cxims[i]->header.length; +handler_arg (h, arg, end); + } +return 0 ; }
@@ -432,6 +570,9 @@ mock_pci_root[NR_CXL_HOST_BRIDGES + NR_CXL_DIRECT_HOST_BRIDGES] = { [2 ] = { .bus = &mock_pci_bus[2 ], }, + [3 ] = { + .bus = &mock_pci_bus[3 ], + }, };static bool is_mock_bus (struct pci_bus *bus)
@@ -713,6 +854,16 @@ static __init int cxl_test_init(void)if (rc) goto err_gen_pool_add; +if (interleave_arithmetic ==1 ) { + cfmws_start =CFMWS_XOR_ARRAY_START ; + cfmws_end =CFMWS_XOR_ARRAY_END ; +dev_dbg (NULL ,"cxl_test loading xor_math option\n" ); + }else { + cfmws_start =CFMWS_MOD_ARRAY_START ; + cfmws_end =CFMWS_MOD_ARRAY_END ; +dev_dbg (NULL ,"cxl_test loading default (modulo) option\n" ); + } + rc =populate_cedt ();if (rc) goto err_populate;
@@ -926,6 +1077,8 @@ static __exit void cxl_test_exit(void)unregister_cxl_mock_ops (&cxl_mock_ops); } +module_param (interleave_arithmetic, int,0000 ); +MODULE_PARM_DESC (interleave_arithmetic,"Modulo:0, XOR:1" );module_init (cxl_test_init);module_exit (cxl_test_exit);MODULE_LICENSE ("GPL v2" );
Signature · 1 line
2.31.1
RE: [PATCH v4 1/3] For ACPICA: Add the CXIMS structure definition to the CEDT table
In reply to [email protected]Quoted 4 lines · alison.schofield@ wrote:
From: Alison Schofield <[email protected]>
A linux-ized ACPI patch is included here for reference. The actual ACPICA pull request is pending: https://github.com/acpica/acpica/pull/795
Looks like this is moving along now:
RE: [PATCH v4 2/3] cxl/acpi: Support CXL XOR Interleave Math (CXIMS)
In reply to [email protected]Quoted 70 lines · alison.schofield@ wrote:
From: Alison Schofield <[email protected]>
When the CFMWS is using XOR math, parse the corresponding CXIMS structure and store the xormaps in the root decoder structure. Use the xormaps in a new lookup,
cxl_hb_xor(), to find a targets entry in the host bridge interleave target list.Defined in CXL Specfication 3.0 Section: 9.17.1
Signed-off-by Alison Schofield <[email protected]>--- drivers/cxl/cxl.h | 2 + drivers/cxl/acpi.c | 133 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 130 insertions(+), 5 deletions(-)drivers/cxl/cxl.h +2
-0@@ -330,12 +330,14 @@ struct cxl_switch_decoder {* @res: host / parent resource for region allocations * @region_id: region id for next region provisioning event * @calc_hb: which host bridge covers the n'th position by granularity +* @platform_data: platform specific configuration data * @cxlsd: base cxl switch decoder */ struct cxl_root_decoder {struct resource *res;atomic_t region_id;struct cxl_dport *(*calc_hb)(struct cxl_root_decoder *cxlrd,int pos); +void *platform_data;struct cxl_switch_decoder cxlsd; };drivers/cxl/acpi.c +24
-0@@ -6,9 +6,110 @@#include <linux/kernel.h> #include <linux/acpi.h> #include <linux/pci.h> +#include <asm/div64.h> #include "cxlpci.h" #include "cxl.h" +struct cxims_data { +int nr_maps; +u64 xormaps[]; +}; + +/* +* Find a targets entry (n) in the host bridge interleave list. +* CXL Specfication 3.0 Table 9-22 +*/ +static struct cxl_dport *cxl_hb_xor (struct cxl_root_decoder *cxlrd,int pos) +{ +struct cxl_switch_decoder *cxlsd = &cxlrd->cxlsd; +struct cxims_data *cximsd = cxlrd->platform_data; +struct cxl_decoder *cxld = &cxlsd->cxld; +int ig = cxld->interleave_granularity; +int iw = cxld->interleave_ways; +int i, eiw, n =0 ; +u64 hpa; + +if (dev_WARN_ONCE (&cxld->dev, + cxld->interleave_ways != cxlsd->nr_targets, +"misconfigured root decoder\n" )) +return NULL ;
drivers/cxl/acpi.c line 45
Quoted 10 lines
drivers/cxl/acpi.c +10
-0+ +if (iw ==1 ) +/* Entry is always 0 for no interleave */ +return cxlrd->cxlsd.target[0 ]; + + hpa = cxlrd->res->start + pos * ig; + +if (iw ==3 ) { +/* Initialize 'i' for the modulo calc */ + i =0 ;
From cxl_acpi:
nr_maps = ilog2(cxld->interleave_ways / 3);
...so cximsd->nr_maps is already 0 when iw is 3, so no need for a goto, unless I missed something.
drivers/cxl/acpi.c line 58
Quoted 13 lines
drivers/cxl/acpi.c +13
-0+ goto no_map; + } + +/* IW: 2,4,6,8,12,16 begin building 'n' using xormaps */ +for (i =0 ; i < cximsd->nr_maps; i++) + n |= (hweight64 (hpa & cximsd->xormaps[i]) &1 ) << i; + +no_map: +/* IW: 3,6,12 add a modulo calculation to 'n' */ +if (!is_power_of_2 (iw)) { + eiw =ilog2 (iw /3 ) +8 ; + hpa &=GENMASK_ULL (51 , eiw + ig); + n =do_div (hpa,3 ) << i;
I would just use "<< cxmisd->nr_maps" and move this before the power-of-2 loop with its own early "return cxlrd->cxlsd.target[n];" in the 'if ()' block. No need to force these 2 cases to have a common exit.
Alison Schofield
maintainerRe: [PATCH v4 2/3] cxl/acpi: Support CXL XOR Interleave Math (CXIMS)
Quoted 84 lines · Dan Williams
Quoted 70 lines · alison.schofield@ wrote:
From: Alison Schofield <[email protected]>
When the CFMWS is using XOR math, parse the corresponding CXIMS structure and store the xormaps in the root decoder structure. Use the xormaps in a new lookup,
cxl_hb_xor(), to find a targets entry in the host bridge interleave target list.Defined in CXL Specfication 3.0 Section: 9.17.1
Signed-off-by Alison Schofield <[email protected]>--- drivers/cxl/cxl.h | 2 + drivers/cxl/acpi.c | 133 +++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 130 insertions(+), 5 deletions(-)drivers/cxl/cxl.h +2
-0@@ -330,12 +330,14 @@ struct cxl_switch_decoder {* @res: host / parent resource for region allocations * @region_id: region id for next region provisioning event * @calc_hb: which host bridge covers the n'th position by granularity +* @platform_data: platform specific configuration data * @cxlsd: base cxl switch decoder */ struct cxl_root_decoder {struct resource *res;atomic_t region_id;struct cxl_dport *(*calc_hb)(struct cxl_root_decoder *cxlrd,int pos); +void *platform_data;struct cxl_switch_decoder cxlsd; };drivers/cxl/acpi.c +24
-0@@ -6,9 +6,110 @@#include <linux/kernel.h> #include <linux/acpi.h> #include <linux/pci.h> +#include <asm/div64.h> #include "cxlpci.h" #include "cxl.h" +struct cxims_data { +int nr_maps; +u64 xormaps[]; +}; + +/* +* Find a targets entry (n) in the host bridge interleave list. +* CXL Specfication 3.0 Table 9-22 +*/ +static struct cxl_dport *cxl_hb_xor (struct cxl_root_decoder *cxlrd,int pos) +{ +struct cxl_switch_decoder *cxlsd = &cxlrd->cxlsd; +struct cxims_data *cximsd = cxlrd->platform_data; +struct cxl_decoder *cxld = &cxlsd->cxld; +int ig = cxld->interleave_granularity; +int iw = cxld->interleave_ways; +int i, eiw, n =0 ; +u64 hpa; + +if (dev_WARN_ONCE (&cxld->dev, + cxld->interleave_ways != cxlsd->nr_targets, +"misconfigured root decoder\n" )) +return NULL ;drivers/cxl/acpi.c line 45
Quoted 10 lines
drivers/cxl/acpi.c +10
-0+ +if (iw ==1 ) +/* Entry is always 0 for no interleave */ +return cxlrd->cxlsd.target[0 ]; + + hpa = cxlrd->res->start + pos * ig; + +if (iw ==3 ) { +/* Initialize 'i' for the modulo calc */ + i =0 ;From
cxl_acpi:
Quoted 4 lines
nr_maps=ilog2(cxld->interleave_ways / 3);...so
cximsd->nr_mapsis already 0 when iw is 3, so no need for a goto, unless I missed something.
That'd be a NULL dereference. For x1 and x3, which don't use xormaps, no cximsd is saved.
I kept the goto, but moved the init to the declaration.
Quoted 10 lines
Quoted 10 lines
drivers/cxl/acpi.c +9
-0+ goto no_map; + } + +/* IW: 2,4,6,8,12,16 begin building 'n' using xormaps */ +for (i =0 ; i < cximsd->nr_maps; i++) + n |= (hweight64 (hpa & cximsd->xormaps[i]) &1 ) << i; + +no_map: +/* IW: 3,6,12 add a modulo calculation to 'n' */
Quoted 8 lines
drivers/cxl/acpi.c line 58
Quoted 4 lines
drivers/cxl/acpi.c +4
-0+if (!is_power_of_2 (iw)) { + eiw =ilog2 (iw /3 ) +8 ; + hpa &=GENMASK_ULL (51 , eiw + ig); + n =do_div (hpa,3 ) << i;I would just use "<<
cxmisd->nr_maps" and move this before the power-of-2 loop with its own early "return cxlrd->cxlsd.target[n];" in the 'if ()' block. No need to force these 2 cases to have a common exit.
That makes sense based on what you see above, but the code is wrong. It should add that final calc to the 'n' that was started with the xormaps. Although x3 doesn't use any maps, x6 and x12 do.
this:
Quoted 1 line
+ n =
do_div(hpa, 3)<< i;
it should be
+ n |= do_div(hpa, 3) << i;