2 * Intel 3000/3010 Memory Controller kernel module
3 * Copyright (C) 2007 Akamai Technologies, Inc.
4 * Shamelessly copied from:
5 * Intel D82875P Memory Controller kernel module
6 * (C) 2003 Linux Networx (http://lnxi.com)
8 * This file may be distributed under the terms of the
9 * GNU General Public License.
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/pci.h>
15 #include <linux/pci_ids.h>
16 #include <linux/edac.h>
17 #include "edac_module.h"
19 #define EDAC_MOD_STR "i3000_edac"
22 #define I3000_RANKS_PER_CHANNEL 4
23 #define I3000_CHANNELS 2
25 /* Intel 3000 register addresses - device 0 function 0 - DRAM Controller */
27 #define I3000_MCHBAR 0x44 /* MCH Memory Mapped Register BAR */
28 #define I3000_MCHBAR_MASK 0xffffc000
29 #define I3000_MMR_WINDOW_SIZE 16384
31 #define I3000_EDEAP 0x70 /* Extended DRAM Error Address Pointer (8b)
36 #define I3000_DEAP 0x58 /* DRAM Error Address Pointer (32b)
42 #define I3000_DEAP_GRAIN (1 << 7)
45 * Helper functions to decode the DEAP/EDEAP hardware registers.
47 * The type promotion here is deliberate; we're deriving an
48 * unsigned long pfn and offset from hardware regs which are u8/u32.
51 static inline unsigned long deap_pfn(u8 edeap
, u32 deap
)
54 deap
|= (edeap
& 1) << (32 - PAGE_SHIFT
);
58 static inline unsigned long deap_offset(u32 deap
)
60 return deap
& ~(I3000_DEAP_GRAIN
- 1) & ~PAGE_MASK
;
63 static inline int deap_channel(u32 deap
)
68 #define I3000_DERRSYN 0x5c /* DRAM Error Syndrome (8b)
70 * 7:0 DRAM ECC Syndrome
73 #define I3000_ERRSTS 0xc8 /* Error Status Register (16b)
76 * 11 MCH Thermal Sensor Event
79 * 9 LOCK to non-DRAM Memory Flag (LCKF)
80 * 8 Received Refresh Timeout Flag (RRTOF)
82 * 1 Multi-bit DRAM ECC Error Flag (DMERR)
83 * 0 Single-bit DRAM ECC Error Flag (DSERR)
85 #define I3000_ERRSTS_BITS 0x0b03 /* bits which indicate errors */
86 #define I3000_ERRSTS_UE 0x0002
87 #define I3000_ERRSTS_CE 0x0001
89 #define I3000_ERRCMD 0xca /* Error Command (16b)
92 * 11 SERR on MCH Thermal Sensor Event
95 * 9 SERR on LOCK to non-DRAM Memory
97 * 8 SERR on DRAM Refresh Timeout
100 * 1 SERR Multi-Bit DRAM ECC Error
102 * 0 SERR on Single-Bit ECC Error
106 /* Intel MMIO register space - device 0 function 0 - MMR space */
108 #define I3000_DRB_SHIFT 25 /* 32MiB grain */
110 #define I3000_C0DRB 0x100 /* Channel 0 DRAM Rank Boundary (8b x 4)
112 * 7:0 Channel 0 DRAM Rank Boundary Address
114 #define I3000_C1DRB 0x180 /* Channel 1 DRAM Rank Boundary (8b x 4)
116 * 7:0 Channel 1 DRAM Rank Boundary Address
119 #define I3000_C0DRA 0x108 /* Channel 0 DRAM Rank Attribute (8b x 2)
122 * 6:4 DRAM odd Rank Attribute
124 * 2:0 DRAM even Rank Attribute
126 * Each attribute defines the page
127 * size of the corresponding rank:
135 #define I3000_C1DRA 0x188 /* Channel 1 DRAM Rank Attribute (8b x 2) */
137 static inline unsigned char odd_rank_attrib(unsigned char dra
)
139 return (dra
& 0x70) >> 4;
142 static inline unsigned char even_rank_attrib(unsigned char dra
)
147 #define I3000_C0DRC0 0x120 /* DRAM Controller Mode 0 (32b)
150 * 29 Initialization Complete (IC)
152 * 10:8 Refresh Mode Select (RMS)
154 * 6:4 Mode Select (SMS)
159 #define I3000_C0DRC1 0x124 /* DRAM Controller Mode 1 (32b)
161 * 31 Enhanced Addressing Enable (ENHADE)
169 struct i3000_dev_info
{
170 const char *ctl_name
;
173 struct i3000_error_info
{
181 static const struct i3000_dev_info i3000_devs
[] = {
183 .ctl_name
= "i3000"},
186 static struct pci_dev
*mci_pdev
;
187 static int i3000_registered
= 1;
188 static struct edac_pci_ctl_info
*i3000_pci
;
190 static void i3000_get_error_info(struct mem_ctl_info
*mci
,
191 struct i3000_error_info
*info
)
193 struct pci_dev
*pdev
;
195 pdev
= to_pci_dev(mci
->pdev
);
198 * This is a mess because there is no atomic way to read all the
199 * registers at once and the registers can transition from CE being
202 pci_read_config_word(pdev
, I3000_ERRSTS
, &info
->errsts
);
203 if (!(info
->errsts
& I3000_ERRSTS_BITS
))
205 pci_read_config_byte(pdev
, I3000_EDEAP
, &info
->edeap
);
206 pci_read_config_dword(pdev
, I3000_DEAP
, &info
->deap
);
207 pci_read_config_byte(pdev
, I3000_DERRSYN
, &info
->derrsyn
);
208 pci_read_config_word(pdev
, I3000_ERRSTS
, &info
->errsts2
);
211 * If the error is the same for both reads then the first set
212 * of reads is valid. If there is a change then there is a CE
213 * with no info and the second set of reads is valid and
216 if ((info
->errsts
^ info
->errsts2
) & I3000_ERRSTS_BITS
) {
217 pci_read_config_byte(pdev
, I3000_EDEAP
, &info
->edeap
);
218 pci_read_config_dword(pdev
, I3000_DEAP
, &info
->deap
);
219 pci_read_config_byte(pdev
, I3000_DERRSYN
, &info
->derrsyn
);
223 * Clear any error bits.
224 * (Yes, we really clear bits by writing 1 to them.)
226 pci_write_bits16(pdev
, I3000_ERRSTS
, I3000_ERRSTS_BITS
,
230 static int i3000_process_error_info(struct mem_ctl_info
*mci
,
231 struct i3000_error_info
*info
,
234 int row
, multi_chan
, channel
;
235 unsigned long pfn
, offset
;
237 multi_chan
= mci
->csrows
[0]->nr_channels
- 1;
239 if (!(info
->errsts
& I3000_ERRSTS_BITS
))
245 if ((info
->errsts
^ info
->errsts2
) & I3000_ERRSTS_BITS
) {
246 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, 0, 0, 0,
248 "UE overwrote CE", "");
249 info
->errsts
= info
->errsts2
;
252 pfn
= deap_pfn(info
->edeap
, info
->deap
);
253 offset
= deap_offset(info
->deap
);
254 channel
= deap_channel(info
->deap
);
256 row
= edac_mc_find_csrow_by_page(mci
, pfn
);
258 if (info
->errsts
& I3000_ERRSTS_UE
)
259 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1,
264 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED
, mci
, 1,
265 pfn
, offset
, info
->derrsyn
,
266 row
, multi_chan
? channel
: 0, -1,
272 static void i3000_check(struct mem_ctl_info
*mci
)
274 struct i3000_error_info info
;
276 edac_dbg(1, "MC%d\n", mci
->mc_idx
);
277 i3000_get_error_info(mci
, &info
);
278 i3000_process_error_info(mci
, &info
, 1);
281 static int i3000_is_interleaved(const unsigned char *c0dra
,
282 const unsigned char *c1dra
,
283 const unsigned char *c0drb
,
284 const unsigned char *c1drb
)
289 * If the channels aren't populated identically then
290 * we're not interleaved.
292 for (i
= 0; i
< I3000_RANKS_PER_CHANNEL
/ 2; i
++)
293 if (odd_rank_attrib(c0dra
[i
]) != odd_rank_attrib(c1dra
[i
]) ||
294 even_rank_attrib(c0dra
[i
]) !=
295 even_rank_attrib(c1dra
[i
]))
299 * If the rank boundaries for the two channels are different
300 * then we're not interleaved.
302 for (i
= 0; i
< I3000_RANKS_PER_CHANNEL
; i
++)
303 if (c0drb
[i
] != c1drb
[i
])
309 static int i3000_probe1(struct pci_dev
*pdev
, int dev_idx
)
313 struct mem_ctl_info
*mci
= NULL
;
314 struct edac_mc_layer layers
[2];
315 unsigned long last_cumul_size
, nr_pages
;
316 int interleaved
, nr_channels
;
317 unsigned char dra
[I3000_RANKS
/ 2], drb
[I3000_RANKS
];
318 unsigned char *c0dra
= dra
, *c1dra
= &dra
[I3000_RANKS_PER_CHANNEL
/ 2];
319 unsigned char *c0drb
= drb
, *c1drb
= &drb
[I3000_RANKS_PER_CHANNEL
];
320 unsigned long mchbar
;
321 void __iomem
*window
;
323 edac_dbg(0, "MC:\n");
325 pci_read_config_dword(pdev
, I3000_MCHBAR
, (u32
*) & mchbar
);
326 mchbar
&= I3000_MCHBAR_MASK
;
327 window
= ioremap_nocache(mchbar
, I3000_MMR_WINDOW_SIZE
);
329 printk(KERN_ERR
"i3000: cannot map mmio space at 0x%lx\n",
334 c0dra
[0] = readb(window
+ I3000_C0DRA
+ 0); /* ranks 0,1 */
335 c0dra
[1] = readb(window
+ I3000_C0DRA
+ 1); /* ranks 2,3 */
336 c1dra
[0] = readb(window
+ I3000_C1DRA
+ 0); /* ranks 0,1 */
337 c1dra
[1] = readb(window
+ I3000_C1DRA
+ 1); /* ranks 2,3 */
339 for (i
= 0; i
< I3000_RANKS_PER_CHANNEL
; i
++) {
340 c0drb
[i
] = readb(window
+ I3000_C0DRB
+ i
);
341 c1drb
[i
] = readb(window
+ I3000_C1DRB
+ i
);
347 * Figure out how many channels we have.
349 * If we have what the datasheet calls "asymmetric channels"
350 * (essentially the same as what was called "virtual single
351 * channel mode" in the i82875) then it's a single channel as
352 * far as EDAC is concerned.
354 interleaved
= i3000_is_interleaved(c0dra
, c1dra
, c0drb
, c1drb
);
355 nr_channels
= interleaved
? 2 : 1;
357 layers
[0].type
= EDAC_MC_LAYER_CHIP_SELECT
;
358 layers
[0].size
= I3000_RANKS
/ nr_channels
;
359 layers
[0].is_virt_csrow
= true;
360 layers
[1].type
= EDAC_MC_LAYER_CHANNEL
;
361 layers
[1].size
= nr_channels
;
362 layers
[1].is_virt_csrow
= false;
363 mci
= edac_mc_alloc(0, ARRAY_SIZE(layers
), layers
, 0);
367 edac_dbg(3, "MC: init mci\n");
369 mci
->pdev
= &pdev
->dev
;
370 mci
->mtype_cap
= MEM_FLAG_DDR2
;
372 mci
->edac_ctl_cap
= EDAC_FLAG_SECDED
;
373 mci
->edac_cap
= EDAC_FLAG_SECDED
;
375 mci
->mod_name
= EDAC_MOD_STR
;
376 mci
->ctl_name
= i3000_devs
[dev_idx
].ctl_name
;
377 mci
->dev_name
= pci_name(pdev
);
378 mci
->edac_check
= i3000_check
;
379 mci
->ctl_page_to_phys
= NULL
;
382 * The dram rank boundary (DRB) reg values are boundary addresses
383 * for each DRAM rank with a granularity of 32MB. DRB regs are
384 * cumulative; the last one will contain the total memory
385 * contained in all ranks.
387 * If we're in interleaved mode then we're only walking through
388 * the ranks of controller 0, so we double all the values we see.
390 for (last_cumul_size
= i
= 0; i
< mci
->nr_csrows
; i
++) {
393 struct csrow_info
*csrow
= mci
->csrows
[i
];
396 cumul_size
= value
<< (I3000_DRB_SHIFT
- PAGE_SHIFT
);
399 edac_dbg(3, "MC: (%d) cumul_size 0x%x\n", i
, cumul_size
);
400 if (cumul_size
== last_cumul_size
)
403 csrow
->first_page
= last_cumul_size
;
404 csrow
->last_page
= cumul_size
- 1;
405 nr_pages
= cumul_size
- last_cumul_size
;
406 last_cumul_size
= cumul_size
;
408 for (j
= 0; j
< nr_channels
; j
++) {
409 struct dimm_info
*dimm
= csrow
->channels
[j
]->dimm
;
411 dimm
->nr_pages
= nr_pages
/ nr_channels
;
412 dimm
->grain
= I3000_DEAP_GRAIN
;
413 dimm
->mtype
= MEM_DDR2
;
414 dimm
->dtype
= DEV_UNKNOWN
;
415 dimm
->edac_mode
= EDAC_UNKNOWN
;
420 * Clear any error bits.
421 * (Yes, we really clear bits by writing 1 to them.)
423 pci_write_bits16(pdev
, I3000_ERRSTS
, I3000_ERRSTS_BITS
,
427 if (edac_mc_add_mc(mci
)) {
428 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
432 /* allocating generic PCI control info */
433 i3000_pci
= edac_pci_create_generic_ctl(&pdev
->dev
, EDAC_MOD_STR
);
436 "%s(): Unable to create PCI control\n",
439 "%s(): PCI error report via EDAC not setup\n",
443 /* get this far and it's successful */
444 edac_dbg(3, "MC: success\n");
454 /* returns count (>= 0), or negative on error */
455 static int i3000_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
459 edac_dbg(0, "MC:\n");
461 if (pci_enable_device(pdev
) < 0)
464 rc
= i3000_probe1(pdev
, ent
->driver_data
);
466 mci_pdev
= pci_dev_get(pdev
);
471 static void i3000_remove_one(struct pci_dev
*pdev
)
473 struct mem_ctl_info
*mci
;
478 edac_pci_release_generic_ctl(i3000_pci
);
480 mci
= edac_mc_del_mc(&pdev
->dev
);
487 static const struct pci_device_id i3000_pci_tbl
[] = {
489 PCI_VEND_DEV(INTEL
, 3000_HB
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
493 } /* 0 terminated list. */
496 MODULE_DEVICE_TABLE(pci
, i3000_pci_tbl
);
498 static struct pci_driver i3000_driver
= {
499 .name
= EDAC_MOD_STR
,
500 .probe
= i3000_init_one
,
501 .remove
= i3000_remove_one
,
502 .id_table
= i3000_pci_tbl
,
505 static int __init
i3000_init(void)
509 edac_dbg(3, "MC:\n");
511 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
514 pci_rc
= pci_register_driver(&i3000_driver
);
519 i3000_registered
= 0;
520 mci_pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
521 PCI_DEVICE_ID_INTEL_3000_HB
, NULL
);
523 edac_dbg(0, "i3000 pci_get_device fail\n");
528 pci_rc
= i3000_init_one(mci_pdev
, i3000_pci_tbl
);
530 edac_dbg(0, "i3000 init fail\n");
539 pci_unregister_driver(&i3000_driver
);
542 pci_dev_put(mci_pdev
);
547 static void __exit
i3000_exit(void)
549 edac_dbg(3, "MC:\n");
551 pci_unregister_driver(&i3000_driver
);
552 if (!i3000_registered
) {
553 i3000_remove_one(mci_pdev
);
554 pci_dev_put(mci_pdev
);
558 module_init(i3000_init
);
559 module_exit(i3000_exit
);
561 MODULE_LICENSE("GPL");
562 MODULE_AUTHOR("Akamai Technologies Arthur Ulfeldt/Jason Uhlenkott");
563 MODULE_DESCRIPTION("MC support for Intel 3000 memory hub controllers");
565 module_param(edac_op_state
, int, 0444);
566 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");