2 * Intel 3200/3210 Memory Controller kernel module
3 * Copyright (C) 2008-2009 Akamai Technologies, Inc.
4 * Portions by Hitoshi Mitake <h.mitake@gmail.com>.
6 * This file may be distributed under the terms of the
7 * GNU General Public License.
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/pci.h>
13 #include <linux/pci_ids.h>
14 #include <linux/edac.h>
16 #include "edac_core.h"
18 #include <asm-generic/io-64-nonatomic-lo-hi.h>
20 #define I3200_REVISION "1.1"
22 #define EDAC_MOD_STR "i3200_edac"
24 #define PCI_DEVICE_ID_INTEL_3200_HB 0x29f0
28 #define I3200_RANKS_PER_CHANNEL 4
29 #define I3200_CHANNELS 2
31 /* Intel 3200 register addresses - device 0 function 0 - DRAM Controller */
33 #define I3200_MCHBAR_LOW 0x48 /* MCH Memory Mapped Register BAR */
34 #define I3200_MCHBAR_HIGH 0x4c
35 #define I3200_MCHBAR_MASK 0xfffffc000ULL /* bits 35:14 */
36 #define I3200_MMR_WINDOW_SIZE 16384
38 #define I3200_TOM 0xa0 /* Top of Memory (16b)
41 * 9:0 total populated physical memory
43 #define I3200_TOM_MASK 0x3ff /* bits 9:0 */
44 #define I3200_TOM_SHIFT 26 /* 64MiB grain */
46 #define I3200_ERRSTS 0xc8 /* Error Status Register (16b)
49 * 14 Isochronous TBWRR Run Behind FIFO Full
51 * 13 Isochronous TBWRR Run Behind FIFO Put
54 * 11 MCH Thermal Sensor Event
55 * for SMI/SCI/SERR (GTSE)
57 * 9 LOCK to non-DRAM Memory Flag (LCKF)
59 * 7 DRAM Throttle Flag (DTF)
61 * 1 Multi-bit DRAM ECC Error Flag (DMERR)
62 * 0 Single-bit DRAM ECC Error Flag (DSERR)
64 #define I3200_ERRSTS_UE 0x0002
65 #define I3200_ERRSTS_CE 0x0001
66 #define I3200_ERRSTS_BITS (I3200_ERRSTS_UE | I3200_ERRSTS_CE)
69 /* Intel MMIO register space - device 0 function 0 - MMR space */
71 #define I3200_C0DRB 0x200 /* Channel 0 DRAM Rank Boundary (16b x 4)
74 * 9:0 Channel 0 DRAM Rank Boundary Address
76 #define I3200_C1DRB 0x600 /* Channel 1 DRAM Rank Boundary (16b x 4) */
77 #define I3200_DRB_MASK 0x3ff /* bits 9:0 */
78 #define I3200_DRB_SHIFT 26 /* 64MiB grain */
80 #define I3200_C0ECCERRLOG 0x280 /* Channel 0 ECC Error Log (64b)
82 * 63:48 Error Column Address (ERRCOL)
83 * 47:32 Error Row Address (ERRROW)
84 * 31:29 Error Bank Address (ERRBANK)
85 * 28:27 Error Rank Address (ERRRANK)
87 * 23:16 Error Syndrome (ERRSYND)
89 * 1 Multiple Bit Error Status (MERRSTS)
90 * 0 Correctable Error Status (CERRSTS)
92 #define I3200_C1ECCERRLOG 0x680 /* Chan 1 ECC Error Log (64b) */
93 #define I3200_ECCERRLOG_CE 0x1
94 #define I3200_ECCERRLOG_UE 0x2
95 #define I3200_ECCERRLOG_RANK_BITS 0x18000000
96 #define I3200_ECCERRLOG_RANK_SHIFT 27
97 #define I3200_ECCERRLOG_SYNDROME_BITS 0xff0000
98 #define I3200_ECCERRLOG_SYNDROME_SHIFT 16
99 #define I3200_CAPID0 0xe0 /* P.95 of spec for details */
102 void __iomem
*window
;
105 static int nr_channels
;
107 static int how_many_channels(struct pci_dev
*pdev
)
111 unsigned char capid0_8b
; /* 8th byte of CAPID0 */
113 pci_read_config_byte(pdev
, I3200_CAPID0
+ 8, &capid0_8b
);
115 if (capid0_8b
& 0x20) { /* check DCD: Dual Channel Disable */
116 edac_dbg(0, "In single channel mode\n");
119 edac_dbg(0, "In dual channel mode\n");
123 if (capid0_8b
& 0x10) /* check if both channels are filled */
124 edac_dbg(0, "2 DIMMS per channel disabled\n");
126 edac_dbg(0, "2 DIMMS per channel enabled\n");
131 static unsigned long eccerrlog_syndrome(u64 log
)
133 return (log
& I3200_ECCERRLOG_SYNDROME_BITS
) >>
134 I3200_ECCERRLOG_SYNDROME_SHIFT
;
137 static int eccerrlog_row(int channel
, u64 log
)
139 u64 rank
= ((log
& I3200_ECCERRLOG_RANK_BITS
) >>
140 I3200_ECCERRLOG_RANK_SHIFT
);
141 return rank
| (channel
* I3200_RANKS_PER_CHANNEL
);
148 struct i3200_dev_info
{
149 const char *ctl_name
;
152 struct i3200_error_info
{
155 u64 eccerrlog
[I3200_CHANNELS
];
158 static const struct i3200_dev_info i3200_devs
[] = {
164 static struct pci_dev
*mci_pdev
;
165 static int i3200_registered
= 1;
168 static void i3200_clear_error_info(struct mem_ctl_info
*mci
)
170 struct pci_dev
*pdev
;
172 pdev
= to_pci_dev(mci
->pdev
);
175 * Clear any error bits.
176 * (Yes, we really clear bits by writing 1 to them.)
178 pci_write_bits16(pdev
, I3200_ERRSTS
, I3200_ERRSTS_BITS
,
182 static void i3200_get_and_clear_error_info(struct mem_ctl_info
*mci
,
183 struct i3200_error_info
*info
)
185 struct pci_dev
*pdev
;
186 struct i3200_priv
*priv
= mci
->pvt_info
;
187 void __iomem
*window
= priv
->window
;
189 pdev
= to_pci_dev(mci
->pdev
);
192 * This is a mess because there is no atomic way to read all the
193 * registers at once and the registers can transition from CE being
196 pci_read_config_word(pdev
, I3200_ERRSTS
, &info
->errsts
);
197 if (!(info
->errsts
& I3200_ERRSTS_BITS
))
200 info
->eccerrlog
[0] = readq(window
+ I3200_C0ECCERRLOG
);
201 if (nr_channels
== 2)
202 info
->eccerrlog
[1] = readq(window
+ I3200_C1ECCERRLOG
);
204 pci_read_config_word(pdev
, I3200_ERRSTS
, &info
->errsts2
);
207 * If the error is the same for both reads then the first set
208 * of reads is valid. If there is a change then there is a CE
209 * with no info and the second set of reads is valid and
212 if ((info
->errsts
^ info
->errsts2
) & I3200_ERRSTS_BITS
) {
213 info
->eccerrlog
[0] = readq(window
+ I3200_C0ECCERRLOG
);
214 if (nr_channels
== 2)
215 info
->eccerrlog
[1] = readq(window
+ I3200_C1ECCERRLOG
);
218 i3200_clear_error_info(mci
);
221 static void i3200_process_error_info(struct mem_ctl_info
*mci
,
222 struct i3200_error_info
*info
)
227 if (!(info
->errsts
& I3200_ERRSTS_BITS
))
230 if ((info
->errsts
^ info
->errsts2
) & I3200_ERRSTS_BITS
) {
231 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, 0, 0, 0,
232 -1, -1, -1, "UE overwrote CE", "");
233 info
->errsts
= info
->errsts2
;
236 for (channel
= 0; channel
< nr_channels
; channel
++) {
237 log
= info
->eccerrlog
[channel
];
238 if (log
& I3200_ECCERRLOG_UE
) {
239 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1,
241 eccerrlog_row(channel
, log
),
244 } else if (log
& I3200_ECCERRLOG_CE
) {
245 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1,
246 0, 0, eccerrlog_syndrome(log
),
247 eccerrlog_row(channel
, log
),
254 static void i3200_check(struct mem_ctl_info
*mci
)
256 struct i3200_error_info info
;
258 edac_dbg(1, "MC%d\n", mci
->mc_idx
);
259 i3200_get_and_clear_error_info(mci
, &info
);
260 i3200_process_error_info(mci
, &info
);
263 static void __iomem
*i3200_map_mchbar(struct pci_dev
*pdev
)
272 void __iomem
*window
;
274 pci_read_config_dword(pdev
, I3200_MCHBAR_LOW
, &u
.mchbar_low
);
275 pci_read_config_dword(pdev
, I3200_MCHBAR_HIGH
, &u
.mchbar_high
);
276 u
.mchbar
&= I3200_MCHBAR_MASK
;
278 if (u
.mchbar
!= (resource_size_t
)u
.mchbar
) {
280 "i3200: mmio space beyond accessible range (0x%llx)\n",
281 (unsigned long long)u
.mchbar
);
285 window
= ioremap_nocache(u
.mchbar
, I3200_MMR_WINDOW_SIZE
);
287 printk(KERN_ERR
"i3200: cannot map mmio space at 0x%llx\n",
288 (unsigned long long)u
.mchbar
);
294 static void i3200_get_drbs(void __iomem
*window
,
295 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
])
299 for (i
= 0; i
< I3200_RANKS_PER_CHANNEL
; i
++) {
300 drbs
[0][i
] = readw(window
+ I3200_C0DRB
+ 2*i
) & I3200_DRB_MASK
;
301 drbs
[1][i
] = readw(window
+ I3200_C1DRB
+ 2*i
) & I3200_DRB_MASK
;
303 edac_dbg(0, "drb[0][%d] = %d, drb[1][%d] = %d\n", i
, drbs
[0][i
], i
, drbs
[1][i
]);
307 static bool i3200_is_stacked(struct pci_dev
*pdev
,
308 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
])
312 pci_read_config_word(pdev
, I3200_TOM
, &tom
);
313 tom
&= I3200_TOM_MASK
;
315 return drbs
[I3200_CHANNELS
- 1][I3200_RANKS_PER_CHANNEL
- 1] == tom
;
318 static unsigned long drb_to_nr_pages(
319 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
], bool stacked
,
320 int channel
, int rank
)
324 n
= drbs
[channel
][rank
];
329 n
-= drbs
[channel
][rank
- 1];
330 if (stacked
&& (channel
== 1) &&
331 drbs
[channel
][rank
] == drbs
[channel
][I3200_RANKS_PER_CHANNEL
- 1])
332 n
-= drbs
[0][I3200_RANKS_PER_CHANNEL
- 1];
334 n
<<= (I3200_DRB_SHIFT
- PAGE_SHIFT
);
338 static int i3200_probe1(struct pci_dev
*pdev
, int dev_idx
)
342 struct mem_ctl_info
*mci
= NULL
;
343 struct edac_mc_layer layers
[2];
344 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
];
346 void __iomem
*window
;
347 struct i3200_priv
*priv
;
349 edac_dbg(0, "MC:\n");
351 window
= i3200_map_mchbar(pdev
);
355 i3200_get_drbs(window
, drbs
);
356 nr_channels
= how_many_channels(pdev
);
358 layers
[0].type
= EDAC_MC_LAYER_CHIP_SELECT
;
359 layers
[0].size
= I3200_DIMMS
;
360 layers
[0].is_virt_csrow
= true;
361 layers
[1].type
= EDAC_MC_LAYER_CHANNEL
;
362 layers
[1].size
= nr_channels
;
363 layers
[1].is_virt_csrow
= false;
364 mci
= edac_mc_alloc(0, ARRAY_SIZE(layers
), layers
,
365 sizeof(struct i3200_priv
));
369 edac_dbg(3, "MC: init mci\n");
371 mci
->pdev
= &pdev
->dev
;
372 mci
->mtype_cap
= MEM_FLAG_DDR2
;
374 mci
->edac_ctl_cap
= EDAC_FLAG_SECDED
;
375 mci
->edac_cap
= EDAC_FLAG_SECDED
;
377 mci
->mod_name
= EDAC_MOD_STR
;
378 mci
->mod_ver
= I3200_REVISION
;
379 mci
->ctl_name
= i3200_devs
[dev_idx
].ctl_name
;
380 mci
->dev_name
= pci_name(pdev
);
381 mci
->edac_check
= i3200_check
;
382 mci
->ctl_page_to_phys
= NULL
;
383 priv
= mci
->pvt_info
;
384 priv
->window
= window
;
386 stacked
= i3200_is_stacked(pdev
, drbs
);
389 * The dram rank boundary (DRB) reg values are boundary addresses
390 * for each DRAM rank with a granularity of 64MB. DRB regs are
391 * cumulative; the last one will contain the total memory
392 * contained in all ranks.
394 for (i
= 0; i
< I3200_DIMMS
; i
++) {
395 unsigned long nr_pages
;
397 for (j
= 0; j
< nr_channels
; j
++) {
398 struct dimm_info
*dimm
= EDAC_DIMM_PTR(mci
->layers
, mci
->dimms
,
399 mci
->n_layers
, i
, j
, 0);
401 nr_pages
= drb_to_nr_pages(drbs
, stacked
, j
, i
);
405 edac_dbg(0, "csrow %d, channel %d%s, size = %ld Mb\n", i
, j
,
406 stacked
? " (stacked)" : "", PAGES_TO_MiB(nr_pages
));
408 dimm
->nr_pages
= nr_pages
;
409 dimm
->grain
= nr_pages
<< PAGE_SHIFT
;
410 dimm
->mtype
= MEM_DDR2
;
411 dimm
->dtype
= DEV_UNKNOWN
;
412 dimm
->edac_mode
= EDAC_UNKNOWN
;
416 i3200_clear_error_info(mci
);
419 if (edac_mc_add_mc(mci
)) {
420 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
424 /* get this far and it's successful */
425 edac_dbg(3, "MC: success\n");
436 static int i3200_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
440 edac_dbg(0, "MC:\n");
442 if (pci_enable_device(pdev
) < 0)
445 rc
= i3200_probe1(pdev
, ent
->driver_data
);
447 mci_pdev
= pci_dev_get(pdev
);
452 static void i3200_remove_one(struct pci_dev
*pdev
)
454 struct mem_ctl_info
*mci
;
455 struct i3200_priv
*priv
;
459 mci
= edac_mc_del_mc(&pdev
->dev
);
463 priv
= mci
->pvt_info
;
464 iounmap(priv
->window
);
469 static const struct pci_device_id i3200_pci_tbl
[] = {
471 PCI_VEND_DEV(INTEL
, 3200_HB
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
475 } /* 0 terminated list. */
478 MODULE_DEVICE_TABLE(pci
, i3200_pci_tbl
);
480 static struct pci_driver i3200_driver
= {
481 .name
= EDAC_MOD_STR
,
482 .probe
= i3200_init_one
,
483 .remove
= i3200_remove_one
,
484 .id_table
= i3200_pci_tbl
,
487 static int __init
i3200_init(void)
491 edac_dbg(3, "MC:\n");
493 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
496 pci_rc
= pci_register_driver(&i3200_driver
);
501 i3200_registered
= 0;
502 mci_pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
503 PCI_DEVICE_ID_INTEL_3200_HB
, NULL
);
505 edac_dbg(0, "i3200 pci_get_device fail\n");
510 pci_rc
= i3200_init_one(mci_pdev
, i3200_pci_tbl
);
512 edac_dbg(0, "i3200 init fail\n");
521 pci_unregister_driver(&i3200_driver
);
525 pci_dev_put(mci_pdev
);
530 static void __exit
i3200_exit(void)
532 edac_dbg(3, "MC:\n");
534 pci_unregister_driver(&i3200_driver
);
535 if (!i3200_registered
) {
536 i3200_remove_one(mci_pdev
);
537 pci_dev_put(mci_pdev
);
541 module_init(i3200_init
);
542 module_exit(i3200_exit
);
544 MODULE_LICENSE("GPL");
545 MODULE_AUTHOR("Akamai Technologies, Inc.");
546 MODULE_DESCRIPTION("MC support for Intel 3200 memory hub controllers");
548 module_param(edac_op_state
, int, 0444);
549 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");