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_module.h"
18 #include <linux/io-64-nonatomic-lo-hi.h>
20 #define EDAC_MOD_STR "i3200_edac"
22 #define PCI_DEVICE_ID_INTEL_3200_HB 0x29f0
26 #define I3200_RANKS_PER_CHANNEL 4
27 #define I3200_CHANNELS 2
29 /* Intel 3200 register addresses - device 0 function 0 - DRAM Controller */
31 #define I3200_MCHBAR_LOW 0x48 /* MCH Memory Mapped Register BAR */
32 #define I3200_MCHBAR_HIGH 0x4c
33 #define I3200_MCHBAR_MASK 0xfffffc000ULL /* bits 35:14 */
34 #define I3200_MMR_WINDOW_SIZE 16384
36 #define I3200_TOM 0xa0 /* Top of Memory (16b)
39 * 9:0 total populated physical memory
41 #define I3200_TOM_MASK 0x3ff /* bits 9:0 */
42 #define I3200_TOM_SHIFT 26 /* 64MiB grain */
44 #define I3200_ERRSTS 0xc8 /* Error Status Register (16b)
47 * 14 Isochronous TBWRR Run Behind FIFO Full
49 * 13 Isochronous TBWRR Run Behind FIFO Put
52 * 11 MCH Thermal Sensor Event
53 * for SMI/SCI/SERR (GTSE)
55 * 9 LOCK to non-DRAM Memory Flag (LCKF)
57 * 7 DRAM Throttle Flag (DTF)
59 * 1 Multi-bit DRAM ECC Error Flag (DMERR)
60 * 0 Single-bit DRAM ECC Error Flag (DSERR)
62 #define I3200_ERRSTS_UE 0x0002
63 #define I3200_ERRSTS_CE 0x0001
64 #define I3200_ERRSTS_BITS (I3200_ERRSTS_UE | I3200_ERRSTS_CE)
67 /* Intel MMIO register space - device 0 function 0 - MMR space */
69 #define I3200_C0DRB 0x200 /* Channel 0 DRAM Rank Boundary (16b x 4)
72 * 9:0 Channel 0 DRAM Rank Boundary Address
74 #define I3200_C1DRB 0x600 /* Channel 1 DRAM Rank Boundary (16b x 4) */
75 #define I3200_DRB_MASK 0x3ff /* bits 9:0 */
76 #define I3200_DRB_SHIFT 26 /* 64MiB grain */
78 #define I3200_C0ECCERRLOG 0x280 /* Channel 0 ECC Error Log (64b)
80 * 63:48 Error Column Address (ERRCOL)
81 * 47:32 Error Row Address (ERRROW)
82 * 31:29 Error Bank Address (ERRBANK)
83 * 28:27 Error Rank Address (ERRRANK)
85 * 23:16 Error Syndrome (ERRSYND)
87 * 1 Multiple Bit Error Status (MERRSTS)
88 * 0 Correctable Error Status (CERRSTS)
90 #define I3200_C1ECCERRLOG 0x680 /* Chan 1 ECC Error Log (64b) */
91 #define I3200_ECCERRLOG_CE 0x1
92 #define I3200_ECCERRLOG_UE 0x2
93 #define I3200_ECCERRLOG_RANK_BITS 0x18000000
94 #define I3200_ECCERRLOG_RANK_SHIFT 27
95 #define I3200_ECCERRLOG_SYNDROME_BITS 0xff0000
96 #define I3200_ECCERRLOG_SYNDROME_SHIFT 16
97 #define I3200_CAPID0 0xe0 /* P.95 of spec for details */
100 void __iomem
*window
;
103 static int nr_channels
;
105 static int how_many_channels(struct pci_dev
*pdev
)
109 unsigned char capid0_8b
; /* 8th byte of CAPID0 */
111 pci_read_config_byte(pdev
, I3200_CAPID0
+ 8, &capid0_8b
);
113 if (capid0_8b
& 0x20) { /* check DCD: Dual Channel Disable */
114 edac_dbg(0, "In single channel mode\n");
117 edac_dbg(0, "In dual channel mode\n");
121 if (capid0_8b
& 0x10) /* check if both channels are filled */
122 edac_dbg(0, "2 DIMMS per channel disabled\n");
124 edac_dbg(0, "2 DIMMS per channel enabled\n");
129 static unsigned long eccerrlog_syndrome(u64 log
)
131 return (log
& I3200_ECCERRLOG_SYNDROME_BITS
) >>
132 I3200_ECCERRLOG_SYNDROME_SHIFT
;
135 static int eccerrlog_row(int channel
, u64 log
)
137 u64 rank
= ((log
& I3200_ECCERRLOG_RANK_BITS
) >>
138 I3200_ECCERRLOG_RANK_SHIFT
);
139 return rank
| (channel
* I3200_RANKS_PER_CHANNEL
);
146 struct i3200_dev_info
{
147 const char *ctl_name
;
150 struct i3200_error_info
{
153 u64 eccerrlog
[I3200_CHANNELS
];
156 static const struct i3200_dev_info i3200_devs
[] = {
162 static struct pci_dev
*mci_pdev
;
163 static int i3200_registered
= 1;
166 static void i3200_clear_error_info(struct mem_ctl_info
*mci
)
168 struct pci_dev
*pdev
;
170 pdev
= to_pci_dev(mci
->pdev
);
173 * Clear any error bits.
174 * (Yes, we really clear bits by writing 1 to them.)
176 pci_write_bits16(pdev
, I3200_ERRSTS
, I3200_ERRSTS_BITS
,
180 static void i3200_get_and_clear_error_info(struct mem_ctl_info
*mci
,
181 struct i3200_error_info
*info
)
183 struct pci_dev
*pdev
;
184 struct i3200_priv
*priv
= mci
->pvt_info
;
185 void __iomem
*window
= priv
->window
;
187 pdev
= to_pci_dev(mci
->pdev
);
190 * This is a mess because there is no atomic way to read all the
191 * registers at once and the registers can transition from CE being
194 pci_read_config_word(pdev
, I3200_ERRSTS
, &info
->errsts
);
195 if (!(info
->errsts
& I3200_ERRSTS_BITS
))
198 info
->eccerrlog
[0] = readq(window
+ I3200_C0ECCERRLOG
);
199 if (nr_channels
== 2)
200 info
->eccerrlog
[1] = readq(window
+ I3200_C1ECCERRLOG
);
202 pci_read_config_word(pdev
, I3200_ERRSTS
, &info
->errsts2
);
205 * If the error is the same for both reads then the first set
206 * of reads is valid. If there is a change then there is a CE
207 * with no info and the second set of reads is valid and
210 if ((info
->errsts
^ info
->errsts2
) & I3200_ERRSTS_BITS
) {
211 info
->eccerrlog
[0] = readq(window
+ I3200_C0ECCERRLOG
);
212 if (nr_channels
== 2)
213 info
->eccerrlog
[1] = readq(window
+ I3200_C1ECCERRLOG
);
216 i3200_clear_error_info(mci
);
219 static void i3200_process_error_info(struct mem_ctl_info
*mci
,
220 struct i3200_error_info
*info
)
225 if (!(info
->errsts
& I3200_ERRSTS_BITS
))
228 if ((info
->errsts
^ info
->errsts2
) & I3200_ERRSTS_BITS
) {
229 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1, 0, 0, 0,
230 -1, -1, -1, "UE overwrote CE", "");
231 info
->errsts
= info
->errsts2
;
234 for (channel
= 0; channel
< nr_channels
; channel
++) {
235 log
= info
->eccerrlog
[channel
];
236 if (log
& I3200_ECCERRLOG_UE
) {
237 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED
, mci
, 1,
239 eccerrlog_row(channel
, log
),
242 } else if (log
& I3200_ECCERRLOG_CE
) {
243 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED
, mci
, 1,
244 0, 0, eccerrlog_syndrome(log
),
245 eccerrlog_row(channel
, log
),
252 static void i3200_check(struct mem_ctl_info
*mci
)
254 struct i3200_error_info info
;
256 edac_dbg(1, "MC%d\n", mci
->mc_idx
);
257 i3200_get_and_clear_error_info(mci
, &info
);
258 i3200_process_error_info(mci
, &info
);
261 static void __iomem
*i3200_map_mchbar(struct pci_dev
*pdev
)
270 void __iomem
*window
;
272 pci_read_config_dword(pdev
, I3200_MCHBAR_LOW
, &u
.mchbar_low
);
273 pci_read_config_dword(pdev
, I3200_MCHBAR_HIGH
, &u
.mchbar_high
);
274 u
.mchbar
&= I3200_MCHBAR_MASK
;
276 if (u
.mchbar
!= (resource_size_t
)u
.mchbar
) {
278 "i3200: mmio space beyond accessible range (0x%llx)\n",
279 (unsigned long long)u
.mchbar
);
283 window
= ioremap_nocache(u
.mchbar
, I3200_MMR_WINDOW_SIZE
);
285 printk(KERN_ERR
"i3200: cannot map mmio space at 0x%llx\n",
286 (unsigned long long)u
.mchbar
);
292 static void i3200_get_drbs(void __iomem
*window
,
293 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
])
297 for (i
= 0; i
< I3200_RANKS_PER_CHANNEL
; i
++) {
298 drbs
[0][i
] = readw(window
+ I3200_C0DRB
+ 2*i
) & I3200_DRB_MASK
;
299 drbs
[1][i
] = readw(window
+ I3200_C1DRB
+ 2*i
) & I3200_DRB_MASK
;
301 edac_dbg(0, "drb[0][%d] = %d, drb[1][%d] = %d\n", i
, drbs
[0][i
], i
, drbs
[1][i
]);
305 static bool i3200_is_stacked(struct pci_dev
*pdev
,
306 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
])
310 pci_read_config_word(pdev
, I3200_TOM
, &tom
);
311 tom
&= I3200_TOM_MASK
;
313 return drbs
[I3200_CHANNELS
- 1][I3200_RANKS_PER_CHANNEL
- 1] == tom
;
316 static unsigned long drb_to_nr_pages(
317 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
], bool stacked
,
318 int channel
, int rank
)
322 n
= drbs
[channel
][rank
];
327 n
-= drbs
[channel
][rank
- 1];
328 if (stacked
&& (channel
== 1) &&
329 drbs
[channel
][rank
] == drbs
[channel
][I3200_RANKS_PER_CHANNEL
- 1])
330 n
-= drbs
[0][I3200_RANKS_PER_CHANNEL
- 1];
332 n
<<= (I3200_DRB_SHIFT
- PAGE_SHIFT
);
336 static int i3200_probe1(struct pci_dev
*pdev
, int dev_idx
)
340 struct mem_ctl_info
*mci
= NULL
;
341 struct edac_mc_layer layers
[2];
342 u16 drbs
[I3200_CHANNELS
][I3200_RANKS_PER_CHANNEL
];
344 void __iomem
*window
;
345 struct i3200_priv
*priv
;
347 edac_dbg(0, "MC:\n");
349 window
= i3200_map_mchbar(pdev
);
353 i3200_get_drbs(window
, drbs
);
354 nr_channels
= how_many_channels(pdev
);
356 layers
[0].type
= EDAC_MC_LAYER_CHIP_SELECT
;
357 layers
[0].size
= I3200_DIMMS
;
358 layers
[0].is_virt_csrow
= true;
359 layers
[1].type
= EDAC_MC_LAYER_CHANNEL
;
360 layers
[1].size
= nr_channels
;
361 layers
[1].is_virt_csrow
= false;
362 mci
= edac_mc_alloc(0, ARRAY_SIZE(layers
), layers
,
363 sizeof(struct i3200_priv
));
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
= i3200_devs
[dev_idx
].ctl_name
;
377 mci
->dev_name
= pci_name(pdev
);
378 mci
->edac_check
= i3200_check
;
379 mci
->ctl_page_to_phys
= NULL
;
380 priv
= mci
->pvt_info
;
381 priv
->window
= window
;
383 stacked
= i3200_is_stacked(pdev
, drbs
);
386 * The dram rank boundary (DRB) reg values are boundary addresses
387 * for each DRAM rank with a granularity of 64MB. DRB regs are
388 * cumulative; the last one will contain the total memory
389 * contained in all ranks.
391 for (i
= 0; i
< I3200_DIMMS
; i
++) {
392 unsigned long nr_pages
;
394 for (j
= 0; j
< nr_channels
; j
++) {
395 struct dimm_info
*dimm
= EDAC_DIMM_PTR(mci
->layers
, mci
->dimms
,
396 mci
->n_layers
, i
, j
, 0);
398 nr_pages
= drb_to_nr_pages(drbs
, stacked
, j
, i
);
402 edac_dbg(0, "csrow %d, channel %d%s, size = %ld Mb\n", i
, j
,
403 stacked
? " (stacked)" : "", PAGES_TO_MiB(nr_pages
));
405 dimm
->nr_pages
= nr_pages
;
406 dimm
->grain
= nr_pages
<< PAGE_SHIFT
;
407 dimm
->mtype
= MEM_DDR2
;
408 dimm
->dtype
= DEV_UNKNOWN
;
409 dimm
->edac_mode
= EDAC_UNKNOWN
;
413 i3200_clear_error_info(mci
);
416 if (edac_mc_add_mc(mci
)) {
417 edac_dbg(3, "MC: failed edac_mc_add_mc()\n");
421 /* get this far and it's successful */
422 edac_dbg(3, "MC: success\n");
433 static int i3200_init_one(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
437 edac_dbg(0, "MC:\n");
439 if (pci_enable_device(pdev
) < 0)
442 rc
= i3200_probe1(pdev
, ent
->driver_data
);
444 mci_pdev
= pci_dev_get(pdev
);
449 static void i3200_remove_one(struct pci_dev
*pdev
)
451 struct mem_ctl_info
*mci
;
452 struct i3200_priv
*priv
;
456 mci
= edac_mc_del_mc(&pdev
->dev
);
460 priv
= mci
->pvt_info
;
461 iounmap(priv
->window
);
465 pci_disable_device(pdev
);
468 static const struct pci_device_id i3200_pci_tbl
[] = {
470 PCI_VEND_DEV(INTEL
, 3200_HB
), PCI_ANY_ID
, PCI_ANY_ID
, 0, 0,
474 } /* 0 terminated list. */
477 MODULE_DEVICE_TABLE(pci
, i3200_pci_tbl
);
479 static struct pci_driver i3200_driver
= {
480 .name
= EDAC_MOD_STR
,
481 .probe
= i3200_init_one
,
482 .remove
= i3200_remove_one
,
483 .id_table
= i3200_pci_tbl
,
486 static int __init
i3200_init(void)
490 edac_dbg(3, "MC:\n");
492 /* Ensure that the OPSTATE is set correctly for POLL or NMI */
495 pci_rc
= pci_register_driver(&i3200_driver
);
500 i3200_registered
= 0;
501 mci_pdev
= pci_get_device(PCI_VENDOR_ID_INTEL
,
502 PCI_DEVICE_ID_INTEL_3200_HB
, NULL
);
504 edac_dbg(0, "i3200 pci_get_device fail\n");
509 pci_rc
= i3200_init_one(mci_pdev
, i3200_pci_tbl
);
511 edac_dbg(0, "i3200 init fail\n");
520 pci_unregister_driver(&i3200_driver
);
523 pci_dev_put(mci_pdev
);
528 static void __exit
i3200_exit(void)
530 edac_dbg(3, "MC:\n");
532 pci_unregister_driver(&i3200_driver
);
533 if (!i3200_registered
) {
534 i3200_remove_one(mci_pdev
);
535 pci_dev_put(mci_pdev
);
539 module_init(i3200_init
);
540 module_exit(i3200_exit
);
542 MODULE_LICENSE("GPL");
543 MODULE_AUTHOR("Akamai Technologies, Inc.");
544 MODULE_DESCRIPTION("MC support for Intel 3200 memory hub controllers");
546 module_param(edac_op_state
, int, 0444);
547 MODULE_PARM_DESC(edac_op_state
, "EDAC Error Reporting state: 0=Poll,1=NMI");