Version:  2.0.40 2.2.26 2.4.37 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2

Linux/drivers/ata/sata_mv.c

  1 /*
  2  * sata_mv.c - Marvell SATA support
  3  *
  4  * Copyright 2008-2009: Marvell Corporation, all rights reserved.
  5  * Copyright 2005: EMC Corporation, all rights reserved.
  6  * Copyright 2005 Red Hat, Inc.  All rights reserved.
  7  *
  8  * Originally written by Brett Russ.
  9  * Extensive overhaul and enhancement by Mark Lord <mlord@pobox.com>.
 10  *
 11  * Please ALWAYS copy linux-ide@vger.kernel.org on emails.
 12  *
 13  * This program is free software; you can redistribute it and/or modify
 14  * it under the terms of the GNU General Public License as published by
 15  * the Free Software Foundation; version 2 of the License.
 16  *
 17  * This program is distributed in the hope that it will be useful,
 18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 20  * GNU General Public License for more details.
 21  *
 22  * You should have received a copy of the GNU General Public License
 23  * along with this program; if not, write to the Free Software
 24  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 25  *
 26  */
 27 
 28 /*
 29  * sata_mv TODO list:
 30  *
 31  * --> Develop a low-power-consumption strategy, and implement it.
 32  *
 33  * --> Add sysfs attributes for per-chip / per-HC IRQ coalescing thresholds.
 34  *
 35  * --> [Experiment, Marvell value added] Is it possible to use target
 36  *       mode to cross-connect two Linux boxes with Marvell cards?  If so,
 37  *       creating LibATA target mode support would be very interesting.
 38  *
 39  *       Target mode, for those without docs, is the ability to directly
 40  *       connect two SATA ports.
 41  */
 42 
 43 /*
 44  * 80x1-B2 errata PCI#11:
 45  *
 46  * Users of the 6041/6081 Rev.B2 chips (current is C0)
 47  * should be careful to insert those cards only onto PCI-X bus #0,
 48  * and only in device slots 0..7, not higher.  The chips may not
 49  * work correctly otherwise  (note: this is a pretty rare condition).
 50  */
 51 
 52 #include <linux/kernel.h>
 53 #include <linux/module.h>
 54 #include <linux/pci.h>
 55 #include <linux/init.h>
 56 #include <linux/blkdev.h>
 57 #include <linux/delay.h>
 58 #include <linux/interrupt.h>
 59 #include <linux/dmapool.h>
 60 #include <linux/dma-mapping.h>
 61 #include <linux/device.h>
 62 #include <linux/clk.h>
 63 #include <linux/phy/phy.h>
 64 #include <linux/platform_device.h>
 65 #include <linux/ata_platform.h>
 66 #include <linux/mbus.h>
 67 #include <linux/bitops.h>
 68 #include <linux/gfp.h>
 69 #include <linux/of.h>
 70 #include <linux/of_irq.h>
 71 #include <scsi/scsi_host.h>
 72 #include <scsi/scsi_cmnd.h>
 73 #include <scsi/scsi_device.h>
 74 #include <linux/libata.h>
 75 
 76 #define DRV_NAME        "sata_mv"
 77 #define DRV_VERSION     "1.28"
 78 
 79 /*
 80  * module options
 81  */
 82 
 83 #ifdef CONFIG_PCI
 84 static int msi;
 85 module_param(msi, int, S_IRUGO);
 86 MODULE_PARM_DESC(msi, "Enable use of PCI MSI (0=off, 1=on)");
 87 #endif
 88 
 89 static int irq_coalescing_io_count;
 90 module_param(irq_coalescing_io_count, int, S_IRUGO);
 91 MODULE_PARM_DESC(irq_coalescing_io_count,
 92                  "IRQ coalescing I/O count threshold (0..255)");
 93 
 94 static int irq_coalescing_usecs;
 95 module_param(irq_coalescing_usecs, int, S_IRUGO);
 96 MODULE_PARM_DESC(irq_coalescing_usecs,
 97                  "IRQ coalescing time threshold in usecs");
 98 
 99 enum {
100         /* BAR's are enumerated in terms of pci_resource_start() terms */
101         MV_PRIMARY_BAR          = 0,    /* offset 0x10: memory space */
102         MV_IO_BAR               = 2,    /* offset 0x18: IO space */
103         MV_MISC_BAR             = 3,    /* offset 0x1c: FLASH, NVRAM, SRAM */
104 
105         MV_MAJOR_REG_AREA_SZ    = 0x10000,      /* 64KB */
106         MV_MINOR_REG_AREA_SZ    = 0x2000,       /* 8KB */
107 
108         /* For use with both IRQ coalescing methods ("all ports" or "per-HC" */
109         COAL_CLOCKS_PER_USEC    = 150,          /* for calculating COAL_TIMEs */
110         MAX_COAL_TIME_THRESHOLD = ((1 << 24) - 1), /* internal clocks count */
111         MAX_COAL_IO_COUNT       = 255,          /* completed I/O count */
112 
113         MV_PCI_REG_BASE         = 0,
114 
115         /*
116          * Per-chip ("all ports") interrupt coalescing feature.
117          * This is only for GEN_II / GEN_IIE hardware.
118          *
119          * Coalescing defers the interrupt until either the IO_THRESHOLD
120          * (count of completed I/Os) is met, or the TIME_THRESHOLD is met.
121          */
122         COAL_REG_BASE           = 0x18000,
123         IRQ_COAL_CAUSE          = (COAL_REG_BASE + 0x08),
124         ALL_PORTS_COAL_IRQ      = (1 << 4),     /* all ports irq event */
125 
126         IRQ_COAL_IO_THRESHOLD   = (COAL_REG_BASE + 0xcc),
127         IRQ_COAL_TIME_THRESHOLD = (COAL_REG_BASE + 0xd0),
128 
129         /*
130          * Registers for the (unused here) transaction coalescing feature:
131          */
132         TRAN_COAL_CAUSE_LO      = (COAL_REG_BASE + 0x88),
133         TRAN_COAL_CAUSE_HI      = (COAL_REG_BASE + 0x8c),
134 
135         SATAHC0_REG_BASE        = 0x20000,
136         FLASH_CTL               = 0x1046c,
137         GPIO_PORT_CTL           = 0x104f0,
138         RESET_CFG               = 0x180d8,
139 
140         MV_PCI_REG_SZ           = MV_MAJOR_REG_AREA_SZ,
141         MV_SATAHC_REG_SZ        = MV_MAJOR_REG_AREA_SZ,
142         MV_SATAHC_ARBTR_REG_SZ  = MV_MINOR_REG_AREA_SZ,         /* arbiter */
143         MV_PORT_REG_SZ          = MV_MINOR_REG_AREA_SZ,
144 
145         MV_MAX_Q_DEPTH          = 32,
146         MV_MAX_Q_DEPTH_MASK     = MV_MAX_Q_DEPTH - 1,
147 
148         /* CRQB needs alignment on a 1KB boundary. Size == 1KB
149          * CRPB needs alignment on a 256B boundary. Size == 256B
150          * ePRD (SG) entries need alignment on a 16B boundary. Size == 16B
151          */
152         MV_CRQB_Q_SZ            = (32 * MV_MAX_Q_DEPTH),
153         MV_CRPB_Q_SZ            = (8 * MV_MAX_Q_DEPTH),
154         MV_MAX_SG_CT            = 256,
155         MV_SG_TBL_SZ            = (16 * MV_MAX_SG_CT),
156 
157         /* Determine hc from 0-7 port: hc = port >> MV_PORT_HC_SHIFT */
158         MV_PORT_HC_SHIFT        = 2,
159         MV_PORTS_PER_HC         = (1 << MV_PORT_HC_SHIFT), /* 4 */
160         /* Determine hc port from 0-7 port: hardport = port & MV_PORT_MASK */
161         MV_PORT_MASK            = (MV_PORTS_PER_HC - 1),   /* 3 */
162 
163         /* Host Flags */
164         MV_FLAG_DUAL_HC         = (1 << 30),  /* two SATA Host Controllers */
165 
166         MV_COMMON_FLAGS         = ATA_FLAG_SATA | ATA_FLAG_PIO_POLLING,
167 
168         MV_GEN_I_FLAGS          = MV_COMMON_FLAGS | ATA_FLAG_NO_ATAPI,
169 
170         MV_GEN_II_FLAGS         = MV_COMMON_FLAGS | ATA_FLAG_NCQ |
171                                   ATA_FLAG_PMP | ATA_FLAG_ACPI_SATA,
172 
173         MV_GEN_IIE_FLAGS        = MV_GEN_II_FLAGS | ATA_FLAG_AN,
174 
175         CRQB_FLAG_READ          = (1 << 0),
176         CRQB_TAG_SHIFT          = 1,
177         CRQB_IOID_SHIFT         = 6,    /* CRQB Gen-II/IIE IO Id shift */
178         CRQB_PMP_SHIFT          = 12,   /* CRQB Gen-II/IIE PMP shift */
179         CRQB_HOSTQ_SHIFT        = 17,   /* CRQB Gen-II/IIE HostQueTag shift */
180         CRQB_CMD_ADDR_SHIFT     = 8,
181         CRQB_CMD_CS             = (0x2 << 11),
182         CRQB_CMD_LAST           = (1 << 15),
183 
184         CRPB_FLAG_STATUS_SHIFT  = 8,
185         CRPB_IOID_SHIFT_6       = 5,    /* CRPB Gen-II IO Id shift */
186         CRPB_IOID_SHIFT_7       = 7,    /* CRPB Gen-IIE IO Id shift */
187 
188         EPRD_FLAG_END_OF_TBL    = (1 << 31),
189 
190         /* PCI interface registers */
191 
192         MV_PCI_COMMAND          = 0xc00,
193         MV_PCI_COMMAND_MWRCOM   = (1 << 4),     /* PCI Master Write Combining */
194         MV_PCI_COMMAND_MRDTRIG  = (1 << 7),     /* PCI Master Read Trigger */
195 
196         PCI_MAIN_CMD_STS        = 0xd30,
197         STOP_PCI_MASTER         = (1 << 2),
198         PCI_MASTER_EMPTY        = (1 << 3),
199         GLOB_SFT_RST            = (1 << 4),
200 
201         MV_PCI_MODE             = 0xd00,
202         MV_PCI_MODE_MASK        = 0x30,
203 
204         MV_PCI_EXP_ROM_BAR_CTL  = 0xd2c,
205         MV_PCI_DISC_TIMER       = 0xd04,
206         MV_PCI_MSI_TRIGGER      = 0xc38,
207         MV_PCI_SERR_MASK        = 0xc28,
208         MV_PCI_XBAR_TMOUT       = 0x1d04,
209         MV_PCI_ERR_LOW_ADDRESS  = 0x1d40,
210         MV_PCI_ERR_HIGH_ADDRESS = 0x1d44,
211         MV_PCI_ERR_ATTRIBUTE    = 0x1d48,
212         MV_PCI_ERR_COMMAND      = 0x1d50,
213 
214         PCI_IRQ_CAUSE           = 0x1d58,
215         PCI_IRQ_MASK            = 0x1d5c,
216         PCI_UNMASK_ALL_IRQS     = 0x7fffff,     /* bits 22-0 */
217 
218         PCIE_IRQ_CAUSE          = 0x1900,
219         PCIE_IRQ_MASK           = 0x1910,
220         PCIE_UNMASK_ALL_IRQS    = 0x40a,        /* assorted bits */
221 
222         /* Host Controller Main Interrupt Cause/Mask registers (1 per-chip) */
223         PCI_HC_MAIN_IRQ_CAUSE   = 0x1d60,
224         PCI_HC_MAIN_IRQ_MASK    = 0x1d64,
225         SOC_HC_MAIN_IRQ_CAUSE   = 0x20020,
226         SOC_HC_MAIN_IRQ_MASK    = 0x20024,
227         ERR_IRQ                 = (1 << 0),     /* shift by (2 * port #) */
228         DONE_IRQ                = (1 << 1),     /* shift by (2 * port #) */
229         HC0_IRQ_PEND            = 0x1ff,        /* bits 0-8 = HC0's ports */
230         HC_SHIFT                = 9,            /* bits 9-17 = HC1's ports */
231         DONE_IRQ_0_3            = 0x000000aa,   /* DONE_IRQ ports 0,1,2,3 */
232         DONE_IRQ_4_7            = (DONE_IRQ_0_3 << HC_SHIFT),  /* 4,5,6,7 */
233         PCI_ERR                 = (1 << 18),
234         TRAN_COAL_LO_DONE       = (1 << 19),    /* transaction coalescing */
235         TRAN_COAL_HI_DONE       = (1 << 20),    /* transaction coalescing */
236         PORTS_0_3_COAL_DONE     = (1 << 8),     /* HC0 IRQ coalescing */
237         PORTS_4_7_COAL_DONE     = (1 << 17),    /* HC1 IRQ coalescing */
238         ALL_PORTS_COAL_DONE     = (1 << 21),    /* GEN_II(E) IRQ coalescing */
239         GPIO_INT                = (1 << 22),
240         SELF_INT                = (1 << 23),
241         TWSI_INT                = (1 << 24),
242         HC_MAIN_RSVD            = (0x7f << 25), /* bits 31-25 */
243         HC_MAIN_RSVD_5          = (0x1fff << 19), /* bits 31-19 */
244         HC_MAIN_RSVD_SOC        = (0x3fffffb << 6),     /* bits 31-9, 7-6 */
245 
246         /* SATAHC registers */
247         HC_CFG                  = 0x00,
248 
249         HC_IRQ_CAUSE            = 0x14,
250         DMA_IRQ                 = (1 << 0),     /* shift by port # */
251         HC_COAL_IRQ             = (1 << 4),     /* IRQ coalescing */
252         DEV_IRQ                 = (1 << 8),     /* shift by port # */
253 
254         /*
255          * Per-HC (Host-Controller) interrupt coalescing feature.
256          * This is present on all chip generations.
257          *
258          * Coalescing defers the interrupt until either the IO_THRESHOLD
259          * (count of completed I/Os) is met, or the TIME_THRESHOLD is met.
260          */
261         HC_IRQ_COAL_IO_THRESHOLD        = 0x000c,
262         HC_IRQ_COAL_TIME_THRESHOLD      = 0x0010,
263 
264         SOC_LED_CTRL            = 0x2c,
265         SOC_LED_CTRL_BLINK      = (1 << 0),     /* Active LED blink */
266         SOC_LED_CTRL_ACT_PRESENCE = (1 << 2),   /* Multiplex dev presence */
267                                                 /*  with dev activity LED */
268 
269         /* Shadow block registers */
270         SHD_BLK                 = 0x100,
271         SHD_CTL_AST             = 0x20,         /* ofs from SHD_BLK */
272 
273         /* SATA registers */
274         SATA_STATUS             = 0x300,  /* ctrl, err regs follow status */
275         SATA_ACTIVE             = 0x350,
276         FIS_IRQ_CAUSE           = 0x364,
277         FIS_IRQ_CAUSE_AN        = (1 << 9),     /* async notification */
278 
279         LTMODE                  = 0x30c,        /* requires read-after-write */
280         LTMODE_BIT8             = (1 << 8),     /* unknown, but necessary */
281 
282         PHY_MODE2               = 0x330,
283         PHY_MODE3               = 0x310,
284 
285         PHY_MODE4               = 0x314,        /* requires read-after-write */
286         PHY_MODE4_CFG_MASK      = 0x00000003,   /* phy internal config field */
287         PHY_MODE4_CFG_VALUE     = 0x00000001,   /* phy internal config field */
288         PHY_MODE4_RSVD_ZEROS    = 0x5de3fffa,   /* Gen2e always write zeros */
289         PHY_MODE4_RSVD_ONES     = 0x00000005,   /* Gen2e always write ones */
290 
291         SATA_IFCTL              = 0x344,
292         SATA_TESTCTL            = 0x348,
293         SATA_IFSTAT             = 0x34c,
294         VENDOR_UNIQUE_FIS       = 0x35c,
295 
296         FISCFG                  = 0x360,
297         FISCFG_WAIT_DEV_ERR     = (1 << 8),     /* wait for host on DevErr */
298         FISCFG_SINGLE_SYNC      = (1 << 16),    /* SYNC on DMA activation */
299 
300         PHY_MODE9_GEN2          = 0x398,
301         PHY_MODE9_GEN1          = 0x39c,
302         PHYCFG_OFS              = 0x3a0,        /* only in 65n devices */
303 
304         MV5_PHY_MODE            = 0x74,
305         MV5_LTMODE              = 0x30,
306         MV5_PHY_CTL             = 0x0C,
307         SATA_IFCFG              = 0x050,
308         LP_PHY_CTL              = 0x058,
309         LP_PHY_CTL_PIN_PU_PLL   = (1 << 0),
310         LP_PHY_CTL_PIN_PU_RX    = (1 << 1),
311         LP_PHY_CTL_PIN_PU_TX    = (1 << 2),
312         LP_PHY_CTL_GEN_TX_3G    = (1 << 5),
313         LP_PHY_CTL_GEN_RX_3G    = (1 << 9),
314 
315         MV_M2_PREAMP_MASK       = 0x7e0,
316 
317         /* Port registers */
318         EDMA_CFG                = 0,
319         EDMA_CFG_Q_DEPTH        = 0x1f,         /* max device queue depth */
320         EDMA_CFG_NCQ            = (1 << 5),     /* for R/W FPDMA queued */
321         EDMA_CFG_NCQ_GO_ON_ERR  = (1 << 14),    /* continue on error */
322         EDMA_CFG_RD_BRST_EXT    = (1 << 11),    /* read burst 512B */
323         EDMA_CFG_WR_BUFF_LEN    = (1 << 13),    /* write buffer 512B */
324         EDMA_CFG_EDMA_FBS       = (1 << 16),    /* EDMA FIS-Based Switching */
325         EDMA_CFG_FBS            = (1 << 26),    /* FIS-Based Switching */
326 
327         EDMA_ERR_IRQ_CAUSE      = 0x8,
328         EDMA_ERR_IRQ_MASK       = 0xc,
329         EDMA_ERR_D_PAR          = (1 << 0),     /* UDMA data parity err */
330         EDMA_ERR_PRD_PAR        = (1 << 1),     /* UDMA PRD parity err */
331         EDMA_ERR_DEV            = (1 << 2),     /* device error */
332         EDMA_ERR_DEV_DCON       = (1 << 3),     /* device disconnect */
333         EDMA_ERR_DEV_CON        = (1 << 4),     /* device connected */
334         EDMA_ERR_SERR           = (1 << 5),     /* SError bits [WBDST] raised */
335         EDMA_ERR_SELF_DIS       = (1 << 7),     /* Gen II/IIE self-disable */
336         EDMA_ERR_SELF_DIS_5     = (1 << 8),     /* Gen I self-disable */
337         EDMA_ERR_BIST_ASYNC     = (1 << 8),     /* BIST FIS or Async Notify */
338         EDMA_ERR_TRANS_IRQ_7    = (1 << 8),     /* Gen IIE transprt layer irq */
339         EDMA_ERR_CRQB_PAR       = (1 << 9),     /* CRQB parity error */
340         EDMA_ERR_CRPB_PAR       = (1 << 10),    /* CRPB parity error */
341         EDMA_ERR_INTRL_PAR      = (1 << 11),    /* internal parity error */
342         EDMA_ERR_IORDY          = (1 << 12),    /* IORdy timeout */
343 
344         EDMA_ERR_LNK_CTRL_RX    = (0xf << 13),  /* link ctrl rx error */
345         EDMA_ERR_LNK_CTRL_RX_0  = (1 << 13),    /* transient: CRC err */
346         EDMA_ERR_LNK_CTRL_RX_1  = (1 << 14),    /* transient: FIFO err */
347         EDMA_ERR_LNK_CTRL_RX_2  = (1 << 15),    /* fatal: caught SYNC */
348         EDMA_ERR_LNK_CTRL_RX_3  = (1 << 16),    /* transient: FIS rx err */
349 
350         EDMA_ERR_LNK_DATA_RX    = (0xf << 17),  /* link data rx error */
351 
352         EDMA_ERR_LNK_CTRL_TX    = (0x1f << 21), /* link ctrl tx error */
353         EDMA_ERR_LNK_CTRL_TX_0  = (1 << 21),    /* transient: CRC err */
354         EDMA_ERR_LNK_CTRL_TX_1  = (1 << 22),    /* transient: FIFO err */
355         EDMA_ERR_LNK_CTRL_TX_2  = (1 << 23),    /* transient: caught SYNC */
356         EDMA_ERR_LNK_CTRL_TX_3  = (1 << 24),    /* transient: caught DMAT */
357         EDMA_ERR_LNK_CTRL_TX_4  = (1 << 25),    /* transient: FIS collision */
358 
359         EDMA_ERR_LNK_DATA_TX    = (0x1f << 26), /* link data tx error */
360 
361         EDMA_ERR_TRANS_PROTO    = (1 << 31),    /* transport protocol error */
362         EDMA_ERR_OVERRUN_5      = (1 << 5),
363         EDMA_ERR_UNDERRUN_5     = (1 << 6),
364 
365         EDMA_ERR_IRQ_TRANSIENT  = EDMA_ERR_LNK_CTRL_RX_0 |
366                                   EDMA_ERR_LNK_CTRL_RX_1 |
367                                   EDMA_ERR_LNK_CTRL_RX_3 |
368                                   EDMA_ERR_LNK_CTRL_TX,
369 
370         EDMA_EH_FREEZE          = EDMA_ERR_D_PAR |
371                                   EDMA_ERR_PRD_PAR |
372                                   EDMA_ERR_DEV_DCON |
373                                   EDMA_ERR_DEV_CON |
374                                   EDMA_ERR_SERR |
375                                   EDMA_ERR_SELF_DIS |
376                                   EDMA_ERR_CRQB_PAR |
377                                   EDMA_ERR_CRPB_PAR |
378                                   EDMA_ERR_INTRL_PAR |
379                                   EDMA_ERR_IORDY |
380                                   EDMA_ERR_LNK_CTRL_RX_2 |
381                                   EDMA_ERR_LNK_DATA_RX |
382                                   EDMA_ERR_LNK_DATA_TX |
383                                   EDMA_ERR_TRANS_PROTO,
384 
385         EDMA_EH_FREEZE_5        = EDMA_ERR_D_PAR |
386                                   EDMA_ERR_PRD_PAR |
387                                   EDMA_ERR_DEV_DCON |
388                                   EDMA_ERR_DEV_CON |
389                                   EDMA_ERR_OVERRUN_5 |
390                                   EDMA_ERR_UNDERRUN_5 |
391                                   EDMA_ERR_SELF_DIS_5 |
392                                   EDMA_ERR_CRQB_PAR |
393                                   EDMA_ERR_CRPB_PAR |
394                                   EDMA_ERR_INTRL_PAR |
395                                   EDMA_ERR_IORDY,
396 
397         EDMA_REQ_Q_BASE_HI      = 0x10,
398         EDMA_REQ_Q_IN_PTR       = 0x14,         /* also contains BASE_LO */
399 
400         EDMA_REQ_Q_OUT_PTR      = 0x18,
401         EDMA_REQ_Q_PTR_SHIFT    = 5,
402 
403         EDMA_RSP_Q_BASE_HI      = 0x1c,
404         EDMA_RSP_Q_IN_PTR       = 0x20,
405         EDMA_RSP_Q_OUT_PTR      = 0x24,         /* also contains BASE_LO */
406         EDMA_RSP_Q_PTR_SHIFT    = 3,
407 
408         EDMA_CMD                = 0x28,         /* EDMA command register */
409         EDMA_EN                 = (1 << 0),     /* enable EDMA */
410         EDMA_DS                 = (1 << 1),     /* disable EDMA; self-negated */
411         EDMA_RESET              = (1 << 2),     /* reset eng/trans/link/phy */
412 
413         EDMA_STATUS             = 0x30,         /* EDMA engine status */
414         EDMA_STATUS_CACHE_EMPTY = (1 << 6),     /* GenIIe command cache empty */
415         EDMA_STATUS_IDLE        = (1 << 7),     /* GenIIe EDMA enabled/idle */
416 
417         EDMA_IORDY_TMOUT        = 0x34,
418         EDMA_ARB_CFG            = 0x38,
419 
420         EDMA_HALTCOND           = 0x60,         /* GenIIe halt conditions */
421         EDMA_UNKNOWN_RSVD       = 0x6C,         /* GenIIe unknown/reserved */
422 
423         BMDMA_CMD               = 0x224,        /* bmdma command register */
424         BMDMA_STATUS            = 0x228,        /* bmdma status register */
425         BMDMA_PRD_LOW           = 0x22c,        /* bmdma PRD addr 31:0 */
426         BMDMA_PRD_HIGH          = 0x230,        /* bmdma PRD addr 63:32 */
427 
428         /* Host private flags (hp_flags) */
429         MV_HP_FLAG_MSI          = (1 << 0),
430         MV_HP_ERRATA_50XXB0     = (1 << 1),
431         MV_HP_ERRATA_50XXB2     = (1 << 2),
432         MV_HP_ERRATA_60X1B2     = (1 << 3),
433         MV_HP_ERRATA_60X1C0     = (1 << 4),
434         MV_HP_GEN_I             = (1 << 6),     /* Generation I: 50xx */
435         MV_HP_GEN_II            = (1 << 7),     /* Generation II: 60xx */
436         MV_HP_GEN_IIE           = (1 << 8),     /* Generation IIE: 6042/7042 */
437         MV_HP_PCIE              = (1 << 9),     /* PCIe bus/regs: 7042 */
438         MV_HP_CUT_THROUGH       = (1 << 10),    /* can use EDMA cut-through */
439         MV_HP_FLAG_SOC          = (1 << 11),    /* SystemOnChip, no PCI */
440         MV_HP_QUIRK_LED_BLINK_EN = (1 << 12),   /* is led blinking enabled? */
441         MV_HP_FIX_LP_PHY_CTL    = (1 << 13),    /* fix speed in LP_PHY_CTL ? */
442 
443         /* Port private flags (pp_flags) */
444         MV_PP_FLAG_EDMA_EN      = (1 << 0),     /* is EDMA engine enabled? */
445         MV_PP_FLAG_NCQ_EN       = (1 << 1),     /* is EDMA set up for NCQ? */
446         MV_PP_FLAG_FBS_EN       = (1 << 2),     /* is EDMA set up for FBS? */
447         MV_PP_FLAG_DELAYED_EH   = (1 << 3),     /* delayed dev err handling */
448         MV_PP_FLAG_FAKE_ATA_BUSY = (1 << 4),    /* ignore initial ATA_DRDY */
449 };
450 
451 #define IS_GEN_I(hpriv) ((hpriv)->hp_flags & MV_HP_GEN_I)
452 #define IS_GEN_II(hpriv) ((hpriv)->hp_flags & MV_HP_GEN_II)
453 #define IS_GEN_IIE(hpriv) ((hpriv)->hp_flags & MV_HP_GEN_IIE)
454 #define IS_PCIE(hpriv) ((hpriv)->hp_flags & MV_HP_PCIE)
455 #define IS_SOC(hpriv) ((hpriv)->hp_flags & MV_HP_FLAG_SOC)
456 
457 #define WINDOW_CTRL(i)          (0x20030 + ((i) << 4))
458 #define WINDOW_BASE(i)          (0x20034 + ((i) << 4))
459 
460 enum {
461         /* DMA boundary 0xffff is required by the s/g splitting
462          * we need on /length/ in mv_fill-sg().
463          */
464         MV_DMA_BOUNDARY         = 0xffffU,
465 
466         /* mask of register bits containing lower 32 bits
467          * of EDMA request queue DMA address
468          */
469         EDMA_REQ_Q_BASE_LO_MASK = 0xfffffc00U,
470 
471         /* ditto, for response queue */
472         EDMA_RSP_Q_BASE_LO_MASK = 0xffffff00U,
473 };
474 
475 enum chip_type {
476         chip_504x,
477         chip_508x,
478         chip_5080,
479         chip_604x,
480         chip_608x,
481         chip_6042,
482         chip_7042,
483         chip_soc,
484 };
485 
486 /* Command ReQuest Block: 32B */
487 struct mv_crqb {
488         __le32                  sg_addr;
489         __le32                  sg_addr_hi;
490         __le16                  ctrl_flags;
491         __le16                  ata_cmd[11];
492 };
493 
494 struct mv_crqb_iie {
495         __le32                  addr;
496         __le32                  addr_hi;
497         __le32                  flags;
498         __le32                  len;
499         __le32                  ata_cmd[4];
500 };
501 
502 /* Command ResPonse Block: 8B */
503 struct mv_crpb {
504         __le16                  id;
505         __le16                  flags;
506         __le32                  tmstmp;
507 };
508 
509 /* EDMA Physical Region Descriptor (ePRD); A.K.A. SG */
510 struct mv_sg {
511         __le32                  addr;
512         __le32                  flags_size;
513         __le32                  addr_hi;
514         __le32                  reserved;
515 };
516 
517 /*
518  * We keep a local cache of a few frequently accessed port
519  * registers here, to avoid having to read them (very slow)
520  * when switching between EDMA and non-EDMA modes.
521  */
522 struct mv_cached_regs {
523         u32                     fiscfg;
524         u32                     ltmode;
525         u32                     haltcond;
526         u32                     unknown_rsvd;
527 };
528 
529 struct mv_port_priv {
530         struct mv_crqb          *crqb;
531         dma_addr_t              crqb_dma;
532         struct mv_crpb          *crpb;
533         dma_addr_t              crpb_dma;
534         struct mv_sg            *sg_tbl[MV_MAX_Q_DEPTH];
535         dma_addr_t              sg_tbl_dma[MV_MAX_Q_DEPTH];
536 
537         unsigned int            req_idx;
538         unsigned int            resp_idx;
539 
540         u32                     pp_flags;
541         struct mv_cached_regs   cached;
542         unsigned int            delayed_eh_pmp_map;
543 };
544 
545 struct mv_port_signal {
546         u32                     amps;
547         u32                     pre;
548 };
549 
550 struct mv_host_priv {
551         u32                     hp_flags;
552         unsigned int            board_idx;
553         u32                     main_irq_mask;
554         struct mv_port_signal   signal[8];
555         const struct mv_hw_ops  *ops;
556         int                     n_ports;
557         void __iomem            *base;
558         void __iomem            *main_irq_cause_addr;
559         void __iomem            *main_irq_mask_addr;
560         u32                     irq_cause_offset;
561         u32                     irq_mask_offset;
562         u32                     unmask_all_irqs;
563 
564         /*
565          * Needed on some devices that require their clocks to be enabled.
566          * These are optional: if the platform device does not have any
567          * clocks, they won't be used.  Also, if the underlying hardware
568          * does not support the common clock framework (CONFIG_HAVE_CLK=n),
569          * all the clock operations become no-ops (see clk.h).
570          */
571         struct clk              *clk;
572         struct clk              **port_clks;
573         /*
574          * Some devices have a SATA PHY which can be enabled/disabled
575          * in order to save power. These are optional: if the platform
576          * devices does not have any phy, they won't be used.
577          */
578         struct phy              **port_phys;
579         /*
580          * These consistent DMA memory pools give us guaranteed
581          * alignment for hardware-accessed data structures,
582          * and less memory waste in accomplishing the alignment.
583          */
584         struct dma_pool         *crqb_pool;
585         struct dma_pool         *crpb_pool;
586         struct dma_pool         *sg_tbl_pool;
587 };
588 
589 struct mv_hw_ops {
590         void (*phy_errata)(struct mv_host_priv *hpriv, void __iomem *mmio,
591                            unsigned int port);
592         void (*enable_leds)(struct mv_host_priv *hpriv, void __iomem *mmio);
593         void (*read_preamp)(struct mv_host_priv *hpriv, int idx,
594                            void __iomem *mmio);
595         int (*reset_hc)(struct mv_host_priv *hpriv, void __iomem *mmio,
596                         unsigned int n_hc);
597         void (*reset_flash)(struct mv_host_priv *hpriv, void __iomem *mmio);
598         void (*reset_bus)(struct ata_host *host, void __iomem *mmio);
599 };
600 
601 static int mv_scr_read(struct ata_link *link, unsigned int sc_reg_in, u32 *val);
602 static int mv_scr_write(struct ata_link *link, unsigned int sc_reg_in, u32 val);
603 static int mv5_scr_read(struct ata_link *link, unsigned int sc_reg_in, u32 *val);
604 static int mv5_scr_write(struct ata_link *link, unsigned int sc_reg_in, u32 val);
605 static int mv_port_start(struct ata_port *ap);
606 static void mv_port_stop(struct ata_port *ap);
607 static int mv_qc_defer(struct ata_queued_cmd *qc);
608 static void mv_qc_prep(struct ata_queued_cmd *qc);
609 static void mv_qc_prep_iie(struct ata_queued_cmd *qc);
610 static unsigned int mv_qc_issue(struct ata_queued_cmd *qc);
611 static int mv_hardreset(struct ata_link *link, unsigned int *class,
612                         unsigned long deadline);
613 static void mv_eh_freeze(struct ata_port *ap);
614 static void mv_eh_thaw(struct ata_port *ap);
615 static void mv6_dev_config(struct ata_device *dev);
616 
617 static void mv5_phy_errata(struct mv_host_priv *hpriv, void __iomem *mmio,
618                            unsigned int port);
619 static void mv5_enable_leds(struct mv_host_priv *hpriv, void __iomem *mmio);
620 static void mv5_read_preamp(struct mv_host_priv *hpriv, int idx,
621                            void __iomem *mmio);
622 static int mv5_reset_hc(struct mv_host_priv *hpriv, void __iomem *mmio,
623                         unsigned int n_hc);
624 static void mv5_reset_flash(struct mv_host_priv *hpriv, void __iomem *mmio);
625 static void mv5_reset_bus(struct ata_host *host, void __iomem *mmio);
626 
627 static void mv6_phy_errata(struct mv_host_priv *hpriv, void __iomem *mmio,
628                            unsigned int port);
629 static void mv6_enable_leds(struct mv_host_priv *hpriv, void __iomem *mmio);
630 static void mv6_read_preamp(struct mv_host_priv *hpriv, int idx,
631                            void __iomem *mmio);
632 static int mv6_reset_hc(struct mv_host_priv *hpriv, void __iomem *mmio,
633                         unsigned int n_hc);
634 static void mv6_reset_flash(struct mv_host_priv *hpriv, void __iomem *mmio);
635 static void mv_soc_enable_leds(struct mv_host_priv *hpriv,
636                                       void __iomem *mmio);
637 static void mv_soc_read_preamp(struct mv_host_priv *hpriv, int idx,
638                                       void __iomem *mmio);
639 static int mv_soc_reset_hc(struct mv_host_priv *hpriv,
640                                   void __iomem *mmio, unsigned int n_hc);
641 static void mv_soc_reset_flash(struct mv_host_priv *hpriv,
642                                       void __iomem *mmio);
643 static void mv_soc_reset_bus(struct ata_host *host, void __iomem *mmio);
644 static void mv_soc_65n_phy_errata(struct mv_host_priv *hpriv,
645                                   void __iomem *mmio, unsigned int port);
646 static void mv_reset_pci_bus(struct ata_host *host, void __iomem *mmio);
647 static void mv_reset_channel(struct mv_host_priv *hpriv, void __iomem *mmio,
648                              unsigned int port_no);
649 static int mv_stop_edma(struct ata_port *ap);
650 static int mv_stop_edma_engine(void __iomem *port_mmio);
651 static void mv_edma_cfg(struct ata_port *ap, int want_ncq, int want_edma);
652 
653 static void mv_pmp_select(struct ata_port *ap, int pmp);
654 static int mv_pmp_hardreset(struct ata_link *link, unsigned int *class,
655                                 unsigned long deadline);
656 static int  mv_softreset(struct ata_link *link, unsigned int *class,
657                                 unsigned long deadline);
658 static void mv_pmp_error_handler(struct ata_port *ap);
659 static void mv_process_crpb_entries(struct ata_port *ap,
660                                         struct mv_port_priv *pp);
661 
662 static void mv_sff_irq_clear(struct ata_port *ap);
663 static int mv_check_atapi_dma(struct ata_queued_cmd *qc);
664 static void mv_bmdma_setup(struct ata_queued_cmd *qc);
665 static void mv_bmdma_start(struct ata_queued_cmd *qc);
666 static void mv_bmdma_stop(struct ata_queued_cmd *qc);
667 static u8   mv_bmdma_status(struct ata_port *ap);
668 static u8 mv_sff_check_status(struct ata_port *ap);
669 
670 /* .sg_tablesize is (MV_MAX_SG_CT / 2) in the structures below
671  * because we have to allow room for worst case splitting of
672  * PRDs for 64K boundaries in mv_fill_sg().
673  */
674 #ifdef CONFIG_PCI
675 static struct scsi_host_template mv5_sht = {
676         ATA_BASE_SHT(DRV_NAME),
677         .sg_tablesize           = MV_MAX_SG_CT / 2,
678         .dma_boundary           = MV_DMA_BOUNDARY,
679 };
680 #endif
681 static struct scsi_host_template mv6_sht = {
682         ATA_NCQ_SHT(DRV_NAME),
683         .can_queue              = MV_MAX_Q_DEPTH - 1,
684         .sg_tablesize           = MV_MAX_SG_CT / 2,
685         .dma_boundary           = MV_DMA_BOUNDARY,
686 };
687 
688 static struct ata_port_operations mv5_ops = {
689         .inherits               = &ata_sff_port_ops,
690 
691         .lost_interrupt         = ATA_OP_NULL,
692 
693         .qc_defer               = mv_qc_defer,
694         .qc_prep                = mv_qc_prep,
695         .qc_issue               = mv_qc_issue,
696 
697         .freeze                 = mv_eh_freeze,
698         .thaw                   = mv_eh_thaw,
699         .hardreset              = mv_hardreset,
700 
701         .scr_read               = mv5_scr_read,
702         .scr_write              = mv5_scr_write,
703 
704         .port_start             = mv_port_start,
705         .port_stop              = mv_port_stop,
706 };
707 
708 static struct ata_port_operations mv6_ops = {
709         .inherits               = &ata_bmdma_port_ops,
710 
711         .lost_interrupt         = ATA_OP_NULL,
712 
713         .qc_defer               = mv_qc_defer,
714         .qc_prep                = mv_qc_prep,
715         .qc_issue               = mv_qc_issue,
716 
717         .dev_config             = mv6_dev_config,
718 
719         .freeze                 = mv_eh_freeze,
720         .thaw                   = mv_eh_thaw,
721         .hardreset              = mv_hardreset,
722         .softreset              = mv_softreset,
723         .pmp_hardreset          = mv_pmp_hardreset,
724         .pmp_softreset          = mv_softreset,
725         .error_handler          = mv_pmp_error_handler,
726 
727         .scr_read               = mv_scr_read,
728         .scr_write              = mv_scr_write,
729 
730         .sff_check_status       = mv_sff_check_status,
731         .sff_irq_clear          = mv_sff_irq_clear,
732         .check_atapi_dma        = mv_check_atapi_dma,
733         .bmdma_setup            = mv_bmdma_setup,
734         .bmdma_start            = mv_bmdma_start,
735         .bmdma_stop             = mv_bmdma_stop,
736         .bmdma_status           = mv_bmdma_status,
737 
738         .port_start             = mv_port_start,
739         .port_stop              = mv_port_stop,
740 };
741 
742 static struct ata_port_operations mv_iie_ops = {
743         .inherits               = &mv6_ops,
744         .dev_config             = ATA_OP_NULL,
745         .qc_prep                = mv_qc_prep_iie,
746 };
747 
748 static const struct ata_port_info mv_port_info[] = {
749         {  /* chip_504x */
750                 .flags          = MV_GEN_I_FLAGS,
751                 .pio_mask       = ATA_PIO4,
752                 .udma_mask      = ATA_UDMA6,
753                 .port_ops       = &mv5_ops,
754         },
755         {  /* chip_508x */
756                 .flags          = MV_GEN_I_FLAGS | MV_FLAG_DUAL_HC,
757                 .pio_mask       = ATA_PIO4,
758                 .udma_mask      = ATA_UDMA6,
759                 .port_ops       = &mv5_ops,
760         },
761         {  /* chip_5080 */
762                 .flags          = MV_GEN_I_FLAGS | MV_FLAG_DUAL_HC,
763                 .pio_mask       = ATA_PIO4,
764                 .udma_mask      = ATA_UDMA6,
765                 .port_ops       = &mv5_ops,
766         },
767         {  /* chip_604x */
768                 .flags          = MV_GEN_II_FLAGS,
769                 .pio_mask       = ATA_PIO4,
770                 .udma_mask      = ATA_UDMA6,
771                 .port_ops       = &mv6_ops,
772         },
773         {  /* chip_608x */
774                 .flags          = MV_GEN_II_FLAGS | MV_FLAG_DUAL_HC,
775                 .pio_mask       = ATA_PIO4,
776                 .udma_mask      = ATA_UDMA6,
777                 .port_ops       = &mv6_ops,
778         },
779         {  /* chip_6042 */
780                 .flags          = MV_GEN_IIE_FLAGS,
781                 .pio_mask       = ATA_PIO4,
782                 .udma_mask      = ATA_UDMA6,
783                 .port_ops       = &mv_iie_ops,
784         },
785         {  /* chip_7042 */
786                 .flags          = MV_GEN_IIE_FLAGS,
787                 .pio_mask       = ATA_PIO4,
788                 .udma_mask      = ATA_UDMA6,
789                 .port_ops       = &mv_iie_ops,
790         },
791         {  /* chip_soc */
792                 .flags          = MV_GEN_IIE_FLAGS,
793                 .pio_mask       = ATA_PIO4,
794                 .udma_mask      = ATA_UDMA6,
795                 .port_ops       = &mv_iie_ops,
796         },
797 };
798 
799 static const struct pci_device_id mv_pci_tbl[] = {
800         { PCI_VDEVICE(MARVELL, 0x5040), chip_504x },
801         { PCI_VDEVICE(MARVELL, 0x5041), chip_504x },
802         { PCI_VDEVICE(MARVELL, 0x5080), chip_5080 },
803         { PCI_VDEVICE(MARVELL, 0x5081), chip_508x },
804         /* RocketRAID 1720/174x have different identifiers */
805         { PCI_VDEVICE(TTI, 0x1720), chip_6042 },
806         { PCI_VDEVICE(TTI, 0x1740), chip_6042 },
807         { PCI_VDEVICE(TTI, 0x1742), chip_6042 },
808 
809         { PCI_VDEVICE(MARVELL, 0x6040), chip_604x },
810         { PCI_VDEVICE(MARVELL, 0x6041), chip_604x },
811         { PCI_VDEVICE(MARVELL, 0x6042), chip_6042 },
812         { PCI_VDEVICE(MARVELL, 0x6080), chip_608x },
813         { PCI_VDEVICE(MARVELL, 0x6081), chip_608x },
814 
815         { PCI_VDEVICE(ADAPTEC2, 0x0241), chip_604x },
816 
817         /* Adaptec 1430SA */
818         { PCI_VDEVICE(ADAPTEC2, 0x0243), chip_7042 },
819 
820         /* Marvell 7042 support */
821         { PCI_VDEVICE(MARVELL, 0x7042), chip_7042 },
822 
823         /* Highpoint RocketRAID PCIe series */
824         { PCI_VDEVICE(TTI, 0x2300), chip_7042 },
825         { PCI_VDEVICE(TTI, 0x2310), chip_7042 },
826 
827         { }                     /* terminate list */
828 };
829 
830 static const struct mv_hw_ops mv5xxx_ops = {
831         .phy_errata             = mv5_phy_errata,
832         .enable_leds            = mv5_enable_leds,
833         .read_preamp            = mv5_read_preamp,
834         .reset_hc               = mv5_reset_hc,
835         .reset_flash            = mv5_reset_flash,
836         .reset_bus              = mv5_reset_bus,
837 };
838 
839 static const struct mv_hw_ops mv6xxx_ops = {
840         .phy_errata             = mv6_phy_errata,
841         .enable_leds            = mv6_enable_leds,
842         .read_preamp            = mv6_read_preamp,
843         .reset_hc               = mv6_reset_hc,
844         .reset_flash            = mv6_reset_flash,
845         .reset_bus              = mv_reset_pci_bus,
846 };
847 
848 static const struct mv_hw_ops mv_soc_ops = {
849         .phy_errata             = mv6_phy_errata,
850         .enable_leds            = mv_soc_enable_leds,
851         .read_preamp            = mv_soc_read_preamp,
852         .reset_hc               = mv_soc_reset_hc,
853         .reset_flash            = mv_soc_reset_flash,
854         .reset_bus              = mv_soc_reset_bus,
855 };
856 
857 static const struct mv_hw_ops mv_soc_65n_ops = {
858         .phy_errata             = mv_soc_65n_phy_errata,
859         .enable_leds            = mv_soc_enable_leds,
860         .reset_hc               = mv_soc_reset_hc,
861         .reset_flash            = mv_soc_reset_flash,
862         .reset_bus              = mv_soc_reset_bus,
863 };
864 
865 /*
866  * Functions
867  */
868 
869 static inline void writelfl(unsigned long data, void __iomem *addr)
870 {
871         writel(data, addr);
872         (void) readl(addr);     /* flush to avoid PCI posted write */
873 }
874 
875 static inline unsigned int mv_hc_from_port(unsigned int port)
876 {
877         return port >> MV_PORT_HC_SHIFT;
878 }
879 
880 static inline unsigned int mv_hardport_from_port(unsigned int port)
881 {
882         return port & MV_PORT_MASK;
883 }
884 
885 /*
886  * Consolidate some rather tricky bit shift calculations.
887  * This is hot-path stuff, so not a function.
888  * Simple code, with two return values, so macro rather than inline.
889  *
890  * port is the sole input, in range 0..7.
891  * shift is one output, for use with main_irq_cause / main_irq_mask registers.
892  * hardport is the other output, in range 0..3.
893  *
894  * Note that port and hardport may be the same variable in some cases.
895  */
896 #define MV_PORT_TO_SHIFT_AND_HARDPORT(port, shift, hardport)    \
897 {                                                               \
898         shift    = mv_hc_from_port(port) * HC_SHIFT;            \
899         hardport = mv_hardport_from_port(port);                 \
900         shift   += hardport * 2;                                \
901 }
902 
903 static inline void __iomem *mv_hc_base(void __iomem *base, unsigned int hc)
904 {
905         return (base + SATAHC0_REG_BASE + (hc * MV_SATAHC_REG_SZ));
906 }
907 
908 static inline void __iomem *mv_hc_base_from_port(void __iomem *base,
909                                                  unsigned int port)
910 {
911         return mv_hc_base(base, mv_hc_from_port(port));
912 }
913 
914 static inline void __iomem *mv_port_base(void __iomem *base, unsigned int port)
915 {
916         return  mv_hc_base_from_port(base, port) +
917                 MV_SATAHC_ARBTR_REG_SZ +
918                 (mv_hardport_from_port(port) * MV_PORT_REG_SZ);
919 }
920 
921 static void __iomem *mv5_phy_base(void __iomem *mmio, unsigned int port)
922 {
923         void __iomem *hc_mmio = mv_hc_base_from_port(mmio, port);
924         unsigned long ofs = (mv_hardport_from_port(port) + 1) * 0x100UL;
925 
926         return hc_mmio + ofs;
927 }
928 
929 static inline void __iomem *mv_host_base(struct ata_host *host)
930 {
931         struct mv_host_priv *hpriv = host->private_data;
932         return hpriv->base;
933 }
934 
935 static inline void __iomem *mv_ap_base(struct ata_port *ap)
936 {
937         return mv_port_base(mv_host_base(ap->host), ap->port_no);
938 }
939 
940 static inline int mv_get_hc_count(unsigned long port_flags)
941 {
942         return ((port_flags & MV_FLAG_DUAL_HC) ? 2 : 1);
943 }
944 
945 /**
946  *      mv_save_cached_regs - (re-)initialize cached port registers
947  *      @ap: the port whose registers we are caching
948  *
949  *      Initialize the local cache of port registers,
950  *      so that reading them over and over again can
951  *      be avoided on the hotter paths of this driver.
952  *      This saves a few microseconds each time we switch
953  *      to/from EDMA mode to perform (eg.) a drive cache flush.
954  */
955 static void mv_save_cached_regs(struct ata_port *ap)
956 {
957         void __iomem *port_mmio = mv_ap_base(ap);
958         struct mv_port_priv *pp = ap->private_data;
959 
960         pp->cached.fiscfg = readl(port_mmio + FISCFG);
961         pp->cached.ltmode = readl(port_mmio + LTMODE);
962         pp->cached.haltcond = readl(port_mmio + EDMA_HALTCOND);
963         pp->cached.unknown_rsvd = readl(port_mmio + EDMA_UNKNOWN_RSVD);
964 }
965 
966 /**
967  *      mv_write_cached_reg - write to a cached port register
968  *      @addr: hardware address of the register
969  *      @old: pointer to cached value of the register
970  *      @new: new value for the register
971  *
972  *      Write a new value to a cached register,
973  *      but only if the value is different from before.
974  */
975 static inline void mv_write_cached_reg(void __iomem *addr, u32 *old, u32 new)
976 {
977         if (new != *old) {
978                 unsigned long laddr;
979                 *old = new;
980                 /*
981                  * Workaround for 88SX60x1-B2 FEr SATA#13:
982                  * Read-after-write is needed to prevent generating 64-bit
983                  * write cycles on the PCI bus for SATA interface registers
984                  * at offsets ending in 0x4 or 0xc.
985                  *
986                  * Looks like a lot of fuss, but it avoids an unnecessary
987                  * +1 usec read-after-write delay for unaffected registers.
988                  */
989                 laddr = (long)addr & 0xffff;
990                 if (laddr >= 0x300 && laddr <= 0x33c) {
991                         laddr &= 0x000f;
992                         if (laddr == 0x4 || laddr == 0xc) {
993                                 writelfl(new, addr); /* read after write */
994                                 return;
995                         }
996                 }
997                 writel(new, addr); /* unaffected by the errata */
998         }
999 }
1000 
1001 static void mv_set_edma_ptrs(void __iomem *port_mmio,
1002                              struct mv_host_priv *hpriv,
1003                              struct mv_port_priv *pp)
1004 {
1005         u32 index;
1006 
1007         /*
1008          * initialize request queue
1009          */
1010         pp->req_idx &= MV_MAX_Q_DEPTH_MASK;     /* paranoia */
1011         index = pp->req_idx << EDMA_REQ_Q_PTR_SHIFT;
1012 
1013         WARN_ON(pp->crqb_dma & 0x3ff);
1014         writel((pp->crqb_dma >> 16) >> 16, port_mmio + EDMA_REQ_Q_BASE_HI);
1015         writelfl((pp->crqb_dma & EDMA_REQ_Q_BASE_LO_MASK) | index,
1016                  port_mmio + EDMA_REQ_Q_IN_PTR);
1017         writelfl(index, port_mmio + EDMA_REQ_Q_OUT_PTR);
1018 
1019         /*
1020          * initialize response queue
1021          */
1022         pp->resp_idx &= MV_MAX_Q_DEPTH_MASK;    /* paranoia */
1023         index = pp->resp_idx << EDMA_RSP_Q_PTR_SHIFT;
1024 
1025         WARN_ON(pp->crpb_dma & 0xff);
1026         writel((pp->crpb_dma >> 16) >> 16, port_mmio + EDMA_RSP_Q_BASE_HI);
1027         writelfl(index, port_mmio + EDMA_RSP_Q_IN_PTR);
1028         writelfl((pp->crpb_dma & EDMA_RSP_Q_BASE_LO_MASK) | index,
1029                  port_mmio + EDMA_RSP_Q_OUT_PTR);
1030 }
1031 
1032 static void mv_write_main_irq_mask(u32 mask, struct mv_host_priv *hpriv)
1033 {
1034         /*
1035          * When writing to the main_irq_mask in hardware,
1036          * we must ensure exclusivity between the interrupt coalescing bits
1037          * and the corresponding individual port DONE_IRQ bits.
1038          *
1039          * Note that this register is really an "IRQ enable" register,
1040          * not an "IRQ mask" register as Marvell's naming might suggest.
1041          */
1042         if (mask & (ALL_PORTS_COAL_DONE | PORTS_0_3_COAL_DONE))
1043                 mask &= ~DONE_IRQ_0_3;
1044         if (mask & (ALL_PORTS_COAL_DONE | PORTS_4_7_COAL_DONE))
1045                 mask &= ~DONE_IRQ_4_7;
1046         writelfl(mask, hpriv->main_irq_mask_addr);
1047 }
1048 
1049 static void mv_set_main_irq_mask(struct ata_host *host,
1050                                  u32 disable_bits, u32 enable_bits)
1051 {
1052         struct mv_host_priv *hpriv = host->private_data;
1053         u32 old_mask, new_mask;
1054 
1055         old_mask = hpriv->main_irq_mask;
1056         new_mask = (old_mask & ~disable_bits) | enable_bits;
1057         if (new_mask != old_mask) {
1058                 hpriv->main_irq_mask = new_mask;
1059                 mv_write_main_irq_mask(new_mask, hpriv);
1060         }
1061 }
1062 
1063 static void mv_enable_port_irqs(struct ata_port *ap,
1064                                      unsigned int port_bits)
1065 {
1066         unsigned int shift, hardport, port = ap->port_no;
1067         u32 disable_bits, enable_bits;
1068 
1069         MV_PORT_TO_SHIFT_AND_HARDPORT(port, shift, hardport);
1070 
1071         disable_bits = (DONE_IRQ | ERR_IRQ) << shift;
1072         enable_bits  = port_bits << shift;
1073         mv_set_main_irq_mask(ap->host, disable_bits, enable_bits);
1074 }
1075 
1076 static void mv_clear_and_enable_port_irqs(struct ata_port *ap,
1077                                           void __iomem *port_mmio,
1078                                           unsigned int port_irqs)
1079 {
1080         struct mv_host_priv *hpriv = ap->host->private_data;
1081         int hardport = mv_hardport_from_port(ap->port_no);
1082         void __iomem *hc_mmio = mv_hc_base_from_port(
1083                                 mv_host_base(ap->host), ap->port_no);
1084         u32 hc_irq_cause;
1085 
1086         /* clear EDMA event indicators, if any */
1087         writelfl(0, port_mmio + EDMA_ERR_IRQ_CAUSE);
1088 
1089         /* clear pending irq events */
1090         hc_irq_cause = ~((DEV_IRQ | DMA_IRQ) << hardport);
1091         writelfl(hc_irq_cause, hc_mmio + HC_IRQ_CAUSE);
1092 
1093         /* clear FIS IRQ Cause */
1094         if (IS_GEN_IIE(hpriv))
1095                 writelfl(0, port_mmio + FIS_IRQ_CAUSE);
1096 
1097         mv_enable_port_irqs(ap, port_irqs);
1098 }
1099 
1100 static void mv_set_irq_coalescing(struct ata_host *host,
1101                                   unsigned int count, unsigned int usecs)
1102 {
1103         struct mv_host_priv *hpriv = host->private_data;
1104         void __iomem *mmio = hpriv->base, *hc_mmio;
1105         u32 coal_enable = 0;
1106         unsigned long flags;
1107         unsigned int clks, is_dual_hc = hpriv->n_ports > MV_PORTS_PER_HC;
1108         const u32 coal_disable = PORTS_0_3_COAL_DONE | PORTS_4_7_COAL_DONE |
1109                                                         ALL_PORTS_COAL_DONE;
1110 
1111         /* Disable IRQ coalescing if either threshold is zero */
1112         if (!usecs || !count) {
1113                 clks = count = 0;
1114         } else {
1115                 /* Respect maximum limits of the hardware */
1116                 clks = usecs * COAL_CLOCKS_PER_USEC;
1117                 if (clks > MAX_COAL_TIME_THRESHOLD)
1118                         clks = MAX_COAL_TIME_THRESHOLD;
1119                 if (count > MAX_COAL_IO_COUNT)
1120                         count = MAX_COAL_IO_COUNT;
1121         }
1122 
1123         spin_lock_irqsave(&host->lock, flags);
1124         mv_set_main_irq_mask(host, coal_disable, 0);
1125 
1126         if (is_dual_hc && !IS_GEN_I(hpriv)) {
1127                 /*
1128                  * GEN_II/GEN_IIE with dual host controllers:
1129                  * one set of global thresholds for the entire chip.
1130                  */
1131                 writel(clks,  mmio + IRQ_COAL_TIME_THRESHOLD);
1132                 writel(count, mmio + IRQ_COAL_IO_THRESHOLD);
1133                 /* clear leftover coal IRQ bit */
1134                 writel(~ALL_PORTS_COAL_IRQ, mmio + IRQ_COAL_CAUSE);
1135                 if (count)
1136                         coal_enable = ALL_PORTS_COAL_DONE;
1137                 clks = count = 0; /* force clearing of regular regs below */
1138         }
1139 
1140         /*
1141          * All chips: independent thresholds for each HC on the chip.
1142          */
1143         hc_mmio = mv_hc_base_from_port(mmio, 0);
1144         writel(clks,  hc_mmio + HC_IRQ_COAL_TIME_THRESHOLD);
1145         writel(count, hc_mmio + HC_IRQ_COAL_IO_THRESHOLD);
1146         writel(~HC_COAL_IRQ, hc_mmio + HC_IRQ_CAUSE);
1147         if (count)
1148                 coal_enable |= PORTS_0_3_COAL_DONE;
1149         if (is_dual_hc) {
1150                 hc_mmio = mv_hc_base_from_port(mmio, MV_PORTS_PER_HC);
1151                 writel(clks,  hc_mmio + HC_IRQ_COAL_TIME_THRESHOLD);
1152                 writel(count, hc_mmio + HC_IRQ_COAL_IO_THRESHOLD);
1153                 writel(~HC_COAL_IRQ, hc_mmio + HC_IRQ_CAUSE);
1154                 if (count)
1155                         coal_enable |= PORTS_4_7_COAL_DONE;
1156         }
1157 
1158         mv_set_main_irq_mask(host, 0, coal_enable);
1159         spin_unlock_irqrestore(&host->lock, flags);
1160 }
1161 
1162 /**
1163  *      mv_start_edma - Enable eDMA engine
1164  *      @base: port base address
1165  *      @pp: port private data
1166  *
1167  *      Verify the local cache of the eDMA state is accurate with a
1168  *      WARN_ON.
1169  *
1170  *      LOCKING:
1171  *      Inherited from caller.
1172  */
1173 static void mv_start_edma(struct ata_port *ap, void __iomem *port_mmio,
1174                          struct mv_port_priv *pp, u8 protocol)
1175 {
1176         int want_ncq = (protocol == ATA_PROT_NCQ);
1177 
1178         if (pp->pp_flags & MV_PP_FLAG_EDMA_EN) {
1179                 int using_ncq = ((pp->pp_flags & MV_PP_FLAG_NCQ_EN) != 0);
1180                 if (want_ncq != using_ncq)
1181                         mv_stop_edma(ap);
1182         }
1183         if (!(pp->pp_flags & MV_PP_FLAG_EDMA_EN)) {
1184                 struct mv_host_priv *hpriv = ap->host->private_data;
1185 
1186                 mv_edma_cfg(ap, want_ncq, 1);
1187 
1188                 mv_set_edma_ptrs(port_mmio, hpriv, pp);
1189                 mv_clear_and_enable_port_irqs(ap, port_mmio, DONE_IRQ|ERR_IRQ);
1190 
1191                 writelfl(EDMA_EN, port_mmio + EDMA_CMD);
1192                 pp->pp_flags |= MV_PP_FLAG_EDMA_EN;
1193         }
1194 }
1195 
1196 static void mv_wait_for_edma_empty_idle(struct ata_port *ap)
1197 {
1198         void __iomem *port_mmio = mv_ap_base(ap);
1199         const u32 empty_idle = (EDMA_STATUS_CACHE_EMPTY | EDMA_STATUS_IDLE);
1200         const int per_loop = 5, timeout = (15 * 1000 / per_loop);
1201         int i;
1202 
1203         /*
1204          * Wait for the EDMA engine to finish transactions in progress.
1205          * No idea what a good "timeout" value might be, but measurements
1206          * indicate that it often requires hundreds of microseconds
1207          * with two drives in-use.  So we use the 15msec value above
1208          * as a rough guess at what even more drives might require.
1209          */
1210         for (i = 0; i < timeout; ++i) {
1211                 u32 edma_stat = readl(port_mmio + EDMA_STATUS);
1212                 if ((edma_stat & empty_idle) == empty_idle)
1213                         break;
1214                 udelay(per_loop);
1215         }
1216         /* ata_port_info(ap, "%s: %u+ usecs\n", __func__, i); */
1217 }
1218 
1219 /**
1220  *      mv_stop_edma_engine - Disable eDMA engine
1221  *      @port_mmio: io base address
1222  *
1223  *      LOCKING:
1224  *      Inherited from caller.
1225  */
1226 static int mv_stop_edma_engine(void __iomem *port_mmio)
1227 {
1228         int i;
1229 
1230         /* Disable eDMA.  The disable bit auto clears. */
1231         writelfl(EDMA_DS, port_mmio + EDMA_CMD);
1232 
1233         /* Wait for the chip to confirm eDMA is off. */
1234         for (i = 10000; i > 0; i--) {
1235                 u32 reg = readl(port_mmio + EDMA_CMD);
1236                 if (!(reg & EDMA_EN))
1237                         return 0;
1238                 udelay(10);
1239         }
1240         return -EIO;
1241 }
1242 
1243 static int mv_stop_edma(struct ata_port *ap)
1244 {
1245         void __iomem *port_mmio = mv_ap_base(ap);
1246         struct mv_port_priv *pp = ap->private_data;
1247         int err = 0;
1248 
1249         if (!(pp->pp_flags & MV_PP_FLAG_EDMA_EN))
1250                 return 0;
1251         pp->pp_flags &= ~MV_PP_FLAG_EDMA_EN;
1252         mv_wait_for_edma_empty_idle(ap);
1253         if (mv_stop_edma_engine(port_mmio)) {
1254                 ata_port_err(ap, "Unable to stop eDMA\n");
1255                 err = -EIO;
1256         }
1257         mv_edma_cfg(ap, 0, 0);
1258         return err;
1259 }
1260 
1261 #ifdef ATA_DEBUG
1262 static void mv_dump_mem(void __iomem *start, unsigned bytes)
1263 {
1264         int b, w;
1265         for (b = 0; b < bytes; ) {
1266                 DPRINTK("%p: ", start + b);
1267                 for (w = 0; b < bytes && w < 4; w++) {
1268                         printk("%08x ", readl(start + b));
1269                         b += sizeof(u32);
1270                 }
1271                 printk("\n");
1272         }
1273 }
1274 #endif
1275 #if defined(ATA_DEBUG) || defined(CONFIG_PCI)
1276 static void mv_dump_pci_cfg(struct pci_dev *pdev, unsigned bytes)
1277 {
1278 #ifdef ATA_DEBUG
1279         int b, w;
1280         u32 dw;
1281         for (b = 0; b < bytes; ) {
1282                 DPRINTK("%02x: ", b);
1283                 for (w = 0; b < bytes && w < 4; w++) {
1284                         (void) pci_read_config_dword(pdev, b, &dw);
1285                         printk("%08x ", dw);
1286                         b += sizeof(u32);
1287                 }
1288                 printk("\n");
1289         }
1290 #endif
1291 }
1292 #endif
1293 static void mv_dump_all_regs(void __iomem *mmio_base, int port,
1294                              struct pci_dev *pdev)
1295 {
1296 #ifdef ATA_DEBUG
1297         void __iomem *hc_base = mv_hc_base(mmio_base,
1298                                            port >> MV_PORT_HC_SHIFT);
1299         void __iomem *port_base;
1300         int start_port, num_ports, p, start_hc, num_hcs, hc;
1301 
1302         if (0 > port) {
1303                 start_hc = start_port = 0;
1304                 num_ports = 8;          /* shld be benign for 4 port devs */
1305                 num_hcs = 2;
1306         } else {
1307                 start_hc = port >> MV_PORT_HC_SHIFT;
1308                 start_port = port;
1309                 num_ports = num_hcs = 1;
1310         }
1311         DPRINTK("All registers for port(s) %u-%u:\n", start_port,
1312                 num_ports > 1 ? num_ports - 1 : start_port);
1313 
1314         if (NULL != pdev) {
1315                 DPRINTK("PCI config space regs:\n");
1316                 mv_dump_pci_cfg(pdev, 0x68);
1317         }
1318         DPRINTK("PCI regs:\n");
1319         mv_dump_mem(mmio_base+0xc00, 0x3c);
1320         mv_dump_mem(mmio_base+0xd00, 0x34);
1321         mv_dump_mem(mmio_base+0xf00, 0x4);
1322         mv_dump_mem(mmio_base+0x1d00, 0x6c);
1323         for (hc = start_hc; hc < start_hc + num_hcs; hc++) {
1324                 hc_base = mv_hc_base(mmio_base, hc);
1325                 DPRINTK("HC regs (HC %i):\n", hc);
1326                 mv_dump_mem(hc_base, 0x1c);
1327         }
1328         for (p = start_port; p < start_port + num_ports; p++) {
1329                 port_base = mv_port_base(mmio_base, p);
1330                 DPRINTK("EDMA regs (port %i):\n", p);
1331                 mv_dump_mem(port_base, 0x54);
1332                 DPRINTK("SATA regs (port %i):\n", p);
1333                 mv_dump_mem(port_base+0x300, 0x60);
1334         }
1335 #endif
1336 }
1337 
1338 static unsigned int mv_scr_offset(unsigned int sc_reg_in)
1339 {
1340         unsigned int ofs;
1341 
1342         switch (sc_reg_in) {
1343         case SCR_STATUS:
1344         case SCR_CONTROL:
1345         case SCR_ERROR:
1346                 ofs = SATA_STATUS + (sc_reg_in * sizeof(u32));
1347                 break;
1348         case SCR_ACTIVE:
1349                 ofs = SATA_ACTIVE;   /* active is not with the others */
1350                 break;
1351         default:
1352                 ofs = 0xffffffffU;
1353                 break;
1354         }
1355         return ofs;
1356 }
1357 
1358 static int mv_scr_read(struct ata_link *link, unsigned int sc_reg_in, u32 *val)
1359 {
1360         unsigned int ofs = mv_scr_offset(sc_reg_in);
1361 
1362         if (ofs != 0xffffffffU) {
1363                 *val = readl(mv_ap_base(link->ap) + ofs);
1364                 return 0;
1365         } else
1366                 return -EINVAL;
1367 }
1368 
1369 static int mv_scr_write(struct ata_link *link, unsigned int sc_reg_in, u32 val)
1370 {
1371         unsigned int ofs = mv_scr_offset(sc_reg_in);
1372 
1373         if (ofs != 0xffffffffU) {
1374                 void __iomem *addr = mv_ap_base(link->ap) + ofs;
1375                 struct mv_host_priv *hpriv = link->ap->host->private_data;
1376                 if (sc_reg_in == SCR_CONTROL) {
1377                         /*
1378                          * Workaround for 88SX60x1 FEr SATA#26:
1379                          *
1380                          * COMRESETs have to take care not to accidentally
1381                          * put the drive to sleep when writing SCR_CONTROL.
1382                          * Setting bits 12..15 prevents this problem.
1383                          *
1384                          * So if we see an outbound COMMRESET, set those bits.
1385                          * Ditto for the followup write that clears the reset.
1386                          *
1387                          * The proprietary driver does this for
1388                          * all chip versions, and so do we.
1389                          */
1390                         if ((val & 0xf) == 1 || (readl(addr) & 0xf) == 1)
1391                                 val |= 0xf000;
1392 
1393                         if (hpriv->hp_flags & MV_HP_FIX_LP_PHY_CTL) {
1394                                 void __iomem *lp_phy_addr =
1395                                         mv_ap_base(link->ap) + LP_PHY_CTL;
1396                                 /*
1397                                  * Set PHY speed according to SControl speed.
1398                                  */
1399                                 u32 lp_phy_val =
1400                                         LP_PHY_CTL_PIN_PU_PLL |
1401                                         LP_PHY_CTL_PIN_PU_RX  |
1402                                         LP_PHY_CTL_PIN_PU_TX;
1403 
1404                                 if ((val & 0xf0) != 0x10)
1405                                         lp_phy_val |=
1406                                                 LP_PHY_CTL_GEN_TX_3G |
1407                                                 LP_PHY_CTL_GEN_RX_3G;
1408 
1409                                 writelfl(lp_phy_val, lp_phy_addr);
1410                         }
1411                 }
1412                 writelfl(val, addr);
1413                 return 0;
1414         } else
1415                 return -EINVAL;
1416 }
1417 
1418 static void mv6_dev_config(struct ata_device *adev)
1419 {
1420         /*
1421          * Deal with Gen-II ("mv6") hardware quirks/restrictions:
1422          *
1423          * Gen-II does not support NCQ over a port multiplier
1424          *  (no FIS-based switching).
1425          */
1426         if (adev->flags & ATA_DFLAG_NCQ) {
1427                 if (sata_pmp_attached(adev->link->ap)) {
1428                         adev->flags &= ~ATA_DFLAG_NCQ;
1429                         ata_dev_info(adev,
1430                                 "NCQ disabled for command-based switching\n");
1431                 }
1432         }
1433 }
1434 
1435 static int mv_qc_defer(struct ata_queued_cmd *qc)
1436 {
1437         struct ata_link *link = qc->dev->link;
1438         struct ata_port *ap = link->ap;
1439         struct mv_port_priv *pp = ap->private_data;
1440 
1441         /*
1442          * Don't allow new commands if we're in a delayed EH state
1443          * for NCQ and/or FIS-based switching.
1444          */
1445         if (pp->pp_flags & MV_PP_FLAG_DELAYED_EH)
1446                 return ATA_DEFER_PORT;
1447 
1448         /* PIO commands need exclusive link: no other commands [DMA or PIO]
1449          * can run concurrently.
1450          * set excl_link when we want to send a PIO command in DMA mode
1451          * or a non-NCQ command in NCQ mode.
1452          * When we receive a command from that link, and there are no
1453          * outstanding commands, mark a flag to clear excl_link and let
1454          * the command go through.
1455          */
1456         if (unlikely(ap->excl_link)) {
1457                 if (link == ap->excl_link) {
1458                         if (ap->nr_active_links)
1459                                 return ATA_DEFER_PORT;
1460                         qc->flags |= ATA_QCFLAG_CLEAR_EXCL;
1461                         return 0;
1462                 } else
1463                         return ATA_DEFER_PORT;
1464         }
1465 
1466         /*
1467          * If the port is completely idle, then allow the new qc.
1468          */
1469         if (ap->nr_active_links == 0)
1470                 return 0;
1471 
1472         /*
1473          * The port is operating in host queuing mode (EDMA) with NCQ
1474          * enabled, allow multiple NCQ commands.  EDMA also allows
1475          * queueing multiple DMA commands but libata core currently
1476          * doesn't allow it.
1477          */
1478         if ((pp->pp_flags & MV_PP_FLAG_EDMA_EN) &&
1479             (pp->pp_flags & MV_PP_FLAG_NCQ_EN)) {
1480                 if (ata_is_ncq(qc->tf.protocol))
1481                         return 0;
1482                 else {
1483                         ap->excl_link = link;
1484                         return ATA_DEFER_PORT;
1485                 }
1486         }
1487 
1488         return ATA_DEFER_PORT;
1489 }
1490 
1491 static void mv_config_fbs(struct ata_port *ap, int want_ncq, int want_fbs)
1492 {
1493         struct mv_port_priv *pp = ap->private_data;
1494         void __iomem *port_mmio;
1495 
1496         u32 fiscfg,   *old_fiscfg   = &pp->cached.fiscfg;
1497         u32 ltmode,   *old_ltmode   = &pp->cached.ltmode;
1498         u32 haltcond, *old_haltcond = &pp->cached.haltcond;
1499 
1500         ltmode   = *old_ltmode & ~LTMODE_BIT8;
1501         haltcond = *old_haltcond | EDMA_ERR_DEV;
1502 
1503         if (want_fbs) {
1504                 fiscfg = *old_fiscfg | FISCFG_SINGLE_SYNC;
1505                 ltmode = *old_ltmode | LTMODE_BIT8;
1506                 if (want_ncq)
1507                         haltcond &= ~EDMA_ERR_DEV;
1508                 else
1509                         fiscfg |=  FISCFG_WAIT_DEV_ERR;
1510         } else {
1511                 fiscfg = *old_fiscfg & ~(FISCFG_SINGLE_SYNC | FISCFG_WAIT_DEV_ERR);
1512         }
1513 
1514         port_mmio = mv_ap_base(ap);
1515         mv_write_cached_reg(port_mmio + FISCFG, old_fiscfg, fiscfg);
1516         mv_write_cached_reg(port_mmio + LTMODE, old_ltmode, ltmode);
1517         mv_write_cached_reg(port_mmio + EDMA_HALTCOND, old_haltcond, haltcond);
1518 }
1519 
1520 static void mv_60x1_errata_sata25(struct ata_port *ap, int want_ncq)
1521 {
1522         struct mv_host_priv *hpriv = ap->host->private_data;
1523         u32 old, new;
1524 
1525         /* workaround for 88SX60x1 FEr SATA#25 (part 1) */
1526         old = readl(hpriv->base + GPIO_PORT_CTL);
1527         if (want_ncq)
1528                 new = old | (1 << 22);
1529         else
1530                 new = old & ~(1 << 22);
1531         if (new != old)
1532                 writel(new, hpriv->base + GPIO_PORT_CTL);
1533 }
1534 
1535 /**
1536  *      mv_bmdma_enable - set a magic bit on GEN_IIE to allow bmdma
1537  *      @ap: Port being initialized
1538  *
1539  *      There are two DMA modes on these chips:  basic DMA, and EDMA.
1540  *
1541  *      Bit-0 of the "EDMA RESERVED" register enables/disables use
1542  *      of basic DMA on the GEN_IIE versions of the chips.
1543  *
1544  *      This bit survives EDMA resets, and must be set for basic DMA
1545  *      to function, and should be cleared when EDMA is active.
1546  */
1547 static void mv_bmdma_enable_iie(struct ata_port *ap, int enable_bmdma)
1548 {
1549         struct mv_port_priv *pp = ap->private_data;
1550         u32 new, *old = &pp->cached.unknown_rsvd;
1551 
1552         if (enable_bmdma)
1553                 new = *old | 1;
1554         else
1555                 new = *old & ~1;
1556         mv_write_cached_reg(mv_ap_base(ap) + EDMA_UNKNOWN_RSVD, old, new);
1557 }
1558 
1559 /*
1560  * SOC chips have an issue whereby the HDD LEDs don't always blink
1561  * during I/O when NCQ is enabled. Enabling a special "LED blink" mode
1562  * of the SOC takes care of it, generating a steady blink rate when
1563  * any drive on the chip is active.
1564  *
1565  * Unfortunately, the blink mode is a global hardware setting for the SOC,
1566  * so we must use it whenever at least one port on the SOC has NCQ enabled.
1567  *
1568  * We turn "LED blink" off when NCQ is not in use anywhere, because the normal
1569  * LED operation works then, and provides better (more accurate) feedback.
1570  *
1571  * Note that this code assumes that an SOC never has more than one HC onboard.
1572  */
1573 static void mv_soc_led_blink_enable(struct ata_port *ap)
1574 {
1575         struct ata_host *host = ap->host;
1576         struct mv_host_priv *hpriv = host->private_data;
1577         void __iomem *hc_mmio;
1578         u32 led_ctrl;
1579 
1580         if (hpriv->hp_flags & MV_HP_QUIRK_LED_BLINK_EN)
1581                 return;
1582         hpriv->hp_flags |= MV_HP_QUIRK_LED_BLINK_EN;
1583         hc_mmio = mv_hc_base_from_port(mv_host_base(host), ap->port_no);
1584         led_ctrl = readl(hc_mmio + SOC_LED_CTRL);
1585         writel(led_ctrl | SOC_LED_CTRL_BLINK, hc_mmio + SOC_LED_CTRL);
1586 }
1587 
1588 static void mv_soc_led_blink_disable(struct ata_port *ap)
1589 {
1590         struct ata_host *host = ap->host;
1591         struct mv_host_priv *hpriv = host->private_data;
1592         void __iomem *hc_mmio;
1593         u32 led_ctrl;
1594         unsigned int port;
1595 
1596         if (!(hpriv->hp_flags & MV_HP_QUIRK_LED_BLINK_EN))
1597                 return;
1598 
1599         /* disable led-blink only if no ports are using NCQ */
1600         for (port = 0; port < hpriv->n_ports; port++) {
1601                 struct ata_port *this_ap = host->ports[port];
1602                 struct mv_port_priv *pp = this_ap->private_data;
1603 
1604                 if (pp->pp_flags & MV_PP_FLAG_NCQ_EN)
1605                         return;
1606         }
1607 
1608         hpriv->hp_flags &= ~MV_HP_QUIRK_LED_BLINK_EN;
1609         hc_mmio = mv_hc_base_from_port(mv_host_base(host), ap->port_no);
1610         led_ctrl = readl(hc_mmio + SOC_LED_CTRL);
1611         writel(led_ctrl & ~SOC_LED_CTRL_BLINK, hc_mmio + SOC_LED_CTRL);
1612 }
1613 
1614 static void mv_edma_cfg(struct ata_port *ap, int want_ncq, int want_edma)
1615 {
1616         u32 cfg;
1617         struct mv_port_priv *pp    = ap->private_data;
1618         struct mv_host_priv *hpriv = ap->host->private_data;
1619         void __iomem *port_mmio    = mv_ap_base(ap);
1620 
1621         /* set up non-NCQ EDMA configuration */
1622         cfg = EDMA_CFG_Q_DEPTH;         /* always 0x1f for *all* chips */
1623         pp->pp_flags &=
1624           ~(MV_PP_FLAG_FBS_EN | MV_PP_FLAG_NCQ_EN | MV_PP_FLAG_FAKE_ATA_BUSY);
1625 
1626         if (IS_GEN_I(hpriv))
1627                 cfg |= (1 << 8);        /* enab config burst size mask */
1628 
1629         else if (IS_GEN_II(hpriv)) {
1630                 cfg |= EDMA_CFG_RD_BRST_EXT | EDMA_CFG_WR_BUFF_LEN;
1631                 mv_60x1_errata_sata25(ap, want_ncq);
1632 
1633         } else if (IS_GEN_IIE(hpriv)) {
1634                 int want_fbs = sata_pmp_attached(ap);
1635                 /*
1636                  * Possible future enhancement:
1637                  *
1638                  * The chip can use FBS with non-NCQ, if we allow it,
1639                  * But first we need to have the error handling in place
1640                  * for this mode (datasheet section 7.3.15.4.2.3).
1641                  * So disallow non-NCQ FBS for now.
1642                  */
1643                 want_fbs &= want_ncq;
1644 
1645                 mv_config_fbs(ap, want_ncq, want_fbs);
1646 
1647                 if (want_fbs) {
1648                         pp->pp_flags |= MV_PP_FLAG_FBS_EN;
1649                         cfg |= EDMA_CFG_EDMA_FBS; /* FIS-based switching */
1650                 }
1651 
1652                 cfg |= (1 << 23);       /* do not mask PM field in rx'd FIS */
1653                 if (want_edma) {
1654                         cfg |= (1 << 22); /* enab 4-entry host queue cache */
1655                         if (!IS_SOC(hpriv))
1656                                 cfg |= (1 << 18); /* enab early completion */
1657                 }
1658                 if (hpriv->hp_flags & MV_HP_CUT_THROUGH)
1659                         cfg |= (1 << 17); /* enab cut-thru (dis stor&forwrd) */
1660                 mv_bmdma_enable_iie(ap, !want_edma);
1661 
1662                 if (IS_SOC(hpriv)) {
1663                         if (want_ncq)
1664                                 mv_soc_led_blink_enable(ap);
1665                         else
1666                                 mv_soc_led_blink_disable(ap);
1667                 }
1668         }
1669 
1670         if (want_ncq) {
1671                 cfg |= EDMA_CFG_NCQ;
1672                 pp->pp_flags |=  MV_PP_FLAG_NCQ_EN;
1673         }
1674 
1675         writelfl(cfg, port_mmio + EDMA_CFG);
1676 }
1677 
1678 static void mv_port_free_dma_mem(struct ata_port *ap)
1679 {
1680         struct mv_host_priv *hpriv = ap->host->private_data;
1681         struct mv_port_priv *pp = ap->private_data;
1682         int tag;
1683 
1684         if (pp->crqb) {
1685                 dma_pool_free(hpriv->crqb_pool, pp->crqb, pp->crqb_dma);
1686                 pp->crqb = NULL;
1687         }
1688         if (pp->crpb) {
1689                 dma_pool_free(hpriv->crpb_pool, pp->crpb, pp->crpb_dma);
1690                 pp->crpb = NULL;
1691         }
1692         /*
1693          * For GEN_I, there's no NCQ, so we have only a single sg_tbl.
1694          * For later hardware, we have one unique sg_tbl per NCQ tag.
1695          */
1696         for (tag = 0; tag < MV_MAX_Q_DEPTH; ++tag) {
1697                 if (pp->sg_tbl[tag]) {
1698                         if (tag == 0 || !IS_GEN_I(hpriv))
1699                                 dma_pool_free(hpriv->sg_tbl_pool,
1700                                               pp->sg_tbl[tag],
1701                                               pp->sg_tbl_dma[tag]);
1702                         pp->sg_tbl[tag] = NULL;
1703                 }
1704         }
1705 }
1706 
1707 /**
1708  *      mv_port_start - Port specific init/start routine.
1709  *      @ap: ATA channel to manipulate
1710  *
1711  *      Allocate and point to DMA memory, init port private memory,
1712  *      zero indices.
1713  *
1714  *      LOCKING:
1715  *      Inherited from caller.
1716  */
1717 static int mv_port_start(struct ata_port *ap)
1718 {
1719         struct device *dev = ap->host->dev;
1720         struct mv_host_priv *hpriv = ap->host->private_data;
1721         struct mv_port_priv *pp;
1722         unsigned long flags;
1723         int tag;
1724 
1725         pp = devm_kzalloc(dev, sizeof(*pp), GFP_KERNEL);
1726         if (!pp)
1727                 return -ENOMEM;
1728         ap->private_data = pp;
1729 
1730         pp->crqb = dma_pool_alloc(hpriv->crqb_pool, GFP_KERNEL, &pp->crqb_dma);
1731         if (!pp->crqb)
1732                 return -ENOMEM;
1733         memset(pp->crqb, 0, MV_CRQB_Q_SZ);
1734 
1735         pp->crpb = dma_pool_alloc(hpriv->crpb_pool, GFP_KERNEL, &pp->crpb_dma);
1736         if (!pp->crpb)
1737                 goto out_port_free_dma_mem;
1738         memset(pp->crpb, 0, MV_CRPB_Q_SZ);
1739 
1740         /* 6041/6081 Rev. "C0" (and newer) are okay with async notify */
1741         if (hpriv->hp_flags & MV_HP_ERRATA_60X1C0)
1742                 ap->flags |= ATA_FLAG_AN;
1743         /*
1744          * For GEN_I, there's no NCQ, so we only allocate a single sg_tbl.
1745          * For later hardware, we need one unique sg_tbl per NCQ tag.
1746          */
1747         for (tag = 0; tag < MV_MAX_Q_DEPTH; ++tag) {
1748                 if (tag == 0 || !IS_GEN_I(hpriv)) {
1749                         pp->sg_tbl[tag] = dma_pool_alloc(hpriv->sg_tbl_pool,
1750                                               GFP_KERNEL, &pp->sg_tbl_dma[tag]);
1751                         if (!pp->sg_tbl[tag])
1752                                 goto out_port_free_dma_mem;
1753                 } else {
1754                         pp->sg_tbl[tag]     = pp->sg_tbl[0];
1755                         pp->sg_tbl_dma[tag] = pp->sg_tbl_dma[0];
1756                 }
1757         }
1758 
1759         spin_lock_irqsave(ap->lock, flags);
1760         mv_save_cached_regs(ap);
1761         mv_edma_cfg(ap, 0, 0);
1762         spin_unlock_irqrestore(ap->lock, flags);
1763 
1764         return 0;
1765 
1766 out_port_free_dma_mem:
1767         mv_port_free_dma_mem(ap);
1768         return -ENOMEM;
1769 }
1770 
1771 /**
1772  *      mv_port_stop - Port specific cleanup/stop routine.
1773  *      @ap: ATA channel to manipulate
1774  *
1775  *      Stop DMA, cleanup port memory.
1776  *
1777  *      LOCKING:
1778  *      This routine uses the host lock to protect the DMA stop.
1779  */
1780 static void mv_port_stop(struct ata_port *ap)
1781 {
1782         unsigned long flags;
1783 
1784         spin_lock_irqsave(ap->lock, flags);
1785         mv_stop_edma(ap);
1786         mv_enable_port_irqs(ap, 0);
1787         spin_unlock_irqrestore(ap->lock, flags);
1788         mv_port_free_dma_mem(ap);
1789 }
1790 
1791 /**
1792  *      mv_fill_sg - Fill out the Marvell ePRD (scatter gather) entries
1793  *      @qc: queued command whose SG list to source from
1794  *
1795  *      Populate the SG list and mark the last entry.
1796  *
1797  *      LOCKING:
1798  *      Inherited from caller.
1799  */
1800 static void mv_fill_sg(struct ata_queued_cmd *qc)
1801 {
1802         struct mv_port_priv *pp = qc->ap->private_data;
1803         struct scatterlist *sg;
1804         struct mv_sg *mv_sg, *last_sg = NULL;
1805         unsigned int si;
1806 
1807         mv_sg = pp->sg_tbl[qc->tag];
1808         for_each_sg(qc->sg, sg, qc->n_elem, si) {
1809                 dma_addr_t addr = sg_dma_address(sg);
1810                 u32 sg_len = sg_dma_len(sg);
1811 
1812                 while (sg_len) {
1813                         u32 offset = addr & 0xffff;
1814                         u32 len = sg_len;
1815 
1816                         if (offset + len > 0x10000)
1817                                 len = 0x10000 - offset;
1818 
1819                         mv_sg->addr = cpu_to_le32(addr & 0xffffffff);
1820                         mv_sg->addr_hi = cpu_to_le32((addr >> 16) >> 16);
1821                         mv_sg->flags_size = cpu_to_le32(len & 0xffff);
1822                         mv_sg->reserved = 0;
1823 
1824                         sg_len -= len;
1825                         addr += len;
1826 
1827                         last_sg = mv_sg;
1828                         mv_sg++;
1829                 }
1830         }
1831 
1832         if (likely(last_sg))
1833                 last_sg->flags_size |= cpu_to_le32(EPRD_FLAG_END_OF_TBL);
1834         mb(); /* ensure data structure is visible to the chipset */
1835 }
1836 
1837 static void mv_crqb_pack_cmd(__le16 *cmdw, u8 data, u8 addr, unsigned last)
1838 {
1839         u16 tmp = data | (addr << CRQB_CMD_ADDR_SHIFT) | CRQB_CMD_CS |
1840                 (last ? CRQB_CMD_LAST : 0);
1841         *cmdw = cpu_to_le16(tmp);
1842 }
1843 
1844 /**
1845  *      mv_sff_irq_clear - Clear hardware interrupt after DMA.
1846  *      @ap: Port associated with this ATA transaction.
1847  *
1848  *      We need this only for ATAPI bmdma transactions,
1849  *      as otherwise we experience spurious interrupts
1850  *      after libata-sff handles the bmdma interrupts.
1851  */
1852 static void mv_sff_irq_clear(struct ata_port *ap)
1853 {
1854         mv_clear_and_enable_port_irqs(ap, mv_ap_base(ap), ERR_IRQ);
1855 }
1856 
1857 /**
1858  *      mv_check_atapi_dma - Filter ATAPI cmds which are unsuitable for DMA.
1859  *      @qc: queued command to check for chipset/DMA compatibility.
1860  *
1861  *      The bmdma engines cannot handle speculative data sizes
1862  *      (bytecount under/over flow).  So only allow DMA for
1863  *      data transfer commands with known data sizes.
1864  *
1865  *      LOCKING:
1866  *      Inherited from caller.
1867  */
1868 static int mv_check_atapi_dma(struct ata_queued_cmd *qc)
1869 {
1870         struct scsi_cmnd *scmd = qc->scsicmd;
1871 
1872         if (scmd) {
1873                 switch (scmd->cmnd[0]) {
1874                 case READ_6:
1875                 case READ_10:
1876                 case READ_12:
1877                 case WRITE_6:
1878                 case WRITE_10:
1879                 case WRITE_12:
1880                 case GPCMD_READ_CD:
1881                 case GPCMD_SEND_DVD_STRUCTURE:
1882                 case GPCMD_SEND_CUE_SHEET:
1883                         return 0; /* DMA is safe */
1884                 }
1885         }
1886         return -EOPNOTSUPP; /* use PIO instead */
1887 }
1888 
1889 /**
1890  *      mv_bmdma_setup - Set up BMDMA transaction
1891  *      @qc: queued command to prepare DMA for.
1892  *
1893  *      LOCKING:
1894  *      Inherited from caller.
1895  */
1896 static void mv_bmdma_setup(struct ata_queued_cmd *qc)
1897 {
1898         struct ata_port *ap = qc->ap;
1899         void __iomem *port_mmio = mv_ap_base(ap);
1900         struct mv_port_priv *pp = ap->private_data;
1901 
1902         mv_fill_sg(qc);
1903 
1904         /* clear all DMA cmd bits */
1905         writel(0, port_mmio + BMDMA_CMD);
1906 
1907         /* load PRD table addr. */
1908         writel((pp->sg_tbl_dma[qc->tag] >> 16) >> 16,
1909                 port_mmio + BMDMA_PRD_HIGH);
1910         writelfl(pp->sg_tbl_dma[qc->tag],
1911                 port_mmio + BMDMA_PRD_LOW);
1912 
1913         /* issue r/w command */
1914         ap->ops->sff_exec_command(ap, &qc->tf);
1915 }
1916 
1917 /**
1918  *      mv_bmdma_start - Start a BMDMA transaction
1919  *      @qc: queued command to start DMA on.
1920  *
1921  *      LOCKING:
1922  *      Inherited from caller.
1923  */
1924 static void mv_bmdma_start(struct ata_queued_cmd *qc)
1925 {
1926         struct ata_port *ap = qc->ap;
1927         void __iomem *port_mmio = mv_ap_base(ap);
1928         unsigned int rw = (qc->tf.flags & ATA_TFLAG_WRITE);
1929         u32 cmd = (rw ? 0 : ATA_DMA_WR) | ATA_DMA_START;
1930 
1931         /* start host DMA transaction */
1932         writelfl(cmd, port_mmio + BMDMA_CMD);
1933 }
1934 
1935 /**
1936  *      mv_bmdma_stop - Stop BMDMA transfer
1937  *      @qc: queued command to stop DMA on.
1938  *
1939  *      Clears the ATA_DMA_START flag in the bmdma control register
1940  *
1941  *      LOCKING:
1942  *      Inherited from caller.
1943  */
1944 static void mv_bmdma_stop_ap(struct ata_port *ap)
1945 {
1946         void __iomem *port_mmio = mv_ap_base(ap);
1947         u32 cmd;
1948 
1949         /* clear start/stop bit */
1950         cmd = readl(port_mmio + BMDMA_CMD);
1951         if (cmd & ATA_DMA_START) {
1952                 cmd &= ~ATA_DMA_START;
1953                 writelfl(cmd, port_mmio + BMDMA_CMD);
1954 
1955                 /* one-PIO-cycle guaranteed wait, per spec, for HDMA1:0 transition */
1956                 ata_sff_dma_pause(ap);
1957         }
1958 }
1959 
1960 static void mv_bmdma_stop(struct ata_queued_cmd *qc)
1961 {
1962         mv_bmdma_stop_ap(qc->ap);
1963 }
1964 
1965 /**
1966  *      mv_bmdma_status - Read BMDMA status
1967  *      @ap: port for which to retrieve DMA status.
1968  *
1969  *      Read and return equivalent of the sff BMDMA status register.
1970  *
1971  *      LOCKING:
1972  *      Inherited from caller.
1973  */
1974 static u8 mv_bmdma_status(struct ata_port *ap)
1975 {
1976         void __iomem *port_mmio = mv_ap_base(ap);
1977         u32 reg, status;
1978 
1979         /*
1980          * Other bits are valid only if ATA_DMA_ACTIVE==0,
1981          * and the ATA_DMA_INTR bit doesn't exist.
1982          */
1983         reg = readl(port_mmio + BMDMA_STATUS);
1984         if (reg & ATA_DMA_ACTIVE)
1985                 status = ATA_DMA_ACTIVE;
1986         else if (reg & ATA_DMA_ERR)
1987                 status = (reg & ATA_DMA_ERR) | ATA_DMA_INTR;
1988         else {
1989                 /*
1990                  * Just because DMA_ACTIVE is 0 (DMA completed),
1991                  * this does _not_ mean the device is "done".
1992                  * So we should not yet be signalling ATA_DMA_INTR
1993                  * in some cases.  Eg. DSM/TRIM, and perhaps others.
1994                  */
1995                 mv_bmdma_stop_ap(ap);
1996                 if (ioread8(ap->ioaddr.altstatus_addr) & ATA_BUSY)
1997                         status = 0;
1998                 else
1999                         status = ATA_DMA_INTR;
2000         }
2001         return status;
2002 }
2003 
2004 static void mv_rw_multi_errata_sata24(struct ata_queued_cmd *qc)
2005 {
2006         struct ata_taskfile *tf = &qc->tf;
2007         /*
2008          * Workaround for 88SX60x1 FEr SATA#24.
2009          *
2010          * Chip may corrupt WRITEs if multi_count >= 4kB.
2011          * Note that READs are unaffected.
2012          *
2013          * It's not clear if this errata really means "4K bytes",
2014          * or if it always happens for multi_count > 7
2015          * regardless of device sector_size.
2016          *
2017          * So, for safety, any write with multi_count > 7
2018          * gets converted here into a regular PIO write instead:
2019          */
2020         if ((tf->flags & ATA_TFLAG_WRITE) && is_multi_taskfile(tf)) {
2021                 if (qc->dev->multi_count > 7) {
2022                         switch (tf->command) {
2023                         case ATA_CMD_WRITE_MULTI:
2024                                 tf->command = ATA_CMD_PIO_WRITE;
2025                                 break;
2026                         case ATA_CMD_WRITE_MULTI_FUA_EXT:
2027                                 tf->flags &= ~ATA_TFLAG_FUA; /* ugh */
2028                                 /* fall through */
2029                         case ATA_CMD_WRITE_MULTI_EXT:
2030                                 tf->command = ATA_CMD_PIO_WRITE_EXT;
2031                                 break;
2032                         }
2033                 }
2034         }
2035 }
2036 
2037 /**
2038  *      mv_qc_prep - Host specific command preparation.
2039  *      @qc: queued command to prepare
2040  *
2041  *      This routine simply redirects to the general purpose routine
2042  *      if command is not DMA.  Else, it handles prep of the CRQB
2043  *      (command request block), does some sanity checking, and calls
2044  *      the SG load routine.
2045  *
2046  *      LOCKING:
2047  *      Inherited from caller.
2048  */
2049 static void mv_qc_prep(struct ata_queued_cmd *qc)
2050 {
2051         struct ata_port *ap = qc->ap;
2052         struct mv_port_priv *pp = ap->private_data;
2053         __le16 *cw;
2054         struct ata_taskfile *tf = &qc->tf;
2055         u16 flags = 0;
2056         unsigned in_index;
2057 
2058         switch (tf->protocol) {
2059         case ATA_PROT_DMA:
2060                 if (tf->command == ATA_CMD_DSM)
2061                         return;
2062                 /* fall-thru */
2063         case ATA_PROT_NCQ:
2064                 break;  /* continue below */
2065         case ATA_PROT_PIO:
2066                 mv_rw_multi_errata_sata24(qc);
2067                 return;
2068         default:
2069                 return;
2070         }
2071 
2072         /* Fill in command request block
2073          */
2074         if (!(tf->flags & ATA_TFLAG_WRITE))
2075                 flags |= CRQB_FLAG_READ;
2076         WARN_ON(MV_MAX_Q_DEPTH <= qc->tag);
2077         flags |= qc->tag << CRQB_TAG_SHIFT;
2078         flags |= (qc->dev->link->pmp & 0xf) << CRQB_PMP_SHIFT;
2079 
2080         /* get current queue index from software */
2081         in_index = pp->req_idx;
2082 
2083         pp->crqb[in_index].sg_addr =
2084                 cpu_to_le32(pp->sg_tbl_dma[qc->tag] & 0xffffffff);
2085         pp->crqb[in_index].sg_addr_hi =
2086                 cpu_to_le32((pp->sg_tbl_dma[qc->tag] >> 16) >> 16);
2087         pp->crqb[in_index].ctrl_flags = cpu_to_le16(flags);
2088 
2089         cw = &pp->crqb[in_index].ata_cmd[0];
2090 
2091         /* Sadly, the CRQB cannot accommodate all registers--there are
2092          * only 11 bytes...so we must pick and choose required
2093          * registers based on the command.  So, we drop feature and
2094          * hob_feature for [RW] DMA commands, but they are needed for
2095          * NCQ.  NCQ will drop hob_nsect, which is not needed there
2096          * (nsect is used only for the tag; feat/hob_feat hold true nsect).
2097          */
2098         switch (tf->command) {
2099         case ATA_CMD_READ:
2100         case ATA_CMD_READ_EXT:
2101         case ATA_CMD_WRITE:
2102         case ATA_CMD_WRITE_EXT:
2103         case ATA_CMD_WRITE_FUA_EXT:
2104                 mv_crqb_pack_cmd(cw++, tf->hob_nsect, ATA_REG_NSECT, 0);
2105                 break;
2106         case ATA_CMD_FPDMA_READ:
2107         case ATA_CMD_FPDMA_WRITE:
2108                 mv_crqb_pack_cmd(cw++, tf->hob_feature, ATA_REG_FEATURE, 0);
2109                 mv_crqb_pack_cmd(cw++, tf->feature, ATA_REG_FEATURE, 0);
2110                 break;
2111         default:
2112                 /* The only other commands EDMA supports in non-queued and
2113                  * non-NCQ mode are: [RW] STREAM DMA and W DMA FUA EXT, none
2114                  * of which are defined/used by Linux.  If we get here, this
2115                  * driver needs work.
2116                  *
2117                  * FIXME: modify libata to give qc_prep a return value and
2118                  * return error here.
2119                  */
2120                 BUG_ON(tf->command);
2121                 break;
2122         }
2123         mv_crqb_pack_cmd(cw++, tf->nsect, ATA_REG_NSECT, 0);
2124         mv_crqb_pack_cmd(cw++, tf->hob_lbal, ATA_REG_LBAL, 0);
2125         mv_crqb_pack_cmd(cw++, tf->lbal, ATA_REG_LBAL, 0);
2126         mv_crqb_pack_cmd(cw++, tf->hob_lbam, ATA_REG_LBAM, 0);
2127         mv_crqb_pack_cmd(cw++, tf->lbam, ATA_REG_LBAM, 0);
2128         mv_crqb_pack_cmd(cw++, tf->hob_lbah, ATA_REG_LBAH, 0);
2129         mv_crqb_pack_cmd(cw++, tf->lbah, ATA_REG_LBAH, 0);
2130         mv_crqb_pack_cmd(cw++, tf->device, ATA_REG_DEVICE, 0);
2131         mv_crqb_pack_cmd(cw++, tf->command, ATA_REG_CMD, 1);    /* last */
2132 
2133         if (!(qc->flags & ATA_QCFLAG_DMAMAP))
2134                 return;
2135         mv_fill_sg(qc);
2136 }
2137 
2138 /**
2139  *      mv_qc_prep_iie - Host specific command preparation.
2140  *      @qc: queued command to prepare
2141  *
2142  *      This routine simply redirects to the general purpose routine
2143  *      if command is not DMA.  Else, it handles prep of the CRQB
2144  *      (command request block), does some sanity checking, and calls
2145  *      the SG load routine.
2146  *
2147  *      LOCKING:
2148  *      Inherited from caller.
2149  */
2150 static void mv_qc_prep_iie(struct ata_queued_cmd *qc)
2151 {
2152         struct ata_port *ap = qc->ap;
2153         struct mv_port_priv *pp = ap->private_data;
2154         struct mv_crqb_iie *crqb;
2155         struct ata_taskfile *tf = &qc->tf;
2156         unsigned in_index;
2157         u32 flags = 0;
2158 
2159         if ((tf->protocol != ATA_PROT_DMA) &&
2160             (tf->protocol != ATA_PROT_NCQ))
2161                 return;
2162         if (tf->command == ATA_CMD_DSM)
2163                 return;  /* use bmdma for this */
2164 
2165         /* Fill in Gen IIE command request block */
2166         if (!(tf->flags & ATA_TFLAG_WRITE))
2167                 flags |= CRQB_FLAG_READ;
2168 
2169         WARN_ON(MV_MAX_Q_DEPTH <= qc->tag);
2170         flags |= qc->tag << CRQB_TAG_SHIFT;
2171         flags |= qc->tag << CRQB_HOSTQ_SHIFT;
2172         flags |= (qc->dev->link->pmp & 0xf) << CRQB_PMP_SHIFT;
2173 
2174         /* get current queue index from software */
2175         in_index = pp->req_idx;
2176 
2177         crqb = (struct mv_crqb_iie *) &pp->crqb[in_index];
2178         crqb->addr = cpu_to_le32(pp->sg_tbl_dma[qc->tag] & 0xffffffff);
2179         crqb->addr_hi = cpu_to_le32((pp->sg_tbl_dma[qc->tag] >> 16) >> 16);
2180         crqb->flags = cpu_to_le32(flags);
2181 
2182         crqb->ata_cmd[0] = cpu_to_le32(
2183                         (tf->command << 16) |
2184                         (tf->feature << 24)
2185                 );
2186         crqb->ata_cmd[1] = cpu_to_le32(
2187                         (tf->lbal << 0) |
2188                         (tf->lbam << 8) |
2189                         (tf->lbah << 16) |
2190                         (tf->device << 24)
2191                 );
2192         crqb->ata_cmd[2] = cpu_to_le32(
2193                         (tf->hob_lbal << 0) |
2194                         (tf->hob_lbam << 8) |
2195                         (tf->hob_lbah << 16) |
2196                         (tf->hob_feature << 24)
2197                 );
2198         crqb->ata_cmd[3] = cpu_to_le32(
2199                         (tf->nsect << 0) |
2200                         (tf->hob_nsect << 8)
2201                 );
2202 
2203         if (!(qc->flags & ATA_QCFLAG_DMAMAP))
2204                 return;
2205         mv_fill_sg(qc);
2206 }
2207 
2208 /**
2209  *      mv_sff_check_status - fetch device status, if valid
2210  *      @ap: ATA port to fetch status from
2211  *
2212  *      When using command issue via mv_qc_issue_fis(),
2213  *      the initial ATA_BUSY state does not show up in the
2214  *      ATA status (shadow) register.  This can confuse libata!
2215  *
2216  *      So we have a hook here to fake ATA_BUSY for that situation,
2217  *      until the first time a BUSY, DRQ, or ERR bit is seen.
2218  *
2219  *      The rest of the time, it simply returns the ATA status register.
2220  */
2221 static u8 mv_sff_check_status(struct ata_port *ap)
2222 {
2223         u8 stat = ioread8(ap->ioaddr.status_addr);
2224         struct mv_port_priv *pp = ap->private_data;
2225 
2226         if (pp->pp_flags & MV_PP_FLAG_FAKE_ATA_BUSY) {
2227                 if (stat & (ATA_BUSY | ATA_DRQ | ATA_ERR))
2228                         pp->pp_flags &= ~MV_PP_FLAG_FAKE_ATA_BUSY;
2229                 else
2230                         stat = ATA_BUSY;
2231         }
2232         return stat;
2233 }
2234 
2235 /**
2236  *      mv_send_fis - Send a FIS, using the "Vendor-Unique FIS" register
2237  *      @fis: fis to be sent
2238  *      @nwords: number of 32-bit words in the fis
2239  */
2240 static unsigned int mv_send_fis(struct ata_port *ap, u32 *fis, int nwords)
2241 {
2242         void __iomem *port_mmio = mv_ap_base(ap);
2243         u32 ifctl, old_ifctl, ifstat;
2244         int i, timeout = 200, final_word = nwords - 1;
2245 
2246         /* Initiate FIS transmission mode */
2247         old_ifctl = readl(port_mmio + SATA_IFCTL);
2248         ifctl = 0x100 | (old_ifctl & 0xf);
2249         writelfl(ifctl, port_mmio + SATA_IFCTL);
2250 
2251         /* Send all words of the FIS except for the final word */
2252         for (i = 0; i < final_word; ++i)
2253                 writel(fis[i], port_mmio + VENDOR_UNIQUE_FIS);
2254 
2255         /* Flag end-of-transmission, and then send the final word */
2256         writelfl(ifctl | 0x200, port_mmio + SATA_IFCTL);
2257         writelfl(fis[final_word], port_mmio + VENDOR_UNIQUE_FIS);
2258 
2259         /*
2260          * Wait for FIS transmission to complete.
2261          * This typically takes just a single iteration.
2262          */
2263         do {
2264                 ifstat = readl(port_mmio + SATA_IFSTAT);
2265         } while (!(ifstat & 0x1000) && --timeout);
2266 
2267         /* Restore original port configuration */
2268         writelfl(old_ifctl, port_mmio + SATA_IFCTL);
2269 
2270         /* See if it worked */
2271         if ((ifstat & 0x3000) != 0x1000) {
2272                 ata_port_warn(ap, "%s transmission error, ifstat=%08x\n",
2273                               __func__, ifstat);
2274                 return AC_ERR_OTHER;
2275         }
2276         return 0;
2277 }
2278 
2279 /**
2280  *      mv_qc_issue_fis - Issue a command directly as a FIS
2281  *      @qc: queued command to start
2282  *
2283  *      Note that the ATA shadow registers are not updated
2284  *      after command issue, so the device will appear "READY"
2285  *      if polled, even while it is BUSY processing the command.
2286  *
2287  *      So we use a status hook to fake ATA_BUSY until the drive changes state.
2288  *
2289  *      Note: we don't get updated shadow regs on *completion*
2290  *      of non-data commands. So avoid sending them via this function,
2291  *      as they will appear to have completed immediately.
2292  *
2293  *      GEN_IIE has special registers that we could get the result tf from,
2294  *      but earlier chipsets do not.  For now, we ignore those registers.
2295  */
2296 static unsigned int mv_qc_issue_fis(struct ata_queued_cmd *qc)
2297 {
2298         struct ata_port *ap = qc->ap;
2299         struct mv_port_priv *pp = ap->private_data;
2300         struct ata_link *link = qc->dev->link;
2301         u32 fis[5];
2302         int err = 0;
2303 
2304         ata_tf_to_fis(&qc->tf, link->pmp, 1, (void *)fis);
2305         err = mv_send_fis(ap, fis, ARRAY_SIZE(fis));
2306         if (err)
2307                 return err;
2308 
2309         switch (qc->tf.protocol) {
2310         case ATAPI_PROT_PIO:
2311                 pp->pp_flags |= MV_PP_FLAG_FAKE_ATA_BUSY;
2312                 /* fall through */
2313         case ATAPI_PROT_NODATA:
2314                 ap->hsm_task_state = HSM_ST_FIRST;
2315                 break;
2316         case ATA_PROT_PIO:
2317                 pp->pp_flags |= MV_PP_FLAG_FAKE_ATA_BUSY;
2318                 if (qc->tf.flags & ATA_TFLAG_WRITE)
2319                         ap->hsm_task_state = HSM_ST_FIRST;
2320                 else
2321                         ap->hsm_task_state = HSM_ST;
2322                 break;
2323         default:
2324                 ap->hsm_task_state = HSM_ST_LAST;
2325                 break;
2326         }
2327 
2328         if (qc->tf.flags & ATA_TFLAG_POLLING)
2329                 ata_sff_queue_pio_task(link, 0);
2330         return 0;
2331 }
2332 
2333 /**
2334  *      mv_qc_issue - Initiate a command to the host
2335  *      @qc: queued command to start
2336  *
2337  *      This routine simply redirects to the general purpose routine
2338  *      if command is not DMA.  Else, it sanity checks our local
2339  *      caches of the request producer/consumer indices then enables
2340  *      DMA and bumps the request producer index.
2341  *
2342  *      LOCKING:
2343  *      Inherited from caller.
2344  */
2345 static unsigned int mv_qc_issue(struct ata_queued_cmd *qc)
2346 {
2347         static int limit_warnings = 10;
2348         struct ata_port *ap = qc->ap;
2349         void __iomem *port_mmio = mv_ap_base(ap);
2350         struct mv_port_priv *pp = ap->private_data;
2351         u32 in_index;
2352         unsigned int port_irqs;
2353 
2354         pp->pp_flags &= ~MV_PP_FLAG_FAKE_ATA_BUSY; /* paranoia */
2355 
2356         switch (qc->tf.protocol) {
2357         case ATA_PROT_DMA:
2358                 if (qc->tf.command == ATA_CMD_DSM) {
2359                         if (!ap->ops->bmdma_setup)  /* no bmdma on GEN_I */
2360                                 return AC_ERR_OTHER;
2361                         break;  /* use bmdma for this */
2362                 }
2363                 /* fall thru */
2364         case ATA_PROT_NCQ:
2365                 mv_start_edma(ap, port_mmio, pp, qc->tf.protocol);
2366                 pp->req_idx = (pp->req_idx + 1) & MV_MAX_Q_DEPTH_MASK;
2367                 in_index = pp->req_idx << EDMA_REQ_Q_PTR_SHIFT;
2368 
2369                 /* Write the request in pointer to kick the EDMA to life */
2370                 writelfl((pp->crqb_dma & EDMA_REQ_Q_BASE_LO_MASK) | in_index,
2371                                         port_mmio + EDMA_REQ_Q_IN_PTR);
2372                 return 0;
2373 
2374         case ATA_PROT_PIO:
2375                 /*
2376                  * Errata SATA#16, SATA#24: warn if multiple DRQs expected.
2377                  *
2378                  * Someday, we might implement special polling workarounds
2379                  * for these, but it all seems rather unnecessary since we
2380                  * normally use only DMA for commands which transfer more
2381                  * than a single block of data.
2382                  *
2383                  * Much of the time, this could just work regardless.
2384                  * So for now, just log the incident, and allow the attempt.
2385                  */
2386                 if (limit_warnings > 0 && (qc->nbytes / qc->sect_size) > 1) {
2387                         --limit_warnings;
2388                         ata_link_warn(qc->dev->link, DRV_NAME
2389                                       ": attempting PIO w/multiple DRQ: "
2390                                       "this may fail due to h/w errata\n");
2391                 }
2392                 /* drop through */
2393         case ATA_PROT_NODATA:
2394         case ATAPI_PROT_PIO:
2395         case ATAPI_PROT_NODATA:
2396                 if (ap->flags & ATA_FLAG_PIO_POLLING)
2397                         qc->tf.flags |= ATA_TFLAG_POLLING;
2398                 break;
2399         }
2400 
2401         if (qc->tf.flags & ATA_TFLAG_POLLING)
2402                 port_irqs = ERR_IRQ;    /* mask device interrupt when polling */
2403         else
2404                 port_irqs = ERR_IRQ | DONE_IRQ; /* unmask all interrupts */
2405 
2406         /*
2407          * We're about to send a non-EDMA capable command to the
2408          * port.  Turn off EDMA so there won't be problems accessing
2409          * shadow block, etc registers.
2410          */
2411         mv_stop_edma(ap);
2412         mv_clear_and_enable_port_irqs(ap, mv_ap_base(ap), port_irqs);
2413         mv_pmp_select(ap, qc->dev->link->pmp);
2414 
2415         if (qc->tf.command == ATA_CMD_READ_LOG_EXT) {
2416                 struct mv_host_priv *hpriv = ap->host->private_data;
2417                 /*
2418                  * Workaround for 88SX60x1 FEr SATA#25 (part 2).
2419                  *
2420                  * After any NCQ error, the READ_LOG_EXT command
2421                  * from libata-eh *must* use mv_qc_issue_fis().
2422                  * Otherwise it might fail, due to chip errata.
2423                  *
2424                  * Rather than special-case it, we'll just *always*
2425                  * use this method here for READ_LOG_EXT, making for
2426                  * easier testing.
2427                  */
2428                 if (IS_GEN_II(hpriv))
2429                         return mv_qc_issue_fis(qc);
2430         }
2431         return ata_bmdma_qc_issue(qc);
2432 }
2433 
2434 static struct ata_queued_cmd *mv_get_active_qc(struct ata_port *ap)
2435 {
2436         struct mv_port_priv *pp = ap->private_data;
2437         struct ata_queued_cmd *qc;
2438 
2439         if (pp->pp_flags & MV_PP_FLAG_NCQ_EN)
2440                 return NULL;
2441         qc = ata_qc_from_tag(ap, ap->link.active_tag);
2442         if (qc && !(qc->tf.flags & ATA_TFLAG_POLLING))
2443                 return qc;
2444         return NULL;
2445 }
2446 
2447 static void mv_pmp_error_handler(struct ata_port *ap)
2448 {
2449         unsigned int pmp, pmp_map;
2450         struct mv_port_priv *pp = ap->private_data;
2451 
2452         if (pp->pp_flags & MV_PP_FLAG_DELAYED_EH) {
2453                 /*
2454                  * Perform NCQ error analysis on failed PMPs
2455                  * before we freeze the port entirely.
2456                  *
2457                  * The failed PMPs are marked earlier by mv_pmp_eh_prep().
2458                  */
2459                 pmp_map = pp->delayed_eh_pmp_map;
2460                 pp->pp_flags &= ~MV_PP_FLAG_DELAYED_EH;
2461                 for (pmp = 0; pmp_map != 0; pmp++) {
2462                         unsigned int this_pmp = (1 << pmp);
2463                         if (pmp_map & this_pmp) {
2464                                 struct ata_link *link = &ap->pmp_link[pmp];
2465                                 pmp_map &= ~this_pmp;
2466                                 ata_eh_analyze_ncq_error(link);
2467                         }
2468                 }
2469                 ata_port_freeze(ap);
2470         }
2471         sata_pmp_error_handler(ap);
2472 }
2473 
2474 static unsigned int mv_get_err_pmp_map(struct ata_port *ap)
2475 {
2476         void __iomem *port_mmio = mv_ap_base(ap);
2477 
2478         return readl(port_mmio + SATA_TESTCTL) >> 16;
2479 }
2480 
2481 static void mv_pmp_eh_prep(struct ata_port *ap, unsigned int pmp_map)
2482 {
2483         struct ata_eh_info *ehi;
2484         unsigned int pmp;
2485 
2486         /*
2487          * Initialize EH info for PMPs which saw device errors
2488          */
2489         ehi = &ap->link.eh_info;
2490         for (pmp = 0; pmp_map != 0; pmp++) {
2491                 unsigned int this_pmp = (1 << pmp);
2492                 if (pmp_map & this_pmp) {
2493                         struct ata_link *link = &ap->pmp_link[pmp];
2494 
2495                         pmp_map &= ~this_pmp;
2496                         ehi = &link->eh_info;
2497                         ata_ehi_clear_desc(ehi);
2498                         ata_ehi_push_desc(ehi, "dev err");
2499                         ehi->err_mask |= AC_ERR_DEV;
2500                         ehi->action |= ATA_EH_RESET;
2501                         ata_link_abort(link);
2502                 }
2503         }
2504 }
2505 
2506 static int mv_req_q_empty(struct ata_port *ap)
2507 {
2508         void __iomem *port_mmio = mv_ap_base(ap);
2509         u32 in_ptr, out_ptr;
2510 
2511         in_ptr  = (readl(port_mmio + EDMA_REQ_Q_IN_PTR)
2512                         >> EDMA_REQ_Q_PTR_SHIFT) & MV_MAX_Q_DEPTH_MASK;
2513         out_ptr = (readl(port_mmio + EDMA_REQ_Q_OUT_PTR)
2514                         >> EDMA_REQ_Q_PTR_SHIFT) & MV_MAX_Q_DEPTH_MASK;
2515         return (in_ptr == out_ptr);     /* 1 == queue_is_empty */
2516 }
2517 
2518 static int mv_handle_fbs_ncq_dev_err(struct ata_port *ap)
2519 {
2520         struct mv_port_priv *pp = ap->private_data;
2521         int failed_links;
2522         unsigned int old_map, new_map;
2523 
2524         /*
2525          * Device error during FBS+NCQ operation:
2526          *
2527          * Set a port flag to prevent further I/O being enqueued.
2528          * Leave the EDMA running to drain outstanding commands from this port.
2529          * Perform the post-mortem/EH only when all responses are complete.
2530          * Follow recovery sequence from 6042/7042 datasheet (7.3.15.4.2.2).
2531          */
2532         if (!(pp->pp_flags & MV_PP_FLAG_DELAYED_EH)) {
2533                 pp->pp_flags |= MV_PP_FLAG_DELAYED_EH;
2534                 pp->delayed_eh_pmp_map = 0;
2535         }
2536         old_map = pp->delayed_eh_pmp_map;
2537         new_map = old_map | mv_get_err_pmp_map(ap);
2538 
2539         if (old_map != new_map) {
2540                 pp->delayed_eh_pmp_map = new_map;
2541                 mv_pmp_eh_prep(ap, new_map & ~old_map);
2542         }
2543         failed_links = hweight16(new_map);
2544 
2545         ata_port_info(ap,
2546                       "%s: pmp_map=%04x qc_map=%04x failed_links=%d nr_active_links=%d\n",
2547                       __func__, pp->delayed_eh_pmp_map,
2548                       ap->qc_active, failed_links,
2549                       ap->nr_active_links);
2550 
2551         if (ap->nr_active_links <= failed_links && mv_req_q_empty(ap)) {
2552                 mv_process_crpb_entries(ap, pp);
2553                 mv_stop_edma(ap);
2554                 mv_eh_freeze(ap);
2555                 ata_port_info(ap, "%s: done\n", __func__);
2556                 return 1;       /* handled */
2557         }
2558         ata_port_info(ap, "%s: waiting\n", __func__);
2559         return 1;       /* handled */
2560 }
2561 
2562 static int mv_handle_fbs_non_ncq_dev_err(struct ata_port *ap)
2563 {
2564         /*
2565          * Possible future enhancement:
2566          *
2567          * FBS+non-NCQ operation is not yet implemented.
2568          * See related notes in mv_edma_cfg().
2569          *
2570          * Device error during FBS+non-NCQ operation:
2571          *
2572          * We need to snapshot the shadow registers for each failed command.
2573          * Follow recovery sequence from 6042/7042 datasheet (7.3.15.4.2.3).
2574          */
2575         return 0;       /* not handled */
2576 }
2577 
2578 static int mv_handle_dev_err(struct ata_port *ap, u32 edma_err_cause)
2579 {
2580         struct mv_port_priv *pp = ap->private_data;
2581 
2582         if (!(pp->pp_flags & MV_PP_FLAG_EDMA_EN))
2583                 return 0;       /* EDMA was not active: not handled */
2584         if (!(pp->pp_flags & MV_PP_FLAG_FBS_EN))
2585                 return 0;       /* FBS was not active: not handled */
2586 
2587         if (!(edma_err_cause & EDMA_ERR_DEV))
2588                 return 0;       /* non DEV error: not handled */
2589         edma_err_cause &= ~EDMA_ERR_IRQ_TRANSIENT;
2590         if (edma_err_cause & ~(EDMA_ERR_DEV | EDMA_ERR_SELF_DIS))
2591                 return 0;       /* other problems: not handled */
2592 
2593         if (pp->pp_flags & MV_PP_FLAG_NCQ_EN) {
2594                 /*
2595                  * EDMA should NOT have self-disabled for this case.
2596                  * If it did, then something is wrong elsewhere,
2597                  * and we cannot handle it here.
2598                  */
2599                 if (edma_err_cause & EDMA_ERR_SELF_DIS) {
2600                         ata_port_warn(ap, "%s: err_cause=0x%x pp_flags=0x%x\n",
2601                                       __func__, edma_err_cause, pp->pp_flags);
2602                         return 0; /* not handled */
2603                 }
2604                 return mv_handle_fbs_ncq_dev_err(ap);
2605         } else {
2606                 /*
2607                  * EDMA should have self-disabled for this case.
2608                  * If it did not, then something is wrong elsewhere,
2609                  * and we cannot handle it here.
2610                  */
2611                 if (!(edma_err_cause & EDMA_ERR_SELF_DIS)) {
2612                         ata_port_warn(ap, "%s: err_cause=0x%x pp_flags=0x%x\n",
2613                                       __func__, edma_err_cause, pp->pp_flags);
2614                         return 0; /* not handled */
2615                 }
2616                 return mv_handle_fbs_non_ncq_dev_err(ap);
2617         }
2618         return 0;       /* not handled */
2619 }
2620 
2621 static void mv_unexpected_intr(struct ata_port *ap, int edma_was_enabled)
2622 {
2623         struct ata_eh_info *ehi = &ap->link.eh_info;
2624         char *when = "idle";
2625 
2626         ata_ehi_clear_desc(ehi);
2627         if (edma_was_enabled) {
2628                 when = "EDMA enabled";
2629         } else {
2630                 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, ap->link.active_tag);
2631                 if (qc && (qc->tf.flags & ATA_TFLAG_POLLING))
2632                         when = "polling";
2633         }
2634         ata_ehi_push_desc(ehi, "unexpected device interrupt while %s", when);
2635         ehi->err_mask |= AC_ERR_OTHER;
2636         ehi->action   |= ATA_EH_RESET;
2637         ata_port_freeze(ap);
2638 }
2639 
2640 /**
2641  *      mv_err_intr - Handle error interrupts on the port
2642  *      @ap: ATA channel to manipulate
2643  *
2644  *      Most cases require a full reset of the chip's state machine,
2645  *      which also performs a COMRESET.
2646  *      Also, if the port disabled DMA, update our cached copy to match.
2647  *
2648  *      LOCKING:
2649  *      Inherited from caller.
2650  */
2651 static void mv_err_intr(struct ata_port *ap)
2652 {
2653         void __iomem *port_mmio = mv_ap_base(ap);
2654         u32 edma_err_cause, eh_freeze_mask, serr = 0;
2655         u32 fis_cause = 0;
2656         struct mv_port_priv *pp = ap->private_data;
2657         struct mv_host_priv *hpriv = ap->host->private_data;
2658         unsigned int action = 0, err_mask = 0;
2659         struct ata_eh_info *ehi = &ap->link.eh_info;
2660         struct ata_queued_cmd *qc;
2661         int abort = 0;
2662 
2663         /*
2664          * Read and clear the SError and err_cause bits.
2665          * For GenIIe, if EDMA_ERR_TRANS_IRQ_7 is set, we also must read/clear
2666          * the FIS_IRQ_CAUSE register before clearing edma_err_cause.
2667          */
2668         sata_scr_read(&ap->link, SCR_ERROR, &serr);
2669         sata_scr_write_flush(&ap->link, SCR_ERROR, serr);
2670 
2671         edma_err_cause = readl(port_mmio + EDMA_ERR_IRQ_CAUSE);
2672         if (IS_GEN_IIE(hpriv) && (edma_err_cause & EDMA_ERR_TRANS_IRQ_7)) {
2673                 fis_cause = readl(port_mmio + FIS_IRQ_CAUSE);
2674                 writelfl(~fis_cause, port_mmio + FIS_IRQ_CAUSE);
2675         }
2676         writelfl(~edma_err_cause, port_mmio + EDMA_ERR_IRQ_CAUSE);
2677 
2678         if (edma_err_cause & EDMA_ERR_DEV) {
2679                 /*
2680                  * Device errors during FIS-based switching operation
2681                  * require special handling.
2682                  */
2683                 if (mv_handle_dev_err(ap, edma_err_cause))
2684                         return;
2685         }
2686 
2687         qc = mv_get_active_qc(ap);
2688         ata_ehi_clear_desc(ehi);
2689         ata_ehi_push_desc(ehi, "edma_err_cause=%08x pp_flags=%08x",
2690                           edma_err_cause, pp->pp_flags);
2691 
2692         if (IS_GEN_IIE(hpriv) && (edma_err_cause & EDMA_ERR_TRANS_IRQ_7)) {
2693                 ata_ehi_push_desc(ehi, "fis_cause=%08x", fis_cause);
2694                 if (fis_cause & FIS_IRQ_CAUSE_AN) {
2695                         u32 ec = edma_err_cause &
2696                                ~(EDMA_ERR_TRANS_IRQ_7 | EDMA_ERR_IRQ_TRANSIENT);
2697                         sata_async_notification(ap);
2698                         if (!ec)
2699                                 return; /* Just an AN; no need for the nukes */
2700                         ata_ehi_push_desc(ehi, "SDB notify");
2701                 }
2702         }
2703         /*
2704          * All generations share these EDMA error cause bits:
2705          */
2706         if (edma_err_cause & EDMA_ERR_DEV) {
2707                 err_mask |= AC_ERR_DEV;
2708                 action |= ATA_EH_RESET;
2709                 ata_ehi_push_desc(ehi, "dev error");
2710         }
2711         if (edma_err_cause & (EDMA_ERR_D_PAR | EDMA_ERR_PRD_PAR |
2712                         EDMA_ERR_CRQB_PAR | EDMA_ERR_CRPB_PAR |
2713                         EDMA_ERR_INTRL_PAR)) {
2714                 err_mask |= AC_ERR_ATA_BUS;
2715                 action |= ATA_EH_RESET;
2716                 ata_ehi_push_desc(ehi, "parity error");
2717         }
2718         if (edma_err_cause & (EDMA_ERR_DEV_DCON | EDMA_ERR_DEV_CON)) {
2719                 ata_ehi_hotplugged(ehi);
2720                 ata_ehi_push_desc(ehi, edma_err_cause & EDMA_ERR_DEV_DCON ?
2721                         "dev disconnect" : "dev connect");
2722                 action |= ATA_EH_RESET;
2723         }
2724 
2725         /*
2726          * Gen-I has a different SELF_DIS bit,
2727          * different FREEZE bits, and no SERR bit:
2728          */
2729         if (IS_GEN_I(hpriv)) {
2730                 eh_freeze_mask = EDMA_EH_FREEZE_5;
2731                 if (edma_err_cause & EDMA_ERR_SELF_DIS_5) {
2732                         pp->pp_flags &= ~MV_PP_FLAG_EDMA_EN;
2733                         ata_ehi_push_desc(ehi, "EDMA self-disable");
2734                 }
2735         } else {
2736                 eh_freeze_mask = EDMA_EH_FREEZE;
2737                 if (edma_err_cause & EDMA_ERR_SELF_DIS) {
2738                         pp->pp_flags &= ~MV_PP_FLAG_EDMA_EN;
2739                         ata_ehi_push_desc(ehi, "EDMA self-disable");
2740                 }
2741                 if (edma_err_cause & EDMA_ERR_SERR) {
2742                         ata_ehi_push_desc(ehi, "SError=%08x", serr);
2743                         err_mask |= AC_ERR_ATA_BUS;
2744                         action |= ATA_EH_RESET;
2745                 }
2746         }
2747 
2748         if (!err_mask) {
2749                 err_mask = AC_ERR_OTHER;
2750                 action |= ATA_EH_RESET;
2751         }
2752 
2753         ehi->serror |= serr;
2754         ehi->action |= action;
2755 
2756         if (qc)
2757                 qc->err_mask |= err_mask;
2758         else
2759                 ehi->err_mask |= err_mask;
2760 
2761         if (err_mask == AC_ERR_DEV) {
2762                 /*
2763                  * Cannot do ata_port_freeze() here,
2764                  * because it would kill PIO access,
2765                  * which is needed for further diagnosis.
2766                  */
2767                 mv_eh_freeze(ap);
2768                 abort = 1;
2769         } else if (edma_err_cause & eh_freeze_mask) {
2770                 /*
2771                  * Note to self: ata_port_freeze() calls ata_port_abort()
2772                  */
2773                 ata_port_freeze(ap);
2774         } else {
2775                 abort = 1;
2776         }
2777 
2778         if (abort) {
2779                 if (qc)
2780                         ata_link_abort(qc->dev->link);
2781                 else
2782                         ata_port_abort(ap);
2783         }
2784 }
2785 
2786 static bool mv_process_crpb_response(struct ata_port *ap,
2787                 struct mv_crpb *response, unsigned int tag, int ncq_enabled)
2788 {
2789         u8 ata_status;
2790         u16 edma_status = le16_to_cpu(response->flags);
2791 
2792         /*
2793          * edma_status from a response queue entry:
2794          *   LSB is from EDMA_ERR_IRQ_CAUSE (non-NCQ only).
2795          *   MSB is saved ATA status from command completion.
2796          */
2797         if (!ncq_enabled) {
2798                 u8 err_cause = edma_status & 0xff & ~EDMA_ERR_DEV;
2799                 if (err_cause) {
2800                         /*
2801                          * Error will be seen/handled by
2802                          * mv_err_intr().  So do nothing at all here.
2803                          */
2804                         return false;
2805                 }
2806         }
2807         ata_status = edma_status >> CRPB_FLAG_STATUS_SHIFT;
2808         if (!ac_err_mask(ata_status))
2809                 return true;
2810         /* else: leave it for mv_err_intr() */
2811         return false;
2812 }
2813 
2814 static void mv_process_crpb_entries(struct ata_port *ap, struct mv_port_priv *pp)
2815 {
2816         void __iomem *port_mmio = mv_ap_base(ap);
2817         struct mv_host_priv *hpriv = ap->host->private_data;
2818         u32 in_index;
2819         bool work_done = false;
2820         u32 done_mask = 0;
2821         int ncq_enabled = (pp->pp_flags & MV_PP_FLAG_NCQ_EN);
2822 
2823         /* Get the hardware queue position index */
2824         in_index = (readl(port_mmio + EDMA_RSP_Q_IN_PTR)
2825                         >> EDMA_RSP_Q_PTR_SHIFT) & MV_MAX_Q_DEPTH_MASK;
2826 
2827         /* Process new responses from since the last time we looked */
2828         while (in_index != pp->resp_idx) {
2829                 unsigned int tag;
2830                 struct mv_crpb *response = &pp->crpb[pp->resp_idx];
2831 
2832                 pp->resp_idx = (pp->resp_idx + 1) & MV_MAX_Q_DEPTH_MASK;
2833 
2834                 if (IS_GEN_I(hpriv)) {
2835                         /* 50xx: no NCQ, only one command active at a time */
2836                         tag = ap->link.active_tag;
2837                 } else {
2838                         /* Gen II/IIE: get command tag from CRPB entry */
2839                         tag = le16_to_cpu(response->id) & 0x1f;
2840                 }
2841                 if (mv_process_crpb_response(ap, response, tag, ncq_enabled))
2842                         done_mask |= 1 << tag;
2843                 work_done = true;
2844         }
2845 
2846         if (work_done) {
2847                 ata_qc_complete_multiple(ap, ap->qc_active ^ done_mask);
2848 
2849                 /* Update the software queue position index in hardware */
2850                 writelfl((pp->crpb_dma & EDMA_RSP_Q_BASE_LO_MASK) |
2851                          (pp->resp_idx << EDMA_RSP_Q_PTR_SHIFT),
2852                          port_mmio + EDMA_RSP_Q_OUT_PTR);
2853         }
2854 }
2855 
2856 static void mv_port_intr(struct ata_port *ap, u32 port_cause)
2857 {
2858         struct mv_port_priv *pp;
2859         int edma_was_enabled;
2860 
2861         /*
2862          * Grab a snapshot of the EDMA_EN flag setting,
2863          * so that we have a consistent view for this port,
2864          * even if something we call of our routines changes it.
2865          */
2866         pp = ap->private_data;
2867         edma_was_enabled = (pp->pp_flags & MV_PP_FLAG_EDMA_EN);
2868         /*
2869          * Process completed CRPB response(s) before other events.
2870          */
2871         if (edma_was_enabled && (port_cause & DONE_IRQ)) {
2872                 mv_process_crpb_entries(ap, pp);
2873                 if (pp->pp_flags & MV_PP_FLAG_DELAYED_EH)
2874                         mv_handle_fbs_ncq_dev_err(ap);
2875         }
2876         /*
2877          * Handle chip-reported errors, or continue on to handle PIO.
2878          */
2879         if (unlikely(port_cause & ERR_IRQ)) {
2880                 mv_err_intr(ap);
2881         } else if (!edma_was_enabled) {
2882                 struct ata_