2 * Copyright (c) 2006 Bernd Walter. All rights reserved.
3 * Copyright (c) 2006 M. Warner Losh. All rights reserved.
4 * Copyright (c) 2010 Greg Ansley. All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include "opt_platform.h"
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include <sys/param.h>
34 #include <sys/systm.h>
36 #include <sys/endian.h>
37 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/resource.h>
44 #include <sys/sysctl.h>
46 #include <machine/bus.h>
47 #include <machine/cpu.h>
48 #include <machine/cpufunc.h>
49 #include <machine/resource.h>
50 #include <machine/intr.h>
52 #include <arm/at91/at91var.h>
53 #include <arm/at91/at91_mcireg.h>
54 #include <arm/at91/at91_pdcreg.h>
56 #include <dev/mmc/bridge.h>
57 #include <dev/mmc/mmcbrvar.h>
60 #include <dev/fdt/fdt_common.h>
61 #include <dev/ofw/ofw_bus.h>
62 #include <dev/ofw/ofw_bus_subr.h>
70 * About running the MCI bus above 25MHz
72 * Historically, the MCI bus has been run at 30MHz on systems with a 60MHz
73 * master clock, in part due to a bug in dev/mmc.c making always request
74 * 30MHz, and in part over clocking the bus because 15MHz was too slow.
75 * Fixing that bug causes the mmc driver to request a 25MHz clock (as it
76 * should) and the logic in at91_mci_update_ios() picks the highest speed that
77 * doesn't exceed that limit. With a 60MHz MCK that would be 15MHz, and
78 * that's a real performance buzzkill when you've been getting away with 30MHz
81 * By defining AT91_MCI_ALLOW_OVERCLOCK (or setting the allow_overclock=1
82 * device hint or sysctl) you can enable logic in at91_mci_update_ios() to
83 * overlcock the SD bus a little by running it at MCK / 2 when the requested
84 * speed is 25MHz and the next highest speed is 15MHz or less. This appears
85 * to work on virtually all SD cards, since it is what this driver has been
86 * doing prior to the introduction of this option, where the overclocking vs
87 * underclocking decision was automaticly "overclock". Modern SD cards can
88 * run at 45mhz/1-bit in standard mode (high speed mode enable commands not
89 * sent) without problems.
91 * Speaking of high-speed mode, the rm9200 manual says the MCI device supports
92 * the SD v1.0 specification and can run up to 50MHz. This is interesting in
93 * that the SD v1.0 spec caps the speed at 25MHz; high speed mode was added in
94 * the v1.10 spec. Furthermore, high speed mode doesn't just crank up the
95 * clock, it alters the signal timing. The rm9200 MCI device doesn't support
96 * these altered timings. So while speeds over 25MHz may work, they only work
97 * in what the SD spec calls "default" speed mode, and it amounts to violating
98 * the spec by overclocking the bus.
100 * If you also enable 4-wire mode it's possible transfers faster than 25MHz
101 * will fail. On the AT91RM9200, due to bugs in the bus contention logic, if
102 * you have the USB host device and OHCI driver enabled will fail. Even
103 * underclocking to 15MHz, intermittant overrun and underrun errors occur.
104 * Note that you don't even need to have usb devices attached to the system,
105 * the errors begin to occur as soon as the OHCI driver sets the register bit
106 * to enable periodic transfers. It appears (based on brief investigation)
107 * that the usb host controller uses so much ASB bandwidth that sometimes the
108 * DMA for MCI transfers doesn't get a bus grant in time and data gets
109 * dropped. Adding even a modicum of network activity changes the symptom
110 * from intermittant to very frequent. Members of the AT91SAM9 family have
111 * corrected this problem, or are at least better about their use of the bus.
113 #ifndef AT91_MCI_ALLOW_OVERCLOCK
114 #define AT91_MCI_ALLOW_OVERCLOCK 1
118 * Allocate 2 bounce buffers we'll use to endian-swap the data due to the rm9200
119 * erratum. We use a pair of buffers because when reading that lets us begin
120 * endian-swapping the data in the first buffer while the DMA is reading into
121 * the second buffer. (We can't use the same trick for writing because we might
122 * not get all the data in the 2nd buffer swapped before the hardware needs it;
123 * dealing with that would add complexity to the driver.)
125 * The buffers are sized at 16K each due to the way the busdma cache sync
126 * operations work on arm. A dcache_inv_range() operation on a range larger
127 * than 16K gets turned into a dcache_wbinv_all(). That needlessly flushes the
128 * entire data cache, impacting overall system performance.
131 #define BBSIZE (16*1024)
132 #define MAX_BLOCKS ((BBSIZE*BBCOUNT)/512)
134 static int mci_debug;
136 struct at91_mci_softc {
137 void *intrhand; /* Interrupt handle */
140 #define CAP_HAS_4WIRE 1 /* Has 4 wire bus */
141 #define CAP_NEEDS_BYTESWAP 2 /* broken hardware needing bounce */
142 #define CAP_MCI1_REV2XX 4 /* MCI 1 rev 2.x */
144 #define PENDING_CMD 0x01
145 #define PENDING_STOP 0x02
146 #define CMD_MULTIREAD 0x10
147 #define CMD_MULTIWRITE 0x20
150 struct resource *irq_res; /* IRQ resource */
151 struct resource *mem_res; /* Memory resource */
153 bus_dma_tag_t dmatag;
154 struct mmc_host host;
156 struct mmc_request *req;
157 struct mmc_command *curcmd;
158 bus_dmamap_t bbuf_map[BBCOUNT];
159 char * bbuf_vaddr[BBCOUNT]; /* bounce bufs in KVA space */
160 uint32_t bbuf_len[BBCOUNT]; /* len currently queued for bounce buf */
161 uint32_t bbuf_curidx; /* which bbuf is the active DMA buffer */
162 uint32_t xfer_offset; /* offset so far into caller's buf */
165 /* bus entry points */
166 static int at91_mci_probe(device_t dev);
167 static int at91_mci_attach(device_t dev);
168 static int at91_mci_detach(device_t dev);
169 static void at91_mci_intr(void *);
171 /* helper routines */
172 static int at91_mci_activate(device_t dev);
173 static void at91_mci_deactivate(device_t dev);
174 static int at91_mci_is_mci1rev2xx(void);
176 #define AT91_MCI_LOCK(_sc) mtx_lock(&(_sc)->sc_mtx)
177 #define AT91_MCI_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_mtx)
178 #define AT91_MCI_LOCK_INIT(_sc) \
179 mtx_init(&_sc->sc_mtx, device_get_nameunit(_sc->dev), \
181 #define AT91_MCI_LOCK_DESTROY(_sc) mtx_destroy(&_sc->sc_mtx);
182 #define AT91_MCI_ASSERT_LOCKED(_sc) mtx_assert(&_sc->sc_mtx, MA_OWNED);
183 #define AT91_MCI_ASSERT_UNLOCKED(_sc) mtx_assert(&_sc->sc_mtx, MA_NOTOWNED);
185 static inline uint32_t
186 RD4(struct at91_mci_softc *sc, bus_size_t off)
188 return (bus_read_4(sc->mem_res, off));
192 WR4(struct at91_mci_softc *sc, bus_size_t off, uint32_t val)
194 bus_write_4(sc->mem_res, off, val);
198 at91_bswap_buf(struct at91_mci_softc *sc, void * dptr, void * sptr, uint32_t memsize)
200 uint32_t * dst = (uint32_t *)dptr;
201 uint32_t * src = (uint32_t *)sptr;
205 * If the hardware doesn't need byte-swapping, let bcopy() do the
206 * work. Use bounce buffer even if we don't need byteswap, since
207 * buffer may straddle a page boundry, and we don't handle
208 * multi-segment transfers in hardware. Seen from 'bsdlabel -w' which
209 * uses raw geom access to the volume. Greg Ansley (gja (at)
212 if (!(sc->sc_cap & CAP_NEEDS_BYTESWAP)) {
213 memcpy(dptr, sptr, memsize);
218 * Nice performance boost for slightly unrolling this loop.
219 * (But very little extra boost for further unrolling it.)
221 for (i = 0; i < memsize; i += 16) {
222 *dst++ = bswap32(*src++);
223 *dst++ = bswap32(*src++);
224 *dst++ = bswap32(*src++);
225 *dst++ = bswap32(*src++);
228 /* Mop up the last 1-3 words, if any. */
229 for (i = 0; i < (memsize & 0x0F); i += 4) {
230 *dst++ = bswap32(*src++);
235 at91_mci_getaddr(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
239 *(bus_addr_t *)arg = segs[0].ds_addr;
243 at91_mci_pdc_disable(struct at91_mci_softc *sc)
245 WR4(sc, PDC_PTCR, PDC_PTCR_TXTDIS | PDC_PTCR_RXTDIS);
248 WR4(sc, PDC_RNPR, 0);
249 WR4(sc, PDC_RNCR, 0);
252 WR4(sc, PDC_TNPR, 0);
253 WR4(sc, PDC_TNCR, 0);
257 * Reset the controller, then restore most of the current state.
259 * This is called after detecting an error. It's also called after stopping a
260 * multi-block write, to un-wedge the device so that it will handle the NOTBUSY
261 * signal correctly. See comments in at91_mci_stop_done() for more details.
263 static void at91_mci_reset(struct at91_mci_softc *sc)
270 at91_mci_pdc_disable(sc);
272 /* save current state */
274 imr = RD4(sc, MCI_IMR);
275 mr = RD4(sc, MCI_MR) & 0x7fff;
276 sdcr = RD4(sc, MCI_SDCR);
277 dtor = RD4(sc, MCI_DTOR);
279 /* reset the controller */
281 WR4(sc, MCI_IDR, 0xffffffff);
282 WR4(sc, MCI_CR, MCI_CR_MCIDIS | MCI_CR_SWRST);
286 WR4(sc, MCI_CR, MCI_CR_MCIEN|MCI_CR_PWSEN);
288 WR4(sc, MCI_SDCR, sdcr);
289 WR4(sc, MCI_DTOR, dtor);
290 WR4(sc, MCI_IER, imr);
293 * Make sure sdio interrupts will fire. Not sure why reading
294 * SR ensures that, but this is in the linux driver.
301 at91_mci_init(device_t dev)
303 struct at91_mci_softc *sc = device_get_softc(dev);
306 WR4(sc, MCI_CR, MCI_CR_MCIDIS | MCI_CR_SWRST); /* device into reset */
307 WR4(sc, MCI_IDR, 0xffffffff); /* Turn off interrupts */
308 WR4(sc, MCI_DTOR, MCI_DTOR_DTOMUL_1M | 1);
309 val = MCI_MR_PDCMODE;
310 val |= 0x34a; /* PWSDIV = 3; CLKDIV = 74 */
311 // if (sc->sc_cap & CAP_MCI1_REV2XX)
312 // val |= MCI_MR_RDPROOF | MCI_MR_WRPROOF;
313 WR4(sc, MCI_MR, val);
314 #ifndef AT91_MCI_SLOT_B
315 WR4(sc, MCI_SDCR, 0); /* SLOT A, 1 bit bus */
318 * XXX Really should add second "unit" but nobody using using
319 * a two slot card that we know of. XXX
321 WR4(sc, MCI_SDCR, 1); /* SLOT B, 1 bit bus */
324 * Enable controller, including power-save. The slower clock
325 * of the power-save mode is only in effect when there is no
326 * transfer in progress, so it can be left in this mode all
329 WR4(sc, MCI_CR, MCI_CR_MCIEN|MCI_CR_PWSEN);
333 at91_mci_fini(device_t dev)
335 struct at91_mci_softc *sc = device_get_softc(dev);
337 WR4(sc, MCI_IDR, 0xffffffff); /* Turn off interrupts */
338 at91_mci_pdc_disable(sc);
339 WR4(sc, MCI_CR, MCI_CR_MCIDIS | MCI_CR_SWRST); /* device into reset */
343 at91_mci_probe(device_t dev)
346 if (!ofw_bus_is_compatible(dev, "atmel,hsmci"))
349 device_set_desc(dev, "MCI mmc/sd host bridge");
354 at91_mci_attach(device_t dev)
356 struct at91_mci_softc *sc = device_get_softc(dev);
357 struct sysctl_ctx_list *sctx;
358 struct sysctl_oid *soid;
362 sctx = device_get_sysctl_ctx(dev);
363 soid = device_get_sysctl_tree(dev);
368 sc->sc_cap |= CAP_NEEDS_BYTESWAP;
370 * MCI1 Rev 2 controllers need some workarounds, flag if so.
372 if (at91_mci_is_mci1rev2xx())
373 sc->sc_cap |= CAP_MCI1_REV2XX;
375 err = at91_mci_activate(dev);
379 AT91_MCI_LOCK_INIT(sc);
385 * Allocate DMA tags and maps and bounce buffers.
387 * The parms in the tag_create call cause the dmamem_alloc call to
388 * create each bounce buffer as a single contiguous buffer of BBSIZE
389 * bytes aligned to a 4096 byte boundary.
391 * Do not use DMA_COHERENT for these buffers because that maps the
392 * memory as non-cachable, which prevents cache line burst fills/writes,
393 * which is something we need since we're trying to overlap the
394 * byte-swapping with the DMA operations.
396 err = bus_dma_tag_create(bus_get_dma_tag(dev), 4096, 0,
397 BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
398 BBSIZE, 1, BBSIZE, 0, NULL, NULL, &sc->dmatag);
402 for (i = 0; i < BBCOUNT; ++i) {
403 err = bus_dmamem_alloc(sc->dmatag, (void **)&sc->bbuf_vaddr[i],
404 BUS_DMA_NOWAIT, &sc->bbuf_map[i]);
410 * Activate the interrupt
412 err = bus_setup_intr(dev, sc->irq_res, INTR_TYPE_MISC | INTR_MPSAFE,
413 NULL, at91_mci_intr, sc, &sc->intrhand);
415 AT91_MCI_LOCK_DESTROY(sc);
420 * Allow 4-wire to be initially set via #define.
421 * Allow a device hint to override that.
422 * Allow a sysctl to override that.
424 #if defined(AT91_MCI_HAS_4WIRE) && AT91_MCI_HAS_4WIRE != 0
427 resource_int_value(device_get_name(dev), device_get_unit(dev),
428 "4wire", &sc->has_4wire);
429 SYSCTL_ADD_UINT(sctx, SYSCTL_CHILDREN(soid), OID_AUTO, "4wire",
430 CTLFLAG_RW, &sc->has_4wire, 0, "has 4 wire SD Card bus");
432 sc->sc_cap |= CAP_HAS_4WIRE;
434 sc->allow_overclock = AT91_MCI_ALLOW_OVERCLOCK;
435 resource_int_value(device_get_name(dev), device_get_unit(dev),
436 "allow_overclock", &sc->allow_overclock);
437 SYSCTL_ADD_UINT(sctx, SYSCTL_CHILDREN(soid), OID_AUTO, "allow_overclock",
438 CTLFLAG_RW, &sc->allow_overclock, 0,
439 "Allow up to 30MHz clock for 25MHz request when next highest speed 15MHz or less.");
441 SYSCTL_ADD_UINT(sctx, SYSCTL_CHILDREN(soid), OID_AUTO, "debug",
442 CTLFLAG_RWTUN, &mci_debug, 0, "enable debug output");
445 * Our real min freq is master_clock/512, but upper driver layers are
446 * going to set the min speed during card discovery, and the right speed
447 * for that is 400kHz, so advertise a safe value just under that.
449 * For max speed, while the rm9200 manual says the max is 50mhz, it also
450 * says it supports only the SD v1.0 spec, which means the real limit is
451 * 25mhz. On the other hand, historical use has been to slightly violate
452 * the standard by running the bus at 30MHz. For more information on
453 * that, see the comments at the top of this file.
455 sc->host.f_min = 375000;
456 sc->host.f_max = at91_master_clock / 2;
457 if (sc->host.f_max > 25000000)
458 sc->host.f_max = 25000000;
459 sc->host.host_ocr = MMC_OCR_320_330 | MMC_OCR_330_340;
461 if (sc->sc_cap & CAP_HAS_4WIRE)
462 sc->host.caps |= MMC_CAP_4_BIT_DATA;
464 child = device_add_child(dev, "mmc", 0);
465 device_set_ivars(dev, &sc->host);
466 err = bus_generic_attach(dev);
469 at91_mci_deactivate(dev);
474 at91_mci_detach(device_t dev)
476 struct at91_mci_softc *sc = device_get_softc(dev);
479 at91_mci_deactivate(dev);
481 bus_dmamem_free(sc->dmatag, sc->bbuf_vaddr[0], sc->bbuf_map[0]);
482 bus_dmamem_free(sc->dmatag, sc->bbuf_vaddr[1], sc->bbuf_map[1]);
483 bus_dma_tag_destroy(sc->dmatag);
485 return (EBUSY); /* XXX */
489 at91_mci_activate(device_t dev)
491 struct at91_mci_softc *sc;
494 sc = device_get_softc(dev);
496 sc->mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
498 if (sc->mem_res == NULL)
502 sc->irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
504 if (sc->irq_res == NULL)
509 at91_mci_deactivate(dev);
514 at91_mci_deactivate(device_t dev)
516 struct at91_mci_softc *sc;
518 sc = device_get_softc(dev);
520 bus_teardown_intr(dev, sc->irq_res, sc->intrhand);
522 bus_generic_detach(sc->dev);
524 bus_release_resource(dev, SYS_RES_MEMORY,
525 rman_get_rid(sc->mem_res), sc->mem_res);
528 bus_release_resource(dev, SYS_RES_IRQ,
529 rman_get_rid(sc->irq_res), sc->irq_res);
535 at91_mci_is_mci1rev2xx(void)
538 switch (soc_info.type) {
552 at91_mci_update_ios(device_t brdev, device_t reqdev)
554 struct at91_mci_softc *sc;
559 sc = device_get_softc(brdev);
563 * Calculate our closest available clock speed that doesn't exceed the
566 * When overclocking is allowed, the requested clock is 25MHz, the
567 * computed frequency is 15MHz or smaller and clockdiv is 1, use
568 * clockdiv of 0 to double that. If less than 12.5MHz, double
569 * regardless of the overclocking setting.
571 * Whatever we come up with, store it back into ios->clock so that the
572 * upper layer drivers can report the actual speed of the bus.
574 if (ios->clock == 0) {
575 WR4(sc, MCI_CR, MCI_CR_MCIDIS);
578 WR4(sc, MCI_CR, MCI_CR_MCIEN|MCI_CR_PWSEN);
579 if ((at91_master_clock % (ios->clock * 2)) == 0)
580 clkdiv = ((at91_master_clock / ios->clock) / 2) - 1;
582 clkdiv = (at91_master_clock / ios->clock) / 2;
583 freq = at91_master_clock / ((clkdiv+1) * 2);
584 if (clkdiv == 1 && ios->clock == 25000000 && freq <= 15000000) {
585 if (sc->allow_overclock || freq <= 12500000) {
587 freq = at91_master_clock / ((clkdiv+1) * 2);
592 if (ios->bus_width == bus_width_4)
593 WR4(sc, MCI_SDCR, RD4(sc, MCI_SDCR) | MCI_SDCR_SDCBUS);
595 WR4(sc, MCI_SDCR, RD4(sc, MCI_SDCR) & ~MCI_SDCR_SDCBUS);
596 WR4(sc, MCI_MR, (RD4(sc, MCI_MR) & ~MCI_MR_CLKDIV) | clkdiv);
597 /* Do we need a settle time here? */
598 /* XXX We need to turn the device on/off here with a GPIO pin */
603 at91_mci_start_cmd(struct at91_mci_softc *sc, struct mmc_command *cmd)
606 struct mmc_data *data;
611 /* XXX Upper layers don't always set this */
614 /* Begin setting up command register. */
618 if (sc->host.ios.bus_mode == opendrain)
619 cmdr |= MCI_CMDR_OPDCMD;
621 /* Set up response handling. Allow max timeout for responses. */
623 if (MMC_RSP(cmd->flags) == MMC_RSP_NONE)
624 cmdr |= MCI_CMDR_RSPTYP_NO;
626 cmdr |= MCI_CMDR_MAXLAT;
627 if (cmd->flags & MMC_RSP_136)
628 cmdr |= MCI_CMDR_RSPTYP_136;
630 cmdr |= MCI_CMDR_RSPTYP_48;
634 * If there is no data transfer, just set up the right interrupt mask
635 * and start the command.
637 * The interrupt mask needs to be CMDRDY plus all non-data-transfer
638 * errors. It's important to leave the transfer-related errors out, to
639 * avoid spurious timeout or crc errors on a STOP command following a
640 * multiblock read. When a multiblock read is in progress, sending a
641 * STOP in the middle of a block occasionally triggers such errors, but
642 * we're totally disinterested in them because we've already gotten all
643 * the data we wanted without error before sending the STOP command.
647 uint32_t ier = MCI_SR_CMDRDY |
648 MCI_SR_RTOE | MCI_SR_RENDE |
649 MCI_SR_RCRCE | MCI_SR_RDIRE | MCI_SR_RINDE;
651 at91_mci_pdc_disable(sc);
653 if (cmd->opcode == MMC_STOP_TRANSMISSION)
654 cmdr |= MCI_CMDR_TRCMD_STOP;
656 /* Ignore response CRC on CMD2 and ACMD41, per standard. */
658 if (cmd->opcode == MMC_SEND_OP_COND ||
659 cmd->opcode == ACMD_SD_SEND_OP_COND)
660 ier &= ~MCI_SR_RCRCE;
663 printf("CMDR %x (opcode %d) ARGR %x no data\n",
664 cmdr, cmd->opcode, cmd->arg);
666 WR4(sc, MCI_ARGR, cmd->arg);
667 WR4(sc, MCI_CMDR, cmdr);
668 WR4(sc, MCI_IDR, 0xffffffff);
669 WR4(sc, MCI_IER, ier);
673 /* There is data, set up the transfer-related parts of the command. */
675 if (data->flags & MMC_DATA_READ)
676 cmdr |= MCI_CMDR_TRDIR;
678 if (data->flags & (MMC_DATA_READ | MMC_DATA_WRITE))
679 cmdr |= MCI_CMDR_TRCMD_START;
681 if (data->flags & MMC_DATA_STREAM)
682 cmdr |= MCI_CMDR_TRTYP_STREAM;
683 else if (data->flags & MMC_DATA_MULTI) {
684 cmdr |= MCI_CMDR_TRTYP_MULTIPLE;
685 sc->flags |= (data->flags & MMC_DATA_READ) ?
686 CMD_MULTIREAD : CMD_MULTIWRITE;
690 * Disable PDC until we're ready.
692 * Set block size and turn on PDC mode for dma xfer.
693 * Note that the block size is the smaller of the amount of data to be
694 * transferred, or 512 bytes. The 512 size is fixed by the standard;
695 * smaller blocks are possible, but never larger.
698 WR4(sc, PDC_PTCR, PDC_PTCR_RXTDIS | PDC_PTCR_TXTDIS);
700 mr = RD4(sc,MCI_MR) & ~MCI_MR_BLKLEN;
701 mr |= min(data->len, 512) << 16;
702 WR4(sc, MCI_MR, mr | MCI_MR_PDCMODE|MCI_MR_PDCPADV);
707 * Use bounce buffers even if we don't need to byteswap, because doing
708 * multi-block IO with large DMA buffers is way fast (compared to
709 * single-block IO), even after incurring the overhead of also copying
710 * from/to the caller's buffers (which may be in non-contiguous physical
713 * In an ideal non-byteswap world we could create a dma tag that allows
714 * for discontiguous segments and do the IO directly from/to the
715 * caller's buffer(s), using ENDRX/ENDTX interrupts to chain the
716 * discontiguous buffers through the PDC. Someday.
718 * If a read is bigger than 2k, split it in half so that we can start
719 * byte-swapping the first half while the second half is on the wire.
720 * It would be best if we could split it into 8k chunks, but we can't
721 * always keep up with the byte-swapping due to other system activity,
722 * and if an RXBUFF interrupt happens while we're still handling the
723 * byte-swap from the prior buffer (IE, we haven't returned from
724 * handling the prior interrupt yet), then data will get dropped on the
725 * floor and we can't easily recover from that. The right fix for that
726 * would be to have the interrupt handling only keep the DMA flowing and
727 * enqueue filled buffers to be byte-swapped in a non-interrupt context.
728 * Even that won't work on the write side of things though; in that
729 * context we have to have all the data ready to go before starting the
732 * XXX what about stream transfers?
737 if (data->flags & (MMC_DATA_READ | MMC_DATA_WRITE)) {
739 uint32_t remaining = data->len;
743 if (remaining > (BBCOUNT*BBSIZE))
744 panic("IO read size exceeds MAXDATA\n");
746 if (data->flags & MMC_DATA_READ) {
747 if (remaining > 2048) // XXX
751 err = bus_dmamap_load(sc->dmatag, sc->bbuf_map[0],
752 sc->bbuf_vaddr[0], len, at91_mci_getaddr,
753 &paddr, BUS_DMA_NOWAIT);
755 panic("IO read dmamap_load failed\n");
756 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[0],
757 BUS_DMASYNC_PREREAD);
758 WR4(sc, PDC_RPR, paddr);
759 WR4(sc, PDC_RCR, len / 4);
760 sc->bbuf_len[0] = len;
762 if (remaining == 0) {
766 err = bus_dmamap_load(sc->dmatag, sc->bbuf_map[1],
767 sc->bbuf_vaddr[1], len, at91_mci_getaddr,
768 &paddr, BUS_DMA_NOWAIT);
770 panic("IO read dmamap_load failed\n");
771 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[1],
772 BUS_DMASYNC_PREREAD);
773 WR4(sc, PDC_RNPR, paddr);
774 WR4(sc, PDC_RNCR, len / 4);
775 sc->bbuf_len[1] = len;
778 WR4(sc, PDC_PTCR, PDC_PTCR_RXTEN);
780 len = min(BBSIZE, remaining);
781 at91_bswap_buf(sc, sc->bbuf_vaddr[0], data->data, len);
782 err = bus_dmamap_load(sc->dmatag, sc->bbuf_map[0],
783 sc->bbuf_vaddr[0], len, at91_mci_getaddr,
784 &paddr, BUS_DMA_NOWAIT);
786 panic("IO write dmamap_load failed\n");
787 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[0],
788 BUS_DMASYNC_PREWRITE);
790 * Erratum workaround: PDC transfer length on a write
791 * must not be smaller than 12 bytes (3 words); only
792 * blklen bytes (set above) are actually transferred.
794 WR4(sc, PDC_TPR,paddr);
795 WR4(sc, PDC_TCR, (len < 12) ? 3 : len / 4);
796 sc->bbuf_len[0] = len;
798 if (remaining == 0) {
802 at91_bswap_buf(sc, sc->bbuf_vaddr[1],
803 ((char *)data->data)+BBSIZE, len);
804 err = bus_dmamap_load(sc->dmatag, sc->bbuf_map[1],
805 sc->bbuf_vaddr[1], len, at91_mci_getaddr,
806 &paddr, BUS_DMA_NOWAIT);
808 panic("IO write dmamap_load failed\n");
809 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[1],
810 BUS_DMASYNC_PREWRITE);
811 WR4(sc, PDC_TNPR, paddr);
812 WR4(sc, PDC_TNCR, (len < 12) ? 3 : len / 4);
813 sc->bbuf_len[1] = len;
816 /* do not enable PDC xfer until CMDRDY asserted */
818 data->xfer_len = 0; /* XXX what's this? appears to be unused. */
822 printf("CMDR %x (opcode %d) ARGR %x with data len %d\n",
823 cmdr, cmd->opcode, cmd->arg, cmd->data->len);
825 WR4(sc, MCI_ARGR, cmd->arg);
826 WR4(sc, MCI_CMDR, cmdr);
827 WR4(sc, MCI_IER, MCI_SR_ERROR | MCI_SR_CMDRDY);
831 at91_mci_next_operation(struct at91_mci_softc *sc)
833 struct mmc_request *req;
839 if (sc->flags & PENDING_CMD) {
840 sc->flags &= ~PENDING_CMD;
841 at91_mci_start_cmd(sc, req->cmd);
843 } else if (sc->flags & PENDING_STOP) {
844 sc->flags &= ~PENDING_STOP;
845 at91_mci_start_cmd(sc, req->stop);
849 WR4(sc, MCI_IDR, 0xffffffff);
852 //printf("req done\n");
857 at91_mci_request(device_t brdev, device_t reqdev, struct mmc_request *req)
859 struct at91_mci_softc *sc = device_get_softc(brdev);
862 if (sc->req != NULL) {
866 //printf("new req\n");
868 sc->flags = PENDING_CMD;
870 sc->flags |= PENDING_STOP;
871 at91_mci_next_operation(sc);
877 at91_mci_get_ro(device_t brdev, device_t reqdev)
883 at91_mci_acquire_host(device_t brdev, device_t reqdev)
885 struct at91_mci_softc *sc = device_get_softc(brdev);
890 msleep(sc, &sc->sc_mtx, PZERO, "mciah", hz / 5);
897 at91_mci_release_host(device_t brdev, device_t reqdev)
899 struct at91_mci_softc *sc = device_get_softc(brdev);
909 at91_mci_read_done(struct at91_mci_softc *sc, uint32_t sr)
911 struct mmc_command *cmd = sc->curcmd;
912 char * dataptr = (char *)cmd->data->data;
913 uint32_t curidx = sc->bbuf_curidx;
914 uint32_t len = sc->bbuf_len[curidx];
917 * We arrive here when a DMA transfer for a read is done, whether it's
918 * a single or multi-block read.
920 * We byte-swap the buffer that just completed, and if that is the
921 * last buffer that's part of this read then we move on to the next
922 * operation, otherwise we wait for another ENDRX for the next bufer.
925 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[curidx], BUS_DMASYNC_POSTREAD);
926 bus_dmamap_unload(sc->dmatag, sc->bbuf_map[curidx]);
928 at91_bswap_buf(sc, dataptr + sc->xfer_offset, sc->bbuf_vaddr[curidx], len);
931 printf("read done sr %x curidx %d len %d xfer_offset %d\n",
932 sr, curidx, len, sc->xfer_offset);
935 sc->xfer_offset += len;
936 sc->bbuf_curidx = !curidx; /* swap buffers */
939 * If we've transferred all the data, move on to the next operation.
941 * If we're still transferring the last buffer, RNCR is already zero but
942 * we have to write a zero anyway to clear the ENDRX status so we don't
943 * re-interrupt until the last buffer is done.
945 if (sc->xfer_offset == cmd->data->len) {
946 WR4(sc, PDC_PTCR, PDC_PTCR_RXTDIS | PDC_PTCR_TXTDIS);
947 cmd->error = MMC_ERR_NONE;
948 at91_mci_next_operation(sc);
950 WR4(sc, PDC_RNCR, 0);
951 WR4(sc, MCI_IER, MCI_SR_ERROR | MCI_SR_ENDRX);
956 at91_mci_write_done(struct at91_mci_softc *sc, uint32_t sr)
958 struct mmc_command *cmd = sc->curcmd;
961 * We arrive here when the entire DMA transfer for a write is done,
962 * whether it's a single or multi-block write. If it's multi-block we
963 * have to immediately move on to the next operation which is to send
964 * the stop command. If it's a single-block transfer we need to wait
965 * for NOTBUSY, but if that's already asserted we can avoid another
966 * interrupt and just move on to completing the request right away.
969 WR4(sc, PDC_PTCR, PDC_PTCR_RXTDIS | PDC_PTCR_TXTDIS);
971 bus_dmamap_sync(sc->dmatag, sc->bbuf_map[sc->bbuf_curidx],
972 BUS_DMASYNC_POSTWRITE);
973 bus_dmamap_unload(sc->dmatag, sc->bbuf_map[sc->bbuf_curidx]);
975 if ((cmd->data->flags & MMC_DATA_MULTI) || (sr & MCI_SR_NOTBUSY)) {
976 cmd->error = MMC_ERR_NONE;
977 at91_mci_next_operation(sc);
979 WR4(sc, MCI_IER, MCI_SR_ERROR | MCI_SR_NOTBUSY);
984 at91_mci_notbusy(struct at91_mci_softc *sc)
986 struct mmc_command *cmd = sc->curcmd;
989 * We arrive here by either completion of a single-block write, or
990 * completion of the stop command that ended a multi-block write (and,
991 * I suppose, after a card-select or erase, but I haven't tested
992 * those). Anyway, we're done and it's time to move on to the next
996 cmd->error = MMC_ERR_NONE;
997 at91_mci_next_operation(sc);
1001 at91_mci_stop_done(struct at91_mci_softc *sc, uint32_t sr)
1003 struct mmc_command *cmd = sc->curcmd;
1006 * We arrive here after receiving CMDRDY for a MMC_STOP_TRANSMISSION
1007 * command. Depending on the operation being stopped, we may have to
1008 * do some unusual things to work around hardware bugs.
1012 * This is known to be true of at91rm9200 hardware; it may or may not
1013 * apply to more recent chips:
1015 * After stopping a multi-block write, the NOTBUSY bit in MCI_SR does
1016 * not properly reflect the actual busy state of the card as signaled
1017 * on the DAT0 line; it always claims the card is not-busy. If we
1018 * believe that and let operations continue, following commands will
1019 * fail with response timeouts (except of course MMC_SEND_STATUS -- it
1020 * indicates the card is busy in the PRG state, which was the smoking
1021 * gun that showed MCI_SR NOTBUSY was not tracking DAT0 correctly).
1023 * The atmel docs are emphatic: "This flag [NOTBUSY] must be used only
1024 * for Write Operations." I guess technically since we sent a stop
1025 * it's not a write operation anymore. But then just what did they
1026 * think it meant for the stop command to have "...an optional busy
1027 * signal transmitted on the data line" according to the SD spec?
1029 * I tried a variety of things to un-wedge the MCI and get the status
1030 * register to reflect NOTBUSY correctly again, but the only thing
1031 * that worked was a full device reset. It feels like an awfully big
1032 * hammer, but doing a full reset after every multiblock write is
1033 * still faster than doing single-block IO (by almost two orders of
1034 * magnitude: 20KB/sec improves to about 1.8MB/sec best case).
1036 * After doing the reset, wait for a NOTBUSY interrupt before
1037 * continuing with the next operation.
1039 * This workaround breaks multiwrite on the rev2xx parts, but some other
1040 * workaround is needed.
1042 if ((sc->flags & CMD_MULTIWRITE) && (sc->sc_cap & CAP_NEEDS_BYTESWAP)) {
1044 WR4(sc, MCI_IER, MCI_SR_ERROR | MCI_SR_NOTBUSY);
1049 * This is known to be true of at91rm9200 hardware; it may or may not
1050 * apply to more recent chips:
1052 * After stopping a multi-block read, loop to read and discard any
1053 * data that coasts in after we sent the stop command. The docs don't
1054 * say anything about it, but empirical testing shows that 1-3
1055 * additional words of data get buffered up in some unmentioned
1056 * internal fifo and if we don't read and discard them here they end
1057 * up on the front of the next read DMA transfer we do.
1059 * This appears to be unnecessary for rev2xx parts.
1061 if ((sc->flags & CMD_MULTIREAD) && (sc->sc_cap & CAP_NEEDS_BYTESWAP)) {
1066 sr = RD4(sc, MCI_SR);
1067 if (sr & MCI_SR_RXRDY) {
1071 } while (sr & MCI_SR_RXRDY);
1075 cmd->error = MMC_ERR_NONE;
1076 at91_mci_next_operation(sc);
1081 at91_mci_cmdrdy(struct at91_mci_softc *sc, uint32_t sr)
1083 struct mmc_command *cmd = sc->curcmd;
1090 * We get here at the end of EVERY command. We retrieve the command
1091 * response (if any) then decide what to do next based on the command.
1094 if (cmd->flags & MMC_RSP_PRESENT) {
1095 for (i = 0; i < ((cmd->flags & MMC_RSP_136) ? 4 : 1); i++) {
1096 cmd->resp[i] = RD4(sc, MCI_RSPR + i * 4);
1098 printf("RSPR[%d] = %x sr=%x\n", i, cmd->resp[i], sr);
1103 * If this was a stop command, go handle the various special
1104 * conditions (read: bugs) that have to be dealt with following a stop.
1106 if (cmd->opcode == MMC_STOP_TRANSMISSION) {
1107 at91_mci_stop_done(sc, sr);
1112 * If this command can continue to assert BUSY beyond the response then
1113 * we need to wait for NOTBUSY before the command is really done.
1115 * Note that this may not work properly on the at91rm9200. It certainly
1116 * doesn't work for the STOP command that follows a multi-block write,
1117 * so post-stop CMDRDY is handled separately; see the special handling
1118 * in at91_mci_stop_done().
1120 * Beside STOP, there are other R1B-type commands that use the busy
1121 * signal after CMDRDY: CMD7 (card select), CMD28-29 (write protect),
1122 * CMD38 (erase). I haven't tested any of them, but I rather expect
1123 * them all to have the same sort of problem with MCI_SR not actually
1124 * reflecting the state of the DAT0-line busy indicator. So this code
1125 * may need to grow some sort of special handling for them too. (This
1126 * just in: CMD7 isn't a problem right now because dev/mmc.c incorrectly
1127 * sets the response flags to R1 rather than R1B.) XXX
1129 if ((cmd->flags & MMC_RSP_BUSY)) {
1130 WR4(sc, MCI_IER, MCI_SR_ERROR | MCI_SR_NOTBUSY);
1135 * If there is a data transfer with this command, then...
1136 * - If it's a read, we need to wait for ENDRX.
1137 * - If it's a write, now is the time to enable the PDC, and we need
1138 * to wait for a BLKE that follows a TXBUFE, because if we're doing
1139 * a split transfer we get a BLKE after the first half (when TPR/TCR
1140 * get loaded from TNPR/TNCR). So first we wait for the TXBUFE, and
1141 * the handling for that interrupt will then invoke the wait for the
1142 * subsequent BLKE which indicates actual completion.
1146 if (cmd->data->flags & MMC_DATA_READ) {
1149 ier = MCI_SR_TXBUFE;
1150 WR4(sc, PDC_PTCR, PDC_PTCR_TXTEN);
1152 WR4(sc, MCI_IER, MCI_SR_ERROR | ier);
1157 * If we made it to here, we don't need to wait for anything more for
1158 * the current command, move on to the next command (will complete the
1159 * request if there is no next command).
1161 cmd->error = MMC_ERR_NONE;
1162 at91_mci_next_operation(sc);
1166 at91_mci_intr(void *arg)
1168 struct at91_mci_softc *sc = (struct at91_mci_softc*)arg;
1169 struct mmc_command *cmd = sc->curcmd;
1174 sr = RD4(sc, MCI_SR);
1175 isr = sr & RD4(sc, MCI_IMR);
1178 printf("i 0x%x sr 0x%x\n", isr, sr);
1181 * All interrupts are one-shot; disable it now.
1182 * The next operation will re-enable whatever interrupts it wants.
1184 WR4(sc, MCI_IDR, isr);
1185 if (isr & MCI_SR_ERROR) {
1186 if (isr & (MCI_SR_RTOE | MCI_SR_DTOE))
1187 cmd->error = MMC_ERR_TIMEOUT;
1188 else if (isr & (MCI_SR_RCRCE | MCI_SR_DCRCE))
1189 cmd->error = MMC_ERR_BADCRC;
1190 else if (isr & (MCI_SR_OVRE | MCI_SR_UNRE))
1191 cmd->error = MMC_ERR_FIFO;
1193 cmd->error = MMC_ERR_FAILED;
1195 * CMD8 is used to probe for SDHC cards, a standard SD card
1196 * will get a response timeout; don't report it because it's a
1197 * normal and expected condition. One might argue that all
1198 * error reporting should be left to higher levels, but when
1199 * they report at all it's always EIO, which isn't very
1200 * helpful. XXX bootverbose?
1202 if (cmd->opcode != 8) {
1203 device_printf(sc->dev,
1204 "IO error; status MCI_SR = 0x%x cmd opcode = %d%s\n",
1206 (cmd->opcode != 12) ? "" :
1207 (sc->flags & CMD_MULTIREAD) ? " after read" : " after write");
1210 at91_mci_next_operation(sc);
1212 if (isr & MCI_SR_TXBUFE) {
1213 // printf("TXBUFE\n");
1215 * We need to wait for a BLKE that follows TXBUFE
1216 * (intermediate BLKEs might happen after ENDTXes if
1217 * we're chaining multiple buffers). If BLKE is also
1218 * asserted at the time we get TXBUFE, we can avoid
1219 * another interrupt and process it right away, below.
1221 if (sr & MCI_SR_BLKE)
1224 WR4(sc, MCI_IER, MCI_SR_BLKE);
1226 if (isr & MCI_SR_RXBUFF) {
1227 // printf("RXBUFF\n");
1229 if (isr & MCI_SR_ENDTX) {
1230 // printf("ENDTX\n");
1232 if (isr & MCI_SR_ENDRX) {
1233 // printf("ENDRX\n");
1234 at91_mci_read_done(sc, sr);
1236 if (isr & MCI_SR_NOTBUSY) {
1237 // printf("NOTBUSY\n");
1238 at91_mci_notbusy(sc);
1240 if (isr & MCI_SR_DTIP) {
1241 // printf("Data transfer in progress\n");
1243 if (isr & MCI_SR_BLKE) {
1244 // printf("Block transfer end\n");
1245 at91_mci_write_done(sc, sr);
1247 if (isr & MCI_SR_TXRDY) {
1248 // printf("Ready to transmit\n");
1250 if (isr & MCI_SR_RXRDY) {
1251 // printf("Ready to receive\n");
1253 if (isr & MCI_SR_CMDRDY) {
1254 // printf("Command ready\n");
1255 at91_mci_cmdrdy(sc, sr);
1258 AT91_MCI_UNLOCK(sc);
1262 at91_mci_read_ivar(device_t bus, device_t child, int which, uintptr_t *result)
1264 struct at91_mci_softc *sc = device_get_softc(bus);
1269 case MMCBR_IVAR_BUS_MODE:
1270 *(int *)result = sc->host.ios.bus_mode;
1272 case MMCBR_IVAR_BUS_WIDTH:
1273 *(int *)result = sc->host.ios.bus_width;
1275 case MMCBR_IVAR_CHIP_SELECT:
1276 *(int *)result = sc->host.ios.chip_select;
1278 case MMCBR_IVAR_CLOCK:
1279 *(int *)result = sc->host.ios.clock;
1281 case MMCBR_IVAR_F_MIN:
1282 *(int *)result = sc->host.f_min;
1284 case MMCBR_IVAR_F_MAX:
1285 *(int *)result = sc->host.f_max;
1287 case MMCBR_IVAR_HOST_OCR:
1288 *(int *)result = sc->host.host_ocr;
1290 case MMCBR_IVAR_MODE:
1291 *(int *)result = sc->host.mode;
1293 case MMCBR_IVAR_OCR:
1294 *(int *)result = sc->host.ocr;
1296 case MMCBR_IVAR_POWER_MODE:
1297 *(int *)result = sc->host.ios.power_mode;
1299 case MMCBR_IVAR_VDD:
1300 *(int *)result = sc->host.ios.vdd;
1302 case MMCBR_IVAR_CAPS:
1303 if (sc->has_4wire) {
1304 sc->sc_cap |= CAP_HAS_4WIRE;
1305 sc->host.caps |= MMC_CAP_4_BIT_DATA;
1307 sc->sc_cap &= ~CAP_HAS_4WIRE;
1308 sc->host.caps &= ~MMC_CAP_4_BIT_DATA;
1310 *(int *)result = sc->host.caps;
1312 case MMCBR_IVAR_MAX_DATA:
1314 * Something is wrong with the 2x parts and multiblock, so
1315 * just do 1 block at a time for now, which really kills
1318 if (sc->sc_cap & CAP_MCI1_REV2XX)
1321 *(int *)result = MAX_BLOCKS;
1328 at91_mci_write_ivar(device_t bus, device_t child, int which, uintptr_t value)
1330 struct at91_mci_softc *sc = device_get_softc(bus);
1335 case MMCBR_IVAR_BUS_MODE:
1336 sc->host.ios.bus_mode = value;
1338 case MMCBR_IVAR_BUS_WIDTH:
1339 sc->host.ios.bus_width = value;
1341 case MMCBR_IVAR_CHIP_SELECT:
1342 sc->host.ios.chip_select = value;
1344 case MMCBR_IVAR_CLOCK:
1345 sc->host.ios.clock = value;
1347 case MMCBR_IVAR_MODE:
1348 sc->host.mode = value;
1350 case MMCBR_IVAR_OCR:
1351 sc->host.ocr = value;
1353 case MMCBR_IVAR_POWER_MODE:
1354 sc->host.ios.power_mode = value;
1356 case MMCBR_IVAR_VDD:
1357 sc->host.ios.vdd = value;
1359 /* These are read-only */
1360 case MMCBR_IVAR_CAPS:
1361 case MMCBR_IVAR_HOST_OCR:
1362 case MMCBR_IVAR_F_MIN:
1363 case MMCBR_IVAR_F_MAX:
1364 case MMCBR_IVAR_MAX_DATA:
1370 static device_method_t at91_mci_methods[] = {
1372 DEVMETHOD(device_probe, at91_mci_probe),
1373 DEVMETHOD(device_attach, at91_mci_attach),
1374 DEVMETHOD(device_detach, at91_mci_detach),
1377 DEVMETHOD(bus_read_ivar, at91_mci_read_ivar),
1378 DEVMETHOD(bus_write_ivar, at91_mci_write_ivar),
1381 DEVMETHOD(mmcbr_update_ios, at91_mci_update_ios),
1382 DEVMETHOD(mmcbr_request, at91_mci_request),
1383 DEVMETHOD(mmcbr_get_ro, at91_mci_get_ro),
1384 DEVMETHOD(mmcbr_acquire_host, at91_mci_acquire_host),
1385 DEVMETHOD(mmcbr_release_host, at91_mci_release_host),
1390 static driver_t at91_mci_driver = {
1393 sizeof(struct at91_mci_softc),
1396 static devclass_t at91_mci_devclass;
1399 DRIVER_MODULE(at91_mci, simplebus, at91_mci_driver, at91_mci_devclass, NULL,
1402 DRIVER_MODULE(at91_mci, atmelarm, at91_mci_driver, at91_mci_devclass, NULL,
1406 MMC_DECLARE_BRIDGE(at91_mci);