minix/drivers/mmc/mmcblk.c
Kees Jongenburger 0748a95f88 arm:mmc only compile mmchs when targeting arm.
Change-Id: Ie8d0b6461b58c2a7becda98401dece81be607a01
2013-05-24 15:46:59 +02:00

674 lines
19 KiB
C

/*
* Block driver for Multi Media Cards (MMC).
*/
/* kernel headers */
#include <minix/syslib.h>
#include <minix/driver.h>
#include <minix/blockdriver.h>
#include <minix/drvlib.h>
#include <minix/minlib.h>
/* system headers */
#include <sys/ioc_disk.h> /* disk IOCTL's */
/* usr headers */
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <signal.h>
/* local headers */
#include "mmchost.h"
#include "mmclog.h"
/* used for logging */
static struct mmclog log = {
.name = "mmc_block",
.log_level = LEVEL_INFO,
.log_func = default_log
};
/* holding the current host controller */
static struct mmc_host host;
#define SUB_PER_DRIVE (NR_PARTITIONS * NR_PARTITIONS)
#define NR_SUBDEVS (MAX_DRIVES * SUB_PER_DRIVE)
/* When passing data over a grant one needs to pass
* a buffer to sys_safecopy copybuff is used for that*/
#define COPYBUFF_SIZE 0x1000 /* 4k buff */
static unsigned char copybuff[COPYBUFF_SIZE];
static struct sd_slot *get_slot(dev_t minor);
/* Prototypes for the block device */
static int block_open(dev_t minor, int access);
static int block_close(dev_t minor);
static int block_transfer(dev_t minor,
int do_write,
u64_t position,
endpoint_t endpt, iovec_t * iov, unsigned int nr_req, int flags);
static int block_ioctl(dev_t minor,
unsigned int request, endpoint_t endpt, cp_grant_id_t grant);
static struct device *block_part(dev_t minor);
/* System even handling */
static void sef_local_startup();
static int block_system_event_cb(int type, sef_init_info_t * info);
static void block_signal_handler_cb(int signo);
void
bdr_alarm(clock_t stamp)
{
mmc_log_debug(&log, "alarm %d\n", stamp);
}
static int apply_env();
static void hw_intr(unsigned int irqs);
/* set the global logging level */
static void set_log_level(int level);
/* Entry points for the BLOCK driver. */
static struct blockdriver mmc_driver = {
BLOCKDRIVER_TYPE_DISK, /* handle partition requests */
block_open, /* open or mount */
block_close, /* on a close */
block_transfer, /* does the I/O */
block_ioctl, /* ioclt's */
NULL, /* no need to clean up (yet) */
block_part, /* return partition information */
NULL, /* no geometry */
hw_intr, /* left over interrupts */
bdr_alarm, /* no alarm processing */
NULL, /* no processing of other messages */
NULL /* no threading support */
};
static void
hw_intr(unsigned int irqs)
{
mmc_log_debug(&log, "Hardware inter left over\n");
host.hw_intr(irqs);
}
static int
apply_env()
{
long v;
/* apply the env setting passed to this driver parameters accepted
* log_level=[0-4] (NONE,WARN,INFO,DEBUG,TRACE) instance=[0-3]
* instance/bus number to use for this driver Passing these arguments
* is done when starting the driver using the service command in the
* following way service up /sbin/mmc -args "log_level=2 instance=1
* driver=dummy" -dev /dev/c2d0 */
char driver[16];
memset(driver, '\0', 16);
(void) env_get_param("driver", driver, 16);
if (strlen(driver) == 0
|| strncmp(driver, "mmchs", strlen("mmchs") + 1) == 0) {
/* early init of host mmc host controller. This code should
* depend on knowing the hardware that is running bellow. */
#ifdef __arm__
host_initialize_host_structure_mmchs(&host);
#endif
} else if (strncmp(driver, "dummy", strlen("dummy") + 1) == 0) {
host_initialize_host_structure_dummy(&host);
} else {
mmc_log_warn(&log, "Unknown driver %s\n", driver);
}
/* Initialize the verbosity level. */
v = 0;
if (env_parse("log_level", "d", 0, &v, LEVEL_NONE,
LEVEL_TRACE) == EP_SET) {
set_log_level(v);
}
/* Find out which driver instance we are. */
v = 0;
env_parse("instance", "d", 0, &v, 0, 3);
if (host.host_set_instance(&host, v)) {
mmc_log_warn(&log, "Failed to set mmc instance to %d\n", v);
return -1; /* NOT OK */
}
return OK;
}
;
/*===========================================================================*
* block_open *
*===========================================================================*/
static int
block_open(dev_t minor, int access)
{
struct sd_slot *slot;
slot = get_slot(minor);
int i, j;
int part_count, sub_part_count;
i = j = part_count = sub_part_count = 0;
if (!slot) {
mmc_log_debug(&log,
"Not handling open on non existing slot\n");
return EIO;
}
assert(slot->host != NULL);
if (!slot->host->card_detect(slot)) {
mmc_log_debug(&log, "No card inserted in the SD slot\n");
return EIO;
}
/* If we are already open just increase the open count and return */
if (slot->card.state == SD_MODE_DATA_TRANSFER_MODE) {
assert(slot->card.open_ct >= 0);
slot->card.open_ct++;
mmc_log_trace(&log, "increased open count to %d\n",
slot->card.open_ct);
return OK;
}
/* We did not have an sd-card inserted so we are going to probe for it
*/
mmc_log_debug(&log, "First open on (%d)\n", minor);
if (!host.card_initialize(slot)) {
// * TODO: set card state to INVALID until removed? */
return EIO;
}
partition(&mmc_driver, 0 /* first card on bus */ , P_PRIMARY,
0 /* atapi device?? */ );
mmc_log_trace(&log, "descr \toffset(bytes) size(bytes)\n", minor);
mmc_log_trace(&log, "disk %d\t0x%016llx 0x%016llx\n", i,
slot->card.part[0].dv_base, slot->card.part[0].dv_size);
for (i = 1; i < 5; i++) {
if (slot->card.part[i].dv_size == 0)
continue;
part_count++;
mmc_log_trace(&log, "part %d\t0x%016llx 0x%016llx\n", i,
slot->card.part[i].dv_base, slot->card.part[i].dv_size);
for (j = 0; j < 4; j++) {
if (slot->card.subpart[(i - 1) * 4 + j].dv_size == 0)
continue;
sub_part_count++;
mmc_log_trace(&log,
" sub %d/%d\t0x%016llx 0x%016llx\n", i, j,
slot->card.subpart[(i - 1) * 4 + j].dv_base,
slot->card.subpart[(i - 1) * 4 + j].dv_size);
}
}
mmc_log_debug(&log, "Found %d partitions and %d sub partitions\n",
part_count, sub_part_count);
slot->card.open_ct++;
assert(slot->card.open_ct == 1);
return OK;
}
/*===========================================================================*
* block_close *
*===========================================================================*/
static int
block_close(dev_t minor)
{
struct sd_slot *slot;
slot = get_slot(minor);
if (!slot) {
mmc_log_debug(&log,
"Not handling open on non existing slot\n");
return EIO;
}
/* if we arrived here we expect a card to be present, we will need do
* deal with removal later */
assert(slot->host != NULL);
assert(slot->card.open_ct >= 1);
/* If this is not the last open count simply decrease the counter and
* return */
if (slot->card.open_ct > 1) {
slot->card.open_ct--;
mmc_log_trace(&log, "decreased open count to %d\n",
slot->card.open_ct);
return OK;
}
assert(slot->card.open_ct == 1);
mmc_log_debug(&log,
"freeing the block device as it is no longer used\n");
/* release the card as check the open_ct should be 0 */
slot->host->card_release(&slot->card);
assert(slot->card.open_ct == 0);
return OK;
}
static int
copyto(endpoint_t dst_e,
cp_grant_id_t gr_id, vir_bytes offset, vir_bytes address, size_t bytes)
{
/* Helper function that used memcpy to copy data when the endpoint ==
* SELF */
if (dst_e == SELF) {
memcpy((char *) gr_id + offset, (char *) address, bytes);
return OK;
} else {
/* Read io_size bytes from our data at the correct * offset
* and write it to the output buffer at 0 */
return sys_safecopyto(dst_e, gr_id, offset, address, bytes);
}
}
static int
copyfrom(endpoint_t src_e,
cp_grant_id_t gr_id, vir_bytes offset, vir_bytes address, size_t bytes)
{
/* Helper function that used memcpy to copy data when the endpoint ==
* SELF */
if (src_e == SELF) {
memcpy((char *) address, (char *) gr_id + offset, bytes);
return OK;
} else {
return sys_safecopyfrom(src_e, gr_id, offset, address, bytes);
}
}
/*===========================================================================*
* block_transfer *
*===========================================================================*/
static int
block_transfer(dev_t minor, /* minor device number */
int do_write, /* read or write? */
u64_t position, /* offset on device to read or write */
endpoint_t endpt, /* process doing the request */
iovec_t * iov, /* pointer to read or write request vector */
unsigned int nr_req, /* length of request vector */
int flags /* transfer flags */
)
{
unsigned long counter;
iovec_t *ciov; /* Current IO Vector */
struct device *dev; /* The device used */
struct sd_slot *slot; /* The sd slot the requests is pointed to */
vir_bytes io_size; /* Size to read/write to/from the iov */
vir_bytes io_offset; /* Size to read/write to/from the iov */
vir_bytes bytes_written;
int r, blk_size, i;
/* Get the current "device" geometry */
dev = block_part(minor);
if (dev == NULL) {
mmc_log_warn(&log,
"Transfer requested on unknown device minor(%d)\n", minor);
/* Unknown device */
return ENXIO;
}
mmc_log_trace(&log, "I/O on minor(%d) %s at 0x%016llx\n", minor,
(do_write) ? "Write" : "Read", position);
slot = get_slot(minor);
assert(slot);
if (slot->card.blk_size == 0) {
mmc_log_warn(&log, "Request on a card with block size of 0\n");
return EINVAL;
}
if (slot->card.blk_size > COPYBUFF_SIZE) {
mmc_log_warn(&log,
"Card block size (%d) exceeds internal buffer size %d\n",
slot->card.blk_size, COPYBUFF_SIZE);
return EINVAL;
}
/* It is fully up to the driver to decide on restrictions for the
* parameters of transfers, in those cases we return EINVAL */
if (position % slot->card.blk_size != 0) {
/* Starting at a block boundary */
mmc_log_warn(&log,
"Requests must start at a block boundary"
"(start,block size)=(%016llx,%08x)\n", position,
slot->card.blk_size);
return EINVAL;
}
blk_size = slot->card.blk_size;
bytes_written = 0;
/* Are we trying to start reading past the end */
if (position >= dev->dv_size) {
mmc_log_warn(&log, "start reading past drive size\n");
return 0;
};
ciov = iov;
/* do some more validation */
for (counter = 0; counter < nr_req; counter++) {
assert(ciov != NULL);
if (ciov->iov_size % blk_size != 0) {
/* transfer a multiple of blk_size */
mmc_log_warn(&log,
"Requests must start at a block boundary "
"(start,block size)=(%016llx,%08x)\n", position,
slot->card.blk_size);
return EINVAL;
}
if (ciov->iov_size <= 0) {
mmc_log_warn(&log,
"Invalid iov size for iov %d of %d size\n",
counter, nr_req, ciov->iov_size);
return EINVAL;
}
ciov++;
}
ciov = iov;
for (counter = 0; counter < nr_req; counter++) {
/* Assume we are to transfer the amount of data given in the
* input/output vector but ensure we are not doing i/o past
* our own boundaries */
io_size = ciov->iov_size;
io_offset = position + bytes_written;
/* Check we are not reading/writing past the end */
if (position + bytes_written + io_size > dev->dv_size) {
io_size = dev->dv_size - (position + bytes_written);
};
mmc_log_trace(&log,
"I/O %s request(%d/%d) iov(grant,size,iosize,"
"offset)=(%d,%d,%d,%d)\n",
(do_write) ? "write" : "read", counter + 1, nr_req,
ciov->iov_addr, ciov->iov_size, io_size, io_offset);
/* transfer max one block at the time */
for (i = 0; i < io_size / blk_size; i++) {
if (do_write) {
/* Read io_size bytes from i/o vector starting
* at 0 and write it to out buffer at the
* correct offset */
r = copyfrom(endpt, ciov->iov_addr,
i * blk_size, (vir_bytes) copybuff,
blk_size);
if (r != OK) {
mmc_log_warn(&log,
"I/O write error: %s iov(base,size)=(%d,%d)"
" at offset=%d\n",
strerror(_SIGN r), ciov->iov_addr,
ciov->iov_size, io_offset);
return EINVAL;
}
/* write a single block */
slot->host->write(&slot->card,
(dev->dv_base / blk_size) +
(io_offset / blk_size) + i, 1, copybuff);
bytes_written += blk_size;
} else {
/* read a single block info copybuff */
slot->host->read(&slot->card,
(dev->dv_base / blk_size) +
(io_offset / blk_size) + i, 1, copybuff);
/* Read io_size bytes from our data at the
* correct offset and write it to the output
* buffer at 0 */
r = copyto(endpt, ciov->iov_addr, i * blk_size,
(vir_bytes) copybuff, blk_size);
if (r != OK) {
mmc_log_warn(&log,
"I/O read error: %s iov(base,size)=(%d,%d)"
" at offset=%d\n",
strerror(_SIGN r), ciov->iov_addr,
ciov->iov_size, io_offset);
return EINVAL;
}
bytes_written += blk_size;
}
}
ciov++;
}
return bytes_written;
}
/*===========================================================================*
* block_ioctl *
*===========================================================================*/
static int
block_ioctl(dev_t minor,
unsigned int request, endpoint_t endpt, cp_grant_id_t grant)
{
/* IOCTL handling */
struct sd_slot *slot;
mmc_log_trace(&log,
"enter (minor,request,endpoint,grant)=(%d,%lu,%d)\n", minor,
request, endpt, grant);
slot = get_slot(minor);
if (!slot) {
mmc_log_warn(&log,
"Doing ioctl on non existing block device(%d)\n", minor);
return EINVAL;
}
switch (request) {
case DIOCOPENCT:
// TODO: add a check for card validity */
mmc_log_trace(&log, "returning open count %d\n",
slot->card.open_ct);
/* return the current open count */
return sys_safecopyto(endpt, grant, 0,
(vir_bytes) & slot->card.open_ct,
sizeof(slot->card.open_ct));
case DIOCFLUSH:
/* No need to flush but some devices like movinands require
* 500 ms inactivity */
return OK;
}
return EINVAL;
}
/*===========================================================================*
* block_part *
*===========================================================================*/
static struct device *
block_part(dev_t minor)
{
/*
* Reuse the existing MINIX major/minor partitioning scheme.
* - 8 drives
* - 5 devices per drive allowing direct access to the disk and up to 4
* partitions (IBM style partitioning without extended partitions)
* - 4 Minix style sub partitions per partitions
*/
struct device *dev;
struct sd_slot *slot;
dev = NULL;
slot = get_slot(minor);
if (!slot) {
mmc_log_warn(&log,
"Device information requested for non existing partition "
"minor(%d)\n", minor);
return NULL;
}
if (!slot->host->card_detect(slot)) {
mmc_log_warn(&log,
"Device information requested from empty slot(%d)\n",
minor);
return NULL;
}
if (minor < 5) {
/* we are talking about the first disk */
dev = &slot->card.part[minor];
mmc_log_trace(&log,
"returning partition(%d) (base,size)=(0x%016llx,0x%016llx)\n",
minor, dev->dv_base, dev->dv_size);
} else if (minor >= 128 && minor < 128 + 16) {
/* sub partitions of the first disk we don't care about the
* rest */
dev = &slot->card.subpart[minor - 128];
mmc_log_trace(&log,
"returning sub partition(%d) (base,size)=(0x%016llx,0x%016llx)\n",
minor - 128, dev->dv_base, dev->dv_size);
} else {
mmc_log_warn(&log,
"Device information requested for non existing "
"partition minor(%d)\n", minor);
}
return dev;
}
/*===========================================================================*
* sef_local_startup *
*===========================================================================*/
static void
sef_local_startup()
{
mmc_log_info(&log, "Initializing the MMC block device\n");
if (apply_env()) {
mmc_log_warn(&log,
"Failed while applying environment settings\n");
exit(EXIT_FAILURE);
}
if (host.host_init(&host)) {
mmc_log_warn(&log,
"Failed to initialize the host controller\n");
exit(EXIT_FAILURE);
}
/*
* Register callbacks for fresh start, live update and restart.
* Use the same function for all event types
*/
sef_setcb_init_fresh(block_system_event_cb);
sef_setcb_init_lu(block_system_event_cb);
/* Register a signal handler */
sef_setcb_signal_handler(block_signal_handler_cb);
/* SEF startup */
sef_startup();
}
/*===========================================================================*
* block_system_event_cb *
*===========================================================================*/
static int
block_system_event_cb(int type, sef_init_info_t * info)
{
/*
* Callbacks for the System event framework as registered in
* sef_local_startup */
switch (type) {
case SEF_INIT_FRESH:
mmc_log_info(&log, "System event framework fresh start\n");
break;
case SEF_INIT_LU:
/* Restore the state. post update */
mmc_log_info(&log, "System event framework live update\n");
break;
case SEF_INIT_RESTART:
mmc_log_info(&log, "System event framework post restart\n");
break;
}
blockdriver_announce(type);
return OK;
}
/*===========================================================================*
* block_signal_handler_cb *
*===========================================================================*/
static void
block_signal_handler_cb(int signo)
{
struct sd_slot *slot;
mmc_log_debug(&log, "System event framework signal handler sig(%d)\n",
signo);
/* Only check for termination signal, ignore anything else. */
if (signo != SIGTERM)
return;
/* we only have a single slot and need an open count idealy we should
* iterate over the card to determine the open count */
slot = get_slot(0);
assert(slot);
if (slot->card.open_ct > 0) {
mmc_log_debug(&log, "Not responding to SIGTERM (open count=%d)\n",
slot->card.open_ct);
return;
}
mmc_log_info(&log, "MMC driver exit");
exit(0);
}
#define IS_MINIX_SUB_PARTITION_MINOR(minor) (minor >= MINOR_d0p0s0 )
static struct sd_slot *
get_slot(dev_t minor)
{
/*
* Get an sd_slot based on the minor number.
*
* This driver only supports a single card at at time. Also as
* we are following the major/minor scheme of other driver we
* must return a slot for all minors on disk 0 these are 0-5
* for the disk and 4 main partitions and
* number 128 till 144 for sub partitions.
*/
/* If this is a minor for the first disk (e.g. minor 0 till 5) */
if (minor / DEV_PER_DRIVE == 0) {
/* we are talking about the first disk and that is all we
* support */
return &host.slot[0];
} else if (IS_MINIX_SUB_PARTITION_MINOR(minor)
&& (((minor - MINOR_d0p0s0) / SUB_PER_DRIVE) == 0)) {
/* a minor from the first disk */
return &host.slot[0];
} else {
mmc_log_trace(&log,
"Device information requested for non existing partition "
"minor(%d)\n", minor);
return NULL;
}
}
static void
set_log_level(int level)
{
if (level < 0 || level >= 4) {
return;
}
mmc_log_info(&log, "Setting verbosity level to %d\n", level);
log.log_level = level;
if (host.set_log_level) {
host.set_log_level(level);
}
}
int
main(int argc, char **argv)
{
/* Set and apply the environment */
env_setargs(argc, argv);
sef_local_startup();
blockdriver_task(&mmc_driver);
return EXIT_SUCCESS;
}