During rigorous testing of our latest update infrastructure I came
across quite consistent timeouts on certain eMMC parts (e.g. Hynix
H26M21001ECR) when writing big (e.g. in excess of 400 MB) file system
images:
MMC write: dev # 0, block # 40960, count 944128 ...
mmc_send_cmd_bounced: MMC Timeout
Interrupt status 0x00000001
Interrupt status enable 0xdfff003b
Interrupt signal enable 0xdfff0002
Present status 0x01870106
mmc write failed
Comparing the various data sheets I came across the following timeout
specification:
Secure Erase/TRIM Timeout=300ms*2*10=6000ms
Unfortunately empirical testing still failed albeit much more rarely.
Increasing the timeout to 8000ms made it finally disappear entirely.
This patch allows us writing various eMMC parts without seeing any
further issues.
Signed-off-by: Marcel Ziswiler <marcel@ziswiler.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
If CONFIG_API is ever to be enabled on Tegra, this define must be set,
since api/api_storage.c uses it.
A couple of annoyting things about CONFIG_SYS_MMC_MAX_DEVICE
1) It isn't documented in README. The same is true for a lot of similar
defines used by api_storage.c.
2) It doesn't represent MAX_DEVICE but rather NUM_DEVICES, since the
valid values are 0..n-1 not 0..n.
However, I this patch does not address those shortcomings.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The way that struct mmc was implemented was a bit of a mess;
configuration and internal state all jumbled up in a single structure.
On top of that the way initialization is done with mmc_register leads
to a lot of duplicated code in drivers.
Typically the initialization got something like this in every driver.
struct mmc *mmc = malloc(sizeof(struct mmc));
memset(mmc, 0, sizeof(struct mmc);
/* fill in fields of mmc struct */
/* store private data pointer */
mmc_register(mmc);
By using the new mmc_create call one just passes an mmc config struct
and an optional private data pointer like this:
struct mmc = mmc_create(&cfg, priv);
All in tree drivers have been updated to the new form, and expect
mmc_register to go away before long.
Changes since v1:
* Use calloc instead of manually calling memset.
* Mark mmc_register as deprecated.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Using an array is pointless; even more pointless (and scary) is using
sprintf to fill it without a format string.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Remove the in-structure ops and put them in mmc_ops with
a constant pointer to it.
This makes the mmc structure smaller as well as conserving
code space (in theory).
All in-tree drivers are converted as well; this is done in a
single patch in order to not break git bisect.
Changes since V1:
Fix compilation b0rked issue on omap platforms where OMAP_GPIO was
not set.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Tegra124's MMC controller is very similar to earlier SoC generations,
and can be supported by the same driver.
However, there are some non-backwards-compatible HW differences, and
hence a new DT compatible value must be used to describe the HW. This
patch updates the driver to support that new compatible value.
That said, the HW differences are only relevant when enabling certain
high-performance transfer modes. Since the driver is currently very
simple and doesn't enable those modes, we don't actually need to address
any of these HW differences in the code yet, hence the simple nature of
this patch.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Tested-by: Thierry Reding <treding@nvidia.com>
Acked-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Warren <twarren@nvidia.com>
Tegra30 requires the SD Bus Voltage & Power bits be set in the SD
Power Control register. Tegra20 works w/o them set, but do it anyway
for those SoCs as it's part of the SD spec. Also call a common
board pad init routine (pad_init_mmc) in mmc_reset(), used by
Tegra30 only for now.
Note that Tegra20 SD/MMC HW differs enough from Tegra20 that a
new compatible entry is used in the fdt compat_names/id tables.
Signed-off-by: Tom Warren <twarren@nvidia.com>
Reviewed-by: Stephen Warren <swarren@nvidia.com>
tegra_mmc_init() now parses the DT info for bus width, WP/CD GPIOs, etc.
Tested on Seaboard, fully functional.
Tamonten boards (medcom-wide, plutux, and tec) use a different/new
dtsi file w/common settings.
Signed-off-by: Tom Warren <twarren@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@avionic-design.de>
Reviewed-by: Stephen Warren <swarren@nvidia.com>
Tegra's MMC driver does DMA, and hence needs cache-aligned buffers. In
some cases (e.g. user load commands) this cannot be guaranteed by callers
of the MMC APIs. To solve this, modify the Tegra MMC driver to use the
new bounce_buffer_*() APIs.
Note: Ideally, all U-Boot code will always provide address- and size-
aligned buffers, so a bounce buffer will only ever be needed for user-
supplied buffers (e.g. load commands). Ensuring this removes the need
for performance-sucking bounce buffer cache management and memcpy()s.
The one known exception at present is the SCR buffer in sd_change_freq(),
which is only 8 bytes long. Solving this requires enhancing struct
mmc_data to know the difference between buffer size and transferred data
size, or forcing all callers of mmc_send_cmd() to have allocated buffers
using ALLOC_CACHE_ALIGN_BUFFER(), which while true in this case, is not
enforced in any way at present, and so cannot be assumed by the core MMC
code.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Acked-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Andy Fleming <afleming@freescale.com>
If a board has all 8 data lines routed, the SD/MMC controller can still
operate in 4-bit (or presumably even 1-bit) mode. Adjust Tegra's MMC
driver to report the 4-bit capability even for 8-bit slots.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The move is pretty straight-forward. ap20.h and tegra20.h were renamed to ap.h and tegra.h.
Some files remain in arch-tegra20 but 'include' a file in 'arch-tegra' with #defines & structs
that will be common between T20 and T30 HW. HW-specific #defines, etc. stay in the 'arch-tegra20'
'root' file.
All boards build OK w/MAKEALL -s tegra20. Checkpatch.pl runs clean. Seaboard works OK.
Signed-off-by: Tom Warren <twarren@nvidia.com>
Convert TEGRA20_ defines to either TEGRA_ or NV_PA_ where appropriate.
Convert tegra20_ source file and function names to tegra_, also.
Upcoming Tegra30 port will use common code/defines/names where possible.
Signed-off-by: Tom Warren <twarren@nvidia.com>
Acked-by: Stephen Warren <swarren@nvidia.com>
This is make naming consistent with the kernel and devicetree and in
preparation of pulling out the common tegra20 code.
Signed-off-by: Allen Martin <amartin@nvidia.com>
Acked-by: Stephen Warren <swarren@wwwdotorg.org>
Tested-by: Thierry Reding <thierry.reding@avionic-design.de>
Signed-off-by: Tom Warren <twarren@nvidia.com>
In anticipation of Tegra3 support, continue removing/renaming
Tegra2-specific files. No functional changes (yet).
Updated copyrights to 2012.
Signed-off-by: Tom Warren <twarren@nvidia.com>
