| Age | Commit message (Collapse) | Author |
|---|
|
Currently the read throuput test tries:
- 1 page
- 2 pages
- 1 block (64 or more pages, usually)
But it might be interesting to see how the speed gradually increases,
eg. testing all number of pages from 1 to maybe 16, and then
arbitrarilly 32 and 64.
Let's add a -C parameter to enable this additional test.
The 2-page read/write tests are also moved under this new option.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
The read_eraseblock_by_2pages() has been generalized so it became
read_eraseblock_by_npages(), but there is no limitation (besides 0)
regarding the number of pages. Hence, drop the _by_page() helper and
replace it with the _by_npages(), using 'npages = 1'.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Right now there are only 2 pages that may be read continuously, but why
not trying more? At least when the continuous feature is out, this type
of benchmarking will be interesting. In order to facilitate later
additions, lets make this helper more generic and accept a global
'npages' variable as parameter (because this function is called in a
macro, it is simpler like that).
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
So far speed calculations have only be done 1 page at a time or 2 pages
at a time in a block; so basically all the block was always read because
all blocks are multiple of 2. But in the future, if we want to extend
the number of pages in a single read, the final number of pages actually
read might be less than an erase block size, hence failing the throuput
calculations.
Make the number of pages in a set explicit.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
I know no device without a multiple of 2 number of pages in each
block. Even though it might be the case, it is clearly not a big deal
and we don't really care about reading the last page, we are doing a
speed benchmark; so as long as the throughput calculation knows how much
data has been read it's fine. Eitherway, I don't think we ever have
fallen in this block because we would read the content of two pages (so
one past the block) and put it in a page-wide buffer, which would
probably lead to an out-of-bound abort.
Just drop the block.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
This introduces a new feature to the MTD command line utilities that
allows MTD devices to be referenced by name instead of device node. For
example this looks like:
> # Display info for the MTD device with name "data"
> mtdinfo mtd:data
> # Copy file to MTD device with name "data"
> flashcp /my/file mtd:data
This follows the syntax supported by the kernel which allows MTD
device's to be mounted by name[1].
Add the function mtd_find_dev_node() that accepts an MTD "identifier"
and returns the MTD's device node. The function accepts a string
starting with "mtd:" which it treats as the MTD's name. It then attempts
to search for the MTD, and if found maps it back to the /dev/mtdX device
node. If the string does not start with "mtd:", then assume it's the old
style and refers directly to a MTD device node.
The function is then hooked into existing tools like flashcp, mtdinfo,
flash_unlock, etc. To load in the new MTD parsing code in a consistent
way across programs.
[1] http://www.linux-mtd.infradead.org/faq/jffs2.html#L_mtdblock
Signed-off-by: Brandon Maier <brandon.maier@collins.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
The Read While Write (RWW) feature allows to perform reads from the
flash array into cache while a program (from cache) or an erase
operation happens, provided that the two areas are located on different
banks.
The main benefit is the possible reduced latency when requesting to read
a page while a much longer operation is ongoing, like a write or an
erase.
We can try to compare the positive impact of such a feature by enhancing
the flash_speed test tool with the following test:
- Measure the time taken by an eraseblock write in parallel with an
eraseblock read.
- Measure when the read operation ends.
- Compare the two to get the latency saved with the RWW feature.
To be sure the mtd_write actually starts (and acquires the necessary
locks) before the mtd_read does, we use SCHED_FIFO at rather high
(arbitrary) priorities, respectively 42 and 41.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
In order to be able to have interleaved measures, let's not use the
start and finish global variables from the time helpers directly,
provide parameters for these variables so that we can provide either the
global entries, or more specific ones when relevant.
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
The speed test does reads and writes of different sizes:
- eraseblock
- page
- two pages
At least this is the theory because, as opposed to the legacy kernel
module doing the same measurement, the userspace tool uses the subpage
size (hence accessing the same page 4, 8 or 16 times depending on the
subpage setting). Of course if the controller does not support subpages,
this issue is not visible.
Use mtd.min_io_size instead for non-NOR devices in order to get the
right bandwidth (at least one that fits the logs).
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
A common pattern in command line processing is having a usage()
function that prints out how to use the command line options and
then terminates.
The function is typically used inside a switch block for command
line options like `-h' or unknown options. In a lot of places, the
break keyword is omitted, because the function exits anyway. However,
this triggers gcc warnings about implicit fall-through.
Rather than adding a phony "/* fall-through */" this patch flags the
usage() style function with a gcc attribute, indicating that they do
not return and removes further superfluous break statements.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Basically a user space port of the mtd speed test kernel module.
In addition to the block offset and count module parameters, the
utility supports a block stride and can restore the block contents
after test. Furthermore, a flag can be used to disable destructive
tests (i.e. only perform read speed tests).
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
Signed-off-by: Richard Weinberger <richard@nod.at>
|
