Fixes:
* scsi: Fix a bug that caused the device capacity to be set
to an undefined value for some large SCSI devices when
READ CAPACITY (16) was used
* ahci: Fix VPD page reporting so that it does not return
undefined values
* ahci: Set the write bit to true when sending a DATA SET
MANAGEMENT (trim) command to a device. The command would
otherwise fail and time out on some devices.
Improvements:
* scsi: Extend the READ CAPACITY (16) support to also
include logical block provisioning information
* scsi: Prefer READ CAPACITY (16) over READ CAPACITY (10)
on devices that are expected to support this command
* scsi, ahci: Enable trim on SCSI and SATA devices that
are expected to support trim and which correctly report
trim support
* ahci: Redo the implementation of the SCSI UNMAP command
* scsi: Redo UNMAP-related code
* scsi: Add support for UNMAP via WRITE SAME (10) and
WRITE SAME (16) commands
* When copying trim ranges between different data types,
make sure that the values don't change (detect overflows)
* Report the number of trimmed blocks even if the trim
operation fails
Change-Id: Ie5fc993bbbc19546b4308138ba10184bf7b9986a
Reviewed-on: https://review.haiku-os.org/c/haiku/+/4157
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Reviewed-by: Adrien Destugues <pulkomandy@gmail.com>
Store the bus cookie in the mmc_disk driver and pass it to the bus
manager when executing commands. This avoids calling into the device
manager at each read and write operation. The code to get the cookie
from mmc_disk isn't so nice since it needs to access the grandparent
device (the mmc bus root), it would be simpler if this cookie would be
available directly from mmc bus devices.
We can get card removal and card insertion interrupt at the same time
due to insufficient hardware debouncing (the SDHCI spec says we
shouldn't, but it happens on Ricoh controllers. Can't blame them, they
don't advertise themselves as compliant with the spec). So, check the
card status from the interrupt handler and ignore the incorrect
interrupts.
Fix unreliable card initialization: power must be turned on before
starting up the SD clock. Remove a now unneeded delay that was added in
an attempt to avoid initial instability.
Change-Id: Ibd8d051da1a1d859f3924ee535f4a05d9b6398d4
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3639
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
It works, but performance is still unexpectedly low (getting about
50kB/s write speed) with almost no CPU load.
Change-Id: I7da3ee70c8b379c4e6c2250d67f880c78635874f
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3630
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
- Switch to 25MHz clock
- Switch to 4bit transfers mode (the default is 1bit)
Reading and writing SD cards do not seem to work anymore with these
changes. I get invalid data on read, and on write, an interrupt is never
called in some cases.
The main differences:
- The initialization sequence requires an additional command (this was
already done)
- The layout of the CSD register and the way to compute the device
geometry from it changes
- The read and write commands parameter is a sector number instead of a
byte position
Change-Id: Ie729e333c9748f36b37acd70c970adfd425cf0b6
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3512
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
The SDHCI spec also offers an "advanced DMA" mode where we can use
scatter-gather lists. It would allow to remove several of the DMA
restrictions, but hardware support for it is optional, so we need this
version anyway.
The geometry is retrieved on demand in the first read or write or in a
call to the get geometry or get device size ioctl. It is not possible to
retrieve it from the device initialization because that is called as
part of the mmc_bus scanning, which needs a specific sequence of
commands and keeps the bus locked to prevent drivers to insert their own
commands in the middle of that sequence.
TODO:
- Move the DMA restrictions definition to sdhci_pci and forward it up to
mmc_disk (which is the one creating the IOScheduler)
- Decide if we want to keep non-DMA support (probably should, but it
makes things more complex, because it uses virtual addresses)
Change-Id: Ib1dd14eacf62052d747bfb3ef7820bc5a34d3030
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3471
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
First implementation of reading sectors from an SD card.
This is not the best performance for many reasons:
- No DMA
- Reads only one sector at a time
- Cannot read more than 512 bytes per syscall
Also there are major limitations:
- Cannot read less than 512 bytes. The hardware of course works in full
sectors. The mmc_disk driver should go through the io scheduler to
make sure requests have a reasonable size and offset, and nothing
tries to read just a few bytes in the middle of a sector.
- SD cards only (no SDHC, no MMC)
Architecture problems:
I think too much of the implementation is done in sdhci_pci and should
be moved to the upper layers. However it is difficult to say without
having implemented DMA (which indeed will be at the low level of the
sdhci controller). It doesn't help that the order of operations is a
bit different depending on wether there is DMA or not. In DMA mode you
first prepare the buffer, then run the command. In non-DMA mode you
first send the command, then read the data into the buffer. We need an
API at the mmc_bus level that doesn't care about that low-level detail.
There are other things that the MMC bus should be doing however, such
as switching to different clock speeds depending on which card is
activated and how fast it can go.
At least the following should be done:
- The read method for mmc_bus and sdhci_pci should use a scatter-gather
structure as a parameter instead of a single buffer
- See if can be integrated into ExecuteCommand at sdhci level (it's
essentially a command with an additional data phase)
Change-Id: I688b6c694561074535c9c0c2545f06dc04b06e7d
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3466
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
No read and write support for now. But we implement getting SD card
capacity. SDHC is not supported yet (it uses a different layout for the
CSD register which will be rejected by this version of the code)
Change-Id: Ife844a62f3846c0a780259e9a3a08195e2fd965e
Reviewed-on: https://review.haiku-os.org/c/haiku/+/1068
Reviewed-by: Jérôme Duval <jerome.duval@gmail.com>
For now it just forwards the command to the SDHCI controller.
The bus will gain more features and functions as work advances (tracking
which card is active, arbitration of DMA transfers, etc).
Change-Id: I094eb84f27e7789387a3f8fb65fba1e5fcfa3e8a
Reviewed-on: https://review.haiku-os.org/c/haiku/+/3094
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
sdhci:
- Add semaphore for interrupt management
- Add basic operations (setting clock, executing a command)
- Add early initialization (clocks and power up)
- Wrap the bus in a C++ class to ease usage
- Expose API to MMC bus manager
- TODO: manage card insertion and removal interrupts
- TODO: use MSI when available
mmc_bus:
- Implements SD card management independant of the way we access the bus
(later on different drivers can provide the same API as SDHCI)
- Worker thread to do the initialization
- Implement card initialization process up until getting an RCA from the
card. This is the generic part to assign an ID to the card, after this
point commands can be targetted at the specific card so it can be
handed over to the mmc_disk driver.
- TODO: initialization for non-SDHC cards which do not reply to CMD8.
Change-Id: I71950ca3ce206378a68fa7f97c19f638183d6cdd
Reviewed-on: https://review.haiku-os.org/c/1032
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
* This iSCSI implementation only worked on PPC big-endian atm.
* We're pretty sure iSCSI support in haiku_loader doesn't make
much sense anymore. iPXE on (on arm,x86,etc EFI/BIOS platforms)
supports iSCSI boot of disks.
* Haiku could use a iSCSI driver add-on, but it would exist much
higher up and likely use standard drivers vs bare-minimum iSCSI
target impementations.
* Leaving TCP and adding to all arches since it could make sense
for haiku's native network disk subsystem or network debugging?
Change-Id: Ic181b93a1d8ffd77f69e00e372b44b79abbddb42
Reviewed-on: https://review.haiku-os.org/c/899
Reviewed-by: Alex von Gluck IV <kallisti5@unixzen.com>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Some users of MK_ERROR pass in parameters from a variable called "res",
which is obviously not what they want to do, as that will use this "res"
and not theirs.
Spotted by Clang.
* Added B_ prefix.
* Renamed 16 bit variants to B_LENDIAN16_*.
* Added 32 bit variants (albeit only 16 of them for now).
* Adjusted headers that were using them.
* Move to more standardized functions matching AHCI spec
* Don't perform unnecessary double port resets
* Begin implementing a software reset to try first per spec.
Software reset needs more work, falls through to port reset
for the moment which is stable.
* Don't duplicate ATA defines, use what we already provide.
* Tested working on VirtualBox 1-16 AHCI ports, Intel C200,
and AMD FCH.
* The UNMAP command is theoretically much faster, as it can get many block
ranges instead of just a single range.
* Furthermore, the ATA TRIM command resembles it much better.
* Therefore, fs_trim_data now gets an array of ranges, and we use SCSI UNMAP
to trim.
* Updated BFS code to collect array ranges to fully support the new
fs_trim_data possibilities.
* imported asc-num.txt as a reference, was used to generate the asc sense table.
* use the sense asc and key tables to know which action and status codes are
to be applied.
* tested with an hard disk and a dvd reader.
* these tables could be reused by the scsi_periph module.
* The default module is replaced by the Virtio RNG module when found.
* This can have the undesired effect of rendering /dev/urandom slow.
* Tested with the following QEmu command line option:
-device virtio-rng-pci,rng=rng0 -object rng-random,filename=/dev/random,id=rng0
* moved random.h to private/drivers headers.
* added support READ_12/16 and WRITE_12/16 in ata and scsi_periph, this enables read/write on block offsets greater than 2TB
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39278 a95241bf-73f2-0310-859d-f6bbb57e9c96
* ata: don't fail if lba_sector_count is null and lba48_sector_count is not
* scsi_periph: if ReadCapacity() returns 0xffffffff, use ReadCapacity16() instead
* scsi_disk: use a different computation in the struct geometry computation for bigger disks
Tested successfully with a virtual 10TB hard drive.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39252 a95241bf-73f2-0310-859d-f6bbb57e9c96
Define structs for iSCSI messages, to be used by boot loader and kernel add-on.
For now it will refuse to compile for Little Endian systems (e.g., x86).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38528 a95241bf-73f2-0310-859d-f6bbb57e9c96
Based on a SCSI Command Reference Manual by Seagate and Wikipedia.
Tested against OpenSolaris iSCSI target on ppc. It choked on the READ (12) opcode.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38425 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Internally, moved the contents of periph_io() into a static read_write()
function, and use it from the new periph_read_write() as well.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@36988 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Copied ide_adapter.h as ata_adapter.h in attempt to further
separate the two stacks.
* Continued renaming stuff in drivers/bus/ATA.h
* Make all the busses/ata drivers include the new headers,
specifically ata_types.h, ata_adapter.h and bus/ATA.h,
they were all including ide_types and bus/IDE.h still
* Some renaming of global variables for coding style consistency
* Removed the promise driver from the build, it's not used on the
image and I don't believe it compiled even for the old IDE stack.
* There is no more Command Queueing in the new ATA stack, so I
removed the capability indication from the busses/ata drivers
and ata_adapter.h.
The new ATA stack still boots fine on my computer and I proof-read
the diff like two times. Basically, this was a careful search&replace
job only. The only things I am not sure about is renaming some
publishing related strings, but it seems to all work fine.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@30700 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Reworked some of the transfer handling after reading the specs.
* Ensure that the device selection bit is set correctly for all commands.
* Generally disable interrupts and enable them only when expecting a DMA one.
* Renamed disk failure to device fault according to specs.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@30248 a95241bf-73f2-0310-859d-f6bbb57e9c96
but doesn't really do anything more than before.
* It also replaces everything IDE with ATA counterparts and cleans up a lot
of the definitions.
* Cleaning up the old ATA bus_manager as well as some license headers missing.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@30049 a95241bf-73f2-0310-859d-f6bbb57e9c96
ide_mask_sector_count_48, and ide_mask_LBA_*_48 were all wrong.
* Using the high byte in LBA48 mode should work now, too (wasn't written
to the IDE controller before, but that shouldn't have been a problem yet with
today's disks).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@28340 a95241bf-73f2-0310-859d-f6bbb57e9c96
our dma_restrictions structure (but we're using blocks instead of bytes,
since unlike the block size, the restrictions attributes are constant).
* We might want to use blocks for the dma_restrictions structure as well in
the future...
* Fixed another bug in the device_node variant of DMAResource::Init(): the max
segment size was specified in blocks as well.
* Removed the "hardcode" block_io module and header.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26973 a95241bf-73f2-0310-859d-f6bbb57e9c96
architecture: for now, we do this on the lowest layer only, therefore all
requests are handled synchronously (ie. in the scheduler's thread).
* Instead of using the block_io module, scsi_disk (and scsi_cd) are now
exporting a device on their own, and use an I/O scheduler with an appropriate
DMA resource.
* There are still lots of TODOs, and it can easily panic - don't update if
you intend to demo Haiku.
* scsi_periph now only has an io() function that get an io_operation, instead
of the previous read/write functions, moved preferred CCB size from those
functions into the device registration.
* Changed all scsi_periph files to C++.
* scsi_cd ported, too, but untested.
* Removed block_io from image - it will be removed completely soon.
* Temporarily commented an ASSERT() in the ATA bus manager (in case you use
it); it's sometimes triggered by the code now, and I haven't yet looked into
the issue -- doesn't seem to harm, at least.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@26828 a95241bf-73f2-0310-859d-f6bbb57e9c96
