Re: [PATCH tip/core/rcu 04/21] docs/memory-barriers.txt: Rewrite "KERNEL I/O BARRIER EFFECTS" section
From: Akira Yokosawa
Date: Thu Apr 04 2019 - 11:58:49 EST
On Tue, 2 Apr 2019 14:03:46 +0100, Will Deacon wrote:
> On Tue, Mar 26, 2019 at 04:41:16PM -0700, Paul E. McKenney wrote:
>> From: Will Deacon <will.deacon@xxxxxxx>
>> The "KERNEL I/O BARRIER EFFECTS" section of memory-barriers.txt is vague,
>> x86-centric, out-of-date, incomplete and demonstrably incorrect in places.
>> This is largely because I/O ordering is a horrible can of worms, but also
>> because the document has stagnated as our understanding has evolved.
>> Attempt to address some of that, by rewriting the section based on
>> recent(-ish) discussions with Arnd, BenH and others. Maybe one day we'll
>> find a way to formalise this stuff, but for now let's at least try to
>> make the English easier to understand.
>> Cc: "Paul E. McKenney" <paulmck@xxxxxxxxxxxxx>
>> Cc: Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx>
>> Cc: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
>> Cc: Arnd Bergmann <arnd@xxxxxxxx>
>> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
>> Cc: Andrea Parri <andrea.parri@xxxxxxxxxxxxxxxxxxxx>
>> Cc: Palmer Dabbelt <palmer@xxxxxxxxxx>
>> Cc: Daniel Lustig <dlustig@xxxxxxxxxx>
>> Cc: David Howells <dhowells@xxxxxxxxxx>
>> Cc: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>
>> Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
>> Cc: "Maciej W. Rozycki" <macro@xxxxxxxxxxxxxx>
>> Cc: Mikulas Patocka <mpatocka@xxxxxxxxxx>
>> Signed-off-by: Will Deacon <will.deacon@xxxxxxx>
>> Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxx>
>> Documentation/memory-barriers.txt | 115 ++++++++++++++++++------------
>> 1 file changed, 70 insertions(+), 45 deletions(-)
> If somebody could provide an Ack on this patch, I'd really appreciate it,
> please. Whilst the portable ordering guarantees that I've documented are
> fairly conservative, I do think that this change is a big improvement and
> gives you what you need if you're writing a portable device driver for a new
> piece of hardware. I'm tackling the removal of MMIOWB as a separate series.
> I think Paul now requires an Ack before he'll send a patch to mainline,
> hence the grovelling.
I'm afraid I'm not that qualified to provide an Ack to this patch,
but please find a nit fix below.
>> diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
>> index 1c22b21ae922..158947ae78c2 100644
>> --- a/Documentation/memory-barriers.txt
>> +++ b/Documentation/memory-barriers.txt
>> @@ -2599,72 +2599,97 @@ likely, then interrupt-disabling locks should be used to guarantee ordering.
>> KERNEL I/O BARRIER EFFECTS
>> -When accessing I/O memory, drivers should use the appropriate accessor
>> +Interfacing with peripherals via I/O accesses is deeply architecture and device
>> +specific. Therefore, drivers which are inherently non-portable may rely on
>> +specific behaviours of their target systems in order to achieve synchronization
>> +in the most lightweight manner possible. For drivers intending to be portable
>> +between multiple architectures and bus implementations, the kernel offers a
>> +series of accessor functions that provide various degrees of ordering
>> - (*) inX(), outX():
>> + (*) readX(), writeX():
>> - These are intended to talk to I/O space rather than memory space, but
>> - that's primarily a CPU-specific concept. The i386 and x86_64 processors
>> - do indeed have special I/O space access cycles and instructions, but many
>> - CPUs don't have such a concept.
>> + The readX() and writeX() MMIO accessors take a pointer to the peripheral
>> + being accessed as an __iomem * parameter. For pointers mapped with the
>> + default I/O attributes (e.g. those returned by ioremap()), then the
>> + ordering guarantees are as follows:
>> - The PCI bus, amongst others, defines an I/O space concept which - on such
>> - CPUs as i386 and x86_64 - readily maps to the CPU's concept of I/O
>> - space. However, it may also be mapped as a virtual I/O space in the CPU's
>> - memory map, particularly on those CPUs that don't support alternate I/O
>> - spaces.
>> + 1. All readX() and writeX() accesses to the same peripheral are ordered
>> + with respect to each other. For example, this ensures that MMIO register
>> + writes by the CPU to a particular device will arrive in program order.
>> - Accesses to this space may be fully synchronous (as on i386), but
>> - intermediary bridges (such as the PCI host bridge) may not fully honour
>> - that.
>> + 2. A writeX() by the CPU to the peripheral will first wait for the
>> + completion of all prior CPU writes to memory. For example, this ensures
>> + that writes by the CPU to an outbound DMA buffer allocated by
>> + dma_alloc_coherent() will be visible to a DMA engine when the CPU writes
>> + to its MMIO control register to trigger the transfer.
>> - They are guaranteed to be fully ordered with respect to each other.
>> + 3. A readX() by the CPU from the peripheral will complete before any
>> + subsequent CPU reads from memory can begin. For example, this ensures
>> + that reads by the CPU from an incoming DMA buffer allocated by
>> + dma_alloc_coherent() will not see stale data after reading from the DMA
>> + engine's MMIO status register to establish that the DMA transfer has
>> + completed.
>> - They are not guaranteed to be fully ordered with respect to other types of
>> - memory and I/O operation.
>> + 4. A readX() by the CPU from the peripheral will complete before any
>> + subsequent delay() loop can begin execution. For example, this ensures
>> + that two MMIO register writes by the CPU to a peripheral will arrive at
>> + least 1us apart if the first write is immediately read back with readX()
>> + and udelay(1) is called prior to the second writeX().
>> - (*) readX(), writeX():
>> + __iomem pointers obtained with non-default attributes (e.g. those returned
>> + by ioremap_wc()) are unlikely to provide many of these guarantees.
>> - Whether these are guaranteed to be fully ordered and uncombined with
>> - respect to each other on the issuing CPU depends on the characteristics
>> - defined for the memory window through which they're accessing. On later
>> - i386 architecture machines, for example, this is controlled by way of the
>> - MTRR registers.
>> + (*) readX_relaxed(), writeX_relaxed():
>> - Ordinarily, these will be guaranteed to be fully ordered and uncombined,
>> - provided they're not accessing a prefetchable device.
>> + These are similar to readX() and writeX(), but provide weaker memory
>> + ordering guarantees. Specifically, they do not guarantee ordering with
>> + respect to normal memory accesses or delay() loops (i.e bullets 2-4 above)
>> + but they are still guaranteed to be ordered with respect to other accesses
>> + to the same peripheral when operating on __iomem pointers mapped with the
>> + default I/O attributes.
>> - However, intermediary hardware (such as a PCI bridge) may indulge in
>> - deferral if it so wishes; to flush a store, a load from the same location
>> - is preferred[*], but a load from the same device or from configuration
>> - space should suffice for PCI.
>> + (*) readsX(), writesX():
>> - [*] NOTE! attempting to load from the same location as was written to may
>> - cause a malfunction - consider the 16550 Rx/Tx serial registers for
>> - example.
>> + The readsX() and writesX() MMIO accessors are designed for accessing
>> + register-based, memory-mapped FIFOs residing on peripherals that are not
>> + capable of performing DMA. Consequently, they provide only the ordering
>> + guarantees of readX_relaxed() and writeX_relaxed(), as documented above.
>> - Used with prefetchable I/O memory, an mmiowb() barrier may be required to
>> - force stores to be ordered.
>> + (*) inX(), outX():
>> - Please refer to the PCI specification for more information on interactions
>> - between PCI transactions.
>> + The inX() and outX() accessors are intended to access legacy port-mapped
>> + I/O peripherals, which may require special instructions on some
>> + architectures (notably x86). The port number of the peripheral being
>> + accessed is passed as an argument.
>> - (*) readX_relaxed(), writeX_relaxed()
>> + Since many CPU architectures ultimately access these peripherals via an
>> + internal virtual memory mapping, the portable ordering guarantees provided
>> + by inX() and outX() are the same as those provided by readX() and writeX()
>> + respectively when accessing a mapping with the default I/O attributes.
>> - These are similar to readX() and writeX(), but provide weaker memory
>> - ordering guarantees. Specifically, they do not guarantee ordering with
>> - respect to normal memory accesses (e.g. DMA buffers) nor do they guarantee
>> - ordering with respect to LOCK or UNLOCK operations. If the latter is
>> - required, an mmiowb() barrier can be used. Note that relaxed accesses to
>> - the same peripheral are guaranteed to be ordered with respect to each
>> - other.
>> + Device drivers may expect outX() to emit a non-posted write transaction
>> + that waits for a completion response from the I/O peripheral before
>> + returning. This is not guaranteed by all architectures and is therefore
>> + not part of the portable ordering semantics.
>> + (*) insX(), outsX():
>> + As above, the insX() and outX() accessors provide the same ordering
>> + guarantees as readsX() and writesX() respectively when accessing a mapping
>> + with the default I/O attributes.
>> (*) ioreadX(), iowriteX()
>> These will perform appropriately for the type of access they're actually
>> doing, be it inX()/outX() or readX()/writeX().
>> +All of these accessors assume that the underlying peripheral is little-endian,
>> +and will therefore perform byte-swapping operations on big-endian architectures.
>> +Composing I/O ordering barriers with SMP ordering barriers and LOCK/UNLOCK
>> +operations is a dangerous sport which may require the use of mmiowb(). See the
>> +subsection "Acquires vs I/O accesses" for more information.
>> ASSUMED MINIMUM EXECUTION ORDERING MODEL
JFYI, there is another document Documentation/driver-api/device-io.rst,
which is somewhat related to this update. It looks like this one also needs
some update, as Jon commented in transforming to .rst format in commit
8a8a602fdb83 ("docs: Convert the deviceio template to RST"):
Like the rest of our documentation, this one could use some work. There's
no mention of ioremap() and friends, no mention of io_read*() and friends.
But we have nice documentation for all those folks writing new drivers that
do port I/O :).
This commit was merged in v4.11 cycle. And there has been no update whatsoever
since. mmiowb() is lightly mentioned therein. IMHO, just updating
memory-barriers.txt would widen the gap of information.