Use the generic dynamic interrupt moderation (dim) framework to
implement adaptive interrupt coalescing on Rx. With the per-packet
interrupt scheme, a high interrupt rate has been noted for moderate
traffic flows leading to high CPU utilization.
The dpio driver exports new functions to enable/disable adaptive IRQ
coalescing on a DPIO object, to query the state or to update Net DIM
with a new set of bytes and frames dequeued.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In DPAA2 based SoCs, the IRQ coalesing support per software portal has 2
configurable parameters:
- the IRQ timeout period (QBMAN_CINH_SWP_ITPR): how many 256 QBMAN
cycles need to pass until a dequeue interrupt is asserted.
- the IRQ threshold (QBMAN_CINH_SWP_DQRR_ITR): how many dequeue
responses in the DQRR ring would generate an IRQ.
Add support for setting up and querying these IRQ coalescing related
parameters.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Through the dpio_get_attributes() firmware call the dpio driver has
access to the QBMAN clock frequency. Extend the structure which holds
the firmware's response so that we can have access to this information.
This will be needed in the next patches which also add support for
interrupt coalescing which needs to be configured based on the
frequency.
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 6ac9b61786.
This commit was required because at that time, ioread/iowrite
functions were sub-optimal on powerpc/32 compared to the
architecture specific in_/out_ IO accessors.
But there are now equivalent since
commit 894fa235eb ("powerpc: inline iomap accessors").
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
The bd_mem_part member of ucc_geth_info always has the value
MEM_PART_SYSTEM, and AFAICT, there has never been any code setting it
to any other value. Moreover, muram is a somewhat precious resource,
so there's no point using that when normal memory serves just as well.
Apart from removing a lot of dead code, this is also motivated by
wanting to clean up the "store result from kmalloc() in a u32" mess.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a helper that takes a virtual address rather than the muram
offset. This will be used in a couple of places to avoid having to
store both the offset and the virtual address, as well as removing
NULL checks from the callers.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Acked-by: Li Yang <leoyang.li@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
dpaa_eth_napi_schedule() and caam_qi_napi_schedule() schedule NAPI if
invoked from:
- Hard interrupt context
- Any context which is not serving soft interrupts
Any context which is not serving soft interrupts includes hard interrupts
so the in_irq() check is redundant. caam_qi_napi_schedule() has a comment
about this:
/*
* In case of threaded ISR, for RT kernels in_irq() does not return
* appropriate value, so use in_serving_softirq to distinguish between
* softirq and irq contexts.
*/
if (in_irq() || !in_serving_softirq())
This has nothing to do with RT. Even on a non RT kernel force threaded
interrupts run obviously in thread context and therefore in_irq() returns
false when invoked from the handler.
The extension of the in_irq() check with !in_serving_softirq() was there
when the drivers were added, but in the out of tree FSL BSP the original
condition was in_irq() which got extended due to failures on RT.
The usage of in_xxx() in drivers is phased out and Linus clearly requested
that code which changes behaviour depending on context should either be
separated or the context be conveyed in an argument passed by the caller,
which usually knows the context. Right he is, the above construct is
clearly showing why.
The following callchains have been analyzed to end up in
dpaa_eth_napi_schedule():
qman_p_poll_dqrr()
__poll_portal_fast()
fq->cb.dqrr()
dpaa_eth_napi_schedule()
portal_isr()
__poll_portal_fast()
fq->cb.dqrr()
dpaa_eth_napi_schedule()
Both need to schedule NAPI.
The crypto part has another code path leading up to this:
kill_fq()
empty_retired_fq()
qman_p_poll_dqrr()
__poll_portal_fast()
fq->cb.dqrr()
dpaa_eth_napi_schedule()
kill_fq() is called from task context and ends up scheduling NAPI, but
that's pointless and an unintended side effect of the !in_serving_softirq()
check.
The code path:
caam_qi_poll() -> qman_p_poll_dqrr()
is invoked from NAPI and I *assume* from crypto's NAPI device and not
from qbman's NAPI device. I *guess* it is okay to skip scheduling NAPI
(because this is what happens now) but could be changed if it is wrong
due to `budget' handling.
Add an argument to __poll_portal_fast() which is true if NAPI needs to be
scheduled. This requires propagating the value to the caller including
`qman_cb_dqrr' typedef which is used by the dpaa and the crypto driver.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Aymen Sghaier <aymen.sghaier@nxp.com>
Cc: Herbert XS <herbert@gondor.apana.org.au>
Cc: Li Yang <leoyang.li@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Madalin Bucur <madalin.bucur@oss.nxp.com>
Tested-by: Camelia Groza <camelia.groza@nxp.com>
This fixes compile warnings from the -Wpacked-not-aligned option.
In file included from ../drivers/crypto/caam/qi.c:12:
../include/soc/fsl/qman.h:259:1: warning: alignment 1 of ‘struct qm_dqrr_entry’ is less than 8 [-Wpacked-not-aligned]
} __packed;
^
../include/soc/fsl/qman.h:292:2: warning: alignment 1 of ‘struct <anonymous>’ is less than 8 [-Wpacked-not-aligned]
} __packed ern;
^
Reported-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
The current codebase makes use of one-element arrays in the following
form:
struct something {
int length;
u8 data[1];
};
struct something *instance;
instance = kmalloc(sizeof(*instance) + size, GFP_KERNEL);
instance->length = size;
memcpy(instance->data, source, size);
but the preferred mechanism to declare variable-length types such as
these ones is a flexible array member[1][2], introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on. So, replace
the one-element array with a flexible-array member.
Also, make use of the new struct_size() helper to properly calculate the
size of struct qe_firmware.
This issue was found with the help of Coccinelle and, audited and fixed
_manually_.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Reviewed-by: Qiang Zhao <qiang.zhao@nxp.com>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
Update of QMAN the interface to enqueue frame. We now support multiple
enqueue (qbman_swp_enqueue_multiple) and multiple enqueue with
a table of descriptor (qbman_swp_enqueue_multiple_desc).
Signed-off-by: Youri Querry <youri.querry_1@nxp.com>
Acked-by: Roy Pledge <roy.pledge@nxp.com>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
When building this on a 64-bit platform gcc rightly warns that the
error checking is broken (-ENOMEM stored in an u32 does not compare
greater than (unsigned long)-MAX_ERRNO). Instead, change the
ucc_fast_[tr]x_virtual_fifo_base_offset members to s32 and use an
ordinary check-for-negative. Also, this avoids treating 0 as "this
cannot have been returned from qe_muram_alloc() so don't free it".
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
When trying to build this for a 64-bit platform, one gets warnings
from using IS_ERR_VALUE on something which is not sizeof(long).
Instead, change the various *_offset fields to store a signed integer,
and simply check for a negative return from qe_muram_alloc(). Since
qe_muram_free() now accepts and ignores a negative argument, we only
need to make sure these fields are initialized with -1, and we can
just unconditionally call qe_muram_free() in ucc_slow_free().
Note that the error case for us_pram_offset failed to set that field
to 0 (which, as noted earlier, is anyway a bogus sentinel value).
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
Nobody uses the return value from cpm_muram_free, and functions that
free resources usually return void. One could imagine a use for a "how
much have I allocated" a la ksize(), but knowing how much one had
access to after the fact is useless.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
There are a number of problems with cpm_muram_alloc() and its
callers. Most callers assign the return value to some variable and
then use IS_ERR_VALUE to check for allocation failure. However, when
that variable is not sizeof(long), this leads to warnings - and it is
indeed broken to do e.g.
u32 foo = cpm_muram_alloc();
if (IS_ERR_VALUE(foo))
on a 64-bit platform, since the condition
foo >= (unsigned long)-ENOMEM
is tautologically false. There are also callers that ignore the
possibility of error, and then there are those that check for error by
comparing the return value to 0...
One could fix that by changing all callers to store the return value
temporarily in an "unsigned long" and test that. However, use of
IS_ERR_VALUE() is error-prone and should be restricted to things which
are inherently long-sized (stuff in pt_regs etc.). Instead, let's aim
for changing to the standard kernel style
int foo = cpm_muram_alloc();
if (foo < 0)
deal_with_it()
some->where = foo;
Changing the return type from unsigned long to s32 (aka signed int)
doesn't change the value that gets stored into any of the callers'
variables except if the caller was storing the result in a u64 _and_
the allocation failed, so in itself this patch should be a no-op.
Another problem with cpm_muram_alloc() is that it can certainly
validly return 0 - and except if some cpm_muram_alloc_fixed() call
interferes, the very first cpm_muram_alloc() call will return just
that. But that shows that both ucc_slow_free() and ucc_fast_free() are
buggy, since they assume that a value of 0 means "that field was never
allocated". We'll later change cpm_muram_free() to accept (and ignore)
a negative offset, so callers can use a sentinel of -1 instead of 0
and just unconditionally call cpm_muram_free().
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
asm/cpm.h under arch/powerpc is now just a wrapper for including
soc/fsl/cpm.h. In order to make the qe.h header usable on other
architectures, use the latter path directly.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
Some drivers, e.g. ucc_uart, need definitions from cpm.h. In order to
allow building those drivers for non-ppc based SOCs, move the header
to include/soc/fsl. For now, leave a trivial wrapper at the old
location so drivers can be updated one by one.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
The public qe_ic.h header is no longer included by anything but
qe_ic.c. Merge both headers into qe_ic.c, and drop the unused
constants.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
There are no current callers of these functions, and they use the
ppc-specific virq_to_hw(). So removing them gets us one step closer to
building QE support for ARM.
If the functionality is ever actually needed, the code can be dug out
of git and then adapted to work on all architectures, but for future
reference please note that I believe qe_ic_set_priority is buggy: The
"priority < 4" should be "priority <= 4", and in the else branch 24
should be replaced by 28, at least if I'm reading the data sheet right.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
These functions are only ever called through a function pointer, and
therefore it makes no sense for them to be "static inline" - gcc has
no choice but to emit a copy in each translation unit that takes the
address of one of these. Since they are now only referenced from
qe_ic.c, just make them local to that file.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
Having to call qe_ic_init() from platform-specific code makes it
awkward to allow building the QE drivers for ARM. It's also a needless
duplication of code, and slightly error-prone: Instead of the caller
needing to know the details of whether the QUICC Engine High and QUICC
Engine Low are actually the same interrupt (see e.g. the machine_is()
in mpc85xx_mds_qeic_init), just let the init function choose the
appropriate handlers after it has parsed the DT and figured it out. If
the two interrupts are distinct, use separate handlers, otherwise use
the handler which first checks the CHIVEC register (for the high
priority interrupts), then the CIVEC.
All existing callers pass 0 for flags, so continue to do that from the
new single caller. Later cleanups will remove that argument
from qe_ic_init and simplify the body, as well as make qe_ic_init into
a proper init function for an IRQCHIP_DECLARE, eliminating the need to
manually look up the fsl,qe-ic node.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
The *_ipic and *_mpic handlers are almost identical - the only
difference is that the latter end with an unconditional
chip->irq_eoi() call. Since IPIC does not have ->irq_eoi, we can
reduce some code duplication by calling irq_eoi conditionally.
This is similar to what is already done in mpc8xxx_gpio_irq_cascade().
This leaves the functions slightly misnamed, but that will be fixed in
a subsequent patch.
Reviewed-by: Timur Tabi <timur@kernel.org>
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Li Yang <leoyang.li@nxp.com>
