Saeed Mahameed says:
====================
Mellanox mlx5e fixes for 4.11-rc1
This series includes some important bug fixes for mlx5e driver.
Three misc fixes:
From Mohamad, compilation fix on s390 system
From Me, A fix for driver unload when switchdev mode is on.
From Tariq, HW LRO frag size optimization for when build_skb is not used
(striding RQ mode).
Three CQE compression related fixes:
Two fixes from Tariq and I, to correctly setup CQE compression
parameters on driver load and on arbitrary user modifications.
Last patch, fixes a very critical issue that was originally reported
by Tom, where the driver reported csum errors or even page ref issues
for when cqe compression is enabled and rapidly active.
For your convenience this series was generated on top of net-next branch:
005c3490e9 ('Revert "ath10k: Search SMBIOS for OEM board file extension"')
for -stable:
net/mlx5e: Register/unregister vport representors on interface (for kernel >= 4.9)
net/mlx5e: Do not reduce LRO WQE size when not using build_skb (for kernel >= 4.9)
net/mlx5e: Fix broken CQE compression initialization (for kernel >= 4.9)
net/mlx5e: Update MPWQE stride size when modifying CQE compress state (for kernel >= 4.7)
net/mlx5e: Fix wrong CQE decompression (for kernel >= 4.7)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In cqe compression with striding RQ, the decompression of the CQE field
wqe_counter was done with a wrong wraparound value.
This caused handling cqes with a wrong pointer to wqe (rx descriptor)
and creating SKBs with wrong data, pointing to wrong (and already consumed)
strides/pages.
The meaning of the CQE field wqe_counter in striding RQ holds the
stride index instead of the WQE index. Hence, when decompressing
a CQE, wqe_counter should have wrapped-around the number of strides
in a single multi-packet WQE.
We dropped this wrap-around mask at all in CQE decompression of striding
RQ. It is not needed as in such cases the CQE compression session would
break because of different value of wqe_id field, starting a new
compression session.
Tested:
ethtool -K ethxx lro off/on
ethtool --set-priv-flags ethxx rx_cqe_compress on
super_netperf 16 {ipv4,ipv6} -t TCP_STREAM -m 50 -D
verified no csum errors and no page refcount issues.
Fixes: 7219ab34f1 ("net/mlx5e: CQE compression")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reported-by: Tom Herbert <tom@herbertland.com>
Cc: kernel-team@fb.com
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the admin enables/disables cqe compression, updating
mpwqe stride size is required:
CQE compress ON ==> stride size = 256B
CQE compress OFF ==> stride size = 64B
This is already done on driver load via mlx5e_set_rq_type_params, all we
need is just to call it on arbitrary admin changes of cqe compression
state via priv flags or when changing timestamping state
(as it is mutually exclusive with cqe compression).
This bug introduces no functional damage, it only makes cqe compression
occur less often, since in ConnectX4-LX CQE compression is performed
only on packets smaller than stride size.
Tested:
ethtool --set-priv-flags ethxx rx_cqe_compress on
pktgen with 64 < pkt size < 256 and netperf TCP_STREAM (IPv4/IPv6)
verify `ethtool -S ethxx | grep compress` are advancing more often
(rapidly)
Fixes: 7219ab34f1 ("net/mlx5e: CQE compression")
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Cc: kernel-team@fb.com
Signed-off-by: David S. Miller <davem@davemloft.net>
Some of RQ type parameters are derived from CQE compression state flag,
CQE compression flag was initialized only after RQ type parameters
setup. This leads to load RQ with stride size smaller than what we
want for when CQE compression is on.
This bug introduces no functional damage, it only makes CQE compression
occur less often, since in ConnectX4-LX CQE compression is performed
only on packets smaller than stride size.
Fix this by marking default status of CQE compression in PFLAG prior to
calling mlx5e_set_rq_priv_params(), as it inits some fields based on it.
Tested:
load driver on systems where rx CQE compress will be on (MH)
pktgen with 64 < pkt size < 256 and netperf TCP_STREAM (IPv4/IPv6)
verify `ethtool -S ethxx | grep compress` are advancing more often
(rapidly)
Fixes: 2fc4bfb725 ("net/mlx5e: Dynamic RQ type infrastructure")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Cc: kernel-team@fb.com
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When rq_type is Striding RQ, no room of SKB_RESERVE is needed
as SKB allocation is not done via build_skb.
Fixes: e4b8550807 ("net/mlx5e: Slightly reduce hardware LRO size")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently vport representors are added only on driver load and removed on
driver unload. Apparently we forgot to handle them when we added the
seamless reset flow feature. This caused to leave the representors
netdevs alive and active with open HW resources on pci shutdown and on
error reset flows.
To overcome this we move their handling to interface attach/detach, so
they would be cleaned up on shutdown and recreated on reset flows.
Fixes: 26e59d8077 ("net/mlx5e: Implement mlx5e interface attach/detach callbacks")
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Hadar Hen Zion <hadarh@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We must hold the rcu read lock across looking up glocks and trying to
bump their refcount to prevent the glocks from being freed in between.
Cc: <stable@vger.kernel.org> # 4.3+
Signed-off-by: Andrew Price <anprice@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
A request is assigned a global seqno only when it is on the hardware
execution queue. The global seqno can be used to maintain a list of
requests on the same engine in retirement order, for example for
constructing a priority queue for waiting. Prior to its execution, or
if it is subsequently removed in the event of preemption, its global
seqno is zero. As both insertion and removal from the execution queue
may operate in IRQ context, it is not guarded by the usual struct_mutex
BKL. Instead those relying on the global seqno must be prepared for its
value to change between reads. Only when the request is complete can
the global seqno be stable (due to the memory barriers on submitting
the commands to the hardware to write the breadcrumb, if the HWS shows
that it has passed the global seqno and the global seqno is unchanged
after the read, it is indeed complete).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170223074422.4125-9-chris@chris-wilson.co.uk
Replace the global device seqno with one for each engine, and account
for in-flight seqno on each separately. This is consistent with
dma-fence as each timeline has separate fence-contexts for each engine
and a seqno is only ordered within a fence-context (i.e. seqno do not
need to be ordered wrt to other engines, just ordered within a single
engine). This is required to enable request rewinding for preemption on
individual engines (we have to rewind the global seqno to avoid
overflow, and we do not have to rewind all engines just to preempt one.)
v2: Rename active_seqno to inflight_seqnos to more clearly indicate that
it is a counter and not equivalent to the existing seqno. Update
functions that operated on active_seqno similarly.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170223074422.4125-3-chris@chris-wilson.co.uk
When dma_fence_signal() is called, it sets a flag to indicate the fence
is complete. Before the dma_fence is signaled, the seqno check will
first be passed. During an unlocked check (such as inside a waiter), it
is possible for the fence to be signaled even though the seqno has been
reset (by engine wraparound). In this case the waiter will be kicked,
but for an extra layer of protection we can check the persistent
signaled bit from the fence.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170223074422.4125-2-chris@chris-wilson.co.uk
Since the
commit f1c131b454
crypto: xts - Convert to skcipher
the XTS mode is based on ECB, so the mode must select
ECB otherwise it can fail to initialize.
Signed-off-by: Milan Broz <gmazyland@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
pci_enable_msix has been long deprecated, but this driver adds a new
instance. Convert it to pci_alloc_irq_vectors and greatly simplify
the code, and make sure the prope code properly unwinds.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
pci_enable_msix has been long deprecated, but this driver adds a new
instance. Convert it to pci_alloc_irq_vectors and greatly simplify
the code.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Using pci_alloc_irq_vectors() instead of the deprecated
pci_enable_msix() allows to remove the msix_entry from
struct octeon_i2c and thus to get rid of the config symbol check.
Signed-off-by: Jan Glauber <jglauber@cavium.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
In case of arbitration lost adequate interrupt sometimes is not signaled.
As a result transfer timeouts and is not retried, as it should. To avoid
such cases code is added to check transaction status in case of every
interrupt.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Tested-by: Andi Shyti <andi.shyti@samsung.com>
Reviewed-by: Andi Shyti <andi.shyti@samsung.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Fifo-almost-empty irq signal should be disabled as soon as possible,
to avoid unnecessary interrupt storm. The best moment is when there is
no more data to feed fifo.
This patch fixes system stalls caused by IRQ storm.
Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Tested-by: Andi Shyti <andi.shyti@samsung.com>
Reviewed-by: Andi Shyti <andi.shyti@samsung.com>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Previous vGPU type create tried to determine vGPU type name e.g _1, _2
based on the number of mdev devices can be created, but different type
might have very different resource size depending on physical device.
We need to split type name vs. actual mdev resource and create fixed
vGPU type with determined size for consistence.
With this we'd like to fix vGPU types for _1, _2, _4 and _8 now, each
type has fixed defined resource size. Available mdev instances that could
be created is determined by physical resource, and user should query
for that before creating.
Cc: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
The guest VM may initialize the whole GTT table during boot up,
so the warning msg in emulate_gtt_mmio_write is not necessary, it is
the expected behavior and it may confuse the user if error msg is
printed out, so remove the msg from emulate_gtt_mmio_write(),
Signed-off-by: Zhao, Xinda <xinda.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
In GVT-g we always emulate as pcode read/write success and ready for access
anytime, since we don't touch real physical registers here.
Add 'SKL_PCODE_CDCLK_CONTROL' write emulation, without it will cause
skl_set_cdclk fail in guest.
Signed-off-by: Weinan Li <weinan.z.li@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Releasing shadow PPGTT pages is not enough when vGPU reset, the
guest page table tracking data should has same life-cycle with
all the shadow PPGTT pages; Otherwise there is no chance to
re-shadow the PPGTT pages without free the guest page table
tracking data.
This patch clear the PPGTT reset logic and make the vGPU reset in
working order.
v2: refactor some logic to avoid code duplicated.
v3: remove useless macro and add comments from Christophe.
v4: keep reset logic in reset function.
Signed-off-by: Ping Gao <ping.a.gao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
When untracked mmio is visited, too many log info will be printed out,
it may confuse the user, but most of the time, it is not the urgent case,
so use gvt_dbg_mmio() instead.
Signed-off-by: Zhao, Xinda <xinda.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
for a handled mmio reg, its default value is read from hardware, while
for an unhandled mmio regs, its default value would be random if not
explicitly set to 0
Signed-off-by: Zhao Yan <yan.y.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Linux guest is using this MMIO in lri command. Add cmd_access flag
for this mmio in gvt to avoid error log.
v2: change the mmio address to its macro name
Signed-off-by: Pei Zhang <pei.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
add a whitelist to check the content of force-nonpriv registers
v3:
per He Min's comment, modify in_whitelist()'s return type to bool, and use
negative value as the return value for failure for force_nonpriv_write().
v2:
1. split a big patch into two smaller ones per zhenyu's comment.
this patch is the mmio handling part for force-nopriv registers
2. per zhenyu's comment, combine all non-priv registers into a single
MMIO_DFH entry
Signed-off-by: Zhao Yan <yan.y.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
the value of those registers should be applied to hardware on
context restoring
Signed-off-by: Zhao Yan <yan.y.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
send_display_send_uevent() sends two environment variable, and the
first one GVT_DISPLAY_READY is set including a new line at the end of
the string; that is obviously superfluous and wrong -- at least, it
*looks* so when you only read the code.
However, it doesn't appear in the actual output by a (supposedly
unexpected) trick. The code uses snprintf() and truncates the string
in size 20 bytes. This makes the string as GVT_DISPLAY_READY=0 or
...=1 including the trailing NUL-letter. That is, the '\n' found in
the format string is always cut off as a result.
Although the code gives the correct result, it is confusing. This
patch addresses it, just removing the superfluous '\n' from the format
string for avoiding further confusion. If the argument "ready" were
not a bool, the size 20 should be corrected as well. But it's a
bool, so we can leave the magic number 20 as is for now.
FWIW, the bug was spotted by a new GCC7 warning:
drivers/gpu/drm/i915/gvt/handlers.c: In function 'pvinfo_mmio_write':
drivers/gpu/drm/i915/gvt/handlers.c:1042:34: error: 'snprintf' output truncated before the last format character [-Werror=format-truncation=]
snprintf(display_ready_str, 20, "GVT_DISPLAY_READY=%d\n", ready);
^~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/i915/gvt/handlers.c:1042:2: note: 'snprintf' output 21 bytes into a destination of size 20
snprintf(display_ready_str, 20, "GVT_DISPLAY_READY=%d\n", ready);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Fixes: 04d348ae3f ("drm/i915/gvt: vGPU display virtualization")
Bugzilla: https://bugzilla.suse.com/show_bug.cgi?id=1025903
Reported-by: Richard Biener <rguenther@suse.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
some registers were missing or treated as BDW only. This patch is to fix it
avoid unhandled mmio wanrings
v2: update commit message according to zhenyu's comment
Signed-off-by: Zhao Yan <yan.y.zhao@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Due to the request replay, context switch interrupt may come after
gvt free the workload thus can cause a kernel NULL pointer kernel
panic. This patch will add a simple check to avoid this for a short
term.
From long term, gvt workload lifecycle doesn't match with i915 request
and need to find a proper way to manage this.
v4: simplify the NULL pointer check.
v5: add unlikely to optimize.
Signed-off-by: Chuanxiao Dong <chuanxiao.dong@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Windows guest will notitfy GVT-g to request more resources through g2v
interface, when its resources are not enough.
This patch is to handle this case and let vgpu enter failsafe mode to
avoid too many error messages.
Signed-off-by: Min He <min.he@intel.com>
Signed-off-by: Pei Zhang <pei.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
