- 25 Aug, 2020 30 commits
-
-
Aleksey Makarov authored
Precision Timestamping block found on Octeontx2 platform is an independent coprocessor and has internal PTP hardware clock. Once configured PTP runs independently and when a packet arrives CGX hardware block gets the current timestamp from PTP block and forwards the packet to NIX by prepending timestamp to the packet. This patch adds the pci driver for PTP block. The driver gets registered by AF driver and does initial configuration and exposes a mailbox function to read and adjust PTP hardware clock. The mailbox function is called by AF consumers like netdev drivers or userspace drivers. Since PTP being a single block in platform this driver helps in accessing PTP block by any AF consumer. Co-developed-by:
Subbaraya Sundeep <sbhatta@marvell.com> Signed-off-by:
Aleksey Makarov <amakarov@marvell.com> Signed-off-by:
Sunil Goutham <sgoutham@marvell.com> Acked-by:
Richard Cochran <richardcochran@gmail.com> Acked-by:
Jakub Kicinski <kuba@kernel.org> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Zyta Szpak authored
Four new mbox messages ids and handler are added in order to enable or disable timestamping procedure on tx and rx side. Additionally when PTP is enabled, the packet parser must skip over 8 bytes and start analyzing packet data there. To make NPC profiles work seemlesly PTR_ADVANCE of IKPU is set so that parsing can be done as before when all data pointers are shifted by 8 bytes automatically. Co-developed-by:
Zyta Szpak <zyta@marvell.com> Signed-off-by:
-
Miaohe Lin authored
Use helper macro RT_TOS() to get tos in __icmp_send(). Signed-off-by:
Miaohe Lin <linmiaohe@huawei.com> Signed-off-by:
-
Miaohe Lin authored
There is no need to fetch errno and fatal info from icmp_err_convert when icmp code is ICMP_FRAG_NEEDED. Signed-off-by:
-
David S. Miller authored
Igor Russkikh says: ==================== qed: introduce devlink health support This is a followup implementation after series https://patchwork.ozlabs.org/project/netdev/cover/20200514095727.1361-1-irusskikh@marvell.com/ This is an implementation of devlink health infrastructure. With this we are now able to report HW errors to devlink, and it'll take its own actions depending on user configuration to capture and store the dump at the bad moment, and to request the driver to recover the device. So far we do not differentiate global device failures or specific PCI function failures. This means that some errors specific to one physical function will affect an entire device. This is not yet fully designed and verified, will followup in future. Solution was verified with artificial HW errors generated, existing tools for dump analysis could be used. v7: comments from Jesse and Jakub - p2: extra edev check - p9: removed extra indents v6: patch 4: changing serial to board.serial and fw to fw.app v5: improved patch 4 description v4: - commit message and other fixes after Jiri's comments - removed one patch (will send to net) v3: fix uninit var usage in patch 11 v2: fix #include issue from kbuild test robot. ==================== Signed-off-by:
-
Igor Russkikh authored
In case recovery was not successful, netdev still should be present. But we should clear cdev if something bad happens on recovery. We also check cdev for null on dev close. That could be a case if recovery was not successful. Signed-off-by:
Igor Russkikh <irusskikh@marvell.com> Signed-off-by:
Alexander Lobakin <alobakin@marvell.com> Signed-off-by:
Michal Kalderon <michal.kalderon@marvell.com> Signed-off-by:
-
Igor Russkikh authored
Remove extra indent on some of adjacent declarations. Signed-off-by:
-
Igor Russkikh authored
Gather and push out full device dump to devlink. Device dump is the same as with `ethtool -d`, but now its generated exactly at the moment bad thing happens. Signed-off-by:
-
Igor Russkikh authored
Remove forcible recovery trigger and put it as a normal devlink callback. This allows user to enable/disable it via devlink health set pci/0000:03:00.0 reporter fw_fatal auto_recover false Signed-off-by: -
Igor Russkikh authored
Use devlink_health_report to push error indications. We implement this in qede via callback function to make it possible to reuse the same for other drivers sitting on top of qed in future. Signed-off-by:
-
Igor Russkikh authored
Here we declare health reporter ops (empty for now) and register these in qed probe and remove callbacks. This way we get devlink attached to all kind of qed* PCI device entities: networking or storage offload entity. Signed-off-by:
-
Igor Russkikh authored
Here we return existing fw & mfw versions, we also fetch device's serial number: ~$ sudo ~/iproute2/devlink/devlink dev info pci/0000:01:00.1: driver qed board.serial_number REE1915E44552 versions: running: fw.app 8.42.2.0 stored: fw.mgmt 8.52.10.0 MFW and FW are different firmwares on device. Management is a firmware responsible for link configuration and various control plane features. Its permanent and resides in NVM. Running FW (or fastpath FW) is an embedded microprogram implementing all the packet processing, offloads, etc. This FW is being loaded on each start by the driver from FW binary blob. The base device specific structure (qed_dev_info) was not directly available to the base driver before. Thus, here we create and store a private copy of this structure in qed_dev root object to access the data. Signed-off-by: -
Igor Russkikh authored
This patch replaces stubs in kconfig help entries with an actual description. Signed-off-by:
-
Igor Russkikh authored
Devlink instance lifecycle was linked to qed_dev object, that caused devlink to be recreated on each recovery. Changing it by making higher level driver (qede) responsible for its life. This way devlink now survives recoveries. qede now stores devlink structure pointer as a part of its device object, devlink private data contains a linkage structure, qed_devlink. Signed-off-by:
-
Igor Russkikh authored
We are extending devlink infrastructure, thus move the existing stuff into a new file qed_devlink.c Signed-off-by:
Jesse Brandeburg <jesse.brandeburg@intel.com> Signed-off-by:
-
Christophe JAILLET authored
The wrappers in include/linux/pci-dma-compat.h should go away. The patch has been generated with the coccinelle script below and has been hand modified to replace GFP_ with a correct flag. It has been compile tested. When memory is allocated in 'free_rx_resources()' and 'alloc_tx_resources()' (sge.c) GFP_KERNEL can be used because it is already used in these functions. Moreover, they can only be called from a .ndo_open function. So it is guarded by the 'rtnl_lock()', which is a mutex. While at it, a pr_err message in 'init_one()' has been updated accordingly (s/consistent/coherent). @@ @@ - PCI_DMA_BIDIRECTIONAL + DMA_BIDIRECTIONAL @@ @@ - PCI_DMA_TODEVICE + DMA_TO_DEVICE @@ @@ - PCI_DMA_FROMDEVICE + DMA_FROM_DEVICE @@ @@ - PCI_DMA_NONE + DMA_NONE @@ expression e1, e2, e3; @@ - pci_alloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3; @@ - pci_zalloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3, e4; @@ - pci_free_consistent(e1, e2, e3, e4) + dma_free_coherent(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_single(e1, e2, e3, e4) + dma_map_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_single(e1, e2, e3, e4) + dma_unmap_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4, e5; @@ - pci_map_page(e1, e2, e3, e4, e5) + dma_map_page(&e1->dev, e2, e3, e4, e5) @@ expression e1, e2, e3, e4; @@ - pci_unmap_page(e1, e2, e3, e4) + dma_unmap_page(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_sg(e1, e2, e3, e4) + dma_map_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_sg(e1, e2, e3, e4) + dma_unmap_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_cpu(e1, e2, e3, e4) + dma_sync_single_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_device(e1, e2, e3, e4) + dma_sync_single_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_cpu(e1, e2, e3, e4) + dma_sync_sg_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_device(e1, e2, e3, e4) + dma_sync_sg_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2; @@ - pci_dma_mapping_error(e1, e2) + dma_mapping_error(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_dma_mask(e1, e2) + dma_set_mask(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_consistent_dma_mask(e1, e2) + dma_set_coherent_mask(&e1->dev, e2) Signed-off-by:
Christophe JAILLET <christophe.jaillet@wanadoo.fr> Signed-off-by:
-
David S. Miller authored
Ido Schimmel says: ==================== mlxsw: Misc updates This patch set includes various updates for mlxsw. Patches #1-#4 adjust the default burst size of packet trap policers to conform to Spectrum-{2,3} requirements. The corresponding selftest is also adjusted so that it could reliably pass on these platforms. Patch #5 adjusts a selftest so that it could pass with both old and new versions of mausezahn. Patch #6 significantly reduces the runtime of tc-police scale test by changing the preference and masks of the used tc filters. Patch #7 prevents the driver from trying to set invalid ethtool link modes. ==================== Signed-off-by: -
Danielle Ratson authored
The PTYS register is used to report and configure the port type and speed. Currently, internal bits in the register are used the same way other bits are used. Using the internal bits can cause bad parameter firmware errors. For example, trying to write to internal bit 25 returns: EMAD reg access failed (tid=53e2bffa00004310,reg_id=5004(ptys),type=write,status=7(bad parameter)) Remove the internal bits from the PTYS register, so that it is no longer possible to pass them to firmware. Signed-off-by:
Danielle Ratson <danieller@mellanox.com> Reviewed-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Signed-off-by:
-
Ido Schimmel authored
Currently, the test takes about 626 seconds to complete because of an inefficient use of the device's TCAM. Reduce the runtime to 202 seconds by inserting all the flower filters with the same preference and mask, but with a different key. In particular, this reduces the deletion of the qdisc (which triggers the deletion of all the filters) from 66 seconds to 0.2 seconds. This prevents various netlink requests from user space applications (e.g., systemd-networkd) from timing-out because RTNL is not held for too long anymore. Signed-off-by:
-
Danielle Ratson authored
Currently, mausezahn delay parameter in mirror_test() is specified with 'ms' units. mausezahn versions before 0.6.5 interpret 'ms' as seconds and therefore the tests that use mirror_test() take a very long time to complete. Resolve this by specifying 'msec' units. Signed-off-by:
Petr Machata <petrm@mellanox.com> Signed-off-by:
-
Ido Schimmel authored
The current combination of rate and burst size does not adhere to Spectrum-{2,3} limitation which states that the minimum burst size should be 40% of the rate. Increase the burst size in order to honor above mentioned limitation and avoid intermittent failures of this test case on Spectrum-{2,3}. Remove the first sub-test case as the variation in number of received packets is simply too large to reliably test it. Signed-off-by: -
Ido Schimmel authored
The current combination of rate and burst size does not adhere to Spectrum-{2,3} limitation which states that the minimum burst size should be 40% of the rate. Increase the burst size in order to honor above mentioned limitation and avoid intermittent failures of this test case on Spectrum-{2,3}. Signed-off-by: -
Ido Schimmel authored
On Spectrum-{2,3} the required accuracy is +/-10%. Align the test to this requirement so that it can reliably pass on these platforms. Signed-off-by: -
Ido Schimmel authored
On the Spectrum-{2,3} ASICs the minimum burst size of the packet trap policers needs to be 40% of the configured rate. Otherwise, intermittent drops are observed even when the incoming packet rate is slightly lower than the configured policer rate. Adjust the burst size of the registered packet trap policers so that they do not violate above mentioned limitation. Signed-off-by: -
Christophe JAILLET authored
The wrappers in include/linux/pci-dma-compat.h should go away. The patch has been generated with the coccinelle script below and has been hand modified to replace GFP_ with a correct flag. It has been compile tested. When memory is allocated in 'atl1e_setup_ring_resources()' (atl1e_main.c), 'atl1_setup_ring_resources()' (atl1.c) and 'atl2_setup_ring_resources()' (atl2.c) GFP_KERNEL can be used because it can be called from a .ndo_open. 'atl1_setup_ring_resources()' (atl1.c) can also be called from a '.set_ringparam' (see struct ethtool_ops) where sleep is also allowed. Both cases are protected by 'rtnl_lock()' which is a mutex. So these function can sleep. @@ @@ - PCI_DMA_BIDIRECTIONAL + DMA_BIDIRECTIONAL @@ @@ - PCI_DMA_TODEVICE + DMA_TO_DEVICE @@ @@ - PCI_DMA_FROMDEVICE + DMA_FROM_DEVICE @@ @@ - PCI_DMA_NONE + DMA_NONE @@ expression e1, e2, e3; @@ - pci_alloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3; @@ - pci_zalloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3, e4; @@ - pci_free_consistent(e1, e2, e3, e4) + dma_free_coherent(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_single(e1, e2, e3, e4) + dma_map_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_single(e1, e2, e3, e4) + dma_unmap_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4, e5; @@ - pci_map_page(e1, e2, e3, e4, e5) + dma_map_page(&e1->dev, e2, e3, e4, e5) @@ expression e1, e2, e3, e4; @@ - pci_unmap_page(e1, e2, e3, e4) + dma_unmap_page(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_sg(e1, e2, e3, e4) + dma_map_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_sg(e1, e2, e3, e4) + dma_unmap_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_cpu(e1, e2, e3, e4) + dma_sync_single_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_device(e1, e2, e3, e4) + dma_sync_single_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_cpu(e1, e2, e3, e4) + dma_sync_sg_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_device(e1, e2, e3, e4) + dma_sync_sg_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2; @@ - pci_dma_mapping_error(e1, e2) + dma_mapping_error(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_dma_mask(e1, e2) + dma_set_mask(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_consistent_dma_mask(e1, e2) + dma_set_coherent_mask(&e1->dev, e2) Signed-off-by:
-
Christophe JAILLET authored
The wrappers in include/linux/pci-dma-compat.h should go away. The patch has been generated with the coccinelle script below and has been hand modified to replace GFP_ with a correct flag. It has been compile tested. When memory is allocated in 'netdev_open()', GFP_ATOMIC must be used because it can be called from a .ndo_tx_timeout function. So this function can be called with the 'netif_tx_lock' acquired. The call chain is: --> tx_timeout (.ndo_tx_timeout function) --> netdev_open @@ @@ - PCI_DMA_BIDIRECTIONAL + DMA_BIDIRECTIONAL @@ @@ - PCI_DMA_TODEVICE + DMA_TO_DEVICE @@ @@ - PCI_DMA_FROMDEVICE + DMA_FROM_DEVICE @@ @@ - PCI_DMA_NONE + DMA_NONE @@ expression e1, e2, e3; @@ - pci_alloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3; @@ - pci_zalloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3, e4; @@ - pci_free_consistent(e1, e2, e3, e4) + dma_free_coherent(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_single(e1, e2, e3, e4) + dma_map_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_single(e1, e2, e3, e4) + dma_unmap_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4, e5; @@ - pci_map_page(e1, e2, e3, e4, e5) + dma_map_page(&e1->dev, e2, e3, e4, e5) @@ expression e1, e2, e3, e4; @@ - pci_unmap_page(e1, e2, e3, e4) + dma_unmap_page(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_sg(e1, e2, e3, e4) + dma_map_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_sg(e1, e2, e3, e4) + dma_unmap_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_cpu(e1, e2, e3, e4) + dma_sync_single_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_device(e1, e2, e3, e4) + dma_sync_single_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_cpu(e1, e2, e3, e4) + dma_sync_sg_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_device(e1, e2, e3, e4) + dma_sync_sg_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2; @@ - pci_dma_mapping_error(e1, e2) + dma_mapping_error(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_dma_mask(e1, e2) + dma_set_mask(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_consistent_dma_mask(e1, e2) + dma_set_coherent_mask(&e1->dev, e2) Signed-off-by: -
Christophe JAILLET authored
The wrappers in include/linux/pci-dma-compat.h should go away. The patch has been generated with the coccinelle script below and has been hand modified to replace GFP_ with a correct flag. It has been compile tested. When memory is allocated in 'typhoon_init_one()' GFP_KERNEL can be used because it is a probe function and no lock is acquired. When memory is allocated in 'typhoon_download_firmware()', GFP_ATOMIC must be used because it can be called from a .ndo_tx_timeout function. So this function can be called with the 'netif_tx_lock' acquired. The call chain is: --> typhoon_tx_timeout (.ndo_tx_timeout function) --> typhoon_start_runtime --> typhoon_download_firmware While at is, update some comments accordingly. @@ @@ - PCI_DMA_BIDIRECTIONAL + DMA_BIDIRECTIONAL @@ @@ - PCI_DMA_TODEVICE + DMA_TO_DEVICE @@ @@ - PCI_DMA_FROMDEVICE + DMA_FROM_DEVICE @@ @@ - PCI_DMA_NONE + DMA_NONE @@ expression e1, e2, e3; @@ - pci_alloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3; @@ - pci_zalloc_consistent(e1, e2, e3) + dma_alloc_coherent(&e1->dev, e2, e3, GFP_) @@ expression e1, e2, e3, e4; @@ - pci_free_consistent(e1, e2, e3, e4) + dma_free_coherent(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_single(e1, e2, e3, e4) + dma_map_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_single(e1, e2, e3, e4) + dma_unmap_single(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4, e5; @@ - pci_map_page(e1, e2, e3, e4, e5) + dma_map_page(&e1->dev, e2, e3, e4, e5) @@ expression e1, e2, e3, e4; @@ - pci_unmap_page(e1, e2, e3, e4) + dma_unmap_page(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_map_sg(e1, e2, e3, e4) + dma_map_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_unmap_sg(e1, e2, e3, e4) + dma_unmap_sg(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_cpu(e1, e2, e3, e4) + dma_sync_single_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_single_for_device(e1, e2, e3, e4) + dma_sync_single_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_cpu(e1, e2, e3, e4) + dma_sync_sg_for_cpu(&e1->dev, e2, e3, e4) @@ expression e1, e2, e3, e4; @@ - pci_dma_sync_sg_for_device(e1, e2, e3, e4) + dma_sync_sg_for_device(&e1->dev, e2, e3, e4) @@ expression e1, e2; @@ - pci_dma_mapping_error(e1, e2) + dma_mapping_error(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_dma_mask(e1, e2) + dma_set_mask(&e1->dev, e2) @@ expression e1, e2; @@ - pci_set_consistent_dma_mask(e1, e2) + dma_set_coherent_mask(&e1->dev, e2) Signed-off-by:David Dillow <dave@thedillows.org> Signed-off-by:
-
Randy Dunlap authored
Drop duplicated words in /net/dccp/. Signed-off-by:
Randy Dunlap <rdunlap@infradead.org> Cc: Gerrit Renker <gerrit@erg.abdn.ac.uk> Cc: dccp@vger.kernel.org Cc: "David S. Miller" <davem@davemloft.net> Cc: Jakub Kicinski <kuba@kernel.org> Signed-off-by:
-
Randy Dunlap authored
Drop duplicated words in net/netlink/. Signed-off-by:
-
Randy Dunlap authored
Drop duplicate words in comments in net/ipv4/. Signed-off-by:
-
- 24 Aug, 2020 10 commits
-
-
David S. Miller authored
Randy Dunlap says: ==================== net: sctp: delete duplicated words + other fixes Drop or fix repeated words in net/sctp/. ==================== Signed-off-by: -
Randy Dunlap authored
Drop the repeated word "an". Signed-off-by:
-
Randy Dunlap authored
Drop the repeated words "for", "that", and "a". Change "his" to "this". Signed-off-by:
-
Randy Dunlap authored
Drop the repeated words "of" and "that". Add some punctuation for readability. Signed-off-by:
-
Randy Dunlap authored
Drop the repeated word "the". Signed-off-by:
-
Randy Dunlap authored
Drop the repeated word "of". Signed-off-by:
-
Randy Dunlap authored
Drop the repeated word "the" and "now". Signed-off-by:
-
Randy Dunlap authored
Drop the repeated word "the" in two places. Signed-off-by:
-
Luke Hsiao authored
Currently, io_uring's recvmsg subscribes to both POLLERR and POLLIN. In the context of TCP tx zero-copy, this is inefficient since we are only reading the error queue and not using recvmsg to read POLLIN responses. This patch was tested by using a simple sending program to call recvmsg using io_uring with MSG_ERRQUEUE set and verifying with printks that the POLLIN is correctly unset when the msg flags are MSG_ERRQUEUE. Signed-off-by:
Arjun Roy <arjunroy@google.com> Signed-off-by:
Soheil Hassas Yeganeh <soheil@google.com> Acked-by:
Eric Dumazet <edumazet@google.com> Reviewed-by:
Jens Axboe <axboe@kernel.dk> Signed-off-by:
Luke Hsiao <lukehsiao@google.com> Signed-off-by:
-
Luke Hsiao authored
For TCP tx zero-copy, the kernel notifies the process of completions by queuing completion notifications on the socket error queue. This patch allows reading these notifications via recvmsg to support TCP tx zero-copy. Ancillary data was originally disallowed due to privilege escalation via io_uring's offloading of sendmsg() onto a kernel thread with kernel credentials (https://crbug.com/project-zero/1975). So, we must ensure that the socket type is one where the ancillary data types that are delivered on recvmsg are plain data (no file descriptors or values that are translated based on the identity of the calling process). This was tested by using io_uring to call recvmsg on the MSG_ERRQUEUE with tx zero-copy enabled. Before this patch, we received -EINVALID from this specific code path. After this patch, we could read tcp tx zero-copy completion notifications from the MSG_ERRQUEUE. Signed-off-by:
Jann Horn <jannh@google.com> Reviewed-by:
-
