[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
2015-05-13 16:58, Bruce Richardson: > On Wed, May 13, 2015 at 02:40:48PM +0100, Bruce Richardson wrote: > > On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > > > 2015-05-13 11:01, Bruce Richardson: > > > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > > > >> 2015-05-12 19:04, Bruce Richardson: > > > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > > > >> > > > >> As explained in a previous comment, > > > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > > > >> I think this path would be better: > > > >>drivers/net/e1000/base/e1000_hw.h > > > > > > > > Two reasons why I didn't create the "net" subfolder: > > > > 1. I initally forgot to consider it :-( > > > > 2. While we may at some future point have other device driver types, > > > > are we really > > > > needing to start categorising PMDs at this point? > > > > > > > > As for the base driver part, I was viewing that as a something that > > > > should be > > > > a separate patch set, since it's unrelated to moving things to the > > > > drivers > > > > subdir. > > > > > > I understand your points and I partially agree. > > > However, file moves may be perturbing because it change habits > > > and may complicate a bit the git history browsing. > > > So I think it's better to minimize such moves and do altogether. > > > > > Ok. I'll see about renaming the base code directories as part of the overall > > move process [Unless there are objections from any of the driver > > maintainers]. > > > > As for drivers vs drivers/net, I suppose there is no real difference in what > > the path actually is, so I can make that change too. However, I still think > > I > > prefer the shorter path. Anyone else any opinions on this [before I start > > reworking this again]? > > A further thought on the splitting up of drivers. What about devices which > provide > more than one type of offload, how would the PMD for such a device be > classified? Other projects (e.g. Linux) would have the same classification problem. How do they solve it? Is it possible to split code in different directories or different drivers?
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
2015-05-15 13:36, Bruce Richardson: > On Wed, May 13, 2015 at 02:40:48PM +0100, Bruce Richardson wrote: > > On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > > > 2015-05-13 11:01, Bruce Richardson: > > > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > > > >> 2015-05-12 19:04, Bruce Richardson: > > > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > > > >> > > > >> As explained in a previous comment, > > > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > > > >> I think this path would be better: > > > >>drivers/net/e1000/base/e1000_hw.h > > > > > > > > Two reasons why I didn't create the "net" subfolder: > > > > 1. I initally forgot to consider it :-( > > > > 2. While we may at some future point have other device driver types, > > > > are we really > > > > needing to start categorising PMDs at this point? > > > > > > > > As for the base driver part, I was viewing that as a something that > > > > should be > > > > a separate patch set, since it's unrelated to moving things to the > > > > drivers > > > > subdir. > > > > > > I understand your points and I partially agree. > > > However, file moves may be perturbing because it change habits > > > and may complicate a bit the git history browsing. > > > So I think it's better to minimize such moves and do altogether. > > > > > Ok. I'll see about renaming the base code directories as part of the overall > > move process [Unless there are objections from any of the driver > > maintainers]. > > > > As for drivers vs drivers/net, I suppose there is no real difference in what > > the path actually is, so I can make that change too. However, I still think > > I > > prefer the shorter path. Anyone else any opinions on this [before I start > > reworking this again]? > > > > /Bruce > > Hi Thomas, > > a specific question on how you want the rename of the "base" driver > directories. > Can I take it from your question that you want the base driver renaming to be > merged into a single commit with the rest of the move for the driver? For > example, > as part of the patch move from "lib/librte_pmd_e1000" to "drivers/net/e1000" > we > also rename the "e1000" subdir to "base". Yes, to minimize number of patches moving files (and breaking history for some tools). > The slight problem with this approach - of trying to do both moves in a single > patch is that the rename of the subdirectory "e1000" involves changes to > other > files in the librte_pmd_e1000 directory which are being moved, so we can't > have a clean patch that just moves files and doesn't touch those files which > are > being moved. Therefore, from a patch cleanliness point of view, the renames > to the base directory should be a separate patch, which I can make part of > this > series. The changes are only in Makefile and include lines, right? I think it's not a big deal to have small changes. But maybe it will make difficult to read the move with some tools. Feel free to adopt the solution you prefer if nobody else has an experienced opinion.
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
On Fri, May 15, 2015 at 02:54:50PM +0200, Thomas Monjalon wrote: > 2015-05-13 16:58, Bruce Richardson: > > On Wed, May 13, 2015 at 02:40:48PM +0100, Bruce Richardson wrote: > > > On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > > > > 2015-05-13 11:01, Bruce Richardson: > > > > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > > > > >> 2015-05-12 19:04, Bruce Richardson: > > > > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > > > > >> > > > > >> As explained in a previous comment, > > > > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > > > > >> I think this path would be better: > > > > >>drivers/net/e1000/base/e1000_hw.h > > > > > > > > > > Two reasons why I didn't create the "net" subfolder: > > > > > 1. I initally forgot to consider it :-( > > > > > 2. While we may at some future point have other device driver types, > > > > > are we really > > > > > needing to start categorising PMDs at this point? > > > > > > > > > > As for the base driver part, I was viewing that as a something that > > > > > should be > > > > > a separate patch set, since it's unrelated to moving things to the > > > > > drivers > > > > > subdir. > > > > > > > > I understand your points and I partially agree. > > > > However, file moves may be perturbing because it change habits > > > > and may complicate a bit the git history browsing. > > > > So I think it's better to minimize such moves and do altogether. > > > > > > > Ok. I'll see about renaming the base code directories as part of the > > > overall > > > move process [Unless there are objections from any of the driver > > > maintainers]. > > > > > > As for drivers vs drivers/net, I suppose there is no real difference in > > > what > > > the path actually is, so I can make that change too. However, I still > > > think I > > > prefer the shorter path. Anyone else any opinions on this [before I start > > > reworking this again]? > > > > A further thought on the splitting up of drivers. What about devices which > > provide > > more than one type of offload, how would the PMD for such a device be > > classified? > > Other projects (e.g. Linux) would have the same classification problem. > How do they solve it? Is it possible to split code in different directories > or different drivers? > No idea, possibly some of the resident linux kernel experts on-list can inform us. However, this is likely a solved problem, so I'm going to rework the patchset to use drivers/net for now, and thereby allow future-proofing for other device types. /Bruce
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
On Wed, May 13, 2015 at 02:40:48PM +0100, Bruce Richardson wrote: > On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > > 2015-05-13 11:01, Bruce Richardson: > > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > > >> 2015-05-12 19:04, Bruce Richardson: > > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > > >> > > >> As explained in a previous comment, > > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > > >> I think this path would be better: > > >>drivers/net/e1000/base/e1000_hw.h > > > > > > Two reasons why I didn't create the "net" subfolder: > > > 1. I initally forgot to consider it :-( > > > 2. While we may at some future point have other device driver types, are > > > we really > > > needing to start categorising PMDs at this point? > > > > > > As for the base driver part, I was viewing that as a something that > > > should be > > > a separate patch set, since it's unrelated to moving things to the drivers > > > subdir. > > > > I understand your points and I partially agree. > > However, file moves may be perturbing because it change habits > > and may complicate a bit the git history browsing. > > So I think it's better to minimize such moves and do altogether. > > > Ok. I'll see about renaming the base code directories as part of the overall > move process [Unless there are objections from any of the driver maintainers]. > > As for drivers vs drivers/net, I suppose there is no real difference in what > the path actually is, so I can make that change too. However, I still think I > prefer the shorter path. Anyone else any opinions on this [before I start > reworking this again]? > > /Bruce Hi Thomas, a specific question on how you want the rename of the "base" driver directories. Can I take it from your question that you want the base driver renaming to be merged into a single commit with the rest of the move for the driver? For example, as part of the patch move from "lib/librte_pmd_e1000" to "drivers/net/e1000" we also rename the "e1000" subdir to "base". The slight problem with this approach - of trying to do both moves in a single patch is that the rename of the subdirectory "e1000" involves changes to other files in the librte_pmd_e1000 directory which are being moved, so we can't have a clean patch that just moves files and doesn't touch those files which are being moved. Therefore, from a patch cleanliness point of view, the renames to the base directory should be a separate patch, which I can make part of this series. /Bruce
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
On Wed, May 13, 2015 at 02:40:48PM +0100, Bruce Richardson wrote: > On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > > 2015-05-13 11:01, Bruce Richardson: > > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > > >> 2015-05-12 19:04, Bruce Richardson: > > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > > >> > > >> As explained in a previous comment, > > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > > >> I think this path would be better: > > >>drivers/net/e1000/base/e1000_hw.h > > > > > > Two reasons why I didn't create the "net" subfolder: > > > 1. I initally forgot to consider it :-( > > > 2. While we may at some future point have other device driver types, are > > > we really > > > needing to start categorising PMDs at this point? > > > > > > As for the base driver part, I was viewing that as a something that > > > should be > > > a separate patch set, since it's unrelated to moving things to the drivers > > > subdir. > > > > I understand your points and I partially agree. > > However, file moves may be perturbing because it change habits > > and may complicate a bit the git history browsing. > > So I think it's better to minimize such moves and do altogether. > > > Ok. I'll see about renaming the base code directories as part of the overall > move process [Unless there are objections from any of the driver maintainers]. > > As for drivers vs drivers/net, I suppose there is no real difference in what > the path actually is, so I can make that change too. However, I still think I > prefer the shorter path. Anyone else any opinions on this [before I start > reworking this again]? > A further thought on the splitting up of drivers. What about devices which provide more than one type of offload, how would the PMD for such a device be classified? /Bruce
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
2015-05-13 11:01, Bruce Richardson: > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: >> 2015-05-12 19:04, Bruce Richardson: >>> drivers/e1000/e1000/e1000_hw.h | 1026 >> >> As explained in a previous comment, >> http://dpdk.org/ml/archives/dev/2015-May/017509.html >> I think this path would be better: >>drivers/net/e1000/base/e1000_hw.h > > Two reasons why I didn't create the "net" subfolder: > 1. I initally forgot to consider it :-( > 2. While we may at some future point have other device driver types, are we > really > needing to start categorising PMDs at this point? > > As for the base driver part, I was viewing that as a something that should be > a separate patch set, since it's unrelated to moving things to the drivers > subdir. I understand your points and I partially agree. However, file moves may be perturbing because it change habits and may complicate a bit the git history browsing. So I think it's better to minimize such moves and do altogether.
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
On Wed, May 13, 2015 at 03:33:27PM +0200, Thomas Monjalon wrote: > 2015-05-13 11:01, Bruce Richardson: > > On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > >> 2015-05-12 19:04, Bruce Richardson: > >>> drivers/e1000/e1000/e1000_hw.h | 1026 > >> > >> As explained in a previous comment, > >> http://dpdk.org/ml/archives/dev/2015-May/017509.html > >> I think this path would be better: > >>drivers/net/e1000/base/e1000_hw.h > > > > Two reasons why I didn't create the "net" subfolder: > > 1. I initally forgot to consider it :-( > > 2. While we may at some future point have other device driver types, are we > > really > > needing to start categorising PMDs at this point? > > > > As for the base driver part, I was viewing that as a something that should > > be > > a separate patch set, since it's unrelated to moving things to the drivers > > subdir. > > I understand your points and I partially agree. > However, file moves may be perturbing because it change habits > and may complicate a bit the git history browsing. > So I think it's better to minimize such moves and do altogether. > Ok. I'll see about renaming the base code directories as part of the overall move process [Unless there are objections from any of the driver maintainers]. As for drivers vs drivers/net, I suppose there is no real difference in what the path actually is, so I can make that change too. However, I still think I prefer the shorter path. Anyone else any opinions on this [before I start reworking this again]? /Bruce
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
On Tue, May 12, 2015 at 09:30:45PM +0200, Thomas Monjalon wrote: > 2015-05-12 19:04, Bruce Richardson: > > drivers/e1000/e1000/e1000_hw.h | 1026 > > As explained in a previous comment, > http://dpdk.org/ml/archives/dev/2015-May/017509.html > I think this path would be better: >drivers/net/e1000/base/e1000_hw.h > Two reasons why I didn't create the "net" subfolder: 1. I initally forgot to consider it :-( 2. While we may at some future point have other device driver types, are we really needing to start categorising PMDs at this point? As for the base driver part, I was viewing that as a something that should be a separate patch set, since it's unrelated to moving things to the drivers subdir. I realise now, I should have replied to the original email with these points before diving into the new patchset. My bad... /Bruce
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
2015-05-12 19:04, Bruce Richardson: > drivers/e1000/e1000/e1000_hw.h | 1026 As explained in a previous comment, http://dpdk.org/ml/archives/dev/2015-May/017509.html I think this path would be better: drivers/net/e1000/base/e1000_hw.h
[dpdk-dev] [PATCH 00/19] Move PMDs to drivers directory
As covered previously in an RFC, this patchset creates a new top-level drivers folder and moves the PMDs into it. Moving the drivers involves a lot of code - too big for a single patch, so each driver is moved separately in its own patchset. Some doc updates are done as part of this patchset - based on the results of git grep, but may be incomplete, so flag any doc areas that I may have missed, please. Bruce Richardson (19): pmds: Use relative rather than absolute paths drivers: create new top-level drivers directory af_packet: move af_packet pmd to drivers directory bond: Move bonded ethdev pmd to drivers e1000: move e1000 pmd to drivers directory enic: move enic PMD to drivers directory fm10k: move fm10k PMD to drivers directory i40e: move i40e PMD to drivers directory ixbge: move ixgbe PMD to drivers directory mlx4: move mlx4 PMD to drivers directory null pmd: move null PMD to drivers directory pcap: move pcap pmd to drivers directory ring pmd: move ring PMD to the drivers directory virtio: move virtio PMD to drivers directory xmvnet3: move vmxnet3 PMD to drivers directory xenvirt: move xenvirt PMD to drivers directory maintainers: update maintainers with driver paths doc: update GSG doc for new drivers subdirectory doc: update source organisation text for drivers GNUmakefile|2 +- MAINTAINERS| 28 +- app/dump_cfg/Makefile |2 +- app/test-acl/Makefile |2 +- app/test-pipeline/Makefile |2 +- app/test-pmd/Makefile |2 +- app/test/Makefile |2 +- doc/api/doxy-api.conf |2 +- doc/guides/linux_gsg/build_dpdk.rst|8 +- doc/guides/prog_guide/source_org.rst | 37 +- drivers/Makefile | 50 + drivers/af_packet/Makefile | 64 + drivers/af_packet/rte_eth_af_packet.c | 847 +++ drivers/af_packet/rte_eth_af_packet.h | 53 + drivers/af_packet/rte_pmd_af_packet_version.map|7 + drivers/bonding/Makefile | 68 + drivers/bonding/rte_eth_bond.h | 366 ++ drivers/bonding/rte_eth_bond_8023ad.c | 1216 + drivers/bonding/rte_eth_bond_8023ad.h | 222 + drivers/bonding/rte_eth_bond_8023ad_private.h | 308 ++ drivers/bonding/rte_eth_bond_alb.c | 287 + drivers/bonding/rte_eth_bond_alb.h | 142 + drivers/bonding/rte_eth_bond_api.c | 840 +++ drivers/bonding/rte_eth_bond_args.c| 278 + drivers/bonding/rte_eth_bond_pmd.c | 2269 drivers/bonding/rte_eth_bond_private.h | 287 + drivers/bonding/rte_eth_bond_version.map | 22 + drivers/e1000/Makefile | 99 + drivers/e1000/e1000/README | 39 + drivers/e1000/e1000/e1000_80003es2lan.c| 1514 ++ drivers/e1000/e1000/e1000_80003es2lan.h| 100 + drivers/e1000/e1000/e1000_82540.c | 717 +++ drivers/e1000/e1000/e1000_82541.c | 1268 + drivers/e1000/e1000/e1000_82541.h | 91 + drivers/e1000/e1000/e1000_82542.c | 588 ++ drivers/e1000/e1000/e1000_82543.c | 1553 ++ drivers/e1000/e1000/e1000_82543.h | 56 + drivers/e1000/e1000/e1000_82571.c | 2026 +++ drivers/e1000/e1000/e1000_82571.h | 65 + drivers/e1000/e1000/e1000_82575.c | 3639 + drivers/e1000/e1000/e1000_82575.h | 520 ++ drivers/e1000/e1000/e1000_api.c| 1357 + drivers/e1000/e1000/e1000_api.h| 167 + drivers/e1000/e1000/e1000_defines.h| 1498 + drivers/e1000/e1000/e1000_hw.h | 1026 drivers/e1000/e1000/e1000_i210.c | 1000 drivers/e1000/e1000/e1000_i210.h | 110 + drivers/e1000/e1000/e1000_ich8lan.c| 5260 ++ drivers/e1000/e1000/e1000_ich8lan.h| 313 ++ drivers/e1000/e1000/e1000_mac.c| 2247 drivers/e1000/e1000/e1000_mac.h| 95 + drivers/e1000/e1000/e1000_manage.c | 573 ++ drivers/e1000/e1000/e1000_manage.h | 95 + drivers/e1000/e1000/e1000_mbx.c| 777 +++ drivers/e1000/e1000/e1000_mbx.h| 105 + drivers/e1000/e1000/e1000_nvm.c| 1377 + drivers/e1000/e1000/e1000_nvm.h| 98 + drivers/e1000/e1000/e1000_osdep.c | 83 +
