introduce HAVE_EFFICIENT_UNALIGNED_ACCESS Kconfig symbol
Subsystems (3)
PATCH 1/1 introduce HAVE_EFFICIENT_UNALIGNED_ACCESS Kconfig symbol
In many cases, especially in networking, it can be beneficial to know at compile time whether the architecture can do unaligned accesses efficiently. This patch introduces a new Kconfig symbol
HAVE_EFFICIENT_UNALIGNED_ACCESS
for that purpose and adds it to the powerpc and x86 architectures. Also add some documentation about alignment and networking, and especially one intended use of this symbol.
---
v5: rename to HAVE_EFFICIENT_UNALIGNED_ACCESS
I have opted to not introduce any symbol associated with the cost of unaligned accesses, David Woodhouse once suggested that such a cost should be combined with a probability of (un-)alignment even in the get_unaligned/put_unaligned macros and I think this should be combined with a patch introducing a global cost constant.
Also, this would require architecture changes because if some code knew the likelyhood of unaligned accesses was small enough over the cost of them, architectures that complain in the trap handle would still complain in that unlikely case although the code explicitly made a trade-off based on the cost/probability.
Documentation/unaligned-memory-access.txt | 32 +++++++++++++++++++++++++++--- arch/Kconfig | 19 +++++++++++++++++ arch/powerpc/Kconfig | 1 arch/x86/Kconfig | 1 4 files changed, 50 insertions(+), 3 deletions(-)
Documentation/unaligned-memory-access.txt +29 -3
@@ -218,9 +218,35 @@ If use of such macros is not convenient, where the source or destination (or both) are of type u8* or unsigned char*. Due to the byte-wise nature of this operation, unaligned accesses are avoided. + +Alignment vs. Networking +======================== + +On architectures that require aligned loads, networking requires that the IP +header is aligned on a four-byte boundary to optimise the IP stack. For +regular ethernet hardware, the constant NET_IP_ALIGN is used, on most +architectures this constant has the value 2 because the normal ethernet +header is 14 bytes long, so in order to get proper alignment one needs to +DMA to an address that is can be expressed as 4*n + 2. One notable exception +here is powerpc which defines NET_IP_ALIGN to 0 because DMA to unaligned +addresses can be very expensive and dwarf the cost of unaligned loads. + +For some ethernet hardware that cannot DMA to unaligned addresses like +4*n+2 or non-ethernet hardware, this can be a problem, and it is then +required to copy the incoming frame into an aligned buffer. Because this is +unnecessary on architectures that can do unaligned accesses, the code can be +made depend on CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS like so: + +#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS + skb = original skb +#else + skb = copy skb +#endif + -- -Author: Daniel Drake <[email protected]> +Authors: Daniel Drake <[email protected]>, + Johannes Berg <[email protected]> With help from: Alan Cox, Avuton Olrich, Heikki Orsila, Jan Engelhardt, -Johannes Berg, Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, -Uli Kunitz, Vadim Lobanov +Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, Uli Kunitz, +Vadim Lobanov
arch/powerpc/Kconfig +1 -0
@@ -103,6 +103,7 @@ config PPC select HAVE_OPROFILE select HAVE_KPROBES select HAVE_KRETPROBES + select HAVE_EFFICIENT_UNALIGNED_ACCESS config EARLY_PRINTK bool
arch/x86/Kconfig +1 -0
@@ -23,6 +23,7 @@ config X86 select HAVE_KPROBES select HAVE_KRETPROBES select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64) + select HAVE_EFFICIENT_UNALIGNED_ACCESS config GENERIC_LOCKBREAK
arch/Kconfig +19 -0
@@ -36,3 +36,22 @@ config HAVE_KPROBES config HAVE_KRETPROBES def_bool n + +config HAVE_EFFICIENT_UNALIGNED_ACCESS + def_bool n + help + Some architectures are unable to perform unaligned accesses + without the use of get_unaligned/put_unaligned. Others are + unable to perform such accesses efficiently (e.g. trap on + unaligned access and require fixing it up in the exception + handler.) + + This symbol should be selected by an architecture if it can + perform unaligned accesses efficiently to allow different + code paths to be selected for these cases. Some network + drivers, for example, could opt to not fix up alignment + problems with received packets if doing so would not help + much. + + See Documentation/unaligned-memory-access.txt for more + information on the topic of unaligned memory accesses.
Documentation/unaligned-memory-access.txt If use of such macros is not convenient, · line 227
Quoted 13 lines · Johannes Berg
Documentation/unaligned-memory-access.txt +7
-0@@ -218,9 +218,35 @@ If use of such macros is not convenient, where the source or destination (or both) are of type u8* or unsigned char*. Due to the byte-wise nature of this operation, unaligned accesses are avoided. + +Alignment vs. Networking +======================== + +On architectures that require aligned loads, networking requires that the IP +header is aligned on a four-byte boundary to optimise the IP stack. For +regular ethernet hardware, the constant NET_IP_ALIGN is used, on most
. On
Documentation/unaligned-memory-access.txt If use of such macros is not convenient, · line 229
Quoted 2 lines
Documentation/unaligned-memory-access.txt +2
-0+architectures this constant has the value 2 because the normal ethernet +header is 14 bytes long, so in order to get proper alignment one needs to
. So
Documentation/unaligned-memory-access.txt If use of such macros is not convenient, · line 230
Quoted 1 line
Documentation/unaligned-memory-access.txt +1
-0+DMA to an address that is can be expressed as 4*n + 2. One notable exception
which can
Documentation/unaligned-memory-access.txt If use of such macros is not convenient, · line 238
Quoted 8 lines
Documentation/unaligned-memory-access.txt +8
-0+here is powerpc which defines NET_IP_ALIGN to 0 because DMA to unaligned +addresses can be very expensive and dwarf the cost of unaligned loads. + +For some ethernet hardware that cannot DMA to unaligned addresses like +4*n+2 or non-ethernet hardware, this can be a problem, and it is then +required to copy the incoming frame into an aligned buffer. Because this is +unnecessary on architectures that can do unaligned accesses, the code can be +made depend on CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS like so:
made dependent on
Quoted 40 lines
Documentation/unaligned-memory-access.txt +11
-3+ +#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS + skb = original skb +#else + skb = copy skb +#endif + -- -Author: Daniel Drake <[email protected]> +Authors: Daniel Drake <[email protected]>, + Johannes Berg <[email protected]> With help from: Alan Cox, Avuton Olrich, Heikki Orsila, Jan Engelhardt, -Johannes Berg, Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, -Uli Kunitz, Vadim Lobanov +Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, Uli Kunitz, +Vadim Lobanovarch/powerpc/Kconfig +1
-0@@ -103,6 +103,7 @@ config PPC select HAVE_OPROFILE select HAVE_KPROBES select HAVE_KRETPROBES + select HAVE_EFFICIENT_UNALIGNED_ACCESS config EARLY_PRINTK boolarch/x86/Kconfig +1
-0@@ -23,6 +23,7 @@ config X86 select HAVE_KPROBES select HAVE_KRETPROBES select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64) + select HAVE_EFFICIENT_UNALIGNED_ACCESS config GENERIC_LOCKBREAK
arch/Kconfig config HAVE_KPROBES · line 45
Quoted 10 lines
arch/Kconfig +7
-0@@ -36,3 +36,22 @@ config HAVE_KPROBES config HAVE_KRETPROBES def_bool n + +config HAVE_EFFICIENT_UNALIGNED_ACCESS + def_bool n + help + Some architectures are unable to perform unaligned accesses + without the use of get_unaligned/put_unaligned. Others are + unable to perform such accesses efficiently (e.g. trap on
Wouldn't a comment be more appropriate instead of a help text? This option shouldn't be a visible make XYZconfig prompt.
I have no comment on the option itself; it's not my domain.
Signature · 3 lines
Stefan Richter -=====-==--- -=-- -=--- http://arcgraph.de/sr/
Stefan,
Thanks for the text review. I'll follow up with a new version in a minute.
Quoted 17 lines
arch/Kconfig config HAVE_KPROBES · line 45
Quoted 13 lines
arch/Kconfig +7
-0@@ -36,3 +36,22 @@ config HAVE_KPROBES config HAVE_KRETPROBES def_bool n + +config HAVE_EFFICIENT_UNALIGNED_ACCESS + def_bool n + help + Some architectures are unable to perform unaligned accesses + without the use of get_unaligned/put_unaligned. Others are + unable to perform such accesses efficiently (e.g. trap onWouldn't a comment be more appropriate instead of a help text? This option shouldn't be a visible make XYZconfig prompt.
Quoted 1 line
I have no comment on the option itself; it's not my domain.
Note that the help text doesn't mean the option gets to be visible, only the one-line description does, as in
config XYZ
bool "description"
vs.
config XYZ
bool
johannes
PATCH v6 1/1 introduce HAVE_EFFICIENT_UNALIGNED_ACCESS Kconfig symbol
In reply to Stefan RichterIn many cases, especially in networking, it can be beneficial to know at compile time whether the architecture can do unaligned accesses efficiently. This patch introduces a new Kconfig symbol
HAVE_EFFICIENT_UNALIGNED_ACCESS
for that purpose and adds it to the powerpc and x86 architectures. Also add some documentation about alignment and networking, and especially one intended use of this symbol.
---
v5: rename to HAVE_EFFICIENT_UNALIGNED_ACCESS
v6: small text fixes
I have opted to not introduce any symbol associated with the cost of unaligned accesses, David Woodhouse once suggested that such a cost should be combined with a probability of (un-)alignment even in the get_unaligned/put_unaligned macros and I think this should be combined with a patch introducing a global cost constant.
Also, this would require architecture changes because if some code knew the likelyhood of unaligned accesses was small enough over the cost of them, architectures that complain in the trap handle would still complain in that unlikely case although the code explicitly made a trade-off based on the cost/probability.
Documentation/unaligned-memory-access.txt | 32 +++++++++++++++++++++++++++--- arch/Kconfig | 19 +++++++++++++++++ arch/powerpc/Kconfig | 1 arch/x86/Kconfig | 1 4 files changed, 50 insertions(+), 3 deletions(-)
Documentation/unaligned-memory-access.txt +29 -3
@@ -218,9 +218,35 @@ If use of such macros is not convenient, where the source or destination (or both) are of type u8* or unsigned char*. Due to the byte-wise nature of this operation, unaligned accesses are avoided. + +Alignment vs. Networking +======================== + +On architectures that require aligned loads, networking requires that the IP +header is aligned on a four-byte boundary to optimise the IP stack. For +regular ethernet hardware, the constant NET_IP_ALIGN is used. On most +architectures this constant has the value 2 because the normal ethernet +header is 14 bytes long, so in order to get proper alignment one needs to +DMA to an address which can be expressed as 4*n + 2. One notable exception +here is powerpc which defines NET_IP_ALIGN to 0 because DMA to unaligned +addresses can be very expensive and dwarf the cost of unaligned loads. + +For some ethernet hardware that cannot DMA to unaligned addresses like +4*n+2 or non-ethernet hardware, this can be a problem, and it is then +required to copy the incoming frame into an aligned buffer. Because this is +unnecessary on architectures that can do unaligned accesses, the code can be +made dependent on CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS like so: + +#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS + skb = original skb +#else + skb = copy skb +#endif + -- -Author: Daniel Drake <[email protected]> +Authors: Daniel Drake <[email protected]>, + Johannes Berg <[email protected]> With help from: Alan Cox, Avuton Olrich, Heikki Orsila, Jan Engelhardt, -Johannes Berg, Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, -Uli Kunitz, Vadim Lobanov +Kyle McMartin, Kyle Moffett, Randy Dunlap, Robert Hancock, Uli Kunitz, +Vadim Lobanov
arch/powerpc/Kconfig +1 -0
@@ -103,6 +103,7 @@ config PPC select HAVE_OPROFILE select HAVE_KPROBES select HAVE_KRETPROBES + select HAVE_EFFICIENT_UNALIGNED_ACCESS config EARLY_PRINTK bool
arch/x86/Kconfig +1 -0
@@ -23,6 +23,7 @@ config X86 select HAVE_KPROBES select HAVE_KRETPROBES select HAVE_KVM if ((X86_32 && !X86_VOYAGER && !X86_VISWS && !X86_NUMAQ) || X86_64) + select HAVE_EFFICIENT_UNALIGNED_ACCESS config GENERIC_LOCKBREAK
arch/Kconfig +19 -0
@@ -36,3 +36,22 @@ config HAVE_KPROBES config HAVE_KRETPROBES def_bool n + +config HAVE_EFFICIENT_UNALIGNED_ACCESS + def_bool n + help + Some architectures are unable to perform unaligned accesses + without the use of get_unaligned/put_unaligned. Others are + unable to perform such accesses efficiently (e.g. trap on + unaligned access and require fixing it up in the exception + handler.) + + This symbol should be selected by an architecture if it can + perform unaligned accesses efficiently to allow different + code paths to be selected for these cases. Some network + drivers, for example, could opt to not fix up alignment + problems with received packets if doing so would not help + much. + + See Documentation/unaligned-memory-access.txt for more + information on the topic of unaligned memory accesses.
Hi,
Quoted 7 lines · Johannes Berg
In many cases, especially in networking, it can be beneficial to know at compile time whether the architecture can do unaligned accesses efficiently. This patch introduces a new
Kconfigsymbol
HAVE_EFFICIENT_UNALIGNED_ACCESS
for that purpose and adds it to the powerpc and x86 architectures. Also add some documentation about alignment and networking, and especially one intended use of this symbol.
Please CC [email protected] for such changes, so it's more likely noticed by more arch maintainers, so a list can be compiled what is appropriate for the various archs.
The scope of this symbol is IMO a little unclear, is it intended for one-time accesses to small object or also for repeated accesses to larger objects? Anyway, this probably should be set for m68k as well.
bye, Roman
Hi,
Quoted 3 lines
Please CC [email protected] for such changes, so it's more likely noticed by more arch maintainers, so a list can be compiled what is appropriate for the various archs.
Hah. I thought there was such a list but it didn't show up in MAINTAINERS so I thought I was wrong.
Quoted 3 lines
The scope of this symbol is IMO a little unclear, is it intended for one-time accesses to small object or also for repeated accesses to larger objects? Anyway, this probably should be set for m68k as well.
Also for repeated accesses, for example network packets where the IPv4 header might not always be aligned. That's why we want it to be "efficient", not just "possible" (as an earlier version of the patch called it.)
johannes