Message ID | 20170710144425.2238584-1-arnd@arndb.de |
---|---|
State | New |
Headers | show |
Series | [RFC,1/2] x86: mark target address as output in 'insb' asm | expand |
On Wed, Jul 12, 2017 at 6:10 AM, tip-bot for Arnd Bergmann <tipbot@zytor.com> wrote: > > Apparently the assember constraints are slightly off here, as marking the > 'addr' argument as a memory output seems appropriate here and gets rid > of the warning. For consistency I'm also adding it as input for outsb(). The new constraints look very questionable to me. > asm volatile("rep; outs" #bwl \ > - : "+S"(addr), "+c"(count) : "d"(port)); \ > + : "+S"(addr), "+c"(count) : "d"(port), "m" (addr));\ > asm volatile("rep; ins" #bwl \ > - : "+D"(addr), "+c"(count) : "d"(port)); \ > + : "+D"(addr), "+c"(count), "=m" (addr) : "d"(port));\ That's not how "m" works, afaik. You're passing in an address, but "m" takes the _value_. So as far as I can tell, what you are really doing is say "this reads/writes the memory that contains the _pointer_". So not only does it not do what you think it does, it probably actually forces "addr" to be loaded into a stack slot, in order for the inline asm to be able to "access" that memory location to get (and set) the value of "addr". So if it hides warnings, it does so by virtue of confusing gcc some more about what is actually going on, rather than by fixing the issue. I do agree that those inline asm things do lack a memory dependency, though. I just think that patch is *completely* wrong. The real fix is probably to just mark them as "clobbers memory" (ie just add "memory" to the clobber list). If you want to be fancy, you can try to do what <asm/uaccess.h> does, which is a disgusting hack, but has traditionally worked; struct __large_struct { unsigned long buf[100]; }; #define __m(x) (*(struct __large_struct __user *)(x)) and then use your approach with "m" and "=m". Linus
* Linus Torvalds <torvalds@linux-foundation.org> wrote: > On Wed, Jul 12, 2017 at 6:10 AM, tip-bot for Arnd Bergmann > <tipbot@zytor.com> wrote: > > > > Apparently the assember constraints are slightly off here, as marking the > > 'addr' argument as a memory output seems appropriate here and gets rid > > of the warning. For consistency I'm also adding it as input for outsb(). > > The new constraints look very questionable to me. Ok, I've removed the commit. > The real fix is probably to just mark them as "clobbers memory" (ie > just add "memory" to the clobber list). > > If you want to be fancy, you can try to do what <asm/uaccess.h> does, > which is a disgusting hack, but has traditionally worked; > > struct __large_struct { unsigned long buf[100]; }; > #define __m(x) (*(struct __large_struct __user *)(x)) > > and then use your approach with "m" and "=m". Arnd, could you please try Linus's suggestions? Thanks, Ingo
On Wed, Jul 12, 2017 at 6:57 PM, Linus Torvalds <torvalds@linux-foundation.org> wrote: > On Wed, Jul 12, 2017 at 6:10 AM, tip-bot for Arnd Bergmann > <tipbot@zytor.com> wrote: >> >> Apparently the assember constraints are slightly off here, as marking the >> 'addr' argument as a memory output seems appropriate here and gets rid >> of the warning. For consistency I'm also adding it as input for outsb(). > > The new constraints look very questionable to me. > >> asm volatile("rep; outs" #bwl \ >> - : "+S"(addr), "+c"(count) : "d"(port)); \ >> + : "+S"(addr), "+c"(count) : "d"(port), "m" (addr));\ > >> asm volatile("rep; ins" #bwl \ >> - : "+D"(addr), "+c"(count) : "d"(port)); \ >> + : "+D"(addr), "+c"(count), "=m" (addr) : "d"(port));\ > > That's not how "m" works, afaik. You're passing in an address, but "m" > takes the _value_. > > So as far as I can tell, what you are really doing is say "this > reads/writes the memory that contains the _pointer_". > > So not only does it not do what you think it does, it probably > actually forces "addr" to be loaded into a stack slot, in order for > the inline asm to be able to "access" that memory location to get (and > set) the value of "addr". Ok, got it, thanks for taking a look! > So if it hides warnings, it does so by virtue of confusing gcc some > more about what is actually going on, rather than by fixing the issue. Right, as far as gcc is concerned, 'addr' might now point to something that was initialized, so it no longer warns even though it still thinks that *addr did not get touched. > I do agree that those inline asm things do lack a memory dependency, > though. I just think that patch is *completely* wrong. > > The real fix is probably to just mark them as "clobbers memory" (ie > just add "memory" to the clobber list). > > If you want to be fancy, you can try to do what <asm/uaccess.h> does, > which is a disgusting hack, but has traditionally worked; > > struct __large_struct { unsigned long buf[100]; }; > #define __m(x) (*(struct __large_struct __user *)(x)) > > and then use your approach with "m" and "=m". Ok, I'll try both tomorrow and see where I end up. I guess it's not urgent to fix it since that code has literally been there since linux-0.1 (first in hd.c, and moved to asm/io.h in 0.99.17k). Would you expect that the missing clobber causes actual runtime bugs and the fix needs to be backported to stable kernels? Arnd
On Wed, Jul 12, 2017 at 2:47 PM, Arnd Bergmann <arnd@arndb.de> wrote: >> >> The real fix is probably to just mark them as "clobbers memory" (ie >> just add "memory" to the clobber list). >> >> If you want to be fancy, you can try to do what <asm/uaccess.h> does, >> which is a disgusting hack, but has traditionally worked; >> >> struct __large_struct { unsigned long buf[100]; }; >> #define __m(x) (*(struct __large_struct __user *)(x)) >> >> and then use your approach with "m" and "=m". > > Ok, I'll try both tomorrow and see where I end up. Note that just adding the "memory" thing to the clobbers likely causes slightly worse code generation (it basically says that the asm can clobber anything at all, so cause re-loads etc that are entirely unrelated to the asm). But for something like "rep in/out", that really doesn't much matter. PIO is very slow due to being fully serialized, and "rep ins/outs" is just about the slowest thing you can do on a machine. So nobody really does it, the main traditional user was the legacy PIO data transfer for ST-506 disks. And, as you noticed a few *really* old network card drivers. So I suspect the big hammer memory clobber is the right thing to do. > Would you expect that the missing clobber causes actual > runtime bugs and the fix needs to be backported to stable > kernels? Probably not. "asm volatile" is already pretty serialized. It's not entirely obvious what gcc will move around it, but almost certainly no operations that matter for the network buffer. And honestly, the wt3501 is basically an ISA card in PCMCIA format. I don't think it's even cardbus (Cardbus aka Yenta is basically "hotplug PCI"). So I don't think anybody even has that hardware. It was a good card for its time. But its time was basically two decades ago. Linus
diff --git a/arch/x86/include/asm/io.h b/arch/x86/include/asm/io.h index 7afb0e2f07f4..d107251eabd9 100644 --- a/arch/x86/include/asm/io.h +++ b/arch/x86/include/asm/io.h @@ -328,13 +328,13 @@ static inline unsigned type in##bwl##_p(int port) \ static inline void outs##bwl(int port, const void *addr, unsigned long count) \ { \ asm volatile("rep; outs" #bwl \ - : "+S"(addr), "+c"(count) : "d"(port)); \ + : "+S"(addr), "+c"(count) : "d"(port), "m" (addr));\ } \ \ static inline void ins##bwl(int port, void *addr, unsigned long count) \ { \ asm volatile("rep; ins" #bwl \ - : "+D"(addr), "+c"(count) : "d"(port)); \ + : "+D"(addr), "+c"(count), "=m" (addr) : "d"(port));\ } BUILDIO(b, b, char)
The -Wmaybe-uninitialized warning triggers for one driver using the output of the 'insb' I/O helper on x86: drivers/net/wireless/wl3501_cs.c: In function ‘wl3501_mgmt_scan_confirm’: drivers/net/wireless/wl3501_cs.c:665:9: error: ‘sig.status’ is used uninitialized in this function [-Werror=uninitialized] drivers/net/wireless/wl3501_cs.c:668:12: error: ‘sig.cap_info’ may be used uninitialized in this function [-Werror=maybe-uninitialized] Apparently the assember constraints are slightly off here, as marking the 'addr' argument as a memory output seems appropriate here and gets rid of the warning. For consistency I'm also adding it as input for outsb(). Unfortunately, this fix triggers another problem when CONFIG_KASAN is set, again only in this one driver: drivers/net/wireless/wl3501_cs.c: In function 'wl3501_rx_interrupt': drivers/net/wireless/wl3501_cs.c:1103:1: error: the frame size of 2232 bytes is larger than 1536 bytes [-Werror=frame-larger-than=] I'm not an x86 person and gcc inline assembly mystifies me all the time, so please review this carefully and suggest a better way if this is not how it should be done. Signed-off-by: Arnd Bergmann <arnd@arndb.de> --- arch/x86/include/asm/io.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) -- 2.9.0