[Xen-ia64-devel] RE: ar.unat[patch] fixed this ar.uant issue.[patch] fixed ar.unat save/restore issue

["Magenheimer, Dan (HP Labs Fort Collins)" <dan.magenheimer@xxxxxx>]

> There should be not register nat bit fault when running itp,

Could you explain why this is true? (what is itp?)

> When nat page fault happens, it is usually caused by an 
> instruction which is accessing a page whose page attribute is 
> nat page, so it must be ld or st instruction, it is 

What if a privileged instruction is on a NaT page and
Xen needs to emulate that instruction?

Thanks,
Dan

> -----Original Message-----
> From: Xu, Anthony [mailto:anthony.xu@xxxxxxxxx] 
> Sent: Monday, November 14, 2005 3:37 AM
> To: Magenheimer, Dan (HP Labs Fort Collins)
> Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: ar.unat[patch] fixed this ar.uant issue.[patch] 
> fixed ar.unat save/restore issue
> 
> Yes, this patch may make dom0 go through ltp test, 
> Your logic to handle nat consumption fault is
> If( register nat bit fault)
>       Inject nat consumption fault to guest;
> Else(means this nat page fault)
>       Attempting to handle as privop
>       If( it is privop)
>               Return;
>       Else
>               Inject nat consumption fault to guest
>       
> When nat page fault happens, it is usually caused by an 
> instruction which is accessing a page whose page attribute is 
> nat page, so it must be ld or st instruction, it is 
> definitely not privop instruction. So it is not necessary to 
> attempt to handle nat fault as privop, we should inject it to 
> guest directly.
> There should be not register nat bit fault when running itp,
> So the logic in my mind is,
> If(register nat bit fault)
>       Panic();
> Else
>       Inject nat consumption fault to guest.
> 
> If it panics, there should be some places nearby where 
> ar.unat is not correctly handled. We should take this chance 
> to fix all ar.unat related bugs.
> 
> >I am still not sure about the use of eml_unat.  I commented
> >out your code (in ia64_handle_reflection) that sets it to zero
> 
> yes, you can comment this code, it was used for debugging 
> ar.unat fault.
> 
> 
> 
> Thanks
> -Anthony
> 
> 
> 
> 
> 
> 
> >-----Original Message-----
> >From: Magenheimer, Dan (HP Labs Fort Collins) 
> [mailto:dan.magenheimer@xxxxxx]
> >Sent: 2005å11æ12æ 3:30
> >To: Xu, Anthony
> >Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >Subject: RE: ar.unat[patch] fixed this ar.uant issue.[patch] 
> fixed ar.unat
> >save/restore issue
> >
> >Anthony --
> >
> >I just committed a fix to allow nat consumption faults to
> >be delivered again.  I think this is now necessary after
> >the region0 virtual address fixes needed for ltp-mmap09.
> >Without these nat fixes, ltp-getpeername01 reproducibly
> >goes into an infinite loop reporting NaT errors (because
> >the "return" in the reflection code doesn't result in
> >the NaT getting reflected to the guest).
> >
> >I have left the printfs so any code that results in
> >a inst/data page nat consumption fault (e.g. certain
> >situations where the zero page is accessed) will be
> >very chatty, but I think that's OK for now until we
> >are sure we have fixed all NaT problems.
> >
> >I am still not sure about the use of eml_unat.  I commented
> >out your code (in ia64_handle_reflection) that sets it to zero
> >and Tony's checker program and getpeername01 still work.
> >If this (setting eml_unat to zero) is handling some
> >special case that I am not testing for, please let me
> >know.
> >
> >Thanks,
> >Dan
> >
> >> -----Original Message-----
> >> From: Xu, Anthony [mailto:anthony.xu@xxxxxxxxx]
> >> Sent: Monday, November 07, 2005 6:30 PM
> >> To: Magenheimer, Dan (HP Labs Fort Collins)
> >> Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> Subject: RE: ar.unat[patch] fixed this ar.uant issue.[patch]
> >> fixed ar.unat save/restore issue
> >>
> >> See my comments,
> >>
> >> >-----Original Message-----
> >> >From: Magenheimer, Dan (HP Labs Fort Collins)
> >> [mailto:dan.magenheimer@xxxxxx]
> >> >Sent: 2005å11æ8æ 2:07
> >> >To: Xu, Anthony
> >> >Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> >Subject: RE: ar.unat[patch] fixed this ar.uant issue.[patch]
> >> fixed ar.unat
> >> >save/restore issue
> >> >
> >> >Another NaT question...
> >> >
> >> >>I recall that some time ago (around the time of the merge)
> >> >>you submitted some patches related to fixing ar.unat saving
> >> >>and restoring.
> >> >
> >> >Another part of your earlier patch was a change in
> >> >ia64_handle_reflection.  I still periodically get the
> >> >message:
> >> >
> >> >   NaT fault... attempting to handle as privop
> >> >
> >> >Since your latest fix, Tony's regcheck tool no longer
> >> >reports ar.unat as being saved/restored incorrectly.
> >> >I was hoping that the above message would go away also,
> >> >but it has not.  I see it a couple times at boot and
> >> >a couple times for every linux compile (at the end so
> >> >it is probably the linker or some other link-related
> >> >tool).  I have also seen programs segfault after printing
> >> >this message.  So I went to look at the Xen/ia64 code where
> >> >this is printed.
> >> >
> >>
> >> I have not seen nat consumptions and segmentations faults for
> >> a long time, in your build test and ltp test. Otherwise, I'll
> >> definitely try to fix that.
> >>
> >> >It doesn't look right to me.  There are two issues:
> >> >
> >> >1) Your patch added a "return"... I think this means that
> >> >   NaT faults will never get reflected to a guest (even
> >> >   Register NaT Consumption faults).
> >>
> >> Yes, you are right, we should inject Nat Consumption faults
> >> to guest, but as I know there should be not NaT consumption
> >> faults in linux, so I simply added a "return". I think the
> >> best way is to add "panic" at this place, this will enforce
> >> us to debug this issue rather than temporarily work around.
> >>
> >>
> >> >2) Since a Instruction NaTPage Consumption fault has higher
> >> >   priority than a Privileged Operation fault, I think the
> >> >   original printf/priv_emulate code was intended to catch
> >> >   this case and properly emulate a privileged instruction
> >> >   on a NaTPage.  I think it may also be necessary if a Data
> >> >   NaTPage Consumption fault is incurred when the privop
> >> >   emulation code fetches the instruction.  (The code in
> >> >   ia64_handle_reflection should probably check the ISR to
> >> >   avoid calling priv_emulate for other kinds of NaT
> >> >   Consumption though.)
> >>
> >> I have been being curious why use emulate function to handle
> >> NaT consumption.
> >> Now I understand, thank you for your detailed explain. Maybe
> >> we need to put more comments in the confusing place like this.
> >>
> >>
> >>
> >> >You know more about NaT's than I do... could you recheck
> >> >this code in ia64_handle_reflection please?  Do you have
> >> >any test code that provokes any of these NaT faults?
> >> >
> >>
> >> It' is very kind of you to say that, unfortunately I have not
> >> seen those issues. What I suspect is dom0 does bank switch on
> >> shared page but not consider ar.unat.
> >>
> >> Anyway, I'll try to provoke this fault, If I find, I'll
> >> definitely fix it.
> >>
> >> >Thanks.
> >> >Dan
> >> >
> >> >> -----Original Message-----
> >> >> From: Xu, Anthony [mailto:anthony.xu@xxxxxxxxx]
> >> >> Sent: Friday, November 04, 2005 12:10 AM
> >> >> To: Magenheimer, Dan (HP Labs Fort Collins)
> >> >> Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> >> Subject: RE: ar.unat[patch] fixed this ar.uant issue.[patch]
> >> >> fixed ar.unat save/restore issue
> >> >>
> >> >> >I am curious about the use of B1NATS in the code
> >> >> >around this patch.  Under what circumstances does
> >> >> >this get set/used?
> >> >>
> >> >> 1. emulate  bsw1, bsw0
> >> >> 2. emulate rfi.
> >> >> 3. inject fault to guest.
> >> >>
> >> >> There is similar unat code in
> >> >> >fast_tick (default off) and fast_reflect (default on)
> >> >> >and I am wondering if similar unat changes are needed
> >> >> >and whether it is now OK to turn on HANDLE_AR_UNAT
> >> >> >(which is now default off).
> >> >> You are right, in above two cases you should also save
> >> >> ar.unat to XSI_B1NATS_OFS after spilling the guest bank1to
> >> >> share page. I had handled all this in C code. I didn't look
> >> >> into fast hypercall code, It's hard to read due to I am not
> >> >> good at assembly code. The principle of handling ar.unat is
> >> >> obvious; every time you spill banking register you must save
> >> >> corresponding ar.unat after it, every time you fill banking
> >> >> register you must restore corresponding ar.unat before it.
> >> >>
> >> >> We don't need to clear all guest b0 registers and their's nat
> >> >> bit. Because r16~r23 are preserved regs and r24~r31 are
> >> >> scratch regs, we only need to restore r16~r23 rather than
> >> >> clear r16~r23 to 0.
> >> >>
> >> >> Next time you enable some functions like hyper_ssm_i, when
> >> >> you save bank1 regs you should also save bank1 unat.
> >> >>
> >> >> Below patch enables HANDLE_AR_UNAT.
> >> >>
> >> >>
> >> >>
> >> >> Signed-off-by Anthony Xu <Anthony.xu@xxxxxxxxx>
> >> >>
> >> >> Thanks,
> >> >> Anthony.
> >> >>
> >> >>
> >> >>
> >> >>
> >> >> >-----Original Message-----
> >> >> >From: Magenheimer, Dan (HP Labs Fort Collins)
> >> >> [mailto:dan.magenheimer@xxxxxx]
> >> >> >Sent: 2005å11æ3æ 22:42
> >> >> >To: Xu, Anthony
> >> >> >Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> >> >Subject: RE: ar.unat[patch] fixed this ar.uant issue.
> >> >> >
> >> >> >Hi Anthony --
> >> >> >
> >> >> >I am curious about the use of B1NATS in the code
> >> >> >around this patch.  Under what circumstances does
> >> >> >this get set/used?  There is similar unat code in
> >> >> >fast_tick (default off) and fast_reflect (default on)
> >> >> >and I am wondering if similar unat changes are needed
> >> >> >and whether it is now OK to turn on HANDLE_AR_UNAT
> >> >> >(which is now default off).
> >> >> >
> >> >> >Thanks,
> >> >> >Dan
> >> >> >
> >> >> >> -----Original Message-----
> >> >> >> From: Xu, Anthony [mailto:anthony.xu@xxxxxxxxx]
> >> >> >> Sent: Thursday, November 03, 2005 1:08 AM
> >> >> >> To: Magenheimer, Dan (HP Labs Fort Collins)
> >> >> >> Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> >> >> Subject: RE: ar.unat[patch] fixed this ar.uant issue.
> >> >> >>
> >> >> >> Dan,
> >> >> >> Last time, I used ar.unat register to restore guest general
> >> >> >> register nat bit in hyper_rfi function for eliminating nat
> >> >> >> bit consumption fault,but not restored ar.unat.
> >> >> >>
> >> >> >> Signed-off-by Anthony Xu <Anthony.xu@xxxxxxxxx>
> >> >> >>
> >> >> >> Thanks,
> >> >> >> Anthony.
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >>
> >> >> >> >-----Original Message-----
> >> >> >> >From: Magenheimer, Dan (HP Labs Fort Collins)
> >> >> >> [mailto:dan.magenheimer@xxxxxx]
> >> >> >> >Sent: 2005å11æ3æ 11:54
> >> >> >> >To: Xu, Anthony
> >> >> >> >Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
> >> >> >> >Subject: RE: ar.unat
> >> >> >> >
> >> >> >> >> I can take a look at this, please send me regcheck utilty.
> >> >> >> >>
> >> >> >> >>
> >> >> >> >> Thanks
> >> >> >> >> Anthony
> >> >> >> >
> >> >> >> >Great, thanks!  Here's where I got Tony's regcheck tool.  If
> >> >> >> >it's not still there, perhaps Tony can post it.
> >> >> >> >
> >> >> >> >By the way, if anyone tries this on a domU, Matt Chapman
> >> >> >> >has a pending fix that resolves a FP save/restore issue.
> >> >> >> >
> >> >> >> >Thanks,
> >> >> >> >Dan
> >> >> >> >
> >> >> >> >> -----Original Message-----
> >> >> >> >> From: linux-ia64-owner@xxxxxxxxxxxxxxx
> >> >> >> >> [mailto:linux-ia64-owner@xxxxxxxxxxxxxxx] On Behalf Of
> >> >> Luck, Tony
> >> >> >> >> Sent: Tuesday, March 01, 2005 4:33 PM
> >> >> >> >> To: linux-ia64@xxxxxxxxxxxxxxx
> >> >> >> >> Subject: RE: [patch 2.6.11-rc3-bk4] Correctly dereference
> >> >> >> >> ia64_mca_data
> >> >> >> >>
> >> >> >> >> Back on February 9th, I wrote:
> >> >> >> >> >I wrote a test program that loads up random values
> >> >> into registers
> >> >> >> >> >(just r1-r31, a bunch of stacked registers, and
> >> >> f2-f127 for now)
> >> >> >> >> >and then checks that all the registers haven't
> >> changed value a
> >> >> >> >> >few thousand times, before reloading with a new set
> >> of random
> >> >> >> >> >values.
> >> >> >> >>
> >> >> >> >> A few people asked whether I could post the program
> >> ... it took
> >> >> >> >> a while to get sign-off ... but that gave me time to
> >> >> add "branch",
> >> >> >> >> "predicate" and half a dozen "application" registers
> >> to the mix,
> >> >> >> >> plus make it print the name of the register that was
> >> >> nuked (instead
> >> >> >> >> of a number that required manual translation).
> >> >> >> >>
> >> >> >> >> I've tested it by using a debugger to zap one of 
> each class
> >> >> >> >> of register
> >> >> >> >> that is being monitored to check that it works.
> >> >> >> >>
> >> >> >> >>
> >> >> 
> http://www.kernel.org/pub/linux/kernel/people/aegl/ia64regcheck.tgz
> >> >> >> >>
> >> >> >> >> Usage ... compile, and run a few copies.  If they all
> >> >> >> "exit(0)" (which
> >> >> >> >> may take a couple of days) the test passed.  Otherwise you
> >> >> >> should see
> >> >> >> >> the name of the register printed to stderr, and 
> exit code 1.
> >> >> >> >>
> >> >> >> >> Apart from the MCA case, I haven't seen it report 
> a problem
> >> >> >> >> yet ... but
> >> >> >> >> I've only run a few hours.
> >> >> >> >>
> >> >> >> >> -Tony
> >> >> >>
> >> >>
> >>
> 

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel