| Message ID | 1488329895-11305-1-git-send-email-mdroth@linux.vnet.ibm.com |
|---|---|
| State | Superseded |
| Headers | show |
On Tue, Feb 28, 2017 at 06:58:15PM -0600, Michael Roth wrote: > +: assign-var-min-align ( size var min-align -- al-mem ) > + swap >r \ ( size min-align ) > + over \ ( size min-align size ) > + max \ ( size align ) Why did you change the UMAX to MAX ? They are not the same. Sizes (and alignments) are unsigned numbers. Segher
Quoting Segher Boessenkool (2017-02-28 19:23:18) > On Tue, Feb 28, 2017 at 06:58:15PM -0600, Michael Roth wrote: > > +: assign-var-min-align ( size var min-align -- al-mem ) > > + swap >r \ ( size min-align ) > > + over \ ( size min-align size ) > > + max \ ( size align ) > > Why did you change the UMAX to MAX ? They are not the same. Sizes > (and alignments) are unsigned numbers. umax wasn't defined in the SLOF environment, and with 64-bit signed values it didn't seem likely that min-align/size would overflow. But yah, that's a bit sloppy... Does the following umax implementation seem reasonable? 0 > : same-sign 0< swap 0< = ; ok 0 > : umax 2dup same-sign IF max ELSE dup 0< IF swap THEN drop THEN ; ok 0 > 0 0 umax . 0 ok 0 > 0 1 umax . 1 ok 0 > 1 0 umax . 1 ok 0 > 1 1 umax . 1 ok 0 > 70000000 70000000 umax . 70000000 ok 0 > 70000000 80000000 umax . 80000000 ok 0 > 80000000 70000000 umax . 80000000 ok 0 > 80000000 80000000 umax . 80000000 ok 0 > 7000000000000000 7000000000000000 umax . 7000000000000000 ok 0 > 7000000000000000 8000000000000000 umax . -8000000000000000 ok 0 > 8000000000000000 7000000000000000 umax . -8000000000000000 ok 0 > 8000000000000000 8000000000000000 umax . -8000000000000000 ok 0 > ffffffffffffffff fffffffffffffffe umax . -1 ok 0 > fffffffffffffffe ffffffffffffffff umax . -1 ok > > > Segher >
On Wed, Mar 01, 2017 at 10:42:13PM -0600, Michael Roth wrote: > Quoting Segher Boessenkool (2017-02-28 19:23:18) > > On Tue, Feb 28, 2017 at 06:58:15PM -0600, Michael Roth wrote: > > > +: assign-var-min-align ( size var min-align -- al-mem ) > > > + swap >r \ ( size min-align ) > > > + over \ ( size min-align size ) > > > + max \ ( size align ) > > > > Why did you change the UMAX to MAX ? They are not the same. Sizes > > (and alignments) are unsigned numbers. > > umax wasn't defined in the SLOF environment, and with 64-bit signed values > it didn't seem likely that min-align/size would overflow. But yah, > that's a bit sloppy... Yeah exactly :-) > Does the following umax implementation seem reasonable? > > 0 > : same-sign 0< swap 0< = ; ok > 0 > : umax 2dup same-sign IF max ELSE dup 0< IF swap THEN drop THEN ; ok Well you can just use u< of course. We have (in engine.in) : max 2dup < IF swap THEN drop ; so analogous you get : umax 2dup u< IF swap THEN drop ; Cheers, Segher
diff --git a/board-qemu/slof/pci-phb.fs b/board-qemu/slof/pci-phb.fs index 667514e..5289e25 100644 --- a/board-qemu/slof/pci-phb.fs +++ b/board-qemu/slof/pci-phb.fs @@ -331,6 +331,12 @@ setup-puid my-puid TO puid \ Set current puid phb-parse-ranges 1 TO pci-hotplug-enabled + s" qemu,mem-bar-min-align" get-node get-property 0<> IF + 10000 TO pci-mem-bar-min-align \ Force 64k alignment for all mem BARs + ELSE + decode-int TO pci-mem-bar-min-align + 2drop + THEN s" qemu,phb-enumerated" get-node get-property 0<> IF 1 0 (probe-pci-host-bridge) ELSE diff --git a/slof/fs/pci-properties.fs b/slof/fs/pci-properties.fs index 7faa714..01fbc85 100644 --- a/slof/fs/pci-properties.fs +++ b/slof/fs/pci-properties.fs @@ -91,16 +91,35 @@ \ *************************************************************************************** \ align the current mem and set var to next mem \ align with a size of 0 returns 0 !!! +: assign-var-min-align ( size var min-align -- al-mem ) + swap >r \ ( size min-align ) + over \ ( size min-align size ) + max \ ( size align ) + r@ @ \ ( size align cur-mem ) read current free mem + swap #aligned \ ( size al-mem ) align the mem + tuck + \ ( al-mem new-mem ) add size to aligned mem + r> ! \ ( al-mem ) set variable to new mem +; + : assign-var ( size var -- al-mem ) - 2dup @ \ ( size var size cur-mem ) read current free mem - swap #aligned \ ( size var al-mem ) align the mem to the size - dup 2swap -rot + \ ( al-mem var new-mem ) add size to aligned mem - swap ! \ ( al-mem ) set variable to new mem + 0 assign-var-min-align ; \ set bar to current free mem ( in variable ) and set variable to next free mem : assign-bar-value32 ( bar size var -- 4 ) over IF \ IF size > 0 + pci-mem-bar-min-align + assign-var-min-align \ | ( bar al-mem ) set variable to next mem + swap rtas-config-l! \ | ( -- ) set the bar to al-mem + ELSE \ ELSE + 2drop drop \ | clear stack + THEN \ FI + 4 \ size of the base-address-register +; + +\ set bar to current free mem ( in variable ) and set variable to next free mem +: assign-io-bar-value32 ( bar size var -- 4 ) + over IF \ IF size > 0 assign-var \ | ( bar al-mem ) set variable to next mem swap rtas-config-l! \ | ( -- ) set the bar to al-mem ELSE \ ELSE @@ -112,7 +131,8 @@ \ set bar to current free mem ( in variable ) and set variable to next free mem : assign-bar-value64 ( bar size var -- 8 ) over IF \ IF size > 0 - assign-var \ | ( bar al-mem ) set variable to next mem + pci-mem-bar-min-align + assign-var-min-align \ | ( bar al-mem ) set variable to next mem swap \ | ( al-mem addr ) calc config-addr of this bar 2dup rtas-config-l! \ | ( al-mem addr ) set the Lower part of the bar to al-mem 4 + swap 20 rshift \ | ( al-mem>>32 addr ) prepare the upper part of the al-mem @@ -163,7 +183,7 @@ : assign-io-bar ( bar-addr -- 4 ) dup pci-bar-size-io \ fetch size pci-next-io \ var to change - assign-bar-value32 \ and set it all + assign-io-bar-value32 \ and set it all ; \ Setup an Expansion ROM bar diff --git a/slof/fs/pci-scan.fs b/slof/fs/pci-scan.fs index a528c8e..bd8cc0d 100644 --- a/slof/fs/pci-scan.fs +++ b/slof/fs/pci-scan.fs @@ -24,6 +24,9 @@ VARIABLE pci-max-io VARIABLE pci-next-mem64 \ prefetchable 64-bit memory mapped VARIABLE pci-max-mem64 +\ 0 to default to natural alignment +0 VALUE pci-mem-bar-min-align + \ Counter of busses found 0 VALUE pci-bus-number \ Counter of devices found
This is needed to ensure VFIO passthrough devices are able to offload MMIO accesses to KVM. Cc: Segher Boessenkool <segher@kernel.crashing.org> Cc: Nikunj A Dadhania <nikunj@linux.vnet.ibm.com> Cc: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com> --- board-qemu/slof/pci-phb.fs | 6 ++++++ slof/fs/pci-properties.fs | 32 ++++++++++++++++++++++++++------ slof/fs/pci-scan.fs | 3 +++ 3 files changed, 35 insertions(+), 6 deletions(-)