Fix assert in powerpc pmaps after introduction of object busy.

[kib]

Summary

The VM_PAGE_OBJECT_BUSY_ASSERT() in pmap_enter() implementation should be only asserted when the code is executed as result of pmap_enter(), not when the same code is entered from e.g. pmap_enter_quick(). This is relevant for all PowerPC pmap variants, because mmu_*_enter() is used as the backend, and assert is located there.

Add a ppc private pmap_enter() flag to indicate that the call is not from pmap_enter().

[markj]

I don't see a problem with the diff as-is, but I believe we can also assert that m->object is locked in the PMAP_ENTER_QUICK_LOCKED paths.

[bdragon]

sys/powerpc/booke/pmap.c

2282 ↗

(On Diff #63302)

flags is uninitialized in this case.

[bdragon]

sys/powerpc/booke/pmap.c

2282 ↗

(On Diff #63302)

should be pmap_flags maybe?

[bdragon]

booke compile errors.

[kib]

Update flags arg name for booke.
Assert that the object is locked if !quick.

[bdragon]

I don't see an actual change for flags -> pmap_flags in mmu_booke_enter_locked(), did you upload the right diff?

[kib]

Really update used var.

[alc]

Why not use PMAP_ENTER_NOSLEEP instead of introducing a new flag?

[alc]

If my recollection of the sparc64 pmap is correct, it will have the same issue.

[bdragon]

X5000: Gets further, hangs with witness lock held too long on usbd_ctrl_lock during "sh /etc/rc autoboot" (Probably a separate issue)
POWER9: OK
G5: OK
RB800: Not tested yet
G4: Not tested yet

[kib]

In D22041#481448, @alc wrote:

Why not use PMAP_ENTER_NOSLEEP instead of introducing a new flag?

Because PMAP_ENTER_NOSLEEP is used in vm_fault_soft_fast(), where we very much want to assert that the object is busy.

[kib]

In D22041#481451, @alc wrote:

If my recollection of the sparc64 pmap is correct, it will have the same issue.

Thank you for the hint, I will handle this after ppc is done.

[bdragon]

G4: panic: vm_page_free_prep: mapping flags set in page 0xd032c9f8

[markj]

In D22041#481461, @bdragon wrote:

G4: panic: vm_page_free_prep: mapping flags set in page 0xd032c9f8

Can you "show pginfo 0xd032c9f8" from DDB? In particular, which of PGA_EXECUTABLE and PGA_WRITEABLE is set?

[bdragon]

screenshot of the backtrace at http://drop.rtk0.net/IMG_1255.JPG

I don't have keyboard access to actually run commands unless I lock out the framebuffer. Trying to remember the magic boot args to do so...

[bdragon]

ok, rebuilt with nooptions vt so the ofw console stays up.

panic: vm_page_free_prep: mapping flags set in page 0xd1885a98

page 0xd1885a98 obj 0 pidx 0x1e phys 0x4e19d000 q 1 ref 0
  af 0x9 of 0x0 f 0x1 act 5 busy 1 valid 0xff dirty 0xff

[markj]

In D22041#481466, @bdragon wrote:

ok, rebuilt with nooptions vt so the ofw console stays up.

panic: vm_page_free_prep: mapping flags set in page 0xd1885a98

page 0xd1885a98 obj 0 pidx 0x1e phys 0x4e19d000 q 1 ref 0
  af 0x9 of 0x0 f 0x1 act 5 busy 1 valid 0xff dirty 0xff

I believe this is fixed by D22044.

[bdragon]

G4 appears to be working now when using markj's patch on top of this.

[bdragon]

RB800: OK

[bdragon]

X5000: OK with one caveat -- I currently have to disable ehci.