linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] drm/etnaviv: Fix the pm_domain lookup
@ 2020-05-13 22:02 Lubomir Rintel
  2020-05-14 14:58 ` Christian Gmeiner
  0 siblings, 1 reply; 2+ messages in thread
From: Lubomir Rintel @ 2020-05-13 22:02 UTC (permalink / raw)
  To: Lucas Stach
  Cc: Russell King, Christian Gmeiner, etnaviv, dri-devel,
	linux-kernel, Lubomir Rintel

On a GC860 (both 3D and 2D capable) GPU, kmscube crashes:

  # strace -f ~lkundrak/src/kmscube/build/kmscube
  ...
  ioctl(6, DRM_IOCTL_ETNAVIV_PM_QUERY_DOM, 0xbe92b720) = 0
  ioctl(6, DRM_IOCTL_ETNAVIV_PM_QUERY_SIG <unfinished ...>) = ?
  +++ killed by SIGSEGV +++
  Segmentation fault (core dumped)

And triggers an oops:

  8<--- cut here ---
  Unable to handle kernel NULL pointer dereference at virtual address 00000000
  pgd = 40e2c0f7
  [00000000] *pgd=0df6d831, *pte=00000000, *ppte=00000000
  Internal error: Oops: 17 [#1] PREEMPT SMP ARM
  Modules linked in:
  CPU: 0 PID: 346 Comm: kmscube Not tainted 5.7.0-rc4+ #792
  Hardware name: Marvell MMP2 (Device Tree Support)
  PC is at strncpy+0x14/0x30
  LR is at etnaviv_pm_query_sig+0xd0/0x104
  pc : [<c04f35f4>]    lr : [<c05dd878>]    psr: 20010013
  sp : c85f5e00  ip : c85f5eb5  fp : beb58748
  r10: 0000004c  r9 : ca6f9100  r8 : c85f5e6c
  r7 : 00000050  r6 : c85f5e6c  r5 : 00000001  r4 : c0b69ae8
  r3 : c85f5e75  r2 : 0000003f  r1 : 00000000  r0 : c85f5e76
  Flags: nzCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
  Control: 10c5387d  Table: 0df70019  DAC: 00000051
  Process kmscube (pid: 346, stack limit = 0x816fba31)
  Stack: (0xc85f5e00 to 0xc85f6000)
  5e00: 00000000 d90e6000 00000020 c05d5b2c c85f5e6c c059ce90 00000000 c1003f88
  5e20: c04c644b 0000004c c0b69610 c04c644b c85f5e6c 0000004b ca6f9100 c059d0bc
  5e40: 00000001 c0d53ee8 c85f5f18 00000001 c85f5f50 c85f5e6c 0000004c c8454240
  5e60: c05d5b2c 00000051 00000000 00000000 00000001 00000000 00000000 00000000
  5e80: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
  5ea0: 00000000 00000000 00000000 00000000 00000000 00000000 00000274 c011c3fc
  5ec0: 00000100 c0290434 5ebc5632 30e03501 5ebc5632 c8526600 00000274 00100cca
  5ee0: 00000831 b64c5000 cdf72d90 c1003f88 00000000 c04c644b c8454240 beb58748
  5f00: c8454240 00000006 c85f4000 d90ecad8 001c01a0 c02d49ac b64c52cc 80000007
  5f20: da9d6dc0 d9aa4000 d9aa4040 00000000 00000274 c011818c 00000005 0e200080
  5f40: 00000000 000003e5 00000000 00000100 00000000 00000000 00000000 cc78ac40
  5f60: 00000006 00000007 c1009a98 b64c52cc c85f5fb0 c0118080 00000080 c1003f88
  5f80: 00000000 00000001 beb58748 c04c644b 00000036 c0100288 c85f4000 00000036
  5fa0: 001c01a0 c0100060 00000001 beb58748 00000006 c04c644b beb58748 0000004c
  5fc0: 00000001 beb58748 c04c644b 00000036 beb58748 001bd688 beb58700 001c01a0
  5fe0: b6f41f08 beb586d4 b6f2784c b6e16cec 80010010 00000006 00000000 00000000
  [<c04f35f4>] (strncpy) from [<c05dd878>] (etnaviv_pm_query_sig+0xd0/0x104)
  [<c05dd878>] (etnaviv_pm_query_sig) from [<c059ce90>] (drm_ioctl_kernel+0xb4/0xf8)
  [<c059ce90>] (drm_ioctl_kernel) from [<c059d0bc>] (drm_ioctl+0x1e8/0x3b8)
  [<c059d0bc>] (drm_ioctl) from [<c02d49ac>] (ksys_ioctl+0xe0/0xaf0)
  [<c02d49ac>] (ksys_ioctl) from [<c0100060>] (ret_fast_syscall+0x0/0x54)
  Exception stack(0xc85f5fa8 to 0xc85f5ff0)
  5fa0:                   00000001 beb58748 00000006 c04c644b beb58748 0000004c
  5fc0: 00000001 beb58748 c04c644b 00000036 beb58748 001bd688 beb58700 001c01a0
  5fe0: b6f41f08 beb586d4 b6f2784c b6e16cec
  Code: 012fff1e e2422001 e2403001 e080c002 (e5d12000)
  ---[ end trace 387aad33cd9c15ea ]---

Turns out that it's because pm_domain() returns a pointer outside any
any of the etnaviv_pm_domains. Unless I'm mistaken, the algorithm in
pm_domain() is entirely botched when GPU's features match more than one
domain. This tries to remedy it.

Tested with kmscube with mesa 20 on ome machine with GC860 and another
with GC2000 + GC300 pair.

Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
---
 drivers/gpu/drm/etnaviv/etnaviv_perfmon.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_perfmon.c b/drivers/gpu/drm/etnaviv/etnaviv_perfmon.c
index e6795bafcbb9..9dc1bb4d4582 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_perfmon.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_perfmon.c
@@ -444,7 +444,6 @@ static unsigned int num_pm_domains(const struct etnaviv_gpu *gpu)
 static const struct etnaviv_pm_domain *pm_domain(const struct etnaviv_gpu *gpu,
 	unsigned int index)
 {
-	const struct etnaviv_pm_domain *domain = NULL;
 	unsigned int offset = 0, i;
 
 	for (i = 0; i < ARRAY_SIZE(doms_meta); i++) {
@@ -453,15 +452,15 @@ static const struct etnaviv_pm_domain *pm_domain(const struct etnaviv_gpu *gpu,
 		if (!(gpu->identity.features & meta->feature))
 			continue;
 
-		if (meta->nr_domains < (index - offset)) {
+		if (meta->nr_domains <= (index - offset)) {
 			offset += meta->nr_domains;
 			continue;
 		}
 
-		domain = meta->domains + (index - offset);
+		return meta->domains + (index - offset);
 	}
 
-	return domain;
+	return NULL;
 }
 
 int etnaviv_pm_query_dom(struct etnaviv_gpu *gpu,
-- 
2.26.2


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] drm/etnaviv: Fix the pm_domain lookup
  2020-05-13 22:02 [PATCH] drm/etnaviv: Fix the pm_domain lookup Lubomir Rintel
@ 2020-05-14 14:58 ` Christian Gmeiner
  0 siblings, 0 replies; 2+ messages in thread
From: Christian Gmeiner @ 2020-05-14 14:58 UTC (permalink / raw)
  To: Lubomir Rintel
  Cc: Lucas Stach, LKML, DRI mailing list, The etnaviv authors, Russell King

Hi Lubomir


Am Do., 14. Mai 2020 um 00:02 Uhr schrieb Lubomir Rintel <lkundrak@v3.sk>:
>
> On a GC860 (both 3D and 2D capable) GPU, kmscube crashes:
>
>   # strace -f ~lkundrak/src/kmscube/build/kmscube
>   ...
>   ioctl(6, DRM_IOCTL_ETNAVIV_PM_QUERY_DOM, 0xbe92b720) = 0
>   ioctl(6, DRM_IOCTL_ETNAVIV_PM_QUERY_SIG <unfinished ...>) = ?
>   +++ killed by SIGSEGV +++
>   Segmentation fault (core dumped)
>
> And triggers an oops:
>
>   8<--- cut here ---
>   Unable to handle kernel NULL pointer dereference at virtual address 00000000
>   pgd = 40e2c0f7
>   [00000000] *pgd=0df6d831, *pte=00000000, *ppte=00000000
>   Internal error: Oops: 17 [#1] PREEMPT SMP ARM
>   Modules linked in:
>   CPU: 0 PID: 346 Comm: kmscube Not tainted 5.7.0-rc4+ #792
>   Hardware name: Marvell MMP2 (Device Tree Support)
>   PC is at strncpy+0x14/0x30
>   LR is at etnaviv_pm_query_sig+0xd0/0x104
>   pc : [<c04f35f4>]    lr : [<c05dd878>]    psr: 20010013
>   sp : c85f5e00  ip : c85f5eb5  fp : beb58748
>   r10: 0000004c  r9 : ca6f9100  r8 : c85f5e6c
>   r7 : 00000050  r6 : c85f5e6c  r5 : 00000001  r4 : c0b69ae8
>   r3 : c85f5e75  r2 : 0000003f  r1 : 00000000  r0 : c85f5e76
>   Flags: nzCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
>   Control: 10c5387d  Table: 0df70019  DAC: 00000051
>   Process kmscube (pid: 346, stack limit = 0x816fba31)
>   Stack: (0xc85f5e00 to 0xc85f6000)
>   5e00: 00000000 d90e6000 00000020 c05d5b2c c85f5e6c c059ce90 00000000 c1003f88
>   5e20: c04c644b 0000004c c0b69610 c04c644b c85f5e6c 0000004b ca6f9100 c059d0bc
>   5e40: 00000001 c0d53ee8 c85f5f18 00000001 c85f5f50 c85f5e6c 0000004c c8454240
>   5e60: c05d5b2c 00000051 00000000 00000000 00000001 00000000 00000000 00000000
>   5e80: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
>   5ea0: 00000000 00000000 00000000 00000000 00000000 00000000 00000274 c011c3fc
>   5ec0: 00000100 c0290434 5ebc5632 30e03501 5ebc5632 c8526600 00000274 00100cca
>   5ee0: 00000831 b64c5000 cdf72d90 c1003f88 00000000 c04c644b c8454240 beb58748
>   5f00: c8454240 00000006 c85f4000 d90ecad8 001c01a0 c02d49ac b64c52cc 80000007
>   5f20: da9d6dc0 d9aa4000 d9aa4040 00000000 00000274 c011818c 00000005 0e200080
>   5f40: 00000000 000003e5 00000000 00000100 00000000 00000000 00000000 cc78ac40
>   5f60: 00000006 00000007 c1009a98 b64c52cc c85f5fb0 c0118080 00000080 c1003f88
>   5f80: 00000000 00000001 beb58748 c04c644b 00000036 c0100288 c85f4000 00000036
>   5fa0: 001c01a0 c0100060 00000001 beb58748 00000006 c04c644b beb58748 0000004c
>   5fc0: 00000001 beb58748 c04c644b 00000036 beb58748 001bd688 beb58700 001c01a0
>   5fe0: b6f41f08 beb586d4 b6f2784c b6e16cec 80010010 00000006 00000000 00000000
>   [<c04f35f4>] (strncpy) from [<c05dd878>] (etnaviv_pm_query_sig+0xd0/0x104)
>   [<c05dd878>] (etnaviv_pm_query_sig) from [<c059ce90>] (drm_ioctl_kernel+0xb4/0xf8)
>   [<c059ce90>] (drm_ioctl_kernel) from [<c059d0bc>] (drm_ioctl+0x1e8/0x3b8)
>   [<c059d0bc>] (drm_ioctl) from [<c02d49ac>] (ksys_ioctl+0xe0/0xaf0)
>   [<c02d49ac>] (ksys_ioctl) from [<c0100060>] (ret_fast_syscall+0x0/0x54)
>   Exception stack(0xc85f5fa8 to 0xc85f5ff0)
>   5fa0:                   00000001 beb58748 00000006 c04c644b beb58748 0000004c
>   5fc0: 00000001 beb58748 c04c644b 00000036 beb58748 001bd688 beb58700 001c01a0
>   5fe0: b6f41f08 beb586d4 b6f2784c b6e16cec
>   Code: 012fff1e e2422001 e2403001 e080c002 (e5d12000)
>   ---[ end trace 387aad33cd9c15ea ]---
>
> Turns out that it's because pm_domain() returns a pointer outside any
> any of the etnaviv_pm_domains. Unless I'm mistaken, the algorithm in
> pm_domain() is entirely botched when GPU's features match more than one
> domain. This tries to remedy it.
>
> Tested with kmscube with mesa 20 on ome machine with GC860 and another
> with GC2000 + GC300 pair.
>
> Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>

There is already an older/other fix for this problem:
https://patchwork.freedesktop.org/patch/365181/

-- 
greets
--
Christian Gmeiner, MSc

https://christian-gmeiner.info/privacypolicy

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2020-05-14 14:59 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-05-13 22:02 [PATCH] drm/etnaviv: Fix the pm_domain lookup Lubomir Rintel
2020-05-14 14:58 ` Christian Gmeiner

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).