Hi Jarkko, Dirk,

Jarkko,
Thank you for your points - I will try to fix all you have mentioned.
I think it would be good to clarify a few things, before going with
the next version. Regarding use tpm_chip_start/stop() around
tpm2_probe() call - I have followed the similar way it is done in the
probe_itpm() function, where is also a call to tpm_tis_send_data()
guarded by request/release_locality(). I have tested it on the Samsung
Chromebook Pro (which reports TPM 1.2 / Cr50) and it was sufficient
(e.g. I didn't have to enable the clock) to get rid of a trace
mentioned in the commit message....but now writing these words I'm
starting to think that using tpm_chip_start/stop() could be safer from
a point of view of other TPMs... so if you suggest using
tpm_chip_start/stop() shall I also add it to the probe_itpm() (instead
of  request/release_locality()) ?

Dirk,
Thanks for the clarification. Regarding the issue you observe - I
wanted to address at first the one that is generating the trace
(please check the commit message) because it was leading to returning
an error in the tpm_tis_status() function causing TPM module not
initialized at all in the end - requesting locality before the call to
the tpm_tis_send_data() has helped in my case (my test environment). I
am aware of the second issue - "TPM interrupt not working, polling
instead", but as it is not as critical as the first one, I decided to
work on it later.

Thank you once again for all your input and sorry for a confusion with
sending patches.

Best regards,
Lukasz


śr., 3 lut 2021 o 12:56 Dirk Gouders [off-list ref] napisał(a):

Dirk Gouders [off-list ref] writes:

quoted

Lukasz Majczak [off-list ref] writes:

quoted

There are missing calls to tpm_request_locality before the calls to
the tpm_get_timeouts() and tpm_tis_probe_irq_single() - both functions
internally send commands to the tpm. As the current
approach might work for tpm2, it fails for tpm1.x - in that case
call to tpm_get_timeouts() or tpm_tis_probe_irq_single()
without acquired locality fails and in turn causes tpm_tis_core_init()
to fail, it can be observed in the log with the following warning
trace:

[    4.324298] TPM returned invalid status
[    4.324806] WARNING: CPU: 2 PID: 1 at drivers/char/tpm/tpm_tis_core.c:275 tpm_tis_status+0x86/0x8f
[    4.325888] Modules linked in:
[    4.326287] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G        W         5.11.0-rc6-next-20210201-00003-g214461adb2e8 #43
[    4.327406] Hardware name: Google Caroline/Caroline, BIOS Google_Caroline.7820.430.0 07/20/2018
[    4.327918] RIP: 0010:tpm_tis_status+0x86/0x8f
[    4.328323] Code: 28 00 00 00 48 3b 45 f0 75 24 89 d8 48 83 c4 10 5b 5d c3 c6 05 58 d9 28 01 01 31 db 48 c7 c7 73 52 98 9c 31 c0 e8 c2 17 b0 ff <0f> 0b eb cd e8 cf 4f 55 00 0f 1f 44 00 00 55 48 89 e56
[    4.330592] RSP: 0000:ffff88810092f7a0 EFLAGS: 00010246
[    4.331223] RAX: 691ee151166db100 RBX: 0000000000000000 RCX: 0000000000000001
[    4.331860] RDX: 0000000000000006 RSI: ffffffff9c96d302 RDI: 00000000ffffffff
[    4.332272] RBP: ffff88810092f7b8 R08: dffffc0000000000 R09: fffffbfff39c96ce
[    4.332683] R10: fffffbfff39c96ce R11: 0000000000000001 R12: ffff8881053e2000
[    4.333109] R13: 0000000065000000 R14: ffff888105d71000 R15: ffff888105cd2628
[    4.333738] FS:  0000000000000000(0000) GS:ffff88842f200000(0000) knlGS:0000000000000000
[    4.334432] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    4.334783] CR2: 0000000000000000 CR3: 0000000037828001 CR4: 00000000003706e0
[    4.335196] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    4.335886] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[    4.336793] Call Trace:
[    4.337107]  tpm_tis_send_data+0x3d/0x22f
[    4.337506]  tpm_tis_send_main+0x30/0xf5
[    4.337746]  tpm_transmit+0xbf/0x327
[    4.338042]  ? __alloc_pages_nodemask+0x261/0x36d
[    4.338615]  tpm_transmit_cmd+0x2c/0x93
[    4.339109]  tpm1_getcap+0x232/0x285
[    4.339578]  tpm1_get_timeouts+0x48/0x47d
[    4.339964]  ? lockdep_init_map_type+0x71/0x257
[    4.340256]  ? lockdep_init_map_type+0x71/0x257
[    4.340719]  ? __raw_spin_lock_init+0x40/0x69
[    4.341208]  tpm_tis_core_init+0x402/0x5ee
[    4.341629]  tpm_tis_init+0x11d/0x1a2
[    4.341867]  tpm_tis_pnp_init+0x91/0xb5
[    4.342101]  ? tis_int_handler+0x15f/0x15f
[    4.342466]  pnp_device_probe+0x79/0x9f
[    4.342941]  really_probe+0x149/0x4a8
[    4.343412]  driver_probe_device+0xd6/0x144
[    4.343968]  device_driver_attach+0x42/0x5b
[    4.344382]  __driver_attach+0xca/0x139
[    4.344617]  ? driver_attach+0x1f/0x1f
[    4.344860]  bus_for_each_dev+0x85/0xb7
[    4.345096]  bus_add_driver+0x12b/0x228
[    4.345330]  driver_register+0x64/0xed
[    4.345560]  init_tis+0xa5/0xeb
[    4.345784]  ? lock_is_held_type+0x100/0x141
[    4.346044]  ? tpm_init+0x106/0x106
[    4.346259]  ? rcu_read_lock_sched_held+0x41/0x7e
[    4.346542]  ? tpm_init+0x106/0x106
[    4.346678] battery: ACPI: Battery Slot [BAT0] (battery present)
[    4.346754]  do_one_initcall+0x1b9/0x43d
[    4.346776]  ? asm_sysvec_apic_timer_interrupt+0x12/0x20
[    4.347659]  ? lockdep_hardirqs_on+0x8e/0x12e
[    4.347937]  ? lock_is_held_type+0x100/0x141
[    4.348196]  ? rcu_read_lock_sched_held+0x41/0x7e
[    4.348477]  do_initcall_level+0x99/0xa9
[    4.348717]  ? kernel_init+0xe/0x10a
[    4.348954]  do_initcalls+0x4e/0x79
[    4.349170]  kernel_init_freeable+0x15a/0x1ae
[    4.349434]  ? rest_init+0x1d6/0x1d6
[    4.349655]  kernel_init+0xe/0x10a
[    4.349882]  ret_from_fork+0x22/0x30
[    4.350103] irq event stamp: 700039
[    4.350318] hardirqs last  enabled at (700047): [<ffffffff9b735265>] console_unlock+0x4be/0x538
[    4.350836] hardirqs last disabled at (700056): [<ffffffff9b734e84>] console_unlock+0xdd/0x538
[    4.351331] softirqs last  enabled at (699522): [<ffffffff9c4004ec>] __do_softirq+0x4ec/0x539
[    4.351835] softirqs last disabled at (699517): [<ffffffff9c200f62>] asm_call_irq_on_stack+0x12/0x20

Following the trace one can also notice a comment in the tpm_tis_status():

             /*
              * If this trips, the chances are the read is
              * returning 0xff because the locality hasn't been
              * acquired.  Usually because tpm_try_get_ops() hasn't
              * been called before doing a TPM operation.
              */
In this case we don't have to call tpm_try_get_ops()
as both calls (tpm_get_timeouts() and tpm_tis_probe_irq_single()) are
in the tpm_tis_core_init function and don't require any locking or clock
enablement. Similar usage is in the probe_itpm() function also called
inside tpm_tis_core_init().
Tested on Samsung Chromebook Pro (Caroline).

Signed-off-by: Lukasz Majczak <redacted>
---
Hi Jarkko

I have checked the linux-next with James patches, also followed Dirk
suggestion applying remaining ones, although without any luck -
a warning trace was still present. As Guneter mentioned earlier, this
patch[1] doesn't address a lack of acquired locality in the
tpm_get_timeouts() and does it only for tpm_tis_probe_irq_single() but
also without a call to tpm_relinquish_locality().

Here are my logs from the clean linux-next master branch [2]
(with two James' patches present) and with my
patch applied[3]

Best regards,
Lukasz

[1] https://lore.kernel.org/linux-integrity/20201001180925.13808-5-James.Bottomley@HansenPartnership.com/ (local)
[2] https://gist.github.com/semihalf-majczak-lukasz/f588c0684a6cc7d983bb9c4eb4bda586
[3] https://gist.github.com/semihalf-majczak-lukasz/88ede933bc7d28d806e3532850a04054

v2 -> v3:
 - Added braces around if part of if/else statements
 - Rebased to linux-next
 - Updated commit message

 drivers/char/tpm/tpm-chip.c      |  4 ++--
 drivers/char/tpm/tpm-interface.c | 13 ++++++++++---
 drivers/char/tpm/tpm.h           |  2 ++
 drivers/char/tpm/tpm_tis_core.c  | 14 +++++++++++---
 4 files changed, 25 insertions(+), 8 deletions(-)

diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
index ddaeceb7e109..5351963a4b19 100644
--- a/drivers/char/tpm/tpm-chip.c
+++ b/drivers/char/tpm/tpm-chip.c

@@ -32,7 +32,7 @@ struct class *tpm_class;
 struct class *tpmrm_class;
 dev_t tpm_devt;

-static int tpm_request_locality(struct tpm_chip *chip)
+int tpm_request_locality(struct tpm_chip *chip)
 {
     int rc;

@@ -47,7 +47,7 @@ static int tpm_request_locality(struct tpm_chip *chip)
     return 0;
 }

-static void tpm_relinquish_locality(struct tpm_chip *chip)
+void tpm_relinquish_locality(struct tpm_chip *chip)
 {
     int rc;

Here, it seems

+EXPORT_SYMBOL_GPL(tpm_request_locality);

and

+EXPORT_SYMBOL_GPL(tpm_relinquish_locality);

are needed.  Otherwise building tpm* modules fails:

ERROR: modpost: "tpm_request_locality" [drivers/char/tpm/tpm_tis_core.ko] undefined!
ERROR: modpost: "tpm_relinquish_locality" [drivers/char/tpm/tpm_tis_core.ko] undefined!
make[1]: *** [scripts/Makefile.modpost:132: Module.symvers] Error 1
make[1]: *** Deleting file 'Module.symvers'
make: *** [Makefile:1405: modules] Error 2

Otherwise, testing this patch results in no more warning

TPM returned invalid status: 0xff

and also no more warnings:

tpm tpm0: tpm_try_transmit: send(): error -5
tpm tpm0: [Firmware Bug]: TPM interrupt not working, polling instead

Dirk

quoted

diff --git a/drivers/char/tpm/tpm-interface.c b/drivers/char/tpm/tpm-interface.c
index 1621ce818705..2a9001d329f2 100644
--- a/drivers/char/tpm/tpm-interface.c
+++ b/drivers/char/tpm/tpm-interface.c

@@ -241,10 +241,17 @@ int tpm_get_timeouts(struct tpm_chip *chip)
     if (chip->flags & TPM_CHIP_FLAG_HAVE_TIMEOUTS)
             return 0;

-    if (chip->flags & TPM_CHIP_FLAG_TPM2)
+    if (chip->flags & TPM_CHIP_FLAG_TPM2) {
             return tpm2_get_timeouts(chip);
-    else
-            return tpm1_get_timeouts(chip);
+    } else {
+            ssize_t ret = tpm_request_locality(chip);
+
+            if (ret)
+                    return ret;
+            ret = tpm1_get_timeouts(chip);
+            tpm_relinquish_locality(chip);
+            return ret;
+    }
 }
 EXPORT_SYMBOL_GPL(tpm_get_timeouts);

diff --git a/drivers/char/tpm/tpm.h b/drivers/char/tpm/tpm.h
index 947d1db0a5cc..8c13008437dd 100644
--- a/drivers/char/tpm/tpm.h
+++ b/drivers/char/tpm/tpm.h

@@ -193,6 +193,8 @@ static inline void tpm_msleep(unsigned int delay_msec)

 int tpm_chip_start(struct tpm_chip *chip);
 void tpm_chip_stop(struct tpm_chip *chip);
+int tpm_request_locality(struct tpm_chip *chip);
+void tpm_relinquish_locality(struct tpm_chip *chip);
 struct tpm_chip *tpm_find_get_ops(struct tpm_chip *chip);
 __must_check int tpm_try_get_ops(struct tpm_chip *chip);
 void tpm_put_ops(struct tpm_chip *chip);

diff --git a/drivers/char/tpm/tpm_tis_core.c b/drivers/char/tpm/tpm_tis_core.c
index 431919d5f48a..d4f381d6356e 100644
--- a/drivers/char/tpm/tpm_tis_core.c
+++ b/drivers/char/tpm/tpm_tis_core.c

@@ -708,11 +708,19 @@ static int tpm_tis_gen_interrupt(struct tpm_chip *chip)
     u32 cap2;
     cap_t cap;

-    if (chip->flags & TPM_CHIP_FLAG_TPM2)
+    if (chip->flags & TPM_CHIP_FLAG_TPM2) {
             return tpm2_get_tpm_pt(chip, 0x100, &cap2, desc);
-    else
-            return tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc,
+    } else {
+            ssize_t ret = tpm_request_locality(chip);
+
+            if (ret)
+                    return ret;
+            ret = tpm1_getcap(chip, TPM_CAP_PROP_TIS_TIMEOUT, &cap, desc,
                               0);
+            tpm_relinquish_locality(chip);
+            return ret;
+    }
+
 }

 /* Register the IRQ and issue a command that will cause an interrupt. If an

My apologies for just more noise from here.

But I think it could be important that I withdraw my above statement
concerning positive test results on my hardware.

I was now trying to understand Lukasz' fix and started wondering how
changes in the case of !(chip->flags & TPM_CHIP_FLAG_TPM2) could affect
my environment: tpm_tis STM0125:00: 2.0 TPM (device-id 0x0, rev-id 78).

So, I became very nervous and re-did several tests and it
(understandably) turned out that Lukasz' patch does not affect my machine
at all -- nearly: the only effect I noticed is that tpm_tis doesn't get
loaded automatically with his patch applied.  I have to load it manually
but then get the familiar log messages.

But the tests I based my wrong statement on were done with static
tpm_tis, because of symbols not having been exported (V3).
I now noticed that tpm_tis behaves different depending on if it is built
static or as a module (latest tests done with
5.11.0-rc6-next-20210202-x86_64+).

In the static case, all I see in the logs is:

[    2.673818] tpm_tis STM0125:00: 2.0 TPM (device-id 0x0, rev-id 78)

Perhaps there are better ways to access and test TPM but I tested it
using getrandom: no further messages in the kernel log were generated.

If tpm_tis it is built as a module the behavior is the one with warnings
and falling back to polling.

Dirk