Skip to contentSkip to conversation

ARM: exynos: only do SMC_CMD_CPU1BOOT call on Exynos4

RFC PATCH In review Henrik Grimler 2 participants 5 messages Jan 8, 2022 – Jan 10, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover ARM: exynos: only do SMC_CMD_CPU1BOOT call on Exynos4

I am trying to add support in mainline for a tablet based on Exynos 5420 (Samsung Galaxy Tab S). For this tablet, CCI has to be disabled, and with that the device then hangs during

exynos_smc(SMC_CMD_CPU1BOOT, cpu, 0, 0),

which is run in the cpu_boot secure-firmware function. If the call is skipped, then four out of eight CPUs are brought up, which is the same as on exynos5420-arndale-octa [1] which also has disabled CCI.

Based on Samsung's kernel sources, it seems like the SMC_CMD_CPU1BOOT call is only needed on Exynos4 and Exynos5250 devices [2], on other SoC models the call is skipped. To test if the call seem to do anything, I booted exynos5422-odroid-xu4 with disabled CCI and with or without the SMC_CMD_CPU1BOOT call. The result in both cases were the
same: the first four CPUs were brought up.

To further investigate if the SMC_CMD_CPU1BOOT call is handled we need to dig into the trustzone firmware. The trustzone firmware used in exynos5410-odroid-xu has been to a large part reverse engineered [3], and by comparing with sboot/tzsw in mainline devices it can be seen that similar trustzone firmware is used in several. Out of those, the SMC_CMD_CPU1BOOT call seem to be only handled in Exynos4 devices.

Here's a table summarising the findings for the mainline devices that have a secure-firmware node (Exynos3 devices omitted). An extra 4212 device has been included to get a datapoint for that SoC as well.

.----------------------.--------------.--------------.--------------.

| Device (exynos-)     | Similar tzsw | CPU1BOOT SMC | sboot/tzsw   |
|                      |   to 5410?   |   handled?   |   source     |
.----------------------.--------------.--------------.--------------.
  4212 galaxy tab 3    | yes          | yes          | T310XXSBQB2
  4412-i9300           | yes          | yes          | I9300XXUGPE1
  4412-i9305           | yes          | yes          | I9305XXUFPB1
  4412-itop-elite      | not checked  |              |
  4412-n710x           | not checked  |              |
  4412-odroid{x,x2,u3} | yes          | yes          | wiki.odroid.com/_media/en/boot.tar.gz
  4412-origen          | no           | don't know   | Linaro's origen hwpack 20130130
  4412-p4note-n8010    | not checked  |              |
  4412-trats2          | not checked  |              |
  5410-odroidxu        | yes          | no           | github.com/hsnaves/exynos5410-firmware
  5410-smdk5410        | not checked  |              |
  5420-arndale-octa    | yes          | no           | Linaro's arndale-octa hwpack 20140323
  5422-odroid-xu{3,4}  | yes          | no           | Hardkernel's u-boot 2020.01 branch

"Similar tzsw" above means that the secure monitor calls seem to be handled in the same way as for exynos5410-odroid-xu, in a quite easy-to-spot function (see patch for what it looks like).

For the tablet I am actually interested in, the trustzone firmware looks very different, and I have not been able to locate where/how the secure monitor calls are handled.

Nevertheless, since the call seem to only be handled in trustzone firmware for Exynos4, and since exynos5422-odroid-xu4 behaves the same with or without this secure monitor call, I *think* it should be safe to only run it on the devices where we know that it is handled.

[1] https://krzk.eu/#/builders/32
[2] https://github.com/LineageOS/android_kernel_samsung_exynos5420/blob/cm-14.1/arch/arm/mach-exynos/platsmp.c#L225-L229
[3] https://github.com/hsnaves/exynos5410-firmware

Henrik Grimler (1):
ARM: exynos: only do SMC_CMD_CPU1BOOT call on Exynos4

 arch/arm/mach-exynos/firmware.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

base-commit: 3e4d9032d1a223488456f82d05cfe5519962f344
prerequisite-patch-id: f0f30752eb24b3515eee0a8d7bcf043e2cb084ad

Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#1 lore raw

RFC PATCH 1/1 ARM: exynos: only do SMC_CMD_CPU1BOOT call on Exynos4

On Exynos5 the call is simply ignored by most variants of the trustzone firmware. However, on some devices it instead causes the device to hang, so let's avoid the call for the SoCs where it should not be needed.

To see that the call is ignored, we can look into sboot/tzsw. On most of the Exynos{4,5} devices the part of sboot/tzsw that seem to handle the secure monitor calls is quite easy to recognise, the SMC number is compared to known ones, and if equal it branches to the relevant function. In assembly this looks something like:

;-- handle_smc:

0x00000514      650070e3       cmn r0, 0x65
0x00000518      0a00000a       beq loc.smc_cmd_reg
0x0000051c      010070e3       cmn r0, 1
0x00000520      6c00000a       beq loc.smc_cmd_init
0x00000524      020070e3       cmn r0, 2
0x00000528      6b00000a       beq loc.smc_cmd_info
0x0000052c      030070e3       cmn r0, 3
0x00000530      6e00000a       beq loc.smc_cmd_sleep
0x00000534      060070e3       cmn r0, 6
0x00000538      ae00000a       beq loc.smc_cmd_save_state
0x0000053c      070070e3       cmn r0, 7
0x00000540      b400000a       beq loc.smc_cmd_standby
0x00000544      2b01001a       bne loc.smc_return_minus1

where above example is from exynos5420-arndale-octa. As can be seen the case where r0 is 4 (i.e. SMC_CMD_CPU1BOOT) is not handled. The annotations are taken from github.com/hsnaves/exynos5410-firmware, where a large part of the exynos5410 trustzone firmware has been reverse-engineered.

---
 arch/arm/mach-exynos/firmware.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/arm/mach-exynos/firmware.c +3 -1
index 2eaf2dbb8e81..2da5b60b59e2 100644
@@ -60,8 +60,10 @@ static int exynos_cpu_boot(int cpu)
 	/*
 	 * Exynos3250 doesn't need to send smc command for secondary CPU boot
 	 * because Exynos3250 removes WFE in secure mode.
+	 *
+	 * On Exynos5 devices the call is ignored by trustzone firmware.
 	 */
-	if (soc_is_exynos3250())
+	if (!soc_is_exynos4210() && !soc_is_exynos4412())
 		return 0;
 
 	/*
Signature · 7 lines
2.34.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Tested-by on v1 1/1
#2 lore raw
Quoted 26 lines · Henrik Grimler

On Exynos5 the call is simply ignored by most variants of the trustzone firmware. However, on some devices it instead causes the device to hang, so let's avoid the call for the SoCs where it should not be needed.

To see that the call is ignored, we can look into sboot/tzsw. On most of the Exynos{4,5} devices the part of sboot/tzsw that seem to handle the secure monitor calls is quite easy to recognise, the SMC number is compared to known ones, and if equal it branches to the relevant function. In assembly this looks something like:

;-- handle_smc:

0x00000514      650070e3       cmn r0, 0x65
0x00000518      0a00000a       beq loc.smc_cmd_reg
0x0000051c      010070e3       cmn r0, 1
0x00000520      6c00000a       beq loc.smc_cmd_init
0x00000524      020070e3       cmn r0, 2
0x00000528      6b00000a       beq loc.smc_cmd_info
0x0000052c      030070e3       cmn r0, 3
0x00000530      6e00000a       beq loc.smc_cmd_sleep
0x00000534      060070e3       cmn r0, 6
0x00000538      ae00000a       beq loc.smc_cmd_save_state
0x0000053c      070070e3       cmn r0, 7
0x00000540      b400000a       beq loc.smc_cmd_standby
0x00000544      2b01001a       bne loc.smc_return_minus1
Quoted 7 lines

where above example is from exynos5420-arndale-octa. As can be seen the case where r0 is 4 (i.e. SMC_CMD_CPU1BOOT) is not handled. The annotations are taken from github.com/hsnaves/exynos5410-firmware, where a large part of the exynos5410 trustzone firmware has been reverse-engineered.

Works fine on all ARM 32bit Exynos-based boards I have for tests.

Quoted 10 lines
---
  arch/arm/mach-exynos/firmware.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
arch/arm/mach-exynos/firmware.c +0 -0
index 2eaf2dbb8e81..2da5b60b59e2 100644
@@ -60,8 +60,10 @@ static int exynos_cpu_boot(int cpu)
  	/*

arch/arm/mach-exynos/firmware.c static int exynos_cpu_boot(int cpu) · line 69

Quoted 10 lines
arch/arm/mach-exynos/firmware.c +3 -1
  	 * Exynos3250 doesn't need to send smc command for secondary CPU boot
  	 * because Exynos3250 removes WFE in secure mode.
+	 *
+	 * On Exynos5 devices the call is ignored by trustzone firmware.
  	 */
-	if (soc_is_exynos3250())
+	if (!soc_is_exynos4210() && !soc_is_exynos4412())
  		return 0;
 
  	/*

Best regards

Signature · 2 lines
Marek Szyprowski, PhD
Samsung R&D Institute Poland
#4 lore raw
Quoted 6 lines · Henrik Grimler

Hi Marek,

Quoted 3 lines

Works fine on all ARM 32bit Exynos-based boards I have for tests.

Thanks for testing! Would you mind telling me which Exynos boards you have in your test-setup?

Sure:

Exynos3250: Rinato
Exynos4210: Trats
Exynos4412: Trats2, Odroid X2/U3
Exynos5250: Snow Chromebook, Arndale5250
Exynos5410: Odroid XU
Exynos5420: Pit Chromebook
Exynos5422: Odroud Xu3/Xu3-lite/Xu4/HC1
Exynos5800: Pi Chromebook

Best regards

Signature · 2 lines
Marek Szyprowski, PhD
Samsung R&D Institute Poland