Skip to contentSkip to conversation

goldfish: Prevent the trainwreck from doing harm

PATCH In review Thomas Gleixner 5 participants 9 messages Feb 15, 2017

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

cover goldfish: Prevent the trainwreck from doing harm

The goldfish android emulator platform is a complete trainwreck which causes havoc when enabled in Kconfig.

Aside of unconditionally registering a platform device with hardcoded address ranges, it prevents the serial interrupt from being requested and in case of a spurious interrupt the broken interrupt handler of the pdev_bus driver goes into a hard to debug infinite loop.

Thanks to Gabriel and Borislav for providing patiently debug information.

No thanks to the folks @Intel who choked this $corporate trainwreck down our throat and wasted the time of everyone who got bitten by this.

Thanks,

tglx

#1 lore raw

PATCH 1/2 x86/platform/goldfish: Prevent unconditional loading

The goldfish platform code registers the platform device unconditionally which causes havoc in several ways if the goldfish_pdev_bus driver is enabled:

- Access to the hardcoded physical memory region, which is either not available or contains stuff which is completely unrelated.

- Prevents that the interrupt of the serial port can be requested

- In case of a spurious interrupt it goes into a infinite loop in the interrupt handler of the pdev_bus driver (which needs to be fixed seperately).

Add a 'goldfish' command line option to make the registration opt-in when the platform is compiled in.

I'm seriously grumpy about this engineering trainwreck, which has seven SOBs from Intel developers for 50 lines of code. And none of them figured out that this is broken. Impressive fail!

Fixes ddd70cf93d78 ("goldfish: platform device for x86")
---
 Documentation/admin-guide/kernel-parameters.txt |    4 ++++
 arch/x86/platform/goldfish/goldfish.c           |   14 +++++++++++++-
 2 files changed, 17 insertions(+), 1 deletion(-)
Documentation/admin-guide/kernel-parameters.txt +4 -0
@@ -1192,6 +1192,10 @@
 			When zero, profiling data is discarded and associated
 			debugfs files are removed at module unload time.
 
+	goldfish	[X86] Enable the goldfish android emulator platform.
+			Don't use this when you are not running on the
+			android emulator
+
 	gpt		[EFI] Forces disk with valid GPT signature but
 			invalid Protective MBR to be treated as GPT. If the
 			primary GPT is corrupted, it enables the backup/alternate
arch/x86/platform/goldfish/goldfish.c +13 -1
@@ -42,10 +42,22 @@ static struct resource goldfish_pdev_bus
 	}
 };
 
+static bool goldfish_enable __initdata;
+
+static int __init goldfish_setup(char *str)
+{
+	goldfish_enable = true;
+	return 0;
+}
+__setup("goldfish", goldfish_setup);
+
 static int __init goldfish_init(void)
 {
+	if (!goldfish_enable)
+		return -ENODEV;
+
 	platform_device_register_simple("goldfish_pdev_bus", -1,
-						goldfish_pdev_bus_resources, 2);
+					goldfish_pdev_bus_resources, 2);
 	return 0;
 }
 device_initcall(goldfish_init);
#2 lore raw

Re: [patch 1/2] x86/platform/goldfish: Prevent unconditional loading

Quoted 16 lines · Thomas Gleixner

The goldfish platform code registers the platform device unconditionally which causes havoc in several ways if the goldfish_pdev_bus driver is enabled:

- Access to the hardcoded physical memory region, which is either not available or contains stuff which is completely unrelated.

- Prevents that the interrupt of the serial port can be requested

- In case of a spurious interrupt it goes into a infinite loop in the interrupt handler of the pdev_bus driver (which needs to be fixed seperately).

Add a 'goldfish' command line option to make the registration opt-in when the platform is compiled in.

Quoted 8 lines

I'm seriously grumpy about this engineering trainwreck, which has seven SOBs from Intel developers for 50 lines of code. And none of them figured out that this is broken. Impressive fail!

Fixes ddd70cf93d78 ("goldfish: platform device for x86")

Just thinking, could we not simply delete this entire driver and use x86-DT support to setup this platform?

#3 lore raw

Re: [patch 1/2] x86/platform/goldfish: Prevent unconditional loading

Quoted 21 lines · Peter Zijlstra
Quoted 20 lines · Thomas Gleixner

The goldfish platform code registers the platform device unconditionally which causes havoc in several ways if the goldfish_pdev_bus driver is enabled:

- Access to the hardcoded physical memory region, which is either not available or contains stuff which is completely unrelated.

- Prevents that the interrupt of the serial port can be requested

- In case of a spurious interrupt it goes into a infinite loop in the interrupt handler of the pdev_bus driver (which needs to be fixed seperately).

Add a 'goldfish' command line option to make the registration opt-in when the platform is compiled in.

I'm seriously grumpy about this engineering trainwreck, which has seven SOBs from Intel developers for 50 lines of code. And none of them figured out that this is broken. Impressive fail!

Quoted 7 lines
Quoted 4 lines
Fixes ddd70cf93d78 ("goldfish: platform device for x86")

Just thinking, could we not simply delete this entire driver and use x86-DT support to setup this platform?

That would be the proper solution and that's what ARM probably does already.

Thanks,

tglx

#4 lore raw

Re: [patch 1/2] x86/platform/goldfish: Prevent unconditional loading

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

I'm seriously grumpy about this engineering trainwreck, which has seven
SOBs from Intel developers for 50 lines of code. And none of them figured
out that this is broken. Impressive fail!

It was discussed at the time, documented at the time. Unfortunately the people who did the emulator didn't feel the urge to provide a way to detect the platform was Goldfish.

Historically it also used its own custom device discovery scheme. Given the limited use of older versions of Goldfish it might well make sense to remove support for the older emulator versions.

Alan

#5 lore raw

Re: [patch 1/2] x86/platform/goldfish: Prevent unconditional loading

Quoted 7 lines · Alan Cox
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

I'm seriously grumpy about this engineering trainwreck, which has seven
SOBs from Intel developers for 50 lines of code. And none of them figured
out that this is broken. Impressive fail!

It was discussed at the time, documented at the time.

I just have a hard time to find that documentation. It's definitely not in the kernel source, unless you qualify the help text of CONFIG_GOLDFISH as such:

Enable support for the Goldfish virtual platform used primarily for Android development. Unless you are building for the Android Goldfish emulator say N here.

which does not help for randconfig and other builds and does not prevent users from enabling it accidentaly. That all wouldn't be as bad if at least the minimal provisioning of damage prevention would have been done.

Quoted 2 lines

Unfortunately the people who did the emulator didn't feel the urge to provide a way to detect the platform was Goldfish.

Sure, and the people shoving it into the kernel didn't feel the urge to enforce that.

Quoted 3 lines

Historically it also used its own custom device discovery scheme. Given the limited use of older versions of Goldfish it might well make sense to remove support for the older emulator versions.

I'm all for it.

Thanks,

tglx

#6 lore raw

PATCH 2/2 goldfish: Sanitize the broken interrupt handler

In reply to Thomas Gleixner

This interrupt handler is broken in several ways:

- It loops forever when the op code is not decodeable

- It never returns IRQ_HANDLED because the only way to exit the loop returns IRQ_NONE unconditionally.

The whole concept of this is broken. Creating devices in an interrupt handler is beyond any point of sanity.

Make it at least behave halfways sane so accidental users do not have to deal with a hard to debug lockup.

Fixes e809c22b8fb028 ("goldfish: add the goldfish virtual bus")
---
 drivers/platform/goldfish/pdev_bus.c |   13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)
drivers/platform/goldfish/pdev_bus.c +8 -5
@@ -157,23 +157,26 @@ static int goldfish_new_pdev(void)
 static irqreturn_t goldfish_pdev_bus_interrupt(int irq, void *dev_id)
 {
 	irqreturn_t ret = IRQ_NONE;
+
 	while (1) {
 		u32 op = readl(pdev_bus_base + PDEV_BUS_OP);
-		switch (op) {
-		case PDEV_BUS_OP_DONE:
-			return IRQ_NONE;
 
+		switch (op) {
 		case PDEV_BUS_OP_REMOVE_DEV:
 			goldfish_pdev_remove();
+			ret = IRQ_HANDLED;
 			break;
 
 		case PDEV_BUS_OP_ADD_DEV:
 			goldfish_new_pdev();
+			ret = IRQ_HANDLED;
 			break;
+
+		case PDEV_BUS_OP_DONE:
+		default:
+			return ret;
 		}
-		ret = IRQ_HANDLED;
 	}
-	return ret;
 }
 
 static int goldfish_pdev_bus_probe(struct platform_device *pdev)
#8 lore raw
Quoted 2 lines · Linus Torvalds

Ack. I'm assuming Greg will pick this up, in the meantime I hope anybody who has problems just disables the goldfish support.

I've queued them up now for the next merge window, and they are tagged for stable so will show up in 4.10.1.

thanks,

greg k-h