[PATCH 0/2] USB: EHCI: Add register array bounds to HCS ports

[PATCH 0/2] USB: EHCI: Add register array bounds to HCS ports

[Kees Cook]

Hi,

This is cleaning up some of the remaining things to be able to apply
-Warray-bounds and -Wzero-length-bounds globally. Only after doing my
own version of the port_status patch did I find Arnd's earlier
patches, including for the weird Broadcom stuff[1].

Anyway, here's what I got. :) No binary differences.

-Kees

[1] https://lore.kernel.org/lkml/20200527134320.869042-1-arnd@arndb.de/#t

Kees Cook (2):
  USB: EHCI: Add register array bounds to HCS ports
  USB: EHCI: Add alias for Broadcom INSNREG

 drivers/usb/host/ehci-brcm.c | 11 +++++------
 include/linux/usb/ehci_def.h | 38 ++++++++++++++++++++++++------------
 2 files changed, 31 insertions(+), 18 deletions(-)

-- 
2.30.2

[PATCH 1/2] USB: EHCI: Add register array bounds to HCS ports

[Kees Cook]

The original EHCI register struct used a trailing 0-element array for
addressing the N_PORTS-many available registers. However, after
commit a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
the 0-element array started to overlap the USBMODE extension register.

To avoid future compile-time warnings about accessing indexes within a
0-element array, rearrange the struct to actually describe the expected
layout (max 15 registers) with a union. All offsets remain the same, and
bounds checking becomes possible on accesses to port_status and hostpc.

There are no binary differences, and struct offsets continue to match.
"pahole --hex -C ehci_regs" before:

struct ehci_regs {
	u32                        command;              /*     0   0x4 */
	u32                        status;               /*   0x4   0x4 */
	u32                        intr_enable;          /*   0x8   0x4 */
	u32                        frame_index;          /*   0xc   0x4 */
	u32                        segment;              /*  0x10   0x4 */
	u32                        frame_list;           /*  0x14   0x4 */
	u32                        async_next;           /*  0x18   0x4 */
	u32                        reserved1[2];         /*  0x1c   0x8 */
	u32                        txfill_tuning;        /*  0x24   0x4 */
	u32                        reserved2[6];         /*  0x28  0x18 */
	/* --- cacheline 1 boundary (64 bytes) --- */
	u32                        configured_flag;      /*  0x40   0x4 */
	u32                        port_status[0];       /*  0x44     0 */
	u32                        reserved3[9];         /*  0x44  0x24 */
	u32                        usbmode;              /*  0x68   0x4 */
	u32                        reserved4[6];         /*  0x6c  0x18 */
	/* --- cacheline 2 boundary (128 bytes) was 4 bytes ago --- */
	u32                        hostpc[0];            /*  0x84     0 */
	u32                        reserved5[17];        /*  0x84  0x44 */
	/* --- cacheline 3 boundary (192 bytes) was 8 bytes ago --- */
	u32                        usbmode_ex;           /*  0xc8   0x4 */

	/* size: 204, cachelines: 4, members: 18 */
	/* last cacheline: 12 bytes */
};

after:

struct ehci_regs {
	u32                        command;              /*     0   0x4 */
	u32                        status;               /*   0x4   0x4 */
	u32                        intr_enable;          /*   0x8   0x4 */
	u32                        frame_index;          /*   0xc   0x4 */
	u32                        segment;              /*  0x10   0x4 */
	u32                        frame_list;           /*  0x14   0x4 */
	u32                        async_next;           /*  0x18   0x4 */
	u32                        reserved1[2];         /*  0x1c   0x8 */
	u32                        txfill_tuning;        /*  0x24   0x4 */
	u32                        reserved2[6];         /*  0x28  0x18 */
	/* --- cacheline 1 boundary (64 bytes) --- */
	u32                        configured_flag;      /*  0x40   0x4 */
	union {
		u32                port_status[15];      /*  0x44  0x3c */
		struct {
			u32        reserved3[9];         /*  0x44  0x24 */
			u32        usbmode;              /*  0x68   0x4 */
		};                                       /*  0x44  0x28 */
	};                                               /*  0x44  0x3c */
	/* --- cacheline 2 boundary (128 bytes) --- */
	u32                        reserved4;            /*  0x80   0x4 */
	u32                        hostpc[15];           /*  0x84  0x3c */
	/* --- cacheline 3 boundary (192 bytes) --- */
	u32                        reserved5[2];         /*  0xc0   0x8 */
	u32                        usbmode_ex;           /*  0xc8   0x4 */

	/* size: 204, cachelines: 4, members: 16 */
	/* last cacheline: 12 bytes */
};

With this fixed, adding -Wzero-length-bounds to the build no longer
produces several warnings like this:

In file included from drivers/usb/host/ehci-hcd.c:306:
drivers/usb/host/ehci-hub.c: In function 'ehci_port_handed_over':
drivers/usb/host/ehci-hub.c:1194:8: warning: array subscript '<unknown>' is outside the bounds of an interior zero-length array 'u32[0]' {aka 'unsigned int[]'} [-Wzero-length-bounds]
 1194 |  reg = &ehci->regs->port_status[portnum - 1];
      |        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from drivers/usb/host/ehci.h:274,
                 from drivers/usb/host/ehci-hcd.c:97:
./include/linux/usb/ehci_def.h:130:7: note: while referencing 'port_status'
  130 |  u32  port_status[0]; /* up to N_PORTS */
      |       ^~~~~~~~~~~

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Al Cooper <alcooperx@gmail.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: linux-usb@vger.kernel.org
Fixes: a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 include/linux/usb/ehci_def.h | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
index 78e006355557..5398f571113b 100644
--- a/include/linux/usb/ehci_def.h
+++ b/include/linux/usb/ehci_def.h
@@ -45,6 +45,7 @@ struct ehci_caps {
 #define HCS_PORTROUTED(p)	((p)&(1 << 7))	/* true: port routing */
 #define HCS_PPC(p)		((p)&(1 << 4))	/* true: port power control */
 #define HCS_N_PORTS(p)		(((p)>>0)&0xf)	/* bits 3:0, ports on HC */
+#define HCS_N_PORTS_MAX		0xf		/* N_PORTS valid 0x1-0xF */
 
 	u32		hcc_params;      /* HCCPARAMS - offset 0x8 */
 /* EHCI 1.1 addendum */
@@ -126,8 +127,9 @@ struct ehci_regs {
 	u32		configured_flag;
 #define FLAG_CF		(1<<0)		/* true: we'll support "high speed" */
 
-	/* PORTSC: offset 0x44 */
-	u32		port_status[0];	/* up to N_PORTS */
+	union {
+		/* PORTSC: offset 0x44 */
+		u32	port_status[HCS_N_PORTS_MAX];
 /* EHCI 1.1 addendum */
 #define PORTSC_SUSPEND_STS_ACK 0
 #define PORTSC_SUSPEND_STS_NYET 1
@@ -164,28 +166,28 @@ struct ehci_regs {
 #define PORT_CSC	(1<<1)		/* connect status change */
 #define PORT_CONNECT	(1<<0)		/* device connected */
 #define PORT_RWC_BITS   (PORT_CSC | PORT_PEC | PORT_OCC)
-
-	u32		reserved3[9];
-
-	/* USBMODE: offset 0x68 */
-	u32		usbmode;	/* USB Device mode */
+		struct {
+			u32	reserved3[9];
+			/* USBMODE: offset 0x68 */
+			u32	usbmode;	/* USB Device mode */
+		};
 #define USBMODE_SDIS	(1<<3)		/* Stream disable */
 #define USBMODE_BE	(1<<2)		/* BE/LE endianness select */
 #define USBMODE_CM_HC	(3<<0)		/* host controller mode */
 #define USBMODE_CM_IDLE	(0<<0)		/* idle state */
-
-	u32		reserved4[6];
+	};
+	u32		reserved4;
 
 /* Moorestown has some non-standard registers, partially due to the fact that
  * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
  * PORTSCx
  */
 	/* HOSTPC: offset 0x84 */
-	u32		hostpc[0];	/* HOSTPC extension */
+	u32		hostpc[HCS_N_PORTS_MAX];
 #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
 #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
 
-	u32		reserved5[17];
+	u32		reserved5[2];
 
 	/* USBMODE_EX: offset 0xc8 */
 	u32		usbmode_ex;	/* USB Device mode extension */
-- 
2.30.2

[PATCH 2/2] USB: EHCI: Add alias for Broadcom INSNREG

[Kees Cook]

Refactor struct ehci_regs to avoid accessing beyond the end of
port_status. This change results in no difference in the resulting
object code.

Avoids several warnings when building with -Warray-bounds:

drivers/usb/host/ehci-brcm.c: In function 'ehci_brcm_reset':
drivers/usb/host/ehci-brcm.c:113:32: warning: array subscript 16 is above array bounds of 'u32[15]' {aka 'unsigned int[15]'} [-Warray-bounds]
  113 |  ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
      |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from drivers/usb/host/ehci.h:274,
                 from drivers/usb/host/ehci-brcm.c:15:
./include/linux/usb/ehci_def.h:132:7: note: while referencing 'port_status'
  132 |   u32 port_status[HCS_N_PORTS_MAX];
      |       ^~~~~~~~~~~

Note that the documentation around this proprietary register is
confusing. If "USB_EHCI_INSNREG00" is at port_status[0x0f], its offset
would be 0x80 (not 0x90). The code uses port_status[0x10], so is that
not using "USB_EHCI_INSNREG00"?

Perhaps port_status[0x10] is USB_EHCI_INSNREG01 and port_status[0x12]
is USB_EHCI_INSNREG03? If so, the union could be adjusted to better
represent the layout.

Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Al Cooper <alcooperx@gmail.com>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: linux-usb@vger.kernel.org
Cc: bcm-kernel-feedback-list@broadcom.com
Suggested-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 9df231511bd6 ("usb: ehci: Add new EHCI driver for Broadcom STB SoC's")
Signed-off-by: Kees Cook <keescook@chromium.org>
---
 drivers/usb/host/ehci-brcm.c | 11 +++++------
 include/linux/usb/ehci_def.h | 16 ++++++++++++++--
 2 files changed, 19 insertions(+), 8 deletions(-)

diff --git a/drivers/usb/host/ehci-brcm.c b/drivers/usb/host/ehci-brcm.c
index 3e0ebe8cc649..5d232d3701f9 100644
--- a/drivers/usb/host/ehci-brcm.c
+++ b/drivers/usb/host/ehci-brcm.c
@@ -110,8 +110,8 @@ static int ehci_brcm_reset(struct usb_hcd *hcd)
 	 *   bus usage
 	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90
 	 */
-	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
-	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
+	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
+	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
 
 	return ehci_setup(hcd);
 }
@@ -223,11 +223,10 @@ static int __maybe_unused ehci_brcm_resume(struct device *dev)
 	/*
 	 * SWLINUX-1705: Avoid OUT packet underflows during high memory
 	 *   bus usage
-	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00
-	 * @ 0x90
+	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90
 	 */
-	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
-	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
+	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
+	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
 
 	ehci_resume(hcd, false);
 
diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
index 5398f571113b..86f0909cab99 100644
--- a/include/linux/usb/ehci_def.h
+++ b/include/linux/usb/ehci_def.h
@@ -182,11 +182,23 @@ struct ehci_regs {
  * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
  * PORTSCx
  */
-	/* HOSTPC: offset 0x84 */
-	u32		hostpc[HCS_N_PORTS_MAX];
+	union {
+		/* HOSTPC: offset 0x84 */
+		u32	hostpc[HCS_N_PORTS_MAX];
 #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
 #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
 
+		/*
+		 * This was originally documented as:
+		 * "port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90"
+		 * but this doesn't make sense: the code was using
+		 * port_status[0x10]. port_status[0x0f] would be reserved4.
+		 * Also, none of these are near 0x90. port_status[0x10] is
+		 * offset 0x84, and port_status[0x0f] would be 0x80.
+		 */
+		u32	brcm_insnreg[3];
+	};
+
 	u32		reserved5[2];
 
 	/* USBMODE_EX: offset 0xc8 */
-- 
2.30.2

Re: [PATCH 1/2] USB: EHCI: Add register array bounds to HCS ports

[Andy Shevchenko]

On Wed, Aug 18, 2021 at 7:30 AM Kees Cook <keescook@chromium.org> wrote:
>
> The original EHCI register struct used a trailing 0-element array for
> addressing the N_PORTS-many available registers. However, after
> commit a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
> the 0-element array started to overlap the USBMODE extension register.
>
> To avoid future compile-time warnings about accessing indexes within a
> 0-element array, rearrange the struct to actually describe the expected
> layout (max 15 registers) with a union. All offsets remain the same, and
> bounds checking becomes possible on accesses to port_status and hostpc.

...

>         /* HOSTPC: offset 0x84 */
> -       u32             hostpc[0];      /* HOSTPC extension */
> +       u32             hostpc[HCS_N_PORTS_MAX];
>  #define HOSTPC_PHCD    (1<<22)         /* Phy clock disable */
>  #define HOSTPC_PSPD    (3<<25)         /* Port speed detection */
>
> -       u32             reserved5[17];
> +       u32             reserved5[2];

Shouldn't it be rather [17 - PORT_MAX]? for accuracy?
Or also a union approach?

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH 1/2] USB: EHCI: Add register array bounds to HCS ports

[Alan Stern]

On Tue, Aug 17, 2021 at 09:30:33PM -0700, Kees Cook wrote:
> The original EHCI register struct used a trailing 0-element array for
> addressing the N_PORTS-many available registers. However, after
> commit a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
> the 0-element array started to overlap the USBMODE extension register.
> 
> To avoid future compile-time warnings about accessing indexes within a
> 0-element array, rearrange the struct to actually describe the expected
> layout (max 15 registers) with a union. All offsets remain the same, and
> bounds checking becomes possible on accesses to port_status and hostpc.
> 
> There are no binary differences, and struct offsets continue to match.

Two comments...

> ---
>  include/linux/usb/ehci_def.h | 24 +++++++++++++-----------
>  1 file changed, 13 insertions(+), 11 deletions(-)
> 
> diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
> index 78e006355557..5398f571113b 100644
> --- a/include/linux/usb/ehci_def.h
> +++ b/include/linux/usb/ehci_def.h
> @@ -45,6 +45,7 @@ struct ehci_caps {
>  #define HCS_PORTROUTED(p)	((p)&(1 << 7))	/* true: port routing */
>  #define HCS_PPC(p)		((p)&(1 << 4))	/* true: port power control */
>  #define HCS_N_PORTS(p)		(((p)>>0)&0xf)	/* bits 3:0, ports on HC */
> +#define HCS_N_PORTS_MAX		0xf		/* N_PORTS valid 0x1-0xF */

I would prefer to see this value in decimal.  It seems very odd to say 
something like "The maximum number of ports is 0xf".

>  
>  	u32		hcc_params;      /* HCCPARAMS - offset 0x8 */
>  /* EHCI 1.1 addendum */
> @@ -126,8 +127,9 @@ struct ehci_regs {
>  	u32		configured_flag;
>  #define FLAG_CF		(1<<0)		/* true: we'll support "high speed" */
>  
> -	/* PORTSC: offset 0x44 */
> -	u32		port_status[0];	/* up to N_PORTS */
> +	union {
> +		/* PORTSC: offset 0x44 */
> +		u32	port_status[HCS_N_PORTS_MAX];

Please don't lose the second comment.

>  /* EHCI 1.1 addendum */
>  #define PORTSC_SUSPEND_STS_ACK 0
>  #define PORTSC_SUSPEND_STS_NYET 1
> @@ -164,28 +166,28 @@ struct ehci_regs {
>  #define PORT_CSC	(1<<1)		/* connect status change */
>  #define PORT_CONNECT	(1<<0)		/* device connected */
>  #define PORT_RWC_BITS   (PORT_CSC | PORT_PEC | PORT_OCC)
> -
> -	u32		reserved3[9];
> -
> -	/* USBMODE: offset 0x68 */
> -	u32		usbmode;	/* USB Device mode */
> +		struct {
> +			u32	reserved3[9];
> +			/* USBMODE: offset 0x68 */
> +			u32	usbmode;	/* USB Device mode */
> +		};
>  #define USBMODE_SDIS	(1<<3)		/* Stream disable */
>  #define USBMODE_BE	(1<<2)		/* BE/LE endianness select */
>  #define USBMODE_CM_HC	(3<<0)		/* host controller mode */
>  #define USBMODE_CM_IDLE	(0<<0)		/* idle state */
> -
> -	u32		reserved4[6];
> +	};
> +	u32		reserved4;
>  
>  /* Moorestown has some non-standard registers, partially due to the fact that
>   * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
>   * PORTSCx
>   */
>  	/* HOSTPC: offset 0x84 */
> -	u32		hostpc[0];	/* HOSTPC extension */
> +	u32		hostpc[HCS_N_PORTS_MAX];
>  #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
>  #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
>  
> -	u32		reserved5[17];
> +	u32		reserved5[2];
>  
>  	/* USBMODE_EX: offset 0xc8 */
>  	u32		usbmode_ex;	/* USB Device mode extension */

Otherwise okay.

Alan Stern

Re: [PATCH 2/2] USB: EHCI: Add alias for Broadcom INSNREG

[Alan Stern]

On Tue, Aug 17, 2021 at 09:30:34PM -0700, Kees Cook wrote:
> Refactor struct ehci_regs to avoid accessing beyond the end of
> port_status. This change results in no difference in the resulting
> object code.
> 
> Avoids several warnings when building with -Warray-bounds:
> 
> drivers/usb/host/ehci-brcm.c: In function 'ehci_brcm_reset':
> drivers/usb/host/ehci-brcm.c:113:32: warning: array subscript 16 is above array bounds of 'u32[15]' {aka 'unsigned int[15]'} [-Warray-bounds]
>   113 |  ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
>       |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> In file included from drivers/usb/host/ehci.h:274,
>                  from drivers/usb/host/ehci-brcm.c:15:
> ./include/linux/usb/ehci_def.h:132:7: note: while referencing 'port_status'
>   132 |   u32 port_status[HCS_N_PORTS_MAX];
>       |       ^~~~~~~~~~~
> 
> Note that the documentation around this proprietary register is
> confusing. If "USB_EHCI_INSNREG00" is at port_status[0x0f], its offset
> would be 0x80 (not 0x90). The code uses port_status[0x10], so is that
> not using "USB_EHCI_INSNREG00"?

I suspect the 0x90 value in the comment is a typo for 0x80.

> Perhaps port_status[0x10] is USB_EHCI_INSNREG01 and port_status[0x12]
> is USB_EHCI_INSNREG03? If so, the union could be adjusted to better
> represent the layout.
> 
> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> Cc: Al Cooper <alcooperx@gmail.com>
> Cc: Alan Stern <stern@rowland.harvard.edu>
> Cc: linux-usb@vger.kernel.org
> Cc: bcm-kernel-feedback-list@broadcom.com
> Suggested-by: Arnd Bergmann <arnd@arndb.de>
> Fixes: 9df231511bd6 ("usb: ehci: Add new EHCI driver for Broadcom STB SoC's")
> Signed-off-by: Kees Cook <keescook@chromium.org>
> ---
>  drivers/usb/host/ehci-brcm.c | 11 +++++------
>  include/linux/usb/ehci_def.h | 16 ++++++++++++++--
>  2 files changed, 19 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/usb/host/ehci-brcm.c b/drivers/usb/host/ehci-brcm.c
> index 3e0ebe8cc649..5d232d3701f9 100644
> --- a/drivers/usb/host/ehci-brcm.c
> +++ b/drivers/usb/host/ehci-brcm.c
> @@ -110,8 +110,8 @@ static int ehci_brcm_reset(struct usb_hcd *hcd)
>  	 *   bus usage
>  	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90

This last comment line is no longer necessary, thanks to the revised 
port definitions.  And since it is actively misleading, with the 0x90 
instead of 0x80, I think it should be removed entirely.

>  	 */
> -	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> -	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> +	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
> +	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
>  
>  	return ehci_setup(hcd);
>  }
> @@ -223,11 +223,10 @@ static int __maybe_unused ehci_brcm_resume(struct device *dev)
>  	/*
>  	 * SWLINUX-1705: Avoid OUT packet underflows during high memory
>  	 *   bus usage
> -	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00
> -	 * @ 0x90
> +	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90

Same here.

>  	 */
> -	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> -	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> +	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
> +	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
>  
>  	ehci_resume(hcd, false);
>  
> diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
> index 5398f571113b..86f0909cab99 100644
> --- a/include/linux/usb/ehci_def.h
> +++ b/include/linux/usb/ehci_def.h
> @@ -182,11 +182,23 @@ struct ehci_regs {
>   * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
>   * PORTSCx
>   */
> -	/* HOSTPC: offset 0x84 */
> -	u32		hostpc[HCS_N_PORTS_MAX];
> +	union {
> +		/* HOSTPC: offset 0x84 */
> +		u32	hostpc[HCS_N_PORTS_MAX];
>  #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
>  #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
>  
> +		/*
> +		 * This was originally documented as:
> +		 * "port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90"
> +		 * but this doesn't make sense: the code was using
> +		 * port_status[0x10]. port_status[0x0f] would be reserved4.
> +		 * Also, none of these are near 0x90. port_status[0x10] is
> +		 * offset 0x84, and port_status[0x0f] would be 0x80.
> +		 */

This comment is entirely inappropriate.  It's the sort of thing that 
belongs in the git history, not in the code.

> +		u32	brcm_insnreg[3];

Given the notation in the original comments, perhaps it would be better 
to define this as:

		struct {		/* Broadcom proprietary registers */
			u32	brcm_insnreg01;		/* offset 0x84 */
			u32	brcm_insnreg02;
			u32	brcm_insnreg03;
		};

I don't know.  It would be nice to hear from somebody at Broadcom.

Alan Stern

> +	};
> +
>  	u32		reserved5[2];
>  
>  	/* USBMODE_EX: offset 0xc8 */
> -- 
> 2.30.2
> 

Re: [PATCH 1/2] USB: EHCI: Add register array bounds to HCS ports

[Alan Stern]

On Wed, Aug 18, 2021 at 12:48:17PM +0300, Andy Shevchenko wrote:
> On Wed, Aug 18, 2021 at 7:30 AM Kees Cook <keescook@chromium.org> wrote:
> >
> > The original EHCI register struct used a trailing 0-element array for
> > addressing the N_PORTS-many available registers. However, after
> > commit a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
> > the 0-element array started to overlap the USBMODE extension register.
> >
> > To avoid future compile-time warnings about accessing indexes within a
> > 0-element array, rearrange the struct to actually describe the expected
> > layout (max 15 registers) with a union. All offsets remain the same, and
> > bounds checking becomes possible on accesses to port_status and hostpc.
> 
> ...
> 
> >         /* HOSTPC: offset 0x84 */
> > -       u32             hostpc[0];      /* HOSTPC extension */
> > +       u32             hostpc[HCS_N_PORTS_MAX];
> >  #define HOSTPC_PHCD    (1<<22)         /* Phy clock disable */
> >  #define HOSTPC_PSPD    (3<<25)         /* Port speed detection */
> >
> > -       u32             reserved5[17];
> > +       u32             reserved5[2];
> 
> Shouldn't it be rather [17 - PORT_MAX]? for accuracy?
> Or also a union approach?

It's okay to use [2] here.  The only purpose is to ensure that the 
following usbmode_ex field is allocated at offset 0xc8; there's no 
special intrinsic meaning to that 17 value.

Alan Stern

Re: [PATCH 2/2] USB: EHCI: Add alias for Broadcom INSNREG

[Kees Cook]

On Wed, Aug 18, 2021 at 10:57:36AM -0400, Alan Stern wrote:
> On Tue, Aug 17, 2021 at 09:30:34PM -0700, Kees Cook wrote:
> > Refactor struct ehci_regs to avoid accessing beyond the end of
> > port_status. This change results in no difference in the resulting
> > object code.
> > 
> > Avoids several warnings when building with -Warray-bounds:
> > 
> > drivers/usb/host/ehci-brcm.c: In function 'ehci_brcm_reset':
> > drivers/usb/host/ehci-brcm.c:113:32: warning: array subscript 16 is above array bounds of 'u32[15]' {aka 'unsigned int[15]'} [-Warray-bounds]
> >   113 |  ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> >       |                                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > In file included from drivers/usb/host/ehci.h:274,
> >                  from drivers/usb/host/ehci-brcm.c:15:
> > ./include/linux/usb/ehci_def.h:132:7: note: while referencing 'port_status'
> >   132 |   u32 port_status[HCS_N_PORTS_MAX];
> >       |       ^~~~~~~~~~~
> > 
> > Note that the documentation around this proprietary register is
> > confusing. If "USB_EHCI_INSNREG00" is at port_status[0x0f], its offset
> > would be 0x80 (not 0x90). The code uses port_status[0x10], so is that
> > not using "USB_EHCI_INSNREG00"?
> 
> I suspect the 0x90 value in the comment is a typo for 0x80.

That'd be my conclusion too. I've updated this for v2.

> 
> > Perhaps port_status[0x10] is USB_EHCI_INSNREG01 and port_status[0x12]
> > is USB_EHCI_INSNREG03? If so, the union could be adjusted to better
> > represent the layout.
> > 
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: Al Cooper <alcooperx@gmail.com>
> > Cc: Alan Stern <stern@rowland.harvard.edu>
> > Cc: linux-usb@vger.kernel.org
> > Cc: bcm-kernel-feedback-list@broadcom.com
> > Suggested-by: Arnd Bergmann <arnd@arndb.de>
> > Fixes: 9df231511bd6 ("usb: ehci: Add new EHCI driver for Broadcom STB SoC's")
> > Signed-off-by: Kees Cook <keescook@chromium.org>
> > ---
> >  drivers/usb/host/ehci-brcm.c | 11 +++++------
> >  include/linux/usb/ehci_def.h | 16 ++++++++++++++--
> >  2 files changed, 19 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/usb/host/ehci-brcm.c b/drivers/usb/host/ehci-brcm.c
> > index 3e0ebe8cc649..5d232d3701f9 100644
> > --- a/drivers/usb/host/ehci-brcm.c
> > +++ b/drivers/usb/host/ehci-brcm.c
> > @@ -110,8 +110,8 @@ static int ehci_brcm_reset(struct usb_hcd *hcd)
> >  	 *   bus usage
> >  	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90
> 
> This last comment line is no longer necessary, thanks to the revised 
> port definitions.  And since it is actively misleading, with the 0x90 
> instead of 0x80, I think it should be removed entirely.

Done.

> 
> >  	 */
> > -	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> > -	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> > +	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
> > +	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
> >  
> >  	return ehci_setup(hcd);
> >  }
> > @@ -223,11 +223,10 @@ static int __maybe_unused ehci_brcm_resume(struct device *dev)
> >  	/*
> >  	 * SWLINUX-1705: Avoid OUT packet underflows during high memory
> >  	 *   bus usage
> > -	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00
> > -	 * @ 0x90
> > +	 * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90
> 
> Same here.
> 
> >  	 */
> > -	ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> > -	ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> > +	ehci_writel(ehci, 0x00800040, &ehci->regs->brcm_insnreg[0]);
> > +	ehci_writel(ehci, 0x00000001, &ehci->regs->brcm_insnreg[2]);
> >  
> >  	ehci_resume(hcd, false);
> >  
> > diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
> > index 5398f571113b..86f0909cab99 100644
> > --- a/include/linux/usb/ehci_def.h
> > +++ b/include/linux/usb/ehci_def.h
> > @@ -182,11 +182,23 @@ struct ehci_regs {
> >   * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
> >   * PORTSCx
> >   */
> > -	/* HOSTPC: offset 0x84 */
> > -	u32		hostpc[HCS_N_PORTS_MAX];
> > +	union {
> > +		/* HOSTPC: offset 0x84 */
> > +		u32	hostpc[HCS_N_PORTS_MAX];
> >  #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
> >  #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
> >  
> > +		/*
> > +		 * This was originally documented as:
> > +		 * "port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90"
> > +		 * but this doesn't make sense: the code was using
> > +		 * port_status[0x10]. port_status[0x0f] would be reserved4.
> > +		 * Also, none of these are near 0x90. port_status[0x10] is
> > +		 * offset 0x84, and port_status[0x0f] would be 0x80.
> > +		 */
> 
> This comment is entirely inappropriate.  It's the sort of thing that 
> belongs in the git history, not in the code.

I wanted it to be easily discoverable, but since we've got a preferred
result now, I'm dropping this and orienting against 0x80.

> 
> > +		u32	brcm_insnreg[3];
> 
> Given the notation in the original comments, perhaps it would be better 
> to define this as:
> 
> 		struct {		/* Broadcom proprietary registers */
> 			u32	brcm_insnreg01;		/* offset 0x84 */
> 			u32	brcm_insnreg02;
> 			u32	brcm_insnreg03;
> 		};

Following the other register arrays, I'm going to keep an array for
this, but adjust the numbering to start at 0 @ 0x80 so the code will
poke offset 1 and 3.

> I don't know.  It would be nice to hear from somebody at Broadcom.

Agreed. :)

Thanks for the review!

-- 
Kees Cook

Re: [PATCH 1/2] USB: EHCI: Add register array bounds to HCS ports

[Kees Cook]

On Wed, Aug 18, 2021 at 10:44:42AM -0400, Alan Stern wrote:
> On Tue, Aug 17, 2021 at 09:30:33PM -0700, Kees Cook wrote:
> > The original EHCI register struct used a trailing 0-element array for
> > addressing the N_PORTS-many available registers. However, after
> > commit a46af4ebf9ff ("USB: EHCI: define extension registers like normal ones")
> > the 0-element array started to overlap the USBMODE extension register.
> > 
> > To avoid future compile-time warnings about accessing indexes within a
> > 0-element array, rearrange the struct to actually describe the expected
> > layout (max 15 registers) with a union. All offsets remain the same, and
> > bounds checking becomes possible on accesses to port_status and hostpc.
> > 
> > There are no binary differences, and struct offsets continue to match.
> 
> Two comments...
> 
> > ---
> >  include/linux/usb/ehci_def.h | 24 +++++++++++++-----------
> >  1 file changed, 13 insertions(+), 11 deletions(-)
> > 
> > diff --git a/include/linux/usb/ehci_def.h b/include/linux/usb/ehci_def.h
> > index 78e006355557..5398f571113b 100644
> > --- a/include/linux/usb/ehci_def.h
> > +++ b/include/linux/usb/ehci_def.h
> > @@ -45,6 +45,7 @@ struct ehci_caps {
> >  #define HCS_PORTROUTED(p)	((p)&(1 << 7))	/* true: port routing */
> >  #define HCS_PPC(p)		((p)&(1 << 4))	/* true: port power control */
> >  #define HCS_N_PORTS(p)		(((p)>>0)&0xf)	/* bits 3:0, ports on HC */
> > +#define HCS_N_PORTS_MAX		0xf		/* N_PORTS valid 0x1-0xF */
> 
> I would prefer to see this value in decimal.  It seems very odd to say 
> something like "The maximum number of ports is 0xf".

Okay, done.

> 
> >  
> >  	u32		hcc_params;      /* HCCPARAMS - offset 0x8 */
> >  /* EHCI 1.1 addendum */
> > @@ -126,8 +127,9 @@ struct ehci_regs {
> >  	u32		configured_flag;
> >  #define FLAG_CF		(1<<0)		/* true: we'll support "high speed" */
> >  
> > -	/* PORTSC: offset 0x44 */
> > -	u32		port_status[0];	/* up to N_PORTS */
> > +	union {
> > +		/* PORTSC: offset 0x44 */
> > +		u32	port_status[HCS_N_PORTS_MAX];
> 
> Please don't lose the second comment.

I've put it back. It seemed redundant in the face of HCS_N_PORTS_MAX
being there now.

> 
> >  /* EHCI 1.1 addendum */
> >  #define PORTSC_SUSPEND_STS_ACK 0
> >  #define PORTSC_SUSPEND_STS_NYET 1
> > @@ -164,28 +166,28 @@ struct ehci_regs {
> >  #define PORT_CSC	(1<<1)		/* connect status change */
> >  #define PORT_CONNECT	(1<<0)		/* device connected */
> >  #define PORT_RWC_BITS   (PORT_CSC | PORT_PEC | PORT_OCC)
> > -
> > -	u32		reserved3[9];
> > -
> > -	/* USBMODE: offset 0x68 */
> > -	u32		usbmode;	/* USB Device mode */
> > +		struct {
> > +			u32	reserved3[9];
> > +			/* USBMODE: offset 0x68 */
> > +			u32	usbmode;	/* USB Device mode */
> > +		};
> >  #define USBMODE_SDIS	(1<<3)		/* Stream disable */
> >  #define USBMODE_BE	(1<<2)		/* BE/LE endianness select */
> >  #define USBMODE_CM_HC	(3<<0)		/* host controller mode */
> >  #define USBMODE_CM_IDLE	(0<<0)		/* idle state */
> > -
> > -	u32		reserved4[6];
> > +	};
> > +	u32		reserved4;
> >  
> >  /* Moorestown has some non-standard registers, partially due to the fact that
> >   * its EHCI controller has both TT and LPM support. HOSTPCx are extensions to
> >   * PORTSCx
> >   */
> >  	/* HOSTPC: offset 0x84 */
> > -	u32		hostpc[0];	/* HOSTPC extension */
> > +	u32		hostpc[HCS_N_PORTS_MAX];
> >  #define HOSTPC_PHCD	(1<<22)		/* Phy clock disable */
> >  #define HOSTPC_PSPD	(3<<25)		/* Port speed detection */
> >  
> > -	u32		reserved5[17];
> > +	u32		reserved5[2];
> >  
> >  	/* USBMODE_EX: offset 0xc8 */
> >  	u32		usbmode_ex;	/* USB Device mode extension */
> 
> Otherwise okay.

Thanks!

-- 
Kees Cook