diff mbox series

[RESEND,v1] scsi: ufs: Use explicit access size in ufshcd_dump_regs

Message ID 830ab302-7189-1d5f-4b6c-191e5a361852@free.fr (mailing list archive)
State Mainlined
Commit d67247566450cf89a693307c9bc9f05a32d96cea
Headers show
Series [RESEND,v1] scsi: ufs: Use explicit access size in ufshcd_dump_regs | expand

Commit Message

Marc Gonzalez Jan. 22, 2019, 5:29 p.m. UTC 
memcpy_fromio() doesn't provide any control over access size.
For example, on arm64, it is implemented using readb and readq.
This may trigger a synchronous external abort:

[    3.729943] Internal error: synchronous external abort: 96000210 [#1] PREEMPT SMP
[    3.737000] Modules linked in:
[    3.744371] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G S                4.20.0-rc4 #16
[    3.747413] Hardware name: Qualcomm Technologies, Inc. MSM8998 v1 MTP (DT)
[    3.755295] pstate: 00000005 (nzcv daif -PAN -UAO)
[    3.761978] pc : __memcpy_fromio+0x68/0x80
[    3.766718] lr : ufshcd_dump_regs+0x50/0xb0
[    3.770767] sp : ffff00000807ba00
[    3.774830] x29: ffff00000807ba00 x28: 00000000fffffffb
[    3.778344] x27: ffff0000089db068 x26: ffff8000f6e58000
[    3.783728] x25: 000000000000000e x24: 0000000000000800
[    3.789023] x23: ffff8000f6e587c8 x22: 0000000000000800
[    3.794319] x21: ffff000008908368 x20: ffff8000f6e1ab80
[    3.799615] x19: 000000000000006c x18: ffffffffffffffff
[    3.804910] x17: 0000000000000000 x16: 0000000000000000
[    3.810206] x15: ffff000009199648 x14: ffff000089244187
[    3.815502] x13: ffff000009244195 x12: ffff0000091ab000
[    3.820797] x11: 0000000005f5e0ff x10: ffff0000091998a0
[    3.826093] x9 : 0000000000000000 x8 : ffff8000f6e1ac00
[    3.831389] x7 : 0000000000000000 x6 : 0000000000000068
[    3.836676] x5 : ffff8000f6e1abe8 x4 : 0000000000000000
[    3.841971] x3 : ffff00000928c868 x2 : ffff8000f6e1abec
[    3.847267] x1 : ffff00000928c868 x0 : ffff8000f6e1abe8
[    3.852567] Process swapper/0 (pid: 1, stack limit = 0x(____ptrval____))
[    3.857900] Call trace:
[    3.864473]  __memcpy_fromio+0x68/0x80
[    3.866683]  ufs_qcom_dump_dbg_regs+0x1c0/0x370
[    3.870522]  ufshcd_print_host_regs+0x168/0x190
[    3.874946]  ufshcd_init+0xd4c/0xde0
[    3.879459]  ufshcd_pltfrm_init+0x3c8/0x550
[    3.883264]  ufs_qcom_probe+0x24/0x60
[    3.887188]  platform_drv_probe+0x50/0xa0

Assuming aligned 32-bit registers, let's use readl, after making sure
that 'offset' and 'len' are indeed multiples of 4.

Fixes: ba80917d9932d ("scsi: ufs: ufshcd_dump_regs to use memcpy_fromio")
Acked-by: Tomas Winkler <tomas.winkler@intel.com>
Reviewed-by: Jeffrey Hugo <jhugo@codeaurora.org>
Signed-off-by: Marc Gonzalez <marc.w.gonzalez@free.fr>
---
 drivers/scsi/ufs/ufshcd.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

Comments

Bjorn Andersson Jan. 22, 2019, 6:34 p.m. UTC | #1 
On Tue 22 Jan 09:29 PST 2019, Marc Gonzalez wrote:

> memcpy_fromio() doesn't provide any control over access size.
> For example, on arm64, it is implemented using readb and readq.
> This may trigger a synchronous external abort:
> 
> [    3.729943] Internal error: synchronous external abort: 96000210 [#1] PREEMPT SMP
> [    3.737000] Modules linked in:
> [    3.744371] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G S                4.20.0-rc4 #16
> [    3.747413] Hardware name: Qualcomm Technologies, Inc. MSM8998 v1 MTP (DT)
> [    3.755295] pstate: 00000005 (nzcv daif -PAN -UAO)
> [    3.761978] pc : __memcpy_fromio+0x68/0x80
> [    3.766718] lr : ufshcd_dump_regs+0x50/0xb0
> [    3.770767] sp : ffff00000807ba00
> [    3.774830] x29: ffff00000807ba00 x28: 00000000fffffffb
> [    3.778344] x27: ffff0000089db068 x26: ffff8000f6e58000
> [    3.783728] x25: 000000000000000e x24: 0000000000000800
> [    3.789023] x23: ffff8000f6e587c8 x22: 0000000000000800
> [    3.794319] x21: ffff000008908368 x20: ffff8000f6e1ab80
> [    3.799615] x19: 000000000000006c x18: ffffffffffffffff
> [    3.804910] x17: 0000000000000000 x16: 0000000000000000
> [    3.810206] x15: ffff000009199648 x14: ffff000089244187
> [    3.815502] x13: ffff000009244195 x12: ffff0000091ab000
> [    3.820797] x11: 0000000005f5e0ff x10: ffff0000091998a0
> [    3.826093] x9 : 0000000000000000 x8 : ffff8000f6e1ac00
> [    3.831389] x7 : 0000000000000000 x6 : 0000000000000068
> [    3.836676] x5 : ffff8000f6e1abe8 x4 : 0000000000000000
> [    3.841971] x3 : ffff00000928c868 x2 : ffff8000f6e1abec
> [    3.847267] x1 : ffff00000928c868 x0 : ffff8000f6e1abe8
> [    3.852567] Process swapper/0 (pid: 1, stack limit = 0x(____ptrval____))
> [    3.857900] Call trace:
> [    3.864473]  __memcpy_fromio+0x68/0x80
> [    3.866683]  ufs_qcom_dump_dbg_regs+0x1c0/0x370
> [    3.870522]  ufshcd_print_host_regs+0x168/0x190
> [    3.874946]  ufshcd_init+0xd4c/0xde0
> [    3.879459]  ufshcd_pltfrm_init+0x3c8/0x550
> [    3.883264]  ufs_qcom_probe+0x24/0x60
> [    3.887188]  platform_drv_probe+0x50/0xa0
> 
> Assuming aligned 32-bit registers, let's use readl, after making sure
> that 'offset' and 'len' are indeed multiples of 4.
> 
> Fixes: ba80917d9932d ("scsi: ufs: ufshcd_dump_regs to use memcpy_fromio")
> Acked-by: Tomas Winkler <tomas.winkler@intel.com>
> Reviewed-by: Jeffrey Hugo <jhugo@codeaurora.org>
> Signed-off-by: Marc Gonzalez <marc.w.gonzalez@free.fr>

Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>

> ---
>  drivers/scsi/ufs/ufshcd.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> index 535180c01ce8..320bbd9849bc 100644
> --- a/drivers/scsi/ufs/ufshcd.c
> +++ b/drivers/scsi/ufs/ufshcd.c
> @@ -112,13 +112,19 @@
>  int ufshcd_dump_regs(struct ufs_hba *hba, size_t offset, size_t len,
>  		     const char *prefix)
>  {
> -	u8 *regs;
> +	u32 *regs;
> +	size_t pos;
> +
> +	if (offset % 4 != 0 || len % 4 != 0) /* keep readl happy */
> +		return -EINVAL;
>  
>  	regs = kzalloc(len, GFP_KERNEL);
>  	if (!regs)
>  		return -ENOMEM;
>  
> -	memcpy_fromio(regs, hba->mmio_base + offset, len);
> +	for (pos = 0; pos < len; pos += 4)
> +		regs[pos / 4] = ufshcd_readl(hba, offset + pos);
> +
>  	ufshcd_hex_dump(prefix, regs, len);
>  	kfree(regs);
>  
> -- 
> 2.17.1
Greg Kroah-Hartman Jan. 22, 2019, 6:39 p.m. UTC | #2 
On Tue, Jan 22, 2019 at 06:29:22PM +0100, Marc Gonzalez wrote:
> memcpy_fromio() doesn't provide any control over access size.
> For example, on arm64, it is implemented using readb and readq.
> This may trigger a synchronous external abort:
> 
> [    3.729943] Internal error: synchronous external abort: 96000210 [#1] PREEMPT SMP
> [    3.737000] Modules linked in:
> [    3.744371] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G S                4.20.0-rc4 #16
> [    3.747413] Hardware name: Qualcomm Technologies, Inc. MSM8998 v1 MTP (DT)
> [    3.755295] pstate: 00000005 (nzcv daif -PAN -UAO)
> [    3.761978] pc : __memcpy_fromio+0x68/0x80
> [    3.766718] lr : ufshcd_dump_regs+0x50/0xb0
> [    3.770767] sp : ffff00000807ba00
> [    3.774830] x29: ffff00000807ba00 x28: 00000000fffffffb
> [    3.778344] x27: ffff0000089db068 x26: ffff8000f6e58000
> [    3.783728] x25: 000000000000000e x24: 0000000000000800
> [    3.789023] x23: ffff8000f6e587c8 x22: 0000000000000800
> [    3.794319] x21: ffff000008908368 x20: ffff8000f6e1ab80
> [    3.799615] x19: 000000000000006c x18: ffffffffffffffff
> [    3.804910] x17: 0000000000000000 x16: 0000000000000000
> [    3.810206] x15: ffff000009199648 x14: ffff000089244187
> [    3.815502] x13: ffff000009244195 x12: ffff0000091ab000
> [    3.820797] x11: 0000000005f5e0ff x10: ffff0000091998a0
> [    3.826093] x9 : 0000000000000000 x8 : ffff8000f6e1ac00
> [    3.831389] x7 : 0000000000000000 x6 : 0000000000000068
> [    3.836676] x5 : ffff8000f6e1abe8 x4 : 0000000000000000
> [    3.841971] x3 : ffff00000928c868 x2 : ffff8000f6e1abec
> [    3.847267] x1 : ffff00000928c868 x0 : ffff8000f6e1abe8
> [    3.852567] Process swapper/0 (pid: 1, stack limit = 0x(____ptrval____))
> [    3.857900] Call trace:
> [    3.864473]  __memcpy_fromio+0x68/0x80
> [    3.866683]  ufs_qcom_dump_dbg_regs+0x1c0/0x370
> [    3.870522]  ufshcd_print_host_regs+0x168/0x190
> [    3.874946]  ufshcd_init+0xd4c/0xde0
> [    3.879459]  ufshcd_pltfrm_init+0x3c8/0x550
> [    3.883264]  ufs_qcom_probe+0x24/0x60
> [    3.887188]  platform_drv_probe+0x50/0xa0
> 
> Assuming aligned 32-bit registers, let's use readl, after making sure
> that 'offset' and 'len' are indeed multiples of 4.
> 
> Fixes: ba80917d9932d ("scsi: ufs: ufshcd_dump_regs to use memcpy_fromio")
> Acked-by: Tomas Winkler <tomas.winkler@intel.com>
> Reviewed-by: Jeffrey Hugo <jhugo@codeaurora.org>
> Signed-off-by: Marc Gonzalez <marc.w.gonzalez@free.fr>

Also please:

Cc: stable <stable@vger.kernel.org>

thanks,

greg k-h
Evan Green Jan. 22, 2019, 10:50 p.m. UTC | #3 
On Tue, Jan 22, 2019 at 9:29 AM Marc Gonzalez <marc.w.gonzalez@free.fr> wrote:
>
> memcpy_fromio() doesn't provide any control over access size.
> For example, on arm64, it is implemented using readb and readq.
> This may trigger a synchronous external abort:
>
> [    3.729943] Internal error: synchronous external abort: 96000210 [#1] PREEMPT SMP
> [    3.737000] Modules linked in:
> [    3.744371] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G S                4.20.0-rc4 #16
> [    3.747413] Hardware name: Qualcomm Technologies, Inc. MSM8998 v1 MTP (DT)
> [    3.755295] pstate: 00000005 (nzcv daif -PAN -UAO)
> [    3.761978] pc : __memcpy_fromio+0x68/0x80
> [    3.766718] lr : ufshcd_dump_regs+0x50/0xb0
> [    3.770767] sp : ffff00000807ba00
> [    3.774830] x29: ffff00000807ba00 x28: 00000000fffffffb
> [    3.778344] x27: ffff0000089db068 x26: ffff8000f6e58000
> [    3.783728] x25: 000000000000000e x24: 0000000000000800
> [    3.789023] x23: ffff8000f6e587c8 x22: 0000000000000800
> [    3.794319] x21: ffff000008908368 x20: ffff8000f6e1ab80
> [    3.799615] x19: 000000000000006c x18: ffffffffffffffff
> [    3.804910] x17: 0000000000000000 x16: 0000000000000000
> [    3.810206] x15: ffff000009199648 x14: ffff000089244187
> [    3.815502] x13: ffff000009244195 x12: ffff0000091ab000
> [    3.820797] x11: 0000000005f5e0ff x10: ffff0000091998a0
> [    3.826093] x9 : 0000000000000000 x8 : ffff8000f6e1ac00
> [    3.831389] x7 : 0000000000000000 x6 : 0000000000000068
> [    3.836676] x5 : ffff8000f6e1abe8 x4 : 0000000000000000
> [    3.841971] x3 : ffff00000928c868 x2 : ffff8000f6e1abec
> [    3.847267] x1 : ffff00000928c868 x0 : ffff8000f6e1abe8
> [    3.852567] Process swapper/0 (pid: 1, stack limit = 0x(____ptrval____))
> [    3.857900] Call trace:
> [    3.864473]  __memcpy_fromio+0x68/0x80
> [    3.866683]  ufs_qcom_dump_dbg_regs+0x1c0/0x370
> [    3.870522]  ufshcd_print_host_regs+0x168/0x190
> [    3.874946]  ufshcd_init+0xd4c/0xde0
> [    3.879459]  ufshcd_pltfrm_init+0x3c8/0x550
> [    3.883264]  ufs_qcom_probe+0x24/0x60
> [    3.887188]  platform_drv_probe+0x50/0xa0
>
> Assuming aligned 32-bit registers, let's use readl, after making sure
> that 'offset' and 'len' are indeed multiples of 4.
>
> Fixes: ba80917d9932d ("scsi: ufs: ufshcd_dump_regs to use memcpy_fromio")
> Acked-by: Tomas Winkler <tomas.winkler@intel.com>
> Reviewed-by: Jeffrey Hugo <jhugo@codeaurora.org>
> Signed-off-by: Marc Gonzalez <marc.w.gonzalez@free.fr>

Tested-by: Evan Green <evgreen@chromium.org>
diff mbox series

Patch

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 535180c01ce8..320bbd9849bc 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -112,13 +112,19 @@ 
 int ufshcd_dump_regs(struct ufs_hba *hba, size_t offset, size_t len,
 		     const char *prefix)
 {
-	u8 *regs;
+	u32 *regs;
+	size_t pos;
+
+	if (offset % 4 != 0 || len % 4 != 0) /* keep readl happy */
+		return -EINVAL;
 
 	regs = kzalloc(len, GFP_KERNEL);
 	if (!regs)
 		return -ENOMEM;
 
-	memcpy_fromio(regs, hba->mmio_base + offset, len);
+	for (pos = 0; pos < len; pos += 4)
+		regs[pos / 4] = ufshcd_readl(hba, offset + pos);
+
 	ufshcd_hex_dump(prefix, regs, len);
 	kfree(regs);